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/sbuf.h> 80 #include <sys/stat.h> 81 #include <sys/syscallsubr.h> 82 #include <sys/sysctl.h> 83 #include <sys/sysproto.h> 84 #include <sys/systm.h> 85 #include <sys/sx.h> 86 #include <sys/time.h> 87 #include <sys/vnode.h> 88 #include <sys/unistd.h> 89 #include <sys/user.h> 90 91 #include <security/audit/audit.h> 92 #include <security/mac/mac_framework.h> 93 94 #include <vm/vm.h> 95 #include <vm/vm_param.h> 96 #include <vm/pmap.h> 97 #include <vm/vm_extern.h> 98 #include <vm/vm_map.h> 99 #include <vm/vm_kern.h> 100 #include <vm/vm_object.h> 101 #include <vm/vm_page.h> 102 #include <vm/vm_pageout.h> 103 #include <vm/vm_pager.h> 104 #include <vm/swap_pager.h> 105 106 struct shm_mapping { 107 char *sm_path; 108 Fnv32_t sm_fnv; 109 struct shmfd *sm_shmfd; 110 LIST_ENTRY(shm_mapping) sm_link; 111 }; 112 113 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor"); 114 static LIST_HEAD(, shm_mapping) *shm_dictionary; 115 static struct sx shm_dict_lock; 116 static struct mtx shm_timestamp_lock; 117 static u_long shm_hash; 118 static struct unrhdr64 shm_ino_unr; 119 static dev_t shm_dev_ino; 120 121 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash]) 122 123 static void shm_init(void *arg); 124 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd); 125 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv); 126 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred); 127 128 static fo_rdwr_t shm_read; 129 static fo_rdwr_t shm_write; 130 static fo_truncate_t shm_truncate; 131 static fo_ioctl_t shm_ioctl; 132 static fo_stat_t shm_stat; 133 static fo_close_t shm_close; 134 static fo_chmod_t shm_chmod; 135 static fo_chown_t shm_chown; 136 static fo_seek_t shm_seek; 137 static fo_fill_kinfo_t shm_fill_kinfo; 138 static fo_mmap_t shm_mmap; 139 140 /* File descriptor operations. */ 141 struct fileops shm_ops = { 142 .fo_read = shm_read, 143 .fo_write = shm_write, 144 .fo_truncate = shm_truncate, 145 .fo_ioctl = shm_ioctl, 146 .fo_poll = invfo_poll, 147 .fo_kqfilter = invfo_kqfilter, 148 .fo_stat = shm_stat, 149 .fo_close = shm_close, 150 .fo_chmod = shm_chmod, 151 .fo_chown = shm_chown, 152 .fo_sendfile = vn_sendfile, 153 .fo_seek = shm_seek, 154 .fo_fill_kinfo = shm_fill_kinfo, 155 .fo_mmap = shm_mmap, 156 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 157 }; 158 159 FEATURE(posix_shm, "POSIX shared memory"); 160 161 static int 162 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio) 163 { 164 vm_page_t m; 165 vm_pindex_t idx; 166 size_t tlen; 167 int error, offset, rv; 168 169 idx = OFF_TO_IDX(uio->uio_offset); 170 offset = uio->uio_offset & PAGE_MASK; 171 tlen = MIN(PAGE_SIZE - offset, len); 172 173 VM_OBJECT_WLOCK(obj); 174 175 /* 176 * Read I/O without either a corresponding resident page or swap 177 * page: use zero_region. This is intended to avoid instantiating 178 * pages on read from a sparse region. 179 */ 180 if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL && 181 !vm_pager_has_page(obj, idx, NULL, NULL)) { 182 VM_OBJECT_WUNLOCK(obj); 183 return (uiomove(__DECONST(void *, zero_region), tlen, uio)); 184 } 185 186 /* 187 * Parallel reads of the page content from disk are prevented 188 * by exclusive busy. 189 * 190 * Although the tmpfs vnode lock is held here, it is 191 * nonetheless safe to sleep waiting for a free page. The 192 * pageout daemon does not need to acquire the tmpfs vnode 193 * lock to page out tobj's pages because tobj is a OBJT_SWAP 194 * type object. 