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), shm_rename(2), and shm_unlink(2). 37 * While most of the implementation is here, vm_mmap.c contains 38 * mapping logic changes. 39 * 40 * posixshmcontrol(1) allows users to inspect the state of the memory 41 * objects. Per-uid swap resource limit controls total amount of 42 * memory that user can consume for anonymous objects, including 43 * shared. 44 */ 45 46 #include <sys/cdefs.h> 47 __FBSDID("$FreeBSD$"); 48 49 #include "opt_capsicum.h" 50 #include "opt_ktrace.h" 51 52 #include <sys/param.h> 53 #include <sys/capsicum.h> 54 #include <sys/conf.h> 55 #include <sys/fcntl.h> 56 #include <sys/file.h> 57 #include <sys/filedesc.h> 58 #include <sys/filio.h> 59 #include <sys/fnv_hash.h> 60 #include <sys/kernel.h> 61 #include <sys/limits.h> 62 #include <sys/uio.h> 63 #include <sys/signal.h> 64 #include <sys/jail.h> 65 #include <sys/ktrace.h> 66 #include <sys/lock.h> 67 #include <sys/malloc.h> 68 #include <sys/mman.h> 69 #include <sys/mutex.h> 70 #include <sys/priv.h> 71 #include <sys/proc.h> 72 #include <sys/refcount.h> 73 #include <sys/resourcevar.h> 74 #include <sys/rwlock.h> 75 #include <sys/sbuf.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 unrhdr64 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 static int shm_dotruncate_locked(struct shmfd *shmfd, off_t length, 124 void *rl_cookie); 125 static int shm_copyin_path(struct thread *td, const char *userpath_in, 126 char **path_out); 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 static fo_get_seals_t shm_get_seals; 140 static fo_add_seals_t shm_add_seals; 141 142 /* File descriptor operations. */ 143 struct fileops shm_ops = { 144 .fo_read = shm_read, 145 .fo_write = shm_write, 146 .fo_truncate = shm_truncate, 147 .fo_ioctl = shm_ioctl, 148 .fo_poll = invfo_poll, 149 .fo_kqfilter = invfo_kqfilter, 150 .fo_stat = shm_stat, 151 .fo_close = shm_close, 152 .fo_chmod = shm_chmod, 153 .fo_chown = shm_chown, 154 .fo_sendfile = vn_sendfile, 155 .fo_seek = shm_seek, 156 .fo_fill_kinfo = shm_fill_kinfo, 157 .fo_mmap = shm_mmap, 158 .fo_get_seals = shm_get_seals, 159 .fo_add_seals = shm_add_seals, 160 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 161 }; 162 163 FEATURE(posix_shm, "POSIX shared memory"); 164 165 static int 166 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio) 167 { 168 vm_page_t m; 169 vm_pindex_t idx; 170 size_t tlen; 171 int error, offset, rv; 172 173 idx = OFF_TO_IDX(uio->uio_offset); 174 offset = uio->uio_offset & PAGE_MASK; 175 tlen = MIN(PAGE_SIZE - offset, len); 176 177 VM_OBJECT_WLOCK(obj); 178 179 /* 180 * Read I/O without either a corresponding resident page or swap 181 * page: use zero_region. This is intended to avoid instantiating 182 * pages on read from a sparse region. 183 */ 184 if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL && 185 !vm_pager_has_page(obj, idx, NULL, NULL)) { 186 VM_OBJECT_WUNLOCK(obj); 187 return (uiomove(__DECONST(void *, zero_region), tlen, uio)); 188 } 189 190 /* 191 * Parallel reads of the page content from disk are prevented 192 * by exclusive busy. 193 * 194 * Although the tmpfs vnode lock is held here, it is 195 * nonetheless safe to sleep waiting for a free page. The 196 * pageout daemon does not need to acquire the tmpfs vnode 197 * lock to page out tobj's pages because tobj is a OBJT_SWAP 198 * type object. 199 */ 200 rv = vm_page_grab_valid(&m, obj, idx, 201 VM_ALLOC_NORMAL | VM_ALLOC_SBUSY | VM_ALLOC_IGN_SBUSY); 202 if (rv != VM_PAGER_OK) { 203 VM_OBJECT_WUNLOCK(obj); 204 printf("uiomove_object: vm_obj %p idx %jd pager error %d\n", 205 obj, idx, rv); 206 return (EIO); 207 } 208 VM_OBJECT_WUNLOCK(obj); 209 error = uiomove_fromphys(&m, offset, tlen, uio); 210 if (uio->uio_rw == UIO_WRITE && error == 0) 211 vm_page_set_dirty(m); 212 vm_page_activate(m); 213 vm_page_sunbusy(m); 214 215 return (error); 216 } 217 218 int 219 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio) 220 { 221 ssize_t resid; 222 size_t len; 223 int error; 224 225 error = 0; 226 while ((resid = uio->uio_resid) > 0) { 227 if (obj_size <= uio->uio_offset) 228 break; 229 len = MIN(obj_size - uio->uio_offset, resid); 230 if (len == 0) 231 break; 232 error = uiomove_object_page(obj, len, uio); 233 if (error != 0 || resid == uio->uio_resid) 234 break; 235 } 236 return (error); 237 } 238 239 static int 240 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td) 241 { 242 struct shmfd *shmfd; 243 off_t foffset; 244 int error; 245 246 shmfd = fp->f_data; 247 foffset = foffset_lock(fp, 0); 248 error = 0; 249 switch (whence) { 250 case L_INCR: 251 if (foffset < 0 || 252 (offset > 0 && foffset > OFF_MAX - offset)) { 253 error = EOVERFLOW; 254 break; 255 } 256 offset += foffset; 257 break; 258 case L_XTND: 259 if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) { 260 error = EOVERFLOW; 261 break; 262 } 263 offset += shmfd->shm_size; 264 break; 265 case L_SET: 266 break; 267 default: 268 error = EINVAL; 269 } 270 if (error == 0) { 271 if (offset < 0 || offset > shmfd->shm_size) 272 error = EINVAL; 273 else 274 td->td_uretoff.tdu_off = offset; 275 } 276 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); 277 return (error); 278 } 279 280 static int 281 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 282 int flags, struct thread *td) 283 { 284 struct shmfd *shmfd; 285 void *rl_cookie; 286 int error; 287 288 shmfd = fp->f_data; 289 #ifdef MAC 290 error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd); 291 if (error) 292 return (error); 293 #endif 294 foffset_lock_uio(fp, uio, flags); 295 rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset, 296 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx); 297 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio); 298 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 299 foffset_unlock_uio(fp, uio, flags); 300 return (error); 301 } 302 303 static int 304 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 305 int flags, struct thread *td) 306 { 307 struct shmfd *shmfd; 308 void *rl_cookie; 309 int error; 310 311 shmfd = fp->f_data; 312 #ifdef MAC 313 error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd); 314 if (error) 315 return (error); 316 #endif 317 foffset_lock_uio(fp, uio, flags); 318 if ((flags & FOF_OFFSET) == 0) { 319 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX, 320 &shmfd->shm_mtx); 321 } else { 322 rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset, 323 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx); 324 } 325 if ((shmfd->shm_seals & F_SEAL_WRITE) != 0) 326 error = EPERM; 327 else 328 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio); 329 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 330 foffset_unlock_uio(fp, uio, flags); 331 return (error); 332 } 333 334 static int 335 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred, 336 struct thread *td) 337 { 338 struct shmfd *shmfd; 339 #ifdef MAC 340 int error; 341 #endif 342 343 shmfd = fp->f_data; 344 #ifdef MAC 345 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd); 346 if (error) 347 return (error); 348 #endif 349 return (shm_dotruncate(shmfd, length)); 350 } 351 352 int 353 shm_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 354 struct thread *td) 355 { 356 357 switch (com) { 358 case FIONBIO: 359 case FIOASYNC: 360 /* 361 * Allow fcntl(fd, F_SETFL, O_NONBLOCK) to work, 362 * just like it would on an unlinked regular file 363 */ 364 return (0); 365 default: 366 return (ENOTTY); 367 } 368 } 369 370 static int 371 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 372 struct thread *td) 373 { 374 struct shmfd *shmfd; 375 #ifdef MAC 376 int error; 377 #endif 378 379 shmfd = fp->f_data; 380 381 #ifdef MAC 382 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd); 383 if (error) 384 return (error); 385 #endif 386 387 /* 388 * Attempt to return sanish values for fstat() on a memory file 389 * descriptor. 390 */ 391 bzero(sb, sizeof(*sb)); 392 sb->st_blksize = PAGE_SIZE; 393 sb->st_size = shmfd->shm_size; 394 sb->st_blocks = howmany(sb->st_size, sb->st_blksize); 395 mtx_lock(&shm_timestamp_lock); 396 sb->st_atim = shmfd->shm_atime; 397 sb->st_ctim = shmfd->shm_ctime; 398 sb->st_mtim = shmfd->shm_mtime; 399 sb->st_birthtim = shmfd->shm_birthtime; 400 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 401 sb->st_uid = shmfd->shm_uid; 402 sb->st_gid = shmfd->shm_gid; 403 mtx_unlock(&shm_timestamp_lock); 404 sb->st_dev = shm_dev_ino; 405 sb->st_ino = shmfd->shm_ino; 406 sb->st_nlink = shmfd->shm_object->ref_count; 407 408 return (0); 409 } 410 411 static int 412 shm_close(struct file *fp, struct thread *td) 413 { 414 struct shmfd *shmfd; 415 416 shmfd = fp->f_data; 417 fp->f_data = NULL; 418 shm_drop(shmfd); 419 420 return (0); 421 } 422 423 static int 424 shm_copyin_path(struct thread *td, const char *userpath_in, char **path_out) { 425 int error; 426 char *path; 427 const char *pr_path; 428 size_t pr_pathlen; 429 430 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK); 431 pr_path = td->td_ucred->cr_prison->pr_path; 432 433 /* Construct a full pathname for jailed callers. */ 434 pr_pathlen = strcmp(pr_path, "/") == 435 0 ? 