1 /*- 2 * Copyright (c) 2006 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 * (2) 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 * (3) Add support for this file type to fstat(1). 39 * 40 * (4) Resource limits? Does this need its own resource limits or are the 41 * existing limits in mmap(2) sufficient? 42 * 43 * (5) Partial page truncation. vnode_pager_setsize() will zero any parts 44 * of a partially mapped page as a result of ftruncate(2)/truncate(2). 45 * We can do the same (with the same pmap evil), but do we need to 46 * worry about the bits on disk if the page is swapped out or will the 47 * swapper zero the parts of a page that are invalid if the page is 48 * swapped back in for us? 49 * 50 * (6) Add MAC support in mac_biba(4) and mac_mls(4). 51 * 52 * (7) Add a MAC check_create() hook for creating new named objects. 53 */ 54 55 #include <sys/cdefs.h> 56 __FBSDID("$FreeBSD$"); 57 58 #include "opt_mac.h" 59 60 #include <sys/param.h> 61 #include <sys/fcntl.h> 62 #include <sys/file.h> 63 #include <sys/filedesc.h> 64 #include <sys/fnv_hash.h> 65 #include <sys/kernel.h> 66 #include <sys/lock.h> 67 #include <sys/malloc.h> 68 #include <sys/mman.h> 69 #include <sys/mutex.h> 70 #include <sys/proc.h> 71 #include <sys/refcount.h> 72 #include <sys/resourcevar.h> 73 #include <sys/stat.h> 74 #include <sys/sysctl.h> 75 #include <sys/sysproto.h> 76 #include <sys/systm.h> 77 #include <sys/sx.h> 78 #include <sys/time.h> 79 #include <sys/vnode.h> 80 81 #include <security/mac/mac_framework.h> 82 83 #include <vm/vm.h> 84 #include <vm/vm_param.h> 85 #include <vm/pmap.h> 86 #include <vm/vm_map.h> 87 #include <vm/vm_object.h> 88 #include <vm/vm_page.h> 89 #include <vm/vm_pager.h> 90 #include <vm/swap_pager.h> 91 92 struct shm_mapping { 93 char *sm_path; 94 Fnv32_t sm_fnv; 95 struct shmfd *sm_shmfd; 96 LIST_ENTRY(shm_mapping) sm_link; 97 }; 98 99 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor"); 100 static LIST_HEAD(, shm_mapping) *shm_dictionary; 101 static struct sx shm_dict_lock; 102 static struct mtx shm_timestamp_lock; 103 static u_long shm_hash; 104 105 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash]) 106 107 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags); 108 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode); 109 static void shm_dict_init(void *arg); 110 static void shm_drop(struct shmfd *shmfd); 111 static struct shmfd *shm_hold(struct shmfd *shmfd); 112 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd); 113 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv); 114 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred); 115 static void shm_dotruncate(struct shmfd *shmfd, off_t length); 116 117 static fo_rdwr_t shm_read; 118 static fo_rdwr_t shm_write; 119 static fo_truncate_t shm_truncate; 120 static fo_ioctl_t shm_ioctl; 121 static fo_poll_t shm_poll; 122 static fo_kqfilter_t shm_kqfilter; 123 static fo_stat_t shm_stat; 124 static fo_close_t shm_close; 125 126 /* File descriptor operations. */ 127 static struct fileops shm_ops = { 128 .fo_read = shm_read, 129 .fo_write = shm_write, 130 .fo_truncate = shm_truncate, 131 .fo_ioctl = shm_ioctl, 132 .fo_poll = shm_poll, 133 .fo_kqfilter = shm_kqfilter, 134 .fo_stat = shm_stat, 135 .fo_close = shm_close, 136 .fo_flags = DFLAG_PASSABLE 137 }; 138 139 FEATURE(posix_shm, "POSIX shared memory"); 140 141 static int 142 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 143 int flags, struct thread *td) 144 { 145 146 return (EOPNOTSUPP); 147 } 148 149 static int 150 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 151 int flags, struct thread *td) 152 { 153 154 return (EOPNOTSUPP); 155 } 156 157 static int 158 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred, 159 struct thread *td) 160 { 161 struct shmfd *shmfd; 162 #ifdef MAC 163 int error; 164 #endif 165 166 shmfd = fp->f_data; 167 #ifdef MAC 168 