sys/kern/uipc_shm.c

   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
  49 #include <sys/param.h>
  50 #include <sys/capability.h>
  51 #include <sys/fcntl.h>
  52 #include <sys/file.h>
  53 #include <sys/filedesc.h>
  54 #include <sys/fnv_hash.h>
  55 #include <sys/kernel.h>
  56 #include <sys/lock.h>
  57 #include <sys/malloc.h>
  58 #include <sys/mman.h>
  59 #include <sys/mutex.h>
  60 #include <sys/priv.h>
  61 #include <sys/proc.h>
  62 #include <sys/refcount.h>
  63 #include <sys/resourcevar.h>
  64 #include <sys/stat.h>
  65 #include <sys/sysctl.h>
  66 #include <sys/sysproto.h>
  67 #include <sys/systm.h>
  68 #include <sys/sx.h>
  69 #include <sys/time.h>
  70 #include <sys/vnode.h>
  71
  72 #include <security/mac/mac_framework.h>
  73
  74 #include <vm/vm.h>
  75 #include <vm/vm_param.h>
  76 #include <vm/pmap.h>
  77 #include <vm/vm_map.h>
  78 #include <vm/vm_object.h>
  79 #include <vm/vm_page.h>
  80 #include <vm/vm_pageout.h>
  81 #include <vm/vm_pager.h>
  82 #include <vm/swap_pager.h>
  83
  84 struct shm_mapping {
  85         char            *sm_path;
  86         Fnv32_t         sm_fnv;
  87         struct shmfd    *sm_shmfd;
  88         LIST_ENTRY(shm_mapping) sm_link;
  89 };
  90
  91 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
  92 static LIST_HEAD(, shm_mapping) *shm_dictionary;
  93 static struct sx shm_dict_lock;
  94 static struct mtx shm_timestamp_lock;
  95 static u_long shm_hash;
  96
  97 #define SHM_HASH(fnv)   (&shm_dictionary[(fnv) & shm_hash])
  98
  99 static int      shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
 100 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
 101 static void     shm_dict_init(void *arg);
 102 static void     shm_drop(struct shmfd *shmfd);
 103 static struct shmfd *shm_hold(struct shmfd *shmfd);
 104 static void     shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
 105 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
 106 static int      shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
 107 static int      shm_dotruncate(struct shmfd *shmfd, off_t length);
 108
 109 static fo_rdwr_t        shm_read;
 110 static fo_rdwr_t        shm_write;
 111 static fo_truncate_t    shm_truncate;
 112 static fo_ioctl_t       shm_ioctl;
 113 static fo_poll_t        shm_poll;
 114 static fo_kqfilter_t    shm_kqfilter;
 115 static fo_stat_t        shm_stat;
 116 static fo_close_t       shm_close;
 117 static fo_chmod_t       shm_chmod;
 118 static fo_chown_t       shm_chown;
 119
 120 /* File descriptor operations. */
 121 static struct fileops shm_ops = {
 122         .fo_read = shm_read,
 123         .fo_write = shm_write,
 124         .fo_truncate = shm_truncate,
 125         .fo_ioctl = shm_ioctl,
 126         .fo_poll = shm_poll,
 127         .fo_kqfilter = shm_kqfilter,
 128         .fo_stat = shm_stat,
 129         .fo_close = shm_close,
 130         .fo_chmod = shm_chmod,
 131         .fo_chown = shm_chown,
 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         return (shm_dotruncate(shmfd, length));
 169 }
 170
 171 static int
 172 shm_ioctl(struct file *fp, u_long com, void *data,
 173     struct ucred *active_cred, struct thread *td)
 174 {
 175
 176         return (EOPNOTSUPP);
 177 }
 178
 179 static int
 180 shm_poll(struct file *fp, int events, struct ucred *active_cred,
 181     struct thread *td)
 182 {
 183
 184         return (EOPNOTSUPP);
 185 }
 186
 187 static int
 188 shm_kqfilter(struct file *fp, struct knote *kn)
 189 {
 190
 191         return (EOPNOTSUPP);
 192 }
 193
 194 static int
 195 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
 196     struct thread *td)
 197 {
 198         struct shmfd *shmfd;
 199 #ifdef MAC
 200         int error;
 201 #endif
 202
 203         shmfd = fp->f_data;
 204
 205 #ifdef MAC
 206         error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
 207         if (error)
 208                 return (error);
 209 #endif
 210
 211         /*
 212          * Attempt to return sanish values for fstat() on a memory file
 213          * descriptor.
 214          */
 215         bzero(sb, sizeof(*sb));
 216         sb->st_blksize = PAGE_SIZE;
 217         sb->st_size = shmfd->shm_size;
 218         sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
 219         mtx_lock(&shm_timestamp_lock);
 220         sb->st_atim = shmfd->shm_atime;
 221         sb->st_ctim = shmfd->shm_ctime;
 222         sb->st_mtim = shmfd->shm_mtime;
 223         sb->st_birthtim = shmfd->shm_birthtime;
 224         sb->st_mode = S_IFREG | shmfd->shm_mode;                /* XXX */
 225         sb->st_uid = shmfd->shm_uid;
 226         sb->st_gid = shmfd->shm_gid;
 227         mtx_unlock(&shm_timestamp_lock);
 228
 229         return (0);
 230 }
 231
 232 static int
 233 shm_close(struct file *fp, struct thread *td)
 234 {
 235         struct shmfd *shmfd;
 236
 237         shmfd = fp->f_data;
 238         fp->f_data = NULL;
 239         shm_drop(shmfd);
 240
 241         return (0);
 242 }
 243
 244 static int
 245 shm_dotruncate(struct shmfd *shmfd, off_t length)
 246 {
 247         vm_object_t object;
 248         vm_page_t m, ma[1];
 249         vm_pindex_t idx, nobjsize;
 250         vm_ooffset_t delta;
 251         int base, rv;
 252
 253         object = shmfd->shm_object;
 254         VM_OBJECT_LOCK(object);
 255         if (length == shmfd->shm_size) {
 256                 VM_OBJECT_UNLOCK(object);
 257                 return (0);
 258         }
 259         nobjsize = OFF_TO_IDX(length + PAGE_MASK);
 260
 261         /* Are