195 */ 196 m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 197 if (m->valid != VM_PAGE_BITS_ALL) { 198 vm_page_xbusy(m); 199 if (vm_pager_has_page(obj, idx, NULL, NULL)) { 200 rv = vm_pager_get_pages(obj, &m, 1, NULL, NULL); 201 if (rv != VM_PAGER_OK) { 202 printf( 203 "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n", 204 obj, idx, m->valid, rv); 205 vm_page_lock(m); 206 vm_page_free(m); 207 vm_page_unlock(m); 208 VM_OBJECT_WUNLOCK(obj); 209 return (EIO); 210 } 211 } else 212 vm_page_zero_invalid(m, TRUE); 213 vm_page_xunbusy(m); 214 } 215 vm_page_lock(m); 216 vm_page_hold(m); 217 if (vm_page_active(m)) 218 vm_page_reference(m); 219 else 220 vm_page_activate(m); 221 vm_page_unlock(m); 222 VM_OBJECT_WUNLOCK(obj); 223 error = uiomove_fromphys(&m, offset, tlen, uio); 224 if (uio->uio_rw == UIO_WRITE && error == 0) { 225 VM_OBJECT_WLOCK(obj); 226 vm_page_dirty(m); 227 vm_pager_page_unswapped(m); 228 VM_OBJECT_WUNLOCK(obj); 229 } 230 vm_page_lock(m); 231 vm_page_unhold(m); 232 vm_page_unlock(m); 233 234 return (error); 235 } 236 237 int 238 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio) 239 { 240 ssize_t resid; 241 size_t len; 242 int error; 243 244 error = 0; 245 while ((resid = uio->uio_resid) > 0) { 246 if (obj_size <= uio->uio_offset) 247 break; 248 len = MIN(obj_size - uio->uio_offset, resid); 249 if (len == 0) 250 break; 251 error = uiomove_object_page(obj, len, uio); 252 if (error != 0 || resid == uio->uio_resid) 253 break; 254 } 255 return (error); 256 } 257 258 static int 259 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td) 260 { 261 struct shmfd *shmfd; 262 off_t foffset; 263 int error; 264 265 shmfd = fp->f_data; 266 foffset = foffset_lock(fp, 0); 267 error = 0; 268 switch (whence) { 269 case L_INCR: 270 if (foffset < 0 || 271 (offset > 0 && foffset > OFF_MAX - offset)) { 272 error = EOVERFLOW; 273 break; 274 } 275 offset += foffset; 276 break; 277 case L_XTND: 278 if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) { 279 error = EOVERFLOW; 280 break; 281 } 282 offset += shmfd->shm_size; 283 break; 284 case L_SET: 285 break; 286 default: 287 error = EINVAL; 288 } 289 if (error == 0) { 290 if (offset < 0 || offset > shmfd->shm_size) 291 error = EINVAL; 292 else 293 td->td_uretoff.tdu_off = offset; 294 } 295 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); 296 return (error); 297 } 298 299 static int 300 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 301 int flags, struct thread *td) 302 { 303 struct shmfd *shmfd; 304 void *rl_cookie; 305 int error; 306 307 shmfd = fp->f_data; 308 #ifdef MAC 309 error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd); 310 if (error) 311 return (error); 312 #endif 313 foffset_lock_uio(fp, uio, flags); 314 rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset, 315 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx); 316 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio); 317 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 318 foffset_unlock_uio(fp, uio, flags); 319 return (error); 320 } 321 322 static int 323 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 324 int flags, struct thread *td) 325 { 326 struct shmfd *shmfd; 327 void *rl_cookie; 328 int error; 329 330 shmfd = fp->f_data; 331 #ifdef MAC 332 error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd); 333 if (error) 334 return (error); 335 #endif 336 foffset_lock_uio(fp, uio, flags); 337 if ((flags & FOF_OFFSET) == 0) { 338 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX, 339 &shmfd->shm_mtx); 340 } else { 341 rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset, 342 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx); 343 } 344 345 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio); 346 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 347 foffset_unlock_uio(fp, uio, flags); 348 return (error); 349 } 350 351 static int 352 