0 : strlcpy(path, pr_path, MAXPATHLEN); 436 error = copyinstr(userpath_in, path + pr_pathlen, 437 MAXPATHLEN - pr_pathlen, NULL); 438 if (error != 0) 439 goto out; 440 441 #ifdef KTRACE 442 if (KTRPOINT(curthread, KTR_NAMEI)) 443 ktrnamei(path); 444 #endif 445 446 /* Require paths to start with a '/' character. */ 447 if (path[pr_pathlen] != '/') { 448 error = EINVAL; 449 goto out; 450 } 451 452 *path_out = path; 453 454 out: 455 if (error != 0) 456 free(path, M_SHMFD); 457 458 return (error); 459 } 460 461 static int 462 shm_dotruncate_locked(struct shmfd *shmfd, off_t length, void *rl_cookie) 463 { 464 vm_object_t object; 465 vm_page_t m; 466 vm_pindex_t idx, nobjsize; 467 vm_ooffset_t delta; 468 int base, rv; 469 470 KASSERT(length >= 0, ("shm_dotruncate: length < 0")); 471 object = shmfd->shm_object; 472 VM_OBJECT_ASSERT_WLOCKED(object); 473 rangelock_cookie_assert(rl_cookie, RA_WLOCKED); 474 if (length == shmfd->shm_size) 475 return (0); 476 nobjsize = OFF_TO_IDX(length + PAGE_MASK); 477 478 /* Are we shrinking? If so, trim the end. */ 479 if (length < shmfd->shm_size) { 480 if ((shmfd->shm_seals & F_SEAL_SHRINK) != 0) 481 return (EPERM); 482 483 /* 484 * Disallow any requests to shrink the size if this 485 * object is mapped into the kernel. 486 */ 487 if (shmfd->shm_kmappings > 0) 488 return (EBUSY); 489 490 /* 491 * Zero the truncated part of the last page. 492 */ 493 base = length & PAGE_MASK; 494 if (base != 0) { 495 idx = OFF_TO_IDX(length); 496 retry: 497 m = vm_page_grab(object, idx, VM_ALLOC_NOCREAT); 498 if (m != NULL) { 499 MPASS(vm_page_all_valid(m)); 500 } else if (vm_pager_has_page(object, idx, NULL, NULL)) { 501 m = vm_page_alloc(object, idx, 502 VM_ALLOC_NORMAL | VM_ALLOC_WAITFAIL); 503 if (m == NULL) 504 goto retry; 505 rv = vm_pager_get_pages(object, &m, 1, NULL, 506 NULL); 507 if (rv == VM_PAGER_OK) { 508 /* 509 * Since the page was not resident, 510 * and therefore not recently 511 * accessed, immediately enqueue it 512 * for asynchronous laundering. The 513 * current operation is not regarded 514 * as an access. 515 */ 516 vm_page_launder(m); 517 } else { 518 vm_page_free(m); 519 VM_OBJECT_WUNLOCK(object); 520 return (EIO); 521 } 522 } 523 if (m != NULL) { 524 pmap_zero_page_area(m, base, PAGE_SIZE - base); 525 KASSERT(vm_page_all_valid(m), 526 ("shm_dotruncate: page %p is invalid", m)); 527 vm_page_set_dirty(m); 528 vm_page_xunbusy(m); 529 } 530 } 531 delta = IDX_TO_OFF(object->size - nobjsize); 532 533 /* Toss in memory pages. */ 534 if (nobjsize < object->size) 535 vm_object_page_remove(object, nobjsize, object->size, 536 0); 537 538 /* Toss pages from swap. */ 539 if (object->type == OBJT_SWAP) 540 swap_pager_freespace(object, nobjsize, delta); 541 542 /* Free the swap accounted for shm */ 543 swap_release_by_cred(delta, object->cred); 544 object->charge -= delta; 545 } else { 546 if ((shmfd->shm_seals & F_SEAL_GROW) != 0) 547 return (EPERM); 548 549 /* Try to reserve additional swap space. */ 550 delta = IDX_TO_OFF(nobjsize - object->size); 551 if (!swap_reserve_by_cred(delta, object->cred)) 552 return (ENOMEM); 553 object->charge += delta; 554 } 555 shmfd->shm_size = length; 556 mtx_lock(&shm_timestamp_lock); 557 vfs_timestamp(&shmfd->shm_ctime); 558 shmfd->shm_mtime = shmfd->shm_ctime; 559 mtx_unlock(&shm_timestamp_lock); 560 object->size = nobjsize; 561 return (0); 562 } 563 564 int 565 shm_dotruncate(struct shmfd *shmfd, off_t length) 566 { 567 void *rl_cookie; 568 int error; 569 570 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX, 571 &shmfd->shm_mtx); 572 VM_OBJECT_WLOCK(shmfd->shm_object); 573 error = shm_dotruncate_locked(shmfd, length, rl_cookie); 574 VM_OBJECT_WUNLOCK(shmfd->shm_object); 575 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 576 return (error); 577 } 578 579 /* 580 * shmfd object management including creation and reference counting 581 * routines. 582 */ 583 struct shmfd * 584 shm_alloc(struct ucred *ucred, mode_t mode) 585 { 586 struct shmfd *shmfd; 587 588 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO); 589 shmfd->shm_size = 0; 590 shmfd->shm_uid = ucred->cr_uid; 591 shmfd->shm_gid = ucred->cr_gid; 592 shmfd->shm_mode = mode; 593 shmfd->shm_object = vm_pager_allocate(OBJT_SWAP, NULL, 594 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred); 595 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate")); 596 vfs_timestamp(&shmfd->shm_birthtime); 597 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime = 598 shmfd->shm_birthtime; 599 shmfd->shm_ino = alloc_unr64(&shm_ino_unr); 600 refcount_init(&shmfd->shm_refs, 1); 601 mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF); 602 rangelock_init(&shmfd->shm_rl); 603 #ifdef MAC 604 mac_posixshm_init(shmfd); 605 mac_posixshm_create(ucred, shmfd); 606 #endif 607 608 return (shmfd); 609 } 610 611 struct shmfd * 612 shm_hold(struct shmfd *shmfd) 613 { 614 615 refcount_acquire(&shmfd->shm_refs); 616 return (shmfd); 617 } 618 619 void 620 shm_drop(struct shmfd *shmfd) 621 { 622 623 if (refcount_release(&shmfd->shm_refs)) { 624 #ifdef MAC 625 mac_posixshm_destroy(shmfd); 626 #endif 627 rangelock_destroy(&shmfd->shm_rl); 628 mtx_destroy(&shmfd->shm_mtx); 629 vm_object_deallocate(shmfd->shm_object); 630 free(shmfd, M_SHMFD); 631 } 632 } 633 634 /* 635 * Determine if the credentials have sufficient permissions for a 636 * specified combination of FREAD and FWRITE. 