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