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd); 169 if (error) 170 return (error); 171 #endif 172 shm_dotruncate(shmfd, length); 173 return (0); 174 } 175 176 static int 177 shm_ioctl(struct file *fp, u_long com, void *data, 178 struct ucred *active_cred, struct thread *td) 179 { 180 181 return (EOPNOTSUPP); 182 } 183 184 static int 185 shm_poll(struct file *fp, int events, struct ucred *active_cred, 186 struct thread *td) 187 { 188 189 return (EOPNOTSUPP); 190 } 191 192 static int 193 shm_kqfilter(struct file *fp, struct knote *kn) 194 { 195 196 return (EOPNOTSUPP); 197 } 198 199 static int 200 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 201 struct thread *td) 202 { 203 struct shmfd *shmfd; 204 #ifdef MAC 205 int error; 206 #endif 207 208 shmfd = fp->f_data; 209 210 #ifdef MAC 211 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd); 212 if (error) 213 return (error); 214 #endif 215 216 /* 217 * Attempt to return sanish values for fstat() on a memory file 218 * descriptor. 219 */ 220 bzero(sb, sizeof(*sb)); 221 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 222 sb->st_blksize = PAGE_SIZE; 223 sb->st_size = shmfd->shm_size; 224 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize; 225 sb->st_atimespec = shmfd->shm_atime; 226 sb->st_ctimespec = shmfd->shm_ctime; 227 sb->st_mtimespec = shmfd->shm_mtime; 228 sb->st_birthtimespec = shmfd->shm_birthtime; 229 sb->st_uid = shmfd->shm_uid; 230 sb->st_gid = shmfd->shm_gid; 231 232 return (0); 233 } 234 235 static int 236 shm_close(struct file *fp, struct thread *td) 237 { 238 struct shmfd *shmfd; 239 240 shmfd = fp->f_data; 241 fp->f_data = NULL; 242 shm_drop(shmfd); 243 244 return (0); 245 } 246 247 static void 248 shm_dotruncate(struct shmfd *shmfd, off_t length) 249 { 250 vm_object_t object; 251 vm_page_t m; 252 vm_pindex_t nobjsize; 253 254 object = shmfd->shm_object; 255 VM_OBJECT_LOCK(object); 256 if (length == shmfd->shm_size) { 257 VM_OBJECT_UNLOCK(object); 258 return; 259 } 260 nobjsize = OFF_TO_IDX(length + PAGE_MASK); 261 262 /* Are we shrinking? If so, trim the end. */ 263 if (length < shmfd->shm_size) { 264 /* Toss in memory pages. */ 265 if (nobjsize < object->size) 266 vm_object_page_remove(object, nobjsize, object->size, 267 FALSE); 268 269 /* Toss pages from swap. */ 270 if (object->type == OBJT_SWAP) 271 swap_pager_freespace(object, nobjsize, 272 object->size - nobjsize); 273 274 /* 275 * If the last page is partially mapped, then zero out 276 * the garbage at the end of the page. See comments 277 * in vnode_pager_setsize() for more details. 278 * 279 * XXXJHB: This handles in memory pages, but what about 280 * a page swapped out to disk? 281 */ 282 if ((length & PAGE_MASK) && 283 (m = vm_page_lookup(object, OFF_TO_IDX(length))) != NULL && 284 m->valid != 0) { 285 int base = (int)length & PAGE_MASK; 286 int size = PAGE_SIZE - base; 287 288 pmap_zero_page_area(m, base, size); 289 290 /* 291 * Update the valid bits to reflect the blocks that 292 * have been zeroed. Some of these valid bits may 293 * have already been set. 294 */ 295 vm_page_set_valid(m, base, size); 296 297 /* 298 * Round "base" to the next block boundary so that the 299 * dirty bit for a partially zeroed block is not 300 * cleared. 301 */ 302 base = roundup2(base, DEV_BSIZE); 303 304 vm_page_lock_queues(); 305 vm_page_clear_dirty(m, base, PAGE_SIZE - base); 306 vm_page_unlock_queues(); 307 } else if ((length & PAGE_MASK) && 308 __predict_false(object->cache != NULL)) { 309 vm_page_cache_free(object, OFF_TO_IDX(length), 310 nobjsize); 311 } 312 } 313 shmfd->shm_size = length; 314 mtx_lock(&shm_timestamp_lock); 315 vfs_timestamp(&shmfd->shm_ctime); 316 shmfd->shm_mtime = shmfd->shm_ctime; 317 mtx_unlock(&shm_timestamp_lock); 318 object->size = nobjsize; 319 VM_OBJECT_UNLOCK(object); 320 } 321 322 /* 323 * shmfd object management including creation and reference counting 324 * routines. 