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred, 353 struct thread *td) 354 { 355 struct shmfd *shmfd; 356 #ifdef MAC 357 int error; 358 #endif 359 360 shmfd = fp->f_data; 361 #ifdef MAC 362 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd); 363 if (error) 364 return (error); 365 #endif 366 return (shm_dotruncate(shmfd, length)); 367 } 368 369 int 370 shm_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 371 struct thread *td) 372 { 373 374 switch (com) { 375 case FIONBIO: 376 case FIOASYNC: 377 /* 378 * Allow fcntl(fd, F_SETFL, O_NONBLOCK) to work, 379 * just like it would on an unlinked regular file 380 */ 381 return (0); 382 default: 383 return (ENOTTY); 384 } 385 } 386 387 static int 388 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 389 struct thread *td) 390 { 391 struct shmfd *shmfd; 392 #ifdef MAC 393 int error; 394 #endif 395 396 shmfd = fp->f_data; 397 398 #ifdef MAC 399 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd); 400 if (error) 401 return (error); 402 #endif 403 404 /* 405 * Attempt to return sanish values for fstat() on a memory file 406 * descriptor. 407 */ 408 bzero(sb, sizeof(*sb)); 409 sb->st_blksize = PAGE_SIZE; 410 sb->st_size = shmfd->shm_size; 411 sb->st_blocks = howmany(sb->st_size, sb->st_blksize); 412 mtx_lock(&shm_timestamp_lock); 413 sb->st_atim = shmfd->shm_atime; 414 sb->st_ctim = shmfd->shm_ctime; 415 sb->st_mtim = shmfd->shm_mtime; 416 sb->st_birthtim = shmfd->shm_birthtime; 417 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 418 sb->st_uid = shmfd->shm_uid; 419 sb->st_gid = shmfd->shm_gid; 420 mtx_unlock(&shm_timestamp_lock); 421 sb->st_dev = shm_dev_ino; 422 sb->st_ino = shmfd->shm_ino; 423 sb->st_nlink = shmfd->shm_object->ref_count; 424 425 return (0); 426 } 427 428 static int 429 shm_close(struct file *fp, struct thread *td) 430 { 431 struct shmfd *shmfd; 432 433 shmfd = fp->f_data; 434 fp->f_data = NULL; 435 shm_drop(shmfd); 436 437 return (0); 438 } 439 440 int 441 shm_dotruncate(struct shmfd *shmfd, off_t length) 442 { 443 vm_object_t object; 444 vm_page_t m; 445 vm_pindex_t idx, nobjsize; 446 vm_ooffset_t delta; 447 int base, rv; 448 449 KASSERT(length >= 0, ("shm_dotruncate: length < 0")); 450 object = shmfd->shm_object; 451 VM_OBJECT_WLOCK(object); 452 if (length == shmfd->shm_size) { 453 VM_OBJECT_WUNLOCK(object); 454 return (0); 455 } 456 nobjsize = OFF_TO_IDX(length + PAGE_MASK); 457 458 /* Are we shrinking? If so, trim the end. */ 459 if (length < shmfd->shm_size) { 460 /* 461 * Disallow any requests to shrink the size if this 462 * object is mapped into the kernel. 463 */ 464 if (shmfd->shm_kmappings > 0) { 465 VM_OBJECT_WUNLOCK(object); 466 return (EBUSY); 467 } 468 469 /* 470 * Zero the truncated part of the last page. 471 */ 472 base = length & PAGE_MASK; 473 if (base != 0) { 474 idx = OFF_TO_IDX(length); 475 retry: 476 m = vm_page_lookup(object, idx); 477 if (m != NULL) { 478 if (vm_page_sleep_if_busy(m, "shmtrc")) 479 goto retry; 480 } else if (vm_pager_has_page(object, idx, NULL, NULL)) { 481 m = vm_page_alloc(object, idx, 482 VM_ALLOC_NORMAL | VM_ALLOC_WAITFAIL); 483 if (m == NULL) 484 goto retry; 485 rv = vm_pager_get_pages(object, &m, 1, NULL, 486 NULL); 487 vm_page_lock(m); 488 if (rv == VM_PAGER_OK) { 489 /* 490 * Since the page was not resident, 491 * and therefore not recently 492 * accessed, immediately enqueue it 493 * for asynchronous laundering. The 494 * current operation is not regarded 495 * as an access. 