637 */ 638 int 639 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags) 640 { 641 accmode_t accmode; 642 int error; 643 644 accmode = 0; 645 if (flags & FREAD) 646 accmode |= VREAD; 647 if (flags & FWRITE) 648 accmode |= VWRITE; 649 mtx_lock(&shm_timestamp_lock); 650 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid, 651 accmode, ucred, NULL); 652 mtx_unlock(&shm_timestamp_lock); 653 return (error); 654 } 655 656 /* 657 * Dictionary management. We maintain an in-kernel dictionary to map 658 * paths to shmfd objects. We use the FNV hash on the path to store 659 * the mappings in a hash table. 660 */ 661 static void 662 shm_init(void *arg) 663 { 664 665 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF); 666 sx_init(&shm_dict_lock, "shm dictionary"); 667 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash); 668 new_unrhdr64(&shm_ino_unr, 1); 669 shm_dev_ino = devfs_alloc_cdp_inode(); 670 KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized")); 671 } 672 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL); 673 674 static struct shmfd * 675 shm_lookup(char *path, Fnv32_t fnv) 676 { 677 struct shm_mapping *map; 678 679 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 680 if (map->sm_fnv != fnv) 681 continue; 682 if (strcmp(map->sm_path, path) == 0) 683 return (map->sm_shmfd); 684 } 685 686 return (NULL); 687 } 688 689 static void 690 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd) 691 { 692 struct shm_mapping *map; 693 694 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK); 695 map->sm_path = path; 696 map->sm_fnv = fnv; 697 map->sm_shmfd = shm_hold(shmfd); 698 shmfd->shm_path = path; 699 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link); 700 } 701 702 static int 703 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred) 704 { 705 struct shm_mapping *map; 706 int error; 707 708 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 709 if (map->sm_fnv != fnv) 710 continue; 711 if (strcmp(map->sm_path, path) == 0) { 712 #ifdef MAC 713 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd); 714 if (error) 715 return (error); 716 #endif 717 error = shm_access(map->sm_shmfd, ucred, 718 FREAD | FWRITE); 719 if (error) 720 return (error); 721 map->sm_shmfd->shm_path = NULL; 722 LIST_REMOVE(map, sm_link); 723 shm_drop(map->sm_shmfd); 724 free(map->sm_path, M_SHMFD); 725 free(map, M_SHMFD); 726 return (0); 727 } 728 } 729 730 return (ENOENT); 731 } 732 733 int 734 kern_shm_open(struct thread *td, const char *userpath, int flags, mode_t mode, 735 struct filecaps *fcaps, int initial_seals) 736 { 737 struct filedesc *fdp; 738 struct shmfd *shmfd; 739 struct file *fp; 740 char *path; 741 void *rl_cookie; 742 Fnv32_t fnv; 743 mode_t cmode; 744 int fd, error; 745 746 #ifdef CAPABILITY_MODE 747 /* 748 * shm_open(2) is only allowed for anonymous objects. 749 */ 750 if (IN_CAPABILITY_MODE(td) && (userpath != SHM_ANON)) 751 return (ECAPMODE); 752 #endif 753 754 AUDIT_ARG_FFLAGS(flags); 755 AUDIT_ARG_MODE(mode); 756 757 if ((flags & O_ACCMODE) != O_RDONLY && (flags & O_ACCMODE) != O_RDWR) 758 return (EINVAL); 759 760 if ((flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0) 761 return (EINVAL); 762 763 /* 764 * Currently only F_SEAL_SEAL may be set when creating or opening shmfd. 765 * If the decision is made later to allow additional seals, care must be 766 * taken below to ensure that the seals are properly set if the shmfd 767 * already existed -- this currently assumes that only F_SEAL_SEAL can 768 * be set and doesn't take further precautions to ensure the validity of 769 * the seals being added with respect to current mappings. 770 */ 771 if ((initial_seals & ~F_SEAL_SEAL) != 0) 772 return (EINVAL); 773 774 fdp = td->td_proc->p_fd; 775 cmode = (mode & ~fdp->fd_cmask) & ACCESSPERMS; 776 777 /* 778 * shm_open(2) created shm should always have O_CLOEXEC set, as mandated 779 * by POSIX. We allow it to be unset here so that an in-kernel 780 * interface may be written as a thin layer around shm, optionally not 781 * setting CLOEXEC. For shm_open(2), O_CLOEXEC is set unconditionally 782 * in sys_shm_open() to keep this implementation compliant. 