1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2006, 2011, 2016-2017 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Portions of this software were developed by BAE Systems, the University of 8 * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL 9 * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent 10 * Computing (TC) research program. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Support for shared swap-backed anonymous memory objects via 36 * shm_open(2) and shm_unlink(2). While most of the implementation is 37 * here, vm_mmap.c contains mapping logic changes. 38 * 39 * TODO: 40 * 41 * (1) Need to export data to a userland tool via a sysctl. Should ipcs(1) 42 * and ipcrm(1) be expanded or should new tools to manage both POSIX 43 * kernel semaphores and POSIX shared memory be written? 44 * 45 * (2) Add support for this file type to fstat(1). 46 * 47 * (3) Resource limits? Does this need its own resource limits or are the 48 * existing limits in mmap(2) sufficient? 49 */ 50 51 #include <sys/cdefs.h> 52 __FBSDID("$FreeBSD$"); 53 54 #include "opt_capsicum.h" 55 #include "opt_ktrace.h" 56 57 #include <sys/param.h> 58 #include <sys/capsicum.h> 59 #include <sys/conf.h> 60 #include <sys/fcntl.h> 61 #include <sys/file.h> 62 #include <sys/filedesc.h> 63 #include <sys/filio.h> 64 #include <sys/fnv_hash.h> 65 #include <sys/kernel.h> 66 #include <sys/uio.h> 67 #include <sys/signal.h> 68 #include <sys/jail.h> 69 #include <sys/ktrace.h> 70 #include <sys/lock.h> 71 #include <sys/malloc.h> 72 #include <sys/mman.h> 73 #include <sys/mutex.h> 74 #include <sys/priv.h> 75 #include <sys/proc.h> 76 #include <sys/refcount.h> 77 #include <sys/resourcevar.h> 78 #include <sys/rwlock.h> 79 #include <sys/stat.h> 80 #include <sys/syscallsubr.h> 81 #include <sys/sysctl.h> 82 #include <sys/sysproto.h> 83 #include <sys/systm.h> 84 #include <sys/sx.h> 85 #include <sys/time.h> 86 #include <sys/vnode.h> 87 #include <sys/unistd.h> 88 #include <sys/user.h> 89 90 #include <security/audit/audit.h> 91 #include <security/mac/mac_framework.h> 92 93 #include <vm/vm.h> 94 #include <vm/vm_param.h> 95 #include <vm/pmap.h> 96 #include <vm/vm_extern.h> 97 #include <vm/vm_map.h> 98 #include <vm/vm_kern.h> 99 #include <vm/vm_object.h> 100 #include <vm/vm_page.h> 101 #include <vm/vm_pageout.h> 102 #include <vm/vm_pager.h> 103 #include <vm/swap_pager.h> 104 105 struct shm_mapping { 106 char *sm_path; 107 Fnv32_t sm_fnv; 108 struct shmfd *sm_shmfd; 109 LIST_ENTRY(shm_mapping) sm_link; 110 }; 111 112 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor"); 113 static LIST_HEAD(, shm_mapping) *shm_dictionary; 114 static struct sx shm_dict_lock; 115 static struct mtx shm_timestamp_lock; 116 static u_long shm_hash; 117 static struct unrhdr64 shm_ino_unr; 118 static dev_t shm_dev_ino; 119 120 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash]) 121 122 static void shm_init(void *arg); 123 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd); 124 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv); 125 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred); 126 127 static fo_rdwr_t shm_read; 128 static fo_rdwr_t shm_write; 129 static fo_truncate_t shm_truncate; 130 static fo_ioctl_t shm_ioctl; 131 static fo_stat_t shm_stat; 132 static fo_close_t shm_close; 133 static fo_chmod_t shm_chmod; 134 static fo_chown_t shm_chown; 135 static fo_seek_t shm_seek; 136 static fo_fill_kinfo_t shm_fill_kinfo; 137 static fo_mmap_t shm_mmap; 138 139 /* File descriptor operations. */ 140 struct fileops shm_ops = { 141 .fo_read = shm_read, 142 .fo_write = shm_write, 143 .fo_truncate = shm_truncate, 144 .fo_ioctl = shm_ioctl, 145 .fo_poll = invfo_poll, 146 .fo_kqfilter = invfo_kqfilter, 