496 */ 497 vm_page_launder(m); 498 vm_page_unlock(m); 499 vm_page_xunbusy(m); 500 } else { 501 vm_page_free(m); 502 vm_page_unlock(m); 503 VM_OBJECT_WUNLOCK(object); 504 return (EIO); 505 } 506 } 507 if (m != NULL) { 508 pmap_zero_page_area(m, base, PAGE_SIZE - base); 509 KASSERT(m->valid == VM_PAGE_BITS_ALL, 510 ("shm_dotruncate: page %p is invalid", m)); 511 vm_page_dirty(m); 512 vm_pager_page_unswapped(m); 513 } 514 } 515 delta = IDX_TO_OFF(object->size - nobjsize); 516 517 /* Toss in memory pages. */ 518 if (nobjsize < object->size) 519 vm_object_page_remove(object, nobjsize, object->size, 520 0); 521 522 /* Toss pages from swap. */ 523 if (object->type == OBJT_SWAP) 524 swap_pager_freespace(object, nobjsize, delta); 525 526 /* Free the swap accounted for shm */ 527 swap_release_by_cred(delta, object->cred); 528 object->charge -= delta; 529 } else { 530 /* Try to reserve additional swap space. */ 531 delta = IDX_TO_OFF(nobjsize - object->size); 532 if (!swap_reserve_by_cred(delta, object->cred)) { 533 VM_OBJECT_WUNLOCK(object); 534 return (ENOMEM); 535 } 536 object->charge += delta; 537 } 538 shmfd->shm_size = length; 539 mtx_lock(&shm_timestamp_lock); 540 vfs_timestamp(&shmfd->shm_ctime); 541 shmfd->shm_mtime = shmfd->shm_ctime; 542 mtx_unlock(&shm_timestamp_lock); 543 object->size = nobjsize; 544 VM_OBJECT_WUNLOCK(object); 545 return (0); 546 } 547 548 /* 549 * shmfd object management including creation and reference counting 550 * routines. 551 */ 552 struct shmfd * 553 shm_alloc(struct ucred *ucred, mode_t mode) 554 { 555 struct shmfd *shmfd; 556 557 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO); 558 shmfd->shm_size = 0; 559 shmfd->shm_uid = ucred->cr_uid; 560 shmfd->shm_gid = ucred->cr_gid; 561 shmfd->shm_mode = mode; 562 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL, 563 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred); 564 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate")); 565 shmfd->shm_object->pg_color = 0; 566 VM_OBJECT_WLOCK(shmfd->shm_object); 567 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING); 568 vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT); 569 VM_OBJECT_WUNLOCK(shmfd->shm_object); 570 vfs_timestamp(&shmfd->shm_birthtime); 571 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime = 572 shmfd->shm_birthtime; 573 shmfd->shm_ino = alloc_unr64(&shm_ino_unr); 574 refcount_init(&shmfd->shm_refs, 1); 575 mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF); 576 rangelock_init(&shmfd->shm_rl); 577 #ifdef MAC 578 mac_posixshm_init(shmfd); 579 mac_posixshm_create(ucred, shmfd); 580 #endif 581 582 return (shmfd); 583 } 584 585 struct shmfd * 586 shm_hold(struct shmfd *shmfd) 587 { 588 589 refcount_acquire(&shmfd->shm_refs); 590 return (shmfd); 591 } 592 593 void 594 shm_drop(struct shmfd *shmfd) 595 { 596 597 if (refcount_release(&shmfd->shm_refs)) { 598 #ifdef MAC 599 mac_posixshm_destroy(shmfd); 600 #endif 601 rangelock_destroy(&shmfd->shm_rl); 602 mtx_destroy(&shmfd->shm_mtx); 603 vm_object_deallocate(shmfd->shm_object); 604 free(shmfd, M_SHMFD); 605 } 606 } 607 608 /* 609 * Determine if the credentials have sufficient permissions for a 610 * specified combination of FREAD and FWRITE. 611 */ 612 int 613 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags) 614 { 615 accmode_t accmode; 616 int error; 617 618 accmode = 0; 619 if (flags & FREAD) 620 accmode |= VREAD; 621 if (flags & FWRITE) 622 accmode |= VWRITE; 623 mtx_lock(&shm_timestamp_lock); 624 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid, 625 accmode, ucred, NULL); 626 mtx_unlock(&shm_timestamp_lock); 627 return (error); 628 } 629 630 /* 631 * Dictionary management. We maintain an in-kernel dictionary to map 632 * paths to shmfd objects. We use the FNV hash on the path to store 633 * the mappings in a hash table. 