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