325 */ 326 static struct shmfd * 327 shm_alloc(struct ucred *ucred, mode_t mode) 328 { 329 struct shmfd *shmfd; 330 331 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO); 332 shmfd->shm_size = 0; 333 shmfd->shm_uid = ucred->cr_uid; 334 shmfd->shm_gid = ucred->cr_gid; 335 shmfd->shm_mode = mode; 336 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL, 337 shmfd->shm_size, VM_PROT_DEFAULT, 0); 338 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate")); 339 VM_OBJECT_LOCK(shmfd->shm_object); 340 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING); 341 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT); 342 VM_OBJECT_UNLOCK(shmfd->shm_object); 343 vfs_timestamp(&shmfd->shm_birthtime); 344 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime = 345 shmfd->shm_birthtime; 346 refcount_init(&shmfd->shm_refs, 1); 347 #ifdef MAC 348 mac_posixshm_init(shmfd); 349 mac_posixshm_create(ucred, shmfd); 350 #endif 351 352 return (shmfd); 353 } 354 355 static struct shmfd * 356 shm_hold(struct shmfd *shmfd) 357 { 358 359 refcount_acquire(&shmfd->shm_refs); 360 return (shmfd); 361 } 362 363 static void 364 shm_drop(struct shmfd *shmfd) 365 { 366 367 if (refcount_release(&shmfd->shm_refs)) { 368 #ifdef MAC 369 mac_posixshm_destroy(shmfd); 370 #endif 371 vm_object_deallocate(shmfd->shm_object); 372 free(shmfd, M_SHMFD); 373 } 374 } 375 376 /* 377 * Determine if the credentials have sufficient permissions for a 378 * specified combination of FREAD and FWRITE. 379 */ 380 static int 381 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags) 382 { 383 accmode_t accmode; 384 385 accmode = 0; 386 if (flags & FREAD) 387 accmode |= VREAD; 388 if (flags & FWRITE) 389 accmode |= VWRITE; 390 return (vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid, 391 accmode, ucred, NULL)); 392 } 393 394 /* 395 * Dictionary management. We maintain an in-kernel dictionary to map 396 * paths to shmfd objects. We use the FNV hash on the path to store 397 * the mappings in a hash table. 398 */ 399 static void 400 shm_dict_init(void *arg) 401 { 402 403 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF); 404 sx_init(&shm_dict_lock, "shm dictionary"); 405 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash); 406 } 407 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL); 408 409 static struct shmfd * 410 shm_lookup(char *path, Fnv32_t fnv) 411 { 412 struct shm_mapping *map; 413 414 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 415 if (map->sm_fnv != fnv) 416 continue; 417 if (strcmp(map->sm_path, path) == 0) 418 return (map->sm_shmfd); 419 } 420 421 return (NULL); 422 } 423 424 static void 425 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd) 426 { 427 struct shm_mapping *map; 428 429 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK); 430 map->sm_path = path; 431 map->sm_fnv = fnv; 432 map->sm_shmfd = shm_hold(shmfd); 433 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link); 434 } 435 436 static int 437 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred) 438 { 439 struct shm_mapping *map; 440 int error; 441 442 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 443 if (map->sm_fnv != fnv) 444 continue; 445 if (strcmp(map->sm_path, path) == 0) { 446 #ifdef MAC 447 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd); 448 if (error) 449 return (error); 450 #endif 451 error = shm_access(map->sm_shmfd, ucred, 452 FREAD | FWRITE); 453 if (error) 454 return (error); 455 LIST_REMOVE(map, sm_link); 456 shm_drop(map->sm_shmfd); 457 free(map->sm_path, M_SHMFD); 458 free(map, M_SHMFD); 459 return (0); 460 } 461 } 462 463 return (ENOENT); 464 } 465 466 /* System calls. */ 467 int 468 shm_open(struct thread *td, struct shm_open_args *uap) 469 { 470 struct filedesc *fdp; 471 struct shmfd *shmfd; 472 struct file *fp; 473 char *path; 474 Fnv32_t fnv; 475 mode_t cmode; 476 int fd, error; 477 478 if ((uap->flags & O_ACCMODE) != O_RDONLY && 479 (uap->flags & O_ACCMODE) != O_RDWR) 480 return (EINVAL); 481 482 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0) 483 return (EINVAL); 484 485 fdp = td->td_proc->p_fd; 486 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS; 487 488 error = falloc(td, &fp, &fd); 489 if (error) 490 return (error); 491 492 /* A SHM_ANON path pointer creates an anonymous object. */ 493 if (uap->path == SHM_ANON) { 494 /* A read-only anonymous object is pointless. */ 495 if ((uap->flags & O_ACCMODE) == O_RDONLY) { 496 fdclose(fdp, fp, fd, td); 497 fdrop(fp, td); 498 return (EINVAL); 499 } 500 shmfd = shm_alloc(td->td_ucred, cmode); 501 } else { 502 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK); 503 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 504 505 /* Require paths to start with a '/' character. */ 506 if (error == 0 && path[0] != '/') 507 error = EINVAL; 508 if (error) { 509 fdclose(fdp, fp, fd, td); 510 fdrop(fp, td); 511 free(path, M_SHMFD); 512 return (error); 513 } 514 515 fnv = fnv_32_str(path, FNV1_32_INIT); 516 sx_xlock(&shm_dict_lock); 517 shmfd = shm_lookup(path, fnv); 518 if (shmfd == NULL) { 519 /* Object does not yet exist, create it if requested. */ 520 if (uap->flags & O_CREAT) { 521 shmfd = shm_alloc(td->td_ucred, cmode); 522 shm_insert(path, fnv, shmfd); 523 } else { 524 free(path, M_SHMFD); 525 error = ENOENT; 526 } 527 } else { 528 /* 529 * Object already exists, obtain a new 530 * reference if requested and permitted. 531 */ 532 free(path, M_SHMFD); 533 if ((uap->flags & (O_CREAT | O_EXCL)) == 534 (O_CREAT | O_EXCL)) 535 error = EEXIST; 536 else { 537 #ifdef MAC 538 error = mac_posixshm_check_open(td->td_ucred, 539 shmfd); 540 if (error == 0) 541 #endif 542 error = shm_access(shmfd, td->td_ucred, 543 FFLAGS(uap->flags & O_ACCMODE)); 544 } 545 546 /* 547 * Truncate the file back to zero length if 548 * O_TRUNC was specified and the object was 549 * opened with read/write. 550 */ 551 if (error == 0 && 552 (uap->flags & (O_ACCMODE | O_TRUNC)) == 553 (O_RDWR | O_TRUNC)) { 554 #ifdef MAC 555 error = mac_posixshm_check_truncate( 556 td->td_ucred, fp->f_cred, shmfd); 557 if (error == 0) 558 #endif 559 shm_dotruncate(shmfd, 0); 560 } 561 if (error == 0) 562 shm_hold(shmfd); 563 } 564 sx_xunlock(&shm_dict_lock); 565 566 if (error) { 567 fdclose(fdp, fp, fd, td); 568 fdrop(fp, td); 569 return (error); 570 } 571 } 572 573 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops); 574 575 FILEDESC_XLOCK(fdp); 576 if (fdp->fd_ofiles[fd] == fp) 577 fdp->fd_ofileflags[fd] |= UF_EXCLOSE; 578 FILEDESC_XUNLOCK(fdp); 579 td->td_retval[0] = fd; 580 fdrop(fp, td); 581 582 return (0); 583 } 584 585 int 586 shm_unlink(struct thread *td, struct shm_unlink_args *uap) 587 { 588 char *path; 589 Fnv32_t fnv; 590 int error; 591 592 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 593 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 594 if (error) { 595 free(path, M_TEMP); 596 return (error); 597 } 598 599 fnv = fnv_32_str(path, FNV1_32_INIT); 600 sx_xlock(&shm_dict_lock); 601 error = shm_remove(path, fnv, td->td_ucred); 602 sx_xunlock(&shm_dict_lock); 603 free(path, M_TEMP); 604 605 return (error); 606 } 607 608 /* 609 * mmap() helper to validate mmap() requests against shm object state 610 * and give mmap() the vm_object to use for the mapping. 611 */ 612 int 613 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff, 614 vm_object_t *obj) 615 { 616 617 /* 618 * XXXRW: This validation is probably insufficient, and subject to 619 * sign errors. It should be fixed. 620 */ 621 if (foff >= shmfd->shm_size || 622 foff + objsize > round_page(shmfd->shm_size)) 623 return (EINVAL); 624 625 mtx_lock(&shm_timestamp_lock); 626 vfs_timestamp(&shmfd->shm_atime); 627 mtx_unlock(&shm_timestamp_lock); 628 vm_object_reference(shmfd->shm_object); 629 *obj = shmfd->shm_object; 630 return (0); 631 } 632