783 */ 784 error = falloc_caps(td, &fp, &fd, flags & O_CLOEXEC, fcaps); 785 if (error) 786 return (error); 787 788 /* A SHM_ANON path pointer creates an anonymous object. */ 789 if (userpath == SHM_ANON) { 790 /* A read-only anonymous object is pointless. */ 791 if ((flags & O_ACCMODE) == O_RDONLY) { 792 fdclose(td, fp, fd); 793 fdrop(fp, td); 794 return (EINVAL); 795 } 796 shmfd = shm_alloc(td->td_ucred, cmode); 797 shmfd->shm_seals = initial_seals; 798 } else { 799 error = shm_copyin_path(td, userpath, &path); 800 if (error != 0) { 801 fdclose(td, fp, fd); 802 fdrop(fp, td); 803 return (error); 804 } 805 806 AUDIT_ARG_UPATH1_CANON(path); 807 fnv = fnv_32_str(path, FNV1_32_INIT); 808 sx_xlock(&shm_dict_lock); 809 shmfd = shm_lookup(path, fnv); 810 if (shmfd == NULL) { 811 /* Object does not yet exist, create it if requested. */ 812 if (flags & O_CREAT) { 813 #ifdef MAC 814 error = mac_posixshm_check_create(td->td_ucred, 815 path); 816 if (error == 0) { 817 #endif 818 shmfd = shm_alloc(td->td_ucred, cmode); 819 shmfd->shm_seals = initial_seals; 820 shm_insert(path, fnv, shmfd); 821 #ifdef MAC 822 } 823 #endif 824 } else { 825 free(path, M_SHMFD); 826 error = ENOENT; 827 } 828 } else { 829 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX, 830 &shmfd->shm_mtx); 831 832 /* 833 * kern_shm_open() likely shouldn't ever error out on 834 * trying to set a seal that already exists, unlike 835 * F_ADD_SEALS. This would break terribly as 836 * shm_open(2) actually sets F_SEAL_SEAL to maintain 837 * historical behavior where the underlying file could 838 * not be sealed. 839 */ 840 initial_seals &= ~shmfd->shm_seals; 841 842 /* 843 * Object already exists, obtain a new 844 * reference if requested and permitted. 845 */ 846 free(path, M_SHMFD); 847 848 /* 849 * initial_seals can't set additional seals if we've 850 * already been set F_SEAL_SEAL. If F_SEAL_SEAL is set, 851 * then we've already removed that one from 852 * initial_seals. This is currently redundant as we 853 * only allow setting F_SEAL_SEAL at creation time, but 854 * it's cheap to check and decreases the effort required 855 * to allow additional seals. 856 */ 857 if ((shmfd->shm_seals & F_SEAL_SEAL) != 0 && 858 initial_seals != 0) 859 error = EPERM; 860 else if ((flags & (O_CREAT | O_EXCL)) == 861 (O_CREAT | O_EXCL)) 862 error = EEXIST; 863 else { 864 #ifdef MAC 865 error = mac_posixshm_check_open(td->td_ucred, 866 shmfd, FFLAGS(flags & O_ACCMODE)); 867 if (error == 0) 868 #endif 869 error = shm_access(shmfd, td->td_ucred, 870 FFLAGS(flags & O_ACCMODE)); 871 } 872 873 /* 874 * Truncate the file back to zero length if 875 * O_TRUNC was specified and the object was 876 * opened with read/write. 877 */ 878 if (error == 0 && 879 (flags & (O_ACCMODE | O_TRUNC)) == 880 (O_RDWR | O_TRUNC)) { 881 VM_OBJECT_WLOCK(shmfd->shm_object); 882 #ifdef MAC 883 error = mac_posixshm_check_truncate( 884 td->td_ucred, fp->f_cred, shmfd); 885 if (error == 0) 886 #endif 887 error = shm_dotruncate_locked(shmfd, 0, 888 rl_cookie); 889 VM_OBJECT_WUNLOCK(shmfd->shm_object); 890 } 891 if (error == 0) { 892 /* 893 * Currently we only allow F_SEAL_SEAL to be 894 * set initially. As noted above, this would 895 * need to be reworked should that change. 896 */ 897 shmfd->shm_seals |= initial_seals; 898 shm_hold(shmfd); 899 } 900 rangelock_unlock(&shmfd->shm_rl, rl_cookie, 901 &shmfd->shm_mtx); 902 } 903 sx_xunlock(&shm_dict_lock); 904 905 if (error) { 906 fdclose(td, fp, fd); 907 fdrop(fp, td); 908 return (error); 909 } 910 } 911 912 finit(fp, FFLAGS(flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops); 913 914 td->td_retval[0] = fd; 915 fdrop(fp, td); 916 917 return (0); 918 } 919 920 /* System calls. */ 921 #ifdef COMPAT_FREEBSD12 922 int 923 freebsd12_shm_open(struct thread *td, struct freebsd12_shm_open_args *uap) 924 { 925 926 return (kern_shm_open(td, uap->path, uap->flags | O_CLOEXEC, uap->mode, 927 NULL, F_SEAL_SEAL)); 928 } 929 #endif 930 931 int 932 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap) 933 { 934 char *path; 935 Fnv32_t fnv; 936 int error; 937 938 error = shm_copyin_path(td, uap->path, &path); 939 if (error != 0) 940 return (error); 941 942 AUDIT_ARG_UPATH1_CANON(path); 943 fnv = fnv_32_str(path, FNV1_32_INIT); 944 sx_xlock(&shm_dict_lock); 945 error = shm_remove(path, fnv, td->td_ucred); 946 sx_xunlock(&shm_dict_lock); 947 free(path, M_TEMP); 948 949 return (error); 950 } 951 952 int 953 sys_shm_rename(struct thread *td, struct shm_rename_args *uap) 954 { 955 char *path_from = NULL, *path_to = NULL; 956 Fnv32_t fnv_from, fnv_to; 957 struct shmfd *fd_from; 958 struct shmfd *fd_to; 959 int error; 960 int flags; 961 962 flags = uap->flags; 963 AUDIT_ARG_FFLAGS(flags); 964 965 /* 966 * Make sure the user passed only valid flags. 967 * If you add a new flag, please add a new term here. 968 */ 969 if ((flags & ~( 970 SHM_RENAME_NOREPLACE | 971 SHM_RENAME_EXCHANGE 972 )) != 0) { 973 error = EINVAL; 974 goto out; 975 } 976 977 /* 978 * EXCHANGE and NOREPLACE don't quite make sense together. Let's 979 * force the user to choose one or the other. 