634 */ 635 static void 636 shm_init(void *arg) 637 { 638 639 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF); 640 sx_init(&shm_dict_lock, "shm dictionary"); 641 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash); 642 new_unrhdr64(&shm_ino_unr, 1); 643 shm_dev_ino = devfs_alloc_cdp_inode(); 644 KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized")); 645 } 646 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL); 647 648 static struct shmfd * 649 shm_lookup(char *path, Fnv32_t fnv) 650 { 651 struct shm_mapping *map; 652 653 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 654 if (map->sm_fnv != fnv) 655 continue; 656 if (strcmp(map->sm_path, path) == 0) 657 return (map->sm_shmfd); 658 } 659 660 return (NULL); 661 } 662 663 static void 664 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd) 665 { 666 struct shm_mapping *map; 667 668 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK); 669 map->sm_path = path; 670 map->sm_fnv = fnv; 671 map->sm_shmfd = shm_hold(shmfd); 672 shmfd->shm_path = path; 673 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link); 674 } 675 676 static int 677 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred) 678 { 679 struct shm_mapping *map; 680 int error; 681 682 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 683 if (map->sm_fnv != fnv) 684 continue; 685 if (strcmp(map->sm_path, path) == 0) { 686 #ifdef MAC 687 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd); 688 if (error) 689 return (error); 690 #endif 691 error = shm_access(map->sm_shmfd, ucred, 692 FREAD | FWRITE); 693 if (error) 694 return (error); 695 map->sm_shmfd->shm_path = NULL; 696 LIST_REMOVE(map, sm_link); 697 shm_drop(map->sm_shmfd); 698 free(map->sm_path, M_SHMFD); 699 free(map, M_SHMFD); 700 return (0); 701 } 702 } 703 704 return (ENOENT); 705 } 706 707 int 708 kern_shm_open(struct thread *td, const char *userpath, int flags, mode_t mode, 709 struct filecaps *fcaps) 710 { 711 struct filedesc *fdp; 712 struct shmfd *shmfd; 713 struct file *fp; 714 char *path; 715 const char *pr_path; 716 size_t pr_pathlen; 717 Fnv32_t fnv; 718 mode_t cmode; 719 int fd, error; 720 721 #ifdef CAPABILITY_MODE 722 /* 723 * shm_open(2) is only allowed for anonymous objects. 724 */ 725 if (IN_CAPABILITY_MODE(td) && (userpath != SHM_ANON)) 726 return (ECAPMODE); 727 #endif 728 729 AUDIT_ARG_FFLAGS(flags); 730 AUDIT_ARG_MODE(mode); 731 732 if ((flags & O_ACCMODE) != O_RDONLY && (flags & O_ACCMODE) != O_RDWR) 733 return (EINVAL); 734 735 if ((flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0) 736 return (EINVAL); 737 738 fdp = td->td_proc->p_fd; 739 cmode = (mode & ~fdp->fd_cmask) & ACCESSPERMS; 740 741 error = falloc_caps(td, &fp, &fd, O_CLOEXEC, fcaps); 742 if (error) 743 return (error); 744 745 /* A SHM_ANON path pointer creates an anonymous object. */ 746 if (userpath == SHM_ANON) { 747 /* A read-only anonymous object is pointless. */ 748 if ((flags & O_ACCMODE) == O_RDONLY) { 749 fdclose(td, fp, fd); 750 fdrop(fp, td); 751 return (EINVAL); 752 } 753 shmfd = shm_alloc(td->td_ucred, cmode); 754 } else { 755 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK); 756 pr_path = td->td_ucred->cr_prison->pr_path; 757 758 /* Construct a full pathname for jailed callers. */ 759 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0 760 : strlcpy(path, pr_path, MAXPATHLEN); 761 error = copyinstr(userpath, path + pr_pathlen, 762 MAXPATHLEN - pr_pathlen, NULL); 763 #ifdef KTRACE 764 if (error == 0 && KTRPOINT(curthread, KTR_NAMEI)) 765 ktrnamei(path); 766 #endif 767 /* Require paths to start with a '/' character. */ 768 if (error == 0 && path[pr_pathlen] != '/') 769 error = EINVAL; 770 if (error) { 771 fdclose(td, fp, fd); 772 fdrop(fp, td); 773 free(path, M_SHMFD); 774 return (error); 775 } 776 777 