147 .fo_stat = shm_stat, 148 .fo_close = shm_close, 149 .fo_chmod = shm_chmod, 150 .fo_chown = shm_chown, 151 .fo_sendfile = vn_sendfile, 152 .fo_seek = shm_seek, 153 .fo_fill_kinfo = shm_fill_kinfo, 154 .fo_mmap = shm_mmap, 155 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 156 }; 157 158 FEATURE(posix_shm, "POSIX shared memory"); 159 160 static int 161 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio) 162 { 163 vm_page_t m; 164 vm_pindex_t idx; 165 size_t tlen; 166 int error, offset, rv; 167 168 idx = OFF_TO_IDX(uio->uio_offset); 169 offset = uio->uio_offset & PAGE_MASK; 170 tlen = MIN(PAGE_SIZE - offset, len); 171 172 VM_OBJECT_WLOCK(obj); 173 174 /* 175 * Read I/O without either a corresponding resident page or swap 176 * page: use zero_region. This is intended to avoid instantiating 177 * pages on read from a sparse region. 178 */ 179 if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL && 180 !vm_pager_has_page(obj, idx, NULL, NULL)) { 181 VM_OBJECT_WUNLOCK(obj); 182 return (uiomove(__DECONST(void *, zero_region), tlen, uio)); 183 } 184 185 /* 186 * Parallel reads of the page content from disk are prevented 187 * by exclusive busy. 188 * 189 * Although the tmpfs vnode lock is held here, it is 190 * nonetheless safe to sleep waiting for a free page. The 191 * pageout daemon does not need to acquire the tmpfs vnode 192 * lock to page out tobj's pages because tobj is a OBJT_SWAP 193 * type object. 194 */ 195 m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 196 if (m->valid != VM_PAGE_BITS_ALL) { 197 vm_page_xbusy(m); 198 if (vm_pager_has_page(obj, idx, NULL, NULL)) { 199 rv = vm_pager_get_pages(obj, &m, 1, NULL, NULL); 200 if (rv != VM_PAGER_OK) { 201 printf( 202 "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n", 203 obj, idx, m->valid, rv); 204 vm_page_lock(m); 205 vm_page_free(m); 206 vm_page_unlock(m); 207 VM_OBJECT_WUNLOCK(obj); 208 return (EIO); 209 } 210 } else 211 vm_page_zero_invalid(m, TRUE); 212 vm_page_xunbusy(m); 213 } 214 vm_page_lock(m); 215 vm_page_hold(m); 216 if (vm_page_active(m)) 217 vm_page_reference(m); 218 else 219 vm_page_activate(m); 220 vm_page_unlock(m); 221 VM_OBJECT_WUNLOCK(obj); 222 error = uiomove_fromphys(&m, offset, tlen, uio); 223 if (uio->uio_rw == UIO_WRITE && error == 0) { 224 VM_OBJECT_WLOCK(obj); 225 vm_page_dirty(m); 226 vm_pager_page_unswapped(m); 227 VM_OBJECT_WUNLOCK(obj); 228 } 229 vm_page_lock(m); 230 vm_page_unhold(m); 231 vm_page_unlock(m); 232 233 return (error); 234 } 235 236 int 237 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio) 238 { 239 ssize_t resid; 240 size_t len; 241 int error; 242 243 error = 0; 244 while ((resid = uio->uio_resid) > 0) { 245 if (obj_size <= uio->uio_offset) 246 break; 247 len = MIN(obj_size - uio->uio_offset, resid); 248 if (len == 0) 249 break; 250 error = uiomove_object_page(obj, len, uio); 251 if (error != 0 || resid == uio->uio_resid) 252 break; 253 } 254 return (error); 255 } 256 257 static int 258 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td) 259 { 260 struct shmfd *shmfd; 261 off_t foffset; 262 int error; 263 264 shmfd = fp->f_data; 265 foffset = foffset_lock(fp, 0); 266 error = 0; 267 switch (whence) { 268 case L_INCR: 269 if (foffset < 0 || 270 (offset > 0 && foffset > OFF_MAX - offset)) { 271 error = EOVERFLOW; 272 break; 273 } 274 offset += foffset; 275 break; 276 case L_XTND: 277 if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) { 278 error = EOVERFLOW; 279 break; 280 } 281 offset += shmfd->shm_size; 282 break; 283 case L_SET: 284 break; 285 default: 286 error = EINVAL; 287 } 288 if (error == 0) { 289 if (offset < 0 || offset > shmfd->shm_size) 290 error = EINVAL; 291 else 292 td->td_uretoff.tdu_off = offset; 293 } 294 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); 295 return (error); 296 } 297 298 static int 299 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 300 int flags, struct thread *td) 301 { 302 struct shmfd *shmfd; 303 void *rl_cookie; 304 int error; 305 306 shmfd = fp->f_data; 307 #ifdef MAC 308 error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd); 309 if (error) 310 return (error); 311 #endif 312 foffset_lock_uio(fp, uio, flags); 313 rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset, 314 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx); 315 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio); 316 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 317 foffset_unlock_uio(fp, uio, flags); 318 return (error); 319 } 320 321 static int 322 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 323 int flags, struct thread *td) 324 { 325 struct shmfd *shmfd; 326 void *rl_cookie; 327 int error; 328 329 shmfd = fp->f_data; 330 #ifdef MAC 331 error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd); 332 if (error) 333 return (error); 334 #endif 335 foffset_lock_uio(fp, uio, flags); 336 if ((flags & FOF_OFFSET) == 0) { 337 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX, 338 &shmfd->shm_mtx); 339 } else { 340 rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset, 341 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx); 342 } 343 344 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio); 345 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 346 foffset_unlock_uio(fp, uio, flags); 347 return (error); 348 } 349 350 static int 351 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred, 352 struct thread *td) 353 { 354 struct shmfd *shmfd; 355 #ifdef MAC 356 int error; 357 #endif 358 359 shmfd = fp->f_data; 360 #ifdef MAC 361 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd); 362 if (error) 363 return (error); 364 #endif 365 return (shm_dotruncate(shmfd, length)); 366 } 367 368 int 369 shm_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 370 struct thread *td) 371 { 372 373 switch (com) { 374 case FIONBIO: 375 case FIOASYNC: 376 /* 377 * Allow fcntl(fd, F_SETFL, O_NONBLOCK) to work, 378 * just like it would on an unlinked regular file 379 */ 380 return (0); 381 default: 382 return (ENOTTY); 383 } 384 } 385 386 static int 387 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 388 struct thread *td) 389 { 390 struct shmfd *shmfd; 391 #ifdef MAC 392 int error; 393 #endif 394 395 shmfd = fp->f_data; 396 397 #ifdef MAC 398 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd); 399 if (error) 400 return (error); 401 #endif 402 403 /* 404 * Attempt to return sanish values for fstat() on a memory file 405 * descriptor. 406 */ 407 bzero(sb, sizeof(*sb)); 408 sb->st_blksize = PAGE_SIZE; 409 sb->st_size = shmfd->shm_size; 410 sb->st_blocks = howmany(sb->st_size, sb->st_blksize); 411 mtx_lock(&shm_timestamp_lock); 412 sb->st_atim = shmfd->shm_atime; 413 sb->st_ctim = shmfd->shm_ctime; 414 sb->st_mtim = shmfd->shm_mtime; 415 sb->st_birthtim = shmfd->shm_birthtime; 416 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 417 sb->st_uid = shmfd->shm_uid; 418 sb->st_gid = shmfd->shm_gid; 419 mtx_unlock(&shm_timestamp_lock); 420 sb->st_dev = shm_dev_ino; 421 sb->st_ino = shmfd->shm_ino; 422 423 return (0); 424 } 425 426 static int 427 shm_close(struct file *fp, struct thread *td) 428 { 429 struct shmfd *shmfd; 430 431 shmfd = fp->f_data; 432 fp->f_data = NULL; 433 shm_drop(shmfd); 434 435 return (0); 436 } 437 438 int 439 shm_dotruncate(struct shmfd *shmfd, off_t length) 440 { 441 vm_object_t object; 442 vm_page_t m; 443 vm_pindex_t idx, nobjsize; 444 vm_ooffset_t delta; 445 int base, rv; 446 447 KASSERT(length >= 0, ("shm_dotruncate: length < 0")); 448 object = shmfd->shm_object; 449 VM_OBJECT_WLOCK(object); 450 if (length == shmfd->shm_size) { 451 VM_OBJECT_WUNLOCK(object); 452 return (0); 453 } 454 nobjsize = OFF_TO_IDX(length + PAGE_MASK); 455 456 /* Are we shrinking? If so, trim the end. */ 457 if (length < shmfd->shm_size) { 458 /* 459 * Disallow any requests to shrink the size if this 460 * object is mapped into the kernel. 