980 */ 981 if ((flags & SHM_RENAME_NOREPLACE) != 0 && 982 (flags & SHM_RENAME_EXCHANGE) != 0) { 983 error = EINVAL; 984 goto out; 985 } 986 987 /* Renaming to or from anonymous makes no sense */ 988 if (uap->path_from == SHM_ANON || uap->path_to == SHM_ANON) { 989 error = EINVAL; 990 goto out; 991 } 992 993 error = shm_copyin_path(td, uap->path_from, &path_from); 994 if (error != 0) 995 goto out; 996 997 error = shm_copyin_path(td, uap->path_to, &path_to); 998 if (error != 0) 999 goto out; 1000 1001 AUDIT_ARG_UPATH1_CANON(path_from); 1002 AUDIT_ARG_UPATH2_CANON(path_to); 1003 1004 /* Rename with from/to equal is a no-op */ 1005 if (strcmp(path_from, path_to) == 0) 1006 goto out; 1007 1008 fnv_from = fnv_32_str(path_from, FNV1_32_INIT); 1009 fnv_to = fnv_32_str(path_to, FNV1_32_INIT); 1010 1011 sx_xlock(&shm_dict_lock); 1012 1013 fd_from = shm_lookup(path_from, fnv_from); 1014 if (fd_from == NULL) { 1015 error = ENOENT; 1016 goto out_locked; 1017 } 1018 1019 fd_to = shm_lookup(path_to, fnv_to); 1020 if ((flags & SHM_RENAME_NOREPLACE) != 0 && fd_to != NULL) { 1021 error = EEXIST; 1022 goto out_locked; 1023 } 1024 1025 /* 1026 * Unconditionally prevents shm_remove from invalidating the 'from' 1027 * shm's state. 1028 */ 1029 shm_hold(fd_from); 1030 error = shm_remove(path_from, fnv_from, td->td_ucred); 1031 1032 /* 1033 * One of my assumptions failed if ENOENT (e.g. locking didn't 1034 * protect us) 1035 */ 1036 KASSERT(error != ENOENT, ("Our shm disappeared during shm_rename: %s", 1037 path_from)); 1038 if (error != 0) { 1039 shm_drop(fd_from); 1040 goto out_locked; 1041 } 1042 1043 /* 1044 * If we are exchanging, we need to ensure the shm_remove below 1045 * doesn't invalidate the dest shm's state. 1046 */ 1047 if ((flags & SHM_RENAME_EXCHANGE) != 0 && fd_to != NULL) 1048 shm_hold(fd_to); 1049 1050 /* 1051 * NOTE: if path_to is not already in the hash, c'est la vie; 1052 * it simply means we have nothing already at path_to to unlink. 1053 * That is the ENOENT case. 1054 * 1055 * If we somehow don't have access to unlink this guy, but 1056 * did for the shm at path_from, then relink the shm to path_from 1057 * and abort with EACCES. 1058 * 1059 * All other errors: that is weird; let's relink and abort the 1060 * operation. 1061 */ 1062 error = shm_remove(path_to, fnv_to, td->td_ucred); 1063 if (error != 0 && error != ENOENT) { 1064 shm_insert(path_from, fnv_from, fd_from); 1065 shm_drop(fd_from); 1066 /* Don't free path_from now, since the hash references it */ 1067 path_from = NULL; 1068 goto out_locked; 1069 } 1070 1071 error = 0; 1072 1073 shm_insert(path_to, fnv_to, fd_from); 1074 1075 /* Don't free path_to now, since the hash references it */ 1076 path_to = NULL; 1077 1078 /* We kept a ref when we removed, and incremented again in insert */ 1079 shm_drop(fd_from); 1080 KASSERT(fd_from->shm_refs > 0, ("Expected >0 refs; got: %d\n", 1081 fd_from->shm_refs)); 1082 1083 if ((flags & SHM_RENAME_EXCHANGE) != 0 && fd_to != NULL) { 1084 shm_insert(path_from, fnv_from, fd_to); 1085 path_from = NULL; 1086 shm_drop(fd_to); 1087 KASSERT(fd_to->shm_refs > 0, ("Expected >0 refs; got: %d\n", 1088 fd_to->shm_refs)); 1089 } 1090 1091 out_locked: 1092 sx_xunlock(&shm_dict_lock); 1093 1094 out: 1095 free(path_from, M_SHMFD); 1096 free(path_to, M_SHMFD); 1097 return (error); 1098 } 1099 1100 int 1101 shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t objsize, 1102 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, 1103 vm_ooffset_t foff, struct thread *td) 1104 { 1105 struct shmfd *shmfd; 1106 vm_prot_t maxprot; 1107 int error; 1108 bool writecnt; 1109 void *rl_cookie; 1110 1111 shmfd = fp->f_data; 1112 maxprot = VM_PROT_NONE; 1113 1114 rl_cookie = rangelock_rlock(&shmfd->shm_rl, 0, objsize, 1115 &shmfd->shm_mtx); 1116 /* FREAD should always be set. */ 1117 if ((fp->f_flag & FREAD) != 0) 1118 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ; 1119 if ((fp->f_flag & FWRITE) != 0) 1120 maxprot |= VM_PROT_WRITE; 1121 1122 writecnt = (flags & MAP_SHARED) != 0 && (prot & VM_PROT_WRITE) != 0; 1123 1124 if (writecnt && (shmfd->shm_seals & F_SEAL_WRITE) != 0) { 1125 error = EPERM; 1126 goto out; 1127 } 1128 1129 /* Don't permit shared writable mappings on read-only descriptors. */ 1130 if (writecnt && (maxprot & VM_PROT_WRITE) == 0) { 1131 error = EACCES; 1132 goto out; 1133 } 1134 maxprot &= cap_maxprot; 1135 1136 /* See comment in vn_mmap(). */ 1137 if ( 1138 #ifdef _LP64 1139 objsize > OFF_MAX || 1140 #endif 1141 foff < 0 || foff > OFF_MAX - objsize) { 1142 error = EINVAL; 1143 goto out; 1144 } 1145 1146 #ifdef MAC 1147 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, flags); 1148 if (error != 0) 1149 goto out; 1150 #endif 1151 1152 mtx_lock(&shm_timestamp_lock); 1153 vfs_timestamp(&shmfd->shm_atime); 1154 mtx_unlock(&shm_timestamp_lock); 1155 vm_object_reference(shmfd->shm_object); 1156 1157 if (writecnt) 1158 vm_pager_update_writecount(shmfd->shm_object, 0, objsize); 1159 error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags, 1160 shmfd->shm_object, foff, writecnt, td); 1161 if (error != 0) { 1162 if (writecnt) 1163 vm_pager_release_writecount(shmfd->shm_object, 0, 1164 objsize); 1165 vm_object_deallocate(shmfd->shm_object); 1166 } 1167 out: 1168 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 1169 return (error); 1170 } 1171 1172 static int 1173 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 1174 struct thread *td) 1175 { 1176 struct shmfd *shmfd; 1177 int error; 1178 1179 error = 0; 1180 shmfd = fp->f_data; 1181 mtx_lock(&shm_timestamp_lock); 1182 /* 1183 * SUSv4 says that x bits of permission need not be affected. 