AUDIT_ARG_UPATH1_CANON(path); 778 fnv = fnv_32_str(path, FNV1_32_INIT); 779 sx_xlock(&shm_dict_lock); 780 shmfd = shm_lookup(path, fnv); 781 if (shmfd == NULL) { 782 /* Object does not yet exist, create it if requested. */ 783 if (flags & O_CREAT) { 784 #ifdef MAC 785 error = mac_posixshm_check_create(td->td_ucred, 786 path); 787 if (error == 0) { 788 #endif 789 shmfd = shm_alloc(td->td_ucred, cmode); 790 shm_insert(path, fnv, shmfd); 791 #ifdef MAC 792 } 793 #endif 794 } else { 795 free(path, M_SHMFD); 796 error = ENOENT; 797 } 798 } else { 799 /* 800 * Object already exists, obtain a new 801 * reference if requested and permitted. 802 */ 803 free(path, M_SHMFD); 804 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) 805 error = EEXIST; 806 else { 807 #ifdef MAC 808 error = mac_posixshm_check_open(td->td_ucred, 809 shmfd, FFLAGS(flags & O_ACCMODE)); 810 if (error == 0) 811 #endif 812 error = shm_access(shmfd, td->td_ucred, 813 FFLAGS(flags & O_ACCMODE)); 814 } 815 816 /* 817 * Truncate the file back to zero length if 818 * O_TRUNC was specified and the object was 819 * opened with read/write. 820 */ 821 if (error == 0 && 822 (flags & (O_ACCMODE | O_TRUNC)) == 823 (O_RDWR | O_TRUNC)) { 824 #ifdef MAC 825 error = mac_posixshm_check_truncate( 826 td->td_ucred, fp->f_cred, shmfd); 827 if (error == 0) 828 #endif 829 shm_dotruncate(shmfd, 0); 830 } 831 if (error == 0) 832 shm_hold(shmfd); 833 } 834 sx_xunlock(&shm_dict_lock); 835 836 if (error) { 837 fdclose(td, fp, fd); 838 fdrop(fp, td); 839 return (error); 840 } 841 } 842 843 finit(fp, FFLAGS(flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops); 844 845 td->td_retval[0] = fd; 846 fdrop(fp, td); 847 848 return (0); 849 } 850 851 /* System calls. */ 852 int 853 sys_shm_open(struct thread *td, struct shm_open_args *uap) 854 { 855 856 return (kern_shm_open(td, uap->path, uap->flags, uap->mode, NULL)); 857 } 858 859 int 860 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap) 861 { 862 char *path; 863 const char *pr_path; 864 size_t pr_pathlen; 865 Fnv32_t fnv; 866 int error; 867 868 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 869 pr_path = td->td_ucred->cr_prison->pr_path; 870 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0 871 : strlcpy(path, pr_path, MAXPATHLEN); 872 error = copyinstr(uap->path, path + pr_pathlen, MAXPATHLEN - pr_pathlen, 873 NULL); 874 if (error) { 875 free(path, M_TEMP); 876 return (error); 877 } 878 #ifdef KTRACE 879 if (KTRPOINT(curthread, KTR_NAMEI)) 880 ktrnamei(path); 881 #endif 882 AUDIT_ARG_UPATH1_CANON(path); 883 fnv = fnv_32_str(path, FNV1_32_INIT); 884 sx_xlock(&shm_dict_lock); 885 error = shm_remove(path, fnv, td->td_ucred); 886 sx_xunlock(&shm_dict_lock); 887 free(path, M_TEMP); 888 889 return (error); 890 } 891 892 int 893 shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t objsize, 894 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, 895 vm_ooffset_t foff, struct thread *td) 896 { 897 struct shmfd *shmfd; 898 vm_prot_t maxprot; 899 int error; 900 901 shmfd = fp->f_data; 902 maxprot = VM_PROT_NONE; 903 904 /* FREAD should always be set. */ 905 if ((fp->f_flag & FREAD) != 0) 906 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ; 907 if ((fp->f_flag & FWRITE) != 0) 908 maxprot |= VM_PROT_WRITE; 909 910 /* Don't permit shared writable mappings on read-only descriptors. */ 911 if ((flags & MAP_SHARED) != 0 && 912 (maxprot & VM_PROT_WRITE) == 0 && 913 (prot & VM_PROT_WRITE) != 0) 914 return (EACCES); 915 maxprot &= cap_maxprot; 916 917 /* See comment in vn_mmap(). */ 918 if ( 919 #ifdef _LP64 920 objsize > OFF_MAX || 921 #endif 922 foff < 0 || foff > OFF_MAX - objsize) 923 return (EINVAL); 924 925 #ifdef MAC 926 