461 */ 462 if (shmfd->shm_kmappings > 0) { 463 VM_OBJECT_WUNLOCK(object); 464 return (EBUSY); 465 } 466 467 /* 468 * Zero the truncated part of the last page. 469 */ 470 base = length & PAGE_MASK; 471 if (base != 0) { 472 idx = OFF_TO_IDX(length); 473 retry: 474 m = vm_page_lookup(object, idx); 475 if (m != NULL) { 476 if (vm_page_sleep_if_busy(m, "shmtrc")) 477 goto retry; 478 } else if (vm_pager_has_page(object, idx, NULL, NULL)) { 479 m = vm_page_alloc(object, idx, 480 VM_ALLOC_NORMAL | VM_ALLOC_WAITFAIL); 481 if (m == NULL) 482 goto retry; 483 rv = vm_pager_get_pages(object, &m, 1, NULL, 484 NULL); 485 vm_page_lock(m); 486 if (rv == VM_PAGER_OK) { 487 /* 488 * Since the page was not resident, 489 * and therefore not recently 490 * accessed, immediately enqueue it 491 * for asynchronous laundering. The 492 * current operation is not regarded 493 * as an access. 494 */ 495 vm_page_launder(m); 496 vm_page_unlock(m); 497 vm_page_xunbusy(m); 498 } else { 499 vm_page_free(m); 500 vm_page_unlock(m); 501 VM_OBJECT_WUNLOCK(object); 502 return (EIO); 503 } 504 } 505 if (m != NULL) { 506 pmap_zero_page_area(m, base, PAGE_SIZE - base); 507 KASSERT(m->valid == VM_PAGE_BITS_ALL, 508 ("shm_dotruncate: page %p is invalid", m)); 509 vm_page_dirty(m); 510 vm_pager_page_unswapped(m); 511 } 512 } 513 delta = IDX_TO_OFF(object->size - nobjsize); 514 515 /* Toss in memory pages. */ 516 if (nobjsize < object->size) 517 vm_object_page_remove(object, nobjsize, object->size, 518 0); 519 520 /* Toss pages from swap. */ 521 if (object->type == OBJT_SWAP) 522 swap_pager_freespace(object, nobjsize, delta); 523 524 /* Free the swap accounted for shm */ 525 swap_release_by_cred(delta, object->cred); 526 object->charge -= delta; 527 } else { 528 /* Try to reserve additional swap space. */ 529 delta = IDX_TO_OFF(nobjsize - object->size); 530 if (!swap_reserve_by_cred(delta, object->cred)) { 531 VM_OBJECT_WUNLOCK(object); 532 return (ENOMEM); 533 } 534 object->charge += delta; 535 } 536 shmfd->shm_size = length; 537 mtx_lock(&shm_timestamp_lock); 538 vfs_timestamp(&shmfd->shm_ctime); 539 shmfd->shm_mtime = shmfd->shm_ctime; 540 mtx_unlock(&shm_timestamp_lock); 541 object->size = nobjsize; 542 VM_OBJECT_WUNLOCK(object); 543 return (0); 544 } 545 546 /* 547 * shmfd object management including creation and reference counting 548 * routines. 549 */ 550 struct shmfd * 551 shm_alloc(struct ucred *ucred, mode_t mode) 552 { 553 struct shmfd *shmfd; 554 555 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO); 556 shmfd->shm_size = 0; 557 shmfd->shm_uid = ucred->cr_uid; 558 shmfd->shm_gid = ucred->cr_gid; 559 shmfd->shm_mode = mode; 560 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL, 561 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred); 562 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate")); 563 shmfd->shm_object->pg_color = 0; 564 VM_OBJECT_WLOCK(shmfd->shm_object); 565 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING); 566 vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT); 567 VM_OBJECT_WUNLOCK(shmfd->shm_object); 568 vfs_timestamp(&shmfd->shm_birthtime); 569 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime = 570 shmfd->shm_birthtime; 571 shmfd->shm_ino = alloc_unr64(&shm_ino_unr); 