1184 * Be consistent with our shm_open there. 1185 */ 1186 #ifdef MAC 1187 error = mac_posixshm_check_setmode(active_cred, shmfd, mode); 1188 if (error != 0) 1189 goto out; 1190 #endif 1191 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, 1192 shmfd->shm_gid, VADMIN, active_cred, NULL); 1193 if (error != 0) 1194 goto out; 1195 shmfd->shm_mode = mode & ACCESSPERMS; 1196 out: 1197 mtx_unlock(&shm_timestamp_lock); 1198 return (error); 1199 } 1200 1201 static int 1202 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 1203 struct thread *td) 1204 { 1205 struct shmfd *shmfd; 1206 int error; 1207 1208 error = 0; 1209 shmfd = fp->f_data; 1210 mtx_lock(&shm_timestamp_lock); 1211 #ifdef MAC 1212 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid); 1213 if (error != 0) 1214 goto out; 1215 #endif 1216 if (uid == (uid_t)-1) 1217 uid = shmfd->shm_uid; 1218 if (gid == (gid_t)-1) 1219 gid = shmfd->shm_gid; 1220 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) || 1221 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) && 1222 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN))) 1223 goto out; 1224 shmfd->shm_uid = uid; 1225 shmfd->shm_gid = gid; 1226 out: 1227 mtx_unlock(&shm_timestamp_lock); 1228 return (error); 1229 } 1230 1231 /* 1232 * Helper routines to allow the backing object of a shared memory file 1233 * descriptor to be mapped in the kernel. 1234 */ 1235 int 1236 shm_map(struct file *fp, size_t size, off_t offset, void **memp) 1237 { 1238 struct shmfd *shmfd; 1239 vm_offset_t kva, ofs; 1240 vm_object_t obj; 1241 int rv; 1242 1243 if (fp->f_type != DTYPE_SHM) 1244 return (EINVAL); 1245 shmfd = fp->f_data; 1246 obj = shmfd->shm_object; 1247 VM_OBJECT_WLOCK(obj); 1248 /* 1249 * XXXRW: This validation is probably insufficient, and subject to 1250 * sign errors. It should be fixed. 1251 */ 1252 if (offset >= shmfd->shm_size || 1253 offset + size > round_page(shmfd->shm_size)) { 1254 VM_OBJECT_WUNLOCK(obj); 1255 return (EINVAL); 1256 } 1257 1258 shmfd->shm_kmappings++; 1259 vm_object_reference_locked(obj); 1260 VM_OBJECT_WUNLOCK(obj); 1261 1262 /* Map the object into the kernel_map and wire it. */ 1263 kva = vm_map_min(kernel_map); 1264 ofs = offset & PAGE_MASK; 1265 offset = trunc_page(offset); 1266 size = round_page(size + ofs); 1267 rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0, 1268 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE, 1269 VM_PROT_READ | VM_PROT_WRITE, 0); 1270 if (rv == KERN_SUCCESS) { 1271 rv = vm_map_wire(kernel_map, kva, kva + size, 1272 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 1273 if (rv == KERN_SUCCESS) { 1274 *memp = (void *)(kva + ofs); 1275 return (0); 1276 } 1277 vm_map_remove(kernel_map, kva, kva + size); 1278 } else 1279 vm_object_deallocate(obj); 1280 1281 /* On failure, drop our mapping reference. */ 1282 VM_OBJECT_WLOCK(obj); 1283 shmfd->shm_kmappings--; 1284 VM_OBJECT_WUNLOCK(obj); 1285 1286 return (vm_mmap_to_errno(rv)); 1287 } 1288 1289 /* 1290 * We require the caller to unmap the entire entry. This allows us to 1291 * safely decrement shm_kmappings when a mapping is removed. 1292 */ 1293 int 1294 shm_unmap(struct file *fp, void *mem, size_t size) 1295 { 1296 struct shmfd *shmfd; 1297 vm_map_entry_t entry; 1298 vm_offset_t kva, ofs; 1299 vm_object_t obj; 1300 vm_pindex_t pindex; 1301 vm_prot_t prot; 1302 boolean_t wired; 1303 vm_map_t map; 1304 int rv; 1305 1306 if (fp->f_type != DTYPE_SHM) 1307 return (EINVAL); 1308 shmfd = fp->f_data; 1309 kva = (vm_offset_t)mem; 1310 ofs = kva & PAGE_MASK; 1311 kva = trunc_page(kva); 1312 size = round_page(size + ofs); 1313 map = kernel_map; 1314 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry, 1315 &obj, &pindex, &prot, &wired); 1316 if (rv != KERN_SUCCESS) 1317 return (EINVAL); 1318 if (entry->start != kva || entry->end != kva + size) { 1319 vm_map_lookup_done(map, entry); 1320 return (EINVAL); 1321 } 1322 vm_map_lookup_done(map, entry); 1323 if (obj != shmfd->shm_object) 1324 return (EINVAL); 1325 vm_map_remove(map, kva, kva + size); 1326 VM_OBJECT_WLOCK(obj); 1327 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped")); 1328 shmfd->shm_kmappings--; 1329 VM_OBJECT_WUNLOCK(obj); 1330 return (0); 1331 } 1332 1333 static int 1334 shm_fill_kinfo_locked(struct shmfd *shmfd, struct kinfo_file *kif, bool list) 1335 { 1336 const char *path, *pr_path; 1337 size_t pr_pathlen; 1338 bool visible; 1339 1340 sx_assert(&shm_dict_lock, SA_LOCKED); 1341 kif->kf_type = KF_TYPE_SHM; 1342 kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode; 