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, flags); 927 if (error != 0) 928 return (error); 929 #endif 930 931 mtx_lock(&shm_timestamp_lock); 932 vfs_timestamp(&shmfd->shm_atime); 933 mtx_unlock(&shm_timestamp_lock); 934 vm_object_reference(shmfd->shm_object); 935 936 error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags, 937 shmfd->shm_object, foff, FALSE, td); 938 if (error != 0) 939 vm_object_deallocate(shmfd->shm_object); 940 return (error); 941 } 942 943 static int 944 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 945 struct thread *td) 946 { 947 struct shmfd *shmfd; 948 int error; 949 950 error = 0; 951 shmfd = fp->f_data; 952 mtx_lock(&shm_timestamp_lock); 953 /* 954 * SUSv4 says that x bits of permission need not be affected. 955 * Be consistent with our shm_open there. 956 */ 957 #ifdef MAC 958 error = mac_posixshm_check_setmode(active_cred, shmfd, mode); 959 if (error != 0) 960 goto out; 961 #endif 962 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, 963 shmfd->shm_gid, VADMIN, active_cred, NULL); 964 if (error != 0) 965 goto out; 966 shmfd->shm_mode = mode & ACCESSPERMS; 967 out: 968 mtx_unlock(&shm_timestamp_lock); 969 return (error); 970 } 971 972 static int 973 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 974 struct thread *td) 975 { 976 struct shmfd *shmfd; 977 int error; 978 979 error = 0; 980 shmfd = fp->f_data; 981 mtx_lock(&shm_timestamp_lock); 982 #ifdef MAC 983 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid); 984 if (error != 0) 985 goto out; 986 #endif 987 if (uid == (uid_t)-1) 988 uid = shmfd->shm_uid; 989 if (gid == (gid_t)-1) 990 gid = shmfd->shm_gid; 991 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) || 992 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) && 993 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN))) 994 goto out; 995 shmfd->shm_uid = uid; 996 shmfd->shm_gid = gid; 997 out: 998 mtx_unlock(&shm_timestamp_lock); 999 return (error); 1000 } 1001 1002 /* 1003 * Helper routines to allow the backing object of a shared memory file 1004 * descriptor to be mapped in the kernel. 1005 */ 1006 int 1007 shm_map(struct file *fp, size_t size, off_t offset, void **memp) 1008 { 1009 struct shmfd *shmfd; 1010 vm_offset_t kva, ofs; 1011 vm_object_t obj; 1012 int rv; 1013 1014 if (fp->f_type != DTYPE_SHM) 1015 return (EINVAL); 1016 shmfd = fp->f_data; 1017 obj = shmfd->shm_object; 1018 VM_OBJECT_WLOCK(obj); 1019 /* 1020 * XXXRW: This validation is probably insufficient, and subject to 1021 * sign errors. It should be fixed. 1022 */ 1023 if (offset >= shmfd->shm_size || 1024 offset + size > round_page(shmfd->shm_size)) { 1025 VM_OBJECT_WUNLOCK(obj); 1026 return (EINVAL); 1027 } 1028 1029 shmfd->shm_kmappings++; 1030 vm_object_reference_locked(obj); 1031 VM_OBJECT_WUNLOCK(obj); 1032 1033 /* Map the object into the kernel_map and wire it. */ 1034 kva = vm_map_min(kernel_map); 1035 ofs = offset & PAGE_MASK; 1036 offset = trunc_page(offset); 1037 size = round_page(size + ofs); 1038 rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0, 1039 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE, 1040 VM_PROT_READ | VM_PROT_WRITE, 0); 1041 if (rv == KERN_SUCCESS) { 1042 rv = vm_map_wire(kernel_map, kva, kva + size, 1043 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 1044 if (rv == KERN_SUCCESS) { 1045 *memp = (void *)(kva + ofs); 1046 return (0); 1047 } 1048 vm_map_remove(kernel_map, kva, kva + size); 1049 } else 1050 vm_object_deallocate(obj); 1051 1052 /* On failure, drop our mapping reference. */ 1053 VM_OBJECT_WLOCK(obj); 1054 shmfd->shm_kmappings--; 1055 VM_OBJECT_WUNLOCK(obj); 1056 1057 return (vm_mmap_to_errno(rv)); 1058 } 1059 1060 /* 1061 * We require the caller to unmap the entire entry. This allows us to 1062 * safely decrement shm_kmappings when a mapping is removed. 