572 refcount_init(&shmfd->shm_refs, 1); 573 mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF); 574 rangelock_init(&shmfd->shm_rl); 575 #ifdef MAC 576 mac_posixshm_init(shmfd); 577 mac_posixshm_create(ucred, shmfd); 578 #endif 579 580 return (shmfd); 581 } 582 583 struct shmfd * 584 shm_hold(struct shmfd *shmfd) 585 { 586 587 refcount_acquire(&shmfd->shm_refs); 588 return (shmfd); 589 } 590 591 void 592 shm_drop(struct shmfd *shmfd) 593 { 594 595 if (refcount_release(&shmfd->shm_refs)) { 596 #ifdef MAC 597 mac_posixshm_destroy(shmfd); 598 #endif 599 rangelock_destroy(&shmfd->shm_rl); 600 mtx_destroy(&shmfd->shm_mtx); 601 vm_object_deallocate(shmfd->shm_object); 602 free(shmfd, M_SHMFD); 603 } 604 } 605 606 /* 607 * Determine if the credentials have sufficient permissions for a 608 * specified combination of FREAD and FWRITE. 609 */ 610 int 611 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags) 612 { 613 accmode_t accmode; 614 int error; 615 616 accmode = 0; 617 if (flags & FREAD) 618 accmode |= VREAD; 619 if (flags & FWRITE) 620 accmode |= VWRITE; 621 mtx_lock(&shm_timestamp_lock); 622 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid, 623 accmode, ucred, NULL); 624 mtx_unlock(&shm_timestamp_lock); 625 return (error); 626 } 627 628 /* 629 * Dictionary management. We maintain an in-kernel dictionary to map 630 * paths to shmfd objects. We use the FNV hash on the path to store 631 * the mappings in a hash table. 632 */ 633 static void 634 shm_init(void *arg) 635 { 636 637 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF); 638 sx_init(&shm_dict_lock, "shm dictionary"); 639 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash); 640 new_unrhdr64(&shm_ino_unr, 1); 641 shm_dev_ino = devfs_alloc_cdp_inode(); 642 KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized")); 643 } 644 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL); 645 646 static struct shmfd * 647 shm_lookup(char *path, Fnv32_t fnv) 648 { 649 struct shm_mapping *map; 650 651 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 652 if (map->sm_fnv != fnv) 653 continue; 654 if (strcmp(map->sm_path, path) == 0) 655 return (map->sm_shmfd); 656 } 657 658 return (NULL); 659 } 660 661 static void 662 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd) 663 { 664 struct shm_mapping *map; 665 666 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK); 667 map->sm_path = path; 668 map->sm_fnv = fnv; 669 map->sm_shmfd = shm_hold(shmfd); 670 shmfd->shm_path = path; 671 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link); 672 } 673 674 static int 675 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred) 676 { 677 struct shm_mapping *map; 678 int error; 679 680 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 681 if (map->sm_fnv != fnv) 682 continue; 683 if (strcmp(map->sm_path, path) == 0) { 684 #ifdef MAC 685 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd); 686 if (error) 687 return (error); 688 #endif 689 error = shm_access(map->sm_shmfd, ucred, 690 FREAD | FWRITE); 691 if (error) 692 return (error); 693 map->sm_shmfd->shm_path = NULL; 694 LIST_REMOVE(map, sm_link); 695 shm_drop(map->sm_shmfd); 696 free(map->sm_path, M_SHMFD); 697 free(map, M_SHMFD); 698 return (0); 699 } 700 } 701 702 return (ENOENT); 703 } 704 705 int 706 kern_shm_open(struct thread *td, const char *userpath, int flags, mode_t mode, 707 struct filecaps *fcaps) 708 { 709 struct filedesc *fdp; 710 struct shmfd *shmfd; 711 struct file *fp; 712 char *path; 713 const char *pr_path; 714 size_t pr_pathlen; 715 Fnv32_t fnv; 716 mode_t cmode; 717 int fd, error; 718 719 #ifdef CAPABILITY_MODE 720 /* 721 * shm_open(2) is only allowed for anonymous objects. 