1343 kif->kf_un.kf_file.kf_file_size = shmfd->shm_size; 1344 if (shmfd->shm_path != NULL) { 1345 if (shmfd->shm_path != NULL) { 1346 path = shmfd->shm_path; 1347 pr_path = curthread->td_ucred->cr_prison->pr_path; 1348 if (strcmp(pr_path, "/") != 0) { 1349 /* Return the jail-rooted pathname. */ 1350 pr_pathlen = strlen(pr_path); 1351 visible = strncmp(path, pr_path, pr_pathlen) 1352 == 0 && path[pr_pathlen] == '/'; 1353 if (list && !visible) 1354 return (EPERM); 1355 if (visible) 1356 path += pr_pathlen; 1357 } 1358 strlcpy(kif->kf_path, path, sizeof(kif->kf_path)); 1359 } 1360 } 1361 return (0); 1362 } 1363 1364 static int 1365 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif, 1366 struct filedesc *fdp __unused) 1367 { 1368 int res; 1369 1370 sx_slock(&shm_dict_lock); 1371 res = shm_fill_kinfo_locked(fp->f_data, kif, false); 1372 sx_sunlock(&shm_dict_lock); 1373 return (res); 1374 } 1375 1376 static int 1377 shm_add_seals(struct file *fp, int seals) 1378 { 1379 struct shmfd *shmfd; 1380 void *rl_cookie; 1381 vm_ooffset_t writemappings; 1382 int error, nseals; 1383 1384 error = 0; 1385 shmfd = fp->f_data; 1386 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX, 1387 &shmfd->shm_mtx); 1388 1389 /* Even already-set seals should result in EPERM. */ 1390 if ((shmfd->shm_seals & F_SEAL_SEAL) != 0) { 1391 error = EPERM; 1392 goto out; 1393 } 1394 nseals = seals & ~shmfd->shm_seals; 1395 if ((nseals & F_SEAL_WRITE) != 0) { 1396 /* 1397 * The rangelock above prevents writable mappings from being 1398 * added after we've started applying seals. The RLOCK here 1399 * is to avoid torn reads on ILP32 arches as unmapping/reducing 1400 * writemappings will be done without a rangelock. 1401 */ 1402 VM_OBJECT_RLOCK(shmfd->shm_object); 1403 writemappings = shmfd->shm_object->un_pager.swp.writemappings; 1404 VM_OBJECT_RUNLOCK(shmfd->shm_object); 1405 /* kmappings are also writable */ 1406 if (writemappings > 0) { 1407 error = EBUSY; 1408 goto out; 1409 } 1410 } 1411 shmfd->shm_seals |= nseals; 1412 out: 1413 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 1414 return (error); 1415 } 1416 1417 static int 1418 shm_get_seals(struct file *fp, int *seals) 1419 { 1420 struct shmfd *shmfd; 1421 1422 shmfd = fp->f_data; 1423 *seals = shmfd->shm_seals; 1424 return (0); 1425 } 1426 1427 static int 1428 sysctl_posix_shm_list(SYSCTL_HANDLER_ARGS) 1429 { 1430 struct shm_mapping *shmm; 1431 struct sbuf sb; 1432 struct kinfo_file kif; 1433 u_long i; 1434 ssize_t curlen; 1435 int error, error2; 1436 1437 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file) * 5, req); 1438 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 1439 curlen = 0; 1440 error = 0; 1441 sx_slock(&shm_dict_lock); 1442 for (i = 0; i < shm_hash + 1; i++) { 1443 LIST_FOREACH(shmm, &shm_dictionary[i], sm_link) { 1444 error = shm_fill_kinfo_locked(shmm->sm_shmfd, 1445 &kif, true); 1446 if (error == EPERM) 1447 continue; 1448 if (error != 0) 1449 break; 1450 pack_kinfo(&kif); 1451 if (req->oldptr != NULL && 1452 kif.kf_structsize + curlen > req->oldlen) 1453 break; 1454 error = sbuf_bcat(&sb, &kif, kif.kf_structsize) == 0 ? 1455 0 : ENOMEM; 1456 if (error != 0) 1457 break; 1458 curlen += kif.kf_structsize; 1459 } 1460 } 1461 sx_sunlock(&shm_dict_lock); 1462 error2 = sbuf_finish(&sb); 1463 sbuf_delete(&sb); 1464 return (error != 0 ? error : error2); 1465 } 1466 1467 SYSCTL_PROC(_kern_ipc, OID_AUTO, posix_shm_list, 1468 CTLFLAG_RD | CTLFLAG_MPSAFE | CTLTYPE_OPAQUE, 1469 NULL, 0, sysctl_posix_shm_list, "", 1470 "POSIX SHM list"); 1471 1472 int 1473 kern_shm_open2(struct thread *td, const char *path, int flags, mode_t mode, 1474 int shmflags, const char *name __unused) 1475 { 1476 int initial_seals; 1477 1478 if ((shmflags & ~SHM_ALLOW_SEALING) != 0) 1479 return (EINVAL); 1480 1481 initial_seals = F_SEAL_SEAL; 1482 if ((shmflags & SHM_ALLOW_SEALING) != 0) 1483 initial_seals &= ~F_SEAL_SEAL; 1484 return (kern_shm_open(td, path, flags, mode, NULL, initial_seals)); 1485 } 1486 1487 /* 1488 * This version of the shm_open() interface leaves CLOEXEC behavior up to the 1489 * caller, and libc will enforce it for the traditional shm_open() call. This 1490 * allows other consumers, like memfd_create(), to opt-in for CLOEXEC. This 1491 * interface also includes a 'name' argument that is currently unused, but could 1492 * potentially be exported later via some interface for debugging purposes. 1493 * From the kernel's perspective, it is optional. Individual consumers like 1494 * memfd_create() may require it in order to be compatible with other systems 1495 * implementing the same function. 1496 */ 1497 int 1498 sys_shm_open2(struct thread *td, struct shm_open2_args *uap) 1499 { 1500 1501 return (kern_shm_open2(td, uap->path, uap->flags, uap->mode, 1502 uap->shmflags, uap->name)); 1503 } 1504