1063 */ 1064 int 1065 shm_unmap(struct file *fp, void *mem, size_t size) 1066 { 1067 struct shmfd *shmfd; 1068 vm_map_entry_t entry; 1069 vm_offset_t kva, ofs; 1070 vm_object_t obj; 1071 vm_pindex_t pindex; 1072 vm_prot_t prot; 1073 boolean_t wired; 1074 vm_map_t map; 1075 int rv; 1076 1077 if (fp->f_type != DTYPE_SHM) 1078 return (EINVAL); 1079 shmfd = fp->f_data; 1080 kva = (vm_offset_t)mem; 1081 ofs = kva & PAGE_MASK; 1082 kva = trunc_page(kva); 1083 size = round_page(size + ofs); 1084 map = kernel_map; 1085 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry, 1086 &obj, &pindex, &prot, &wired); 1087 if (rv != KERN_SUCCESS) 1088 return (EINVAL); 1089 if (entry->start != kva || entry->end != kva + size) { 1090 vm_map_lookup_done(map, entry); 1091 return (EINVAL); 1092 } 1093 vm_map_lookup_done(map, entry); 1094 if (obj != shmfd->shm_object) 1095 return (EINVAL); 1096 vm_map_remove(map, kva, kva + size); 1097 VM_OBJECT_WLOCK(obj); 1098 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped")); 1099 shmfd->shm_kmappings--; 1100 VM_OBJECT_WUNLOCK(obj); 1101 return (0); 1102 } 1103 1104 static int 1105 shm_fill_kinfo_locked(struct shmfd *shmfd, struct kinfo_file *kif, bool list) 1106 { 1107 const char *path, *pr_path; 1108 size_t pr_pathlen; 1109 bool visible; 1110 1111 sx_assert(&shm_dict_lock, SA_LOCKED); 1112 kif->kf_type = KF_TYPE_SHM; 1113 kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode; 1114 kif->kf_un.kf_file.kf_file_size = shmfd->shm_size; 1115 if (shmfd->shm_path != NULL) { 1116 if (shmfd->shm_path != NULL) { 1117 path = shmfd->shm_path; 1118 pr_path = curthread->td_ucred->cr_prison->pr_path; 1119 if (strcmp(pr_path, "/") != 0) { 1120 /* Return the jail-rooted pathname. */ 1121 pr_pathlen = strlen(pr_path); 1122 visible = strncmp(path, pr_path, pr_pathlen) 1123 == 0 && path[pr_pathlen] == '/'; 1124 if (list && !visible) 1125 return (EPERM); 1126 if (visible) 1127 path += pr_pathlen; 1128 } 1129 strlcpy(kif->kf_path, path, sizeof(kif->kf_path)); 1130 } 1131 } 1132 return (0); 1133 } 1134 1135 static int 1136 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif, 1137 struct filedesc *fdp __unused) 1138 { 1139 int res; 1140 1141 sx_slock(&shm_dict_lock); 1142 res = shm_fill_kinfo_locked(fp->f_data, kif, false); 1143 sx_sunlock(&shm_dict_lock); 1144 return (res); 1145 } 1146 1147 static int 1148 sysctl_posix_shm_list(SYSCTL_HANDLER_ARGS) 1149 { 1150 struct shm_mapping *shmm; 1151 struct sbuf sb; 1152 struct kinfo_file kif; 1153 u_long i; 1154 ssize_t curlen; 1155 int error, error2; 1156 1157 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file) * 5, req); 1158 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 1159 curlen = 0; 1160 error = 0; 1161 sx_slock(&shm_dict_lock); 1162 for (i = 0; i < shm_hash + 1; i++) { 1163 LIST_FOREACH(shmm, &shm_dictionary[i], sm_link) { 1164 error = shm_fill_kinfo_locked(shmm->sm_shmfd, 1165 &kif, true); 1166 if (error == EPERM) 1167 continue; 1168 if (error != 0) 1169 break; 1170 pack_kinfo(&kif); 1171 if (req->oldptr != NULL && 1172 kif.kf_structsize + curlen > req->oldlen) 1173 break; 1174 error = sbuf_bcat(&sb, &kif, kif.kf_structsize) == 0 ? 1175 0 : ENOMEM; 1176 if (error != 0) 1177 break; 1178 curlen += kif.kf_structsize; 1179 } 1180 } 1181 sx_sunlock(&shm_dict_lock); 1182 error2 = sbuf_finish(&sb); 1183 sbuf_delete(&sb); 1184 return (error != 0 ? error : error2); 1185 } 1186 1187 SYSCTL_PROC(_kern_ipc, OID_AUTO, posix_shm_list, 1188 CTLFLAG_RD | CTLFLAG_MPSAFE | CTLTYPE_OPAQUE, 1189 NULL, 0, sysctl_posix_shm_list, "", 1190 "POSIX SHM list"); 1191