722 */ 723 if (IN_CAPABILITY_MODE(td) && (userpath != SHM_ANON)) 724 return (ECAPMODE); 725 #endif 726 727 AUDIT_ARG_FFLAGS(flags); 728 AUDIT_ARG_MODE(mode); 729 730 if ((flags & O_ACCMODE) != O_RDONLY && (flags & O_ACCMODE) != O_RDWR) 731 return (EINVAL); 732 733 if ((flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0) 734 return (EINVAL); 735 736 fdp = td->td_proc->p_fd; 737 cmode = (mode & ~fdp->fd_cmask) & ACCESSPERMS; 738 739 error = falloc_caps(td, &fp, &fd, O_CLOEXEC, fcaps); 740 if (error) 741 return (error); 742 743 /* A SHM_ANON path pointer creates an anonymous object. */ 744 if (userpath == SHM_ANON) { 745 /* A read-only anonymous object is pointless. */ 746 if ((flags & O_ACCMODE) == O_RDONLY) { 747 fdclose(td, fp, fd); 748 fdrop(fp, td); 749 return (EINVAL); 750 } 751 shmfd = shm_alloc(td->td_ucred, cmode); 752 } else { 753 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK); 754 pr_path = td->td_ucred->cr_prison->pr_path; 755 756 /* Construct a full pathname for jailed callers. */ 757 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0 758 : strlcpy(path, pr_path, MAXPATHLEN); 759 error = copyinstr(userpath, path + pr_pathlen, 760 MAXPATHLEN - pr_pathlen, NULL); 761 #ifdef KTRACE 762 if (error == 0 && KTRPOINT(curthread, KTR_NAMEI)) 763 ktrnamei(path); 764 #endif 765 /* Require paths to start with a '/' character. */ 766 if (error == 0 && path[pr_pathlen] != '/') 767 error = EINVAL; 768 if (error) { 769 fdclose(td, fp, fd); 770 fdrop(fp, td); 771 free(path, M_SHMFD); 772 return (error); 773 } 774 775 AUDIT_ARG_UPATH1_CANON(path); 776 fnv = fnv_32_str(path, FNV1_32_INIT); 777 sx_xlock(&shm_dict_lock); 778 shmfd = shm_lookup(path, fnv); 779 if (shmfd == NULL) { 780 /* Object does not yet exist, create it if requested. */ 781 if (flags & O_CREAT) { 782 #ifdef MAC 783 error = mac_posixshm_check_create(td->td_ucred, 784 path); 785 if (error == 0) { 786 #endif 787 shmfd = shm_alloc(td->td_ucred, cmode); 788 shm_insert(path, fnv, shmfd); 789 #ifdef MAC 790 } 791 #endif 792 } else { 793 free(path, M_SHMFD); 794 error = ENOENT; 795 } 796 } else { 797 /* 798 * Object already exists, obtain a new 799 * reference if requested and permitted. 800 */ 801 free(path, M_SHMFD); 802 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) 803 error = EEXIST; 804 else { 805 #ifdef MAC 806 error = mac_posixshm_check_open(td->td_ucred, 807 shmfd, FFLAGS(flags & O_ACCMODE)); 808 if (error == 0) 809 #endif 810 error = shm_access(shmfd, td->td_ucred, 811 FFLAGS(flags & O_ACCMODE)); 812 } 813 814 /* 815 * Truncate the file back to zero length if 816 * O_TRUNC was specified and the object was 817 * opened with read/write. 818 */ 819 if (error == 0 && 820 (flags & (O_ACCMODE | O_TRUNC)) == 821 (O_RDWR | O_TRUNC)) { 822 #ifdef MAC 823 error = mac_posixshm_check_truncate( 824 td->td_ucred, fp->f_cred, shmfd); 825 if (error == 0) 826 #endif 827 shm_dotruncate(shmfd, 0); 828 } 829 if (error == 0) 830 shm_hold(shmfd); 831 } 832 sx_xunlock(&shm_dict_lock); 833 834 if (error) { 835 fdclose(td, fp, fd); 836 fdrop(fp, td); 837 return (error); 838 } 839 } 840 841 finit(fp, FFLAGS(flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops); 842 843 td->td_retval[0] = fd; 844 fdrop(fp, td); 845 846 return (0); 847 } 848 849 /* System calls. */ 850 int 851 sys_shm_open(struct thread *td, struct shm_open_args *uap) 852 { 853 854 return (kern_shm_open(td, uap->path, uap->flags, uap->mode, NULL)); 855 } 856 857 int 858 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap) 859 { 860 char *path; 861 const char *pr_path; 862 size_t pr_pathlen; 863 Fnv32_t fnv; 864 int error; 865 866 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 867 pr_path = td->td_ucred->cr_prison->pr_path; 868 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0 869 : strlcpy(path, pr_path, MAXPATHLEN); 870 error = copyinstr(uap->path, path + pr_pathlen, MAXPATHLEN - pr_pathlen, 871 NULL); 872 if (error) { 873 free(path, M_TEMP); 874 return (error); 875 } 876 #ifdef KTRACE 877 if (KTRPOINT(curthread, KTR_NAMEI)) 878 ktrnamei(path); 879 #endif 880 AUDIT_ARG_UPATH1_CANON(path); 881 fnv = fnv_32_str(path, FNV1_32_INIT); 882 sx_xlock(&shm_dict_lock); 883 error = shm_remove(path, fnv, td->td_ucred); 884 sx_xunlock(&shm_dict_lock); 885 free(path, M_TEMP); 886 887 return (error); 888 } 889 890 int 891 shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t objsize, 892 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, 893 vm_ooffset_t foff, struct thread *td) 894 { 895 struct shmfd *shmfd; 896 vm_prot_t maxprot; 897 int error; 898 899 shmfd = fp->f_data; 900 maxprot = VM_PROT_NONE; 901 902 /* FREAD should always be set. */ 903 if ((fp->f_flag & FREAD) != 0) 904 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ; 905 if ((fp->f_flag & FWRITE) != 0) 906 maxprot |= VM_PROT_WRITE; 907 908 /* Don't permit shared writable mappings on read-only descriptors. */ 909 if ((flags & MAP_SHARED) != 0 && 910 (maxprot & VM_PROT_WRITE) == 0 && 911 (prot & VM_PROT_WRITE) != 0) 912 return (EACCES); 913 maxprot &= cap_maxprot; 914 915 /* See comment in vn_mmap(). */ 916 if ( 917 #ifdef _LP64 918 objsize > OFF_MAX || 919 #endif 920 foff < 0 || foff > OFF_MAX - objsize) 921 return (EINVAL); 922 923 #ifdef MAC 924 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, flags); 925 if (error != 0) 926 return (error); 927 #endif 928 929 mtx_lock(&shm_timestamp_lock); 930 vfs_timestamp(&shmfd->shm_atime); 931 mtx_unlock(&shm_timestamp_lock); 932 vm_object_reference(shmfd->shm_object); 933 934 error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags, 935 shmfd->shm_object, foff, FALSE, td); 936 if (error != 0) 937 vm_object_deallocate(shmfd->shm_object); 938 return (error); 939 } 940 941 static int 942 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 943 struct thread *td) 944 { 945 struct shmfd *shmfd; 946 int error; 947 948 error = 0; 949 shmfd = fp->f_data; 950 mtx_lock(&shm_timestamp_lock); 951 /* 952 * SUSv4 says that x bits of permission need not be affected. 953 * Be consistent with our shm_open there. 954 */ 955 #ifdef MAC 956 error = mac_posixshm_check_setmode(active_cred, shmfd, mode); 957 if (error != 0) 958 goto out; 959 #endif 960 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, 961 shmfd->shm_gid, VADMIN, active_cred, NULL); 962 if (error != 0) 963 goto out; 964 shmfd->shm_mode = mode & ACCESSPERMS; 965 out: 966 mtx_unlock(&shm_timestamp_lock); 967 return (error); 968 } 969 970 static int 971 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 972 struct thread *td) 973 { 974 struct shmfd *shmfd; 975 int error; 976 977 error = 0; 978 shmfd = fp->f_data; 979 mtx_lock(&shm_timestamp_lock); 980 #ifdef MAC 981 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid); 982 if (error != 0) 983 goto out; 984 #endif 985 if (uid == (uid_t)-1) 986 uid = shmfd->shm_uid; 987 if (gid == (gid_t)-1) 988 gid = shmfd->shm_gid; 989 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) || 990 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) && 991 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN))) 992 goto out; 993 shmfd->shm_uid = uid; 994 shmfd->shm_gid = gid; 995 out: 996 mtx_unlock(&shm_timestamp_lock); 997 return (error); 998 } 999 1000 /* 1001 * Helper routines to allow the backing object of a shared memory file 1002 * descriptor to be mapped in the kernel. 1003 */ 1004 int 1005 shm_map(struct file *fp, size_t size, off_t offset, void **memp) 1006 { 1007 struct shmfd *shmfd; 1008 vm_offset_t kva, ofs; 1009 vm_object_t obj; 1010 int rv; 1011 1012 if (fp->f_type != DTYPE_SHM) 1013 return (EINVAL); 1014 shmfd = fp->f_data; 1015 obj = shmfd->shm_object; 1016 VM_OBJECT_WLOCK(obj); 1017 /* 1018 * XXXRW: This validation is probably insufficient, and subject to 1019 * sign errors. It should be fixed. 1020 */ 1021 if (offset >= shmfd->shm_size || 1022 offset + size > round_page(shmfd->shm_size)) { 1023 VM_OBJECT_WUNLOCK(obj); 1024 return (EINVAL); 1025 } 1026 1027 shmfd->shm_kmappings++; 1028 vm_object_reference_locked(obj); 1029 VM_OBJECT_WUNLOCK(obj); 1030 1031 /* Map the object into the kernel_map and wire it. */ 1032 kva = vm_map_min(kernel_map); 1033 ofs = offset & PAGE_MASK; 1034 offset = trunc_page(offset); 1035 size = round_page(size + ofs); 1036 rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0, 1037 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE, 1038 VM_PROT_READ | VM_PROT_WRITE, 0); 1039 if (rv == KERN_SUCCESS) { 1040 rv = vm_map_wire(kernel_map, kva, kva + size, 1041 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 1042 if (rv == KERN_SUCCESS) { 1043 *memp = (void *)(kva + ofs); 1044 return (0); 1045 } 1046 vm_map_remove(kernel_map, kva, kva + size); 1047 } else 1048 vm_object_deallocate(obj); 1049 1050 /* On failure, drop our mapping reference. */ 1051 VM_OBJECT_WLOCK(obj); 1052 shmfd->shm_kmappings--; 1053 VM_OBJECT_WUNLOCK(obj); 1054 1055 return (vm_mmap_to_errno(rv)); 1056 } 1057 1058 /* 1059 * We require the caller to unmap the entire entry. This allows us to 1060 * safely decrement shm_kmappings when a mapping is removed. 1061 */ 1062 int 1063 shm_unmap(struct file *fp, void *mem, size_t size) 1064 { 1065 struct shmfd *shmfd; 1066 vm_map_entry_t entry; 1067 vm_offset_t kva, ofs; 1068 vm_object_t obj; 1069 vm_pindex_t pindex; 1070 vm_prot_t prot; 1071 boolean_t wired; 1072 vm_map_t map; 1073 int rv; 1074 1075 if (fp->f_type != DTYPE_SHM) 1076 return (EINVAL); 1077 shmfd = fp->f_data; 1078 kva = (vm_offset_t)mem; 1079 ofs = kva & PAGE_MASK; 1080 kva = trunc_page(kva); 1081 size = round_page(size + ofs); 1082 map = kernel_map; 1083 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry, 1084 &obj, &pindex, &prot, &wired); 1085 if (rv != KERN_SUCCESS) 1086 return (EINVAL); 1087 if (entry->start != kva || entry->end != kva + size) { 1088 vm_map_lookup_done(map, entry); 1089 return (EINVAL); 1090 } 1091 vm_map_lookup_done(map, entry); 1092 if (obj != shmfd->shm_object) 1093 return (EINVAL); 1094 vm_map_remove(map, kva, kva + size); 1095 VM_OBJECT_WLOCK(obj); 1096 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped")); 1097 shmfd->shm_kmappings--; 1098 VM_OBJECT_WUNLOCK(obj); 1099 return (0); 1100 } 1101 1102 static int 1103 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 1104 { 1105 const char *path, *pr_path; 1106 struct shmfd *shmfd; 1107 size_t pr_pathlen; 1108 1109 kif->kf_type = KF_TYPE_SHM; 1110 shmfd = fp->f_data; 1111 1112 mtx_lock(&shm_timestamp_lock); 1113 kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 1114 mtx_unlock(&shm_timestamp_lock); 1115 kif->kf_un.kf_file.kf_file_size = shmfd->shm_size; 1116 if (shmfd->shm_path != NULL) { 1117 sx_slock(&shm_dict_lock); 1118 if (shmfd->shm_path != NULL) { 1119 path = shmfd->shm_path; 1120 pr_path = curthread->td_ucred->cr_prison->pr_path; 1121 if (strcmp(pr_path, "/") != 0) { 1122 /* Return the jail-rooted pathname. */ 1123 pr_pathlen = strlen(pr_path); 1124 if (strncmp(path, pr_path, pr_pathlen) == 0 && 1125 path[pr_pathlen] == '/') 1126 path += pr_pathlen; 1127 } 1128 strlcpy(kif->kf_path, path, sizeof(kif->kf_path)); 1129 } 1130 sx_sunlock(&shm_dict_lock); 1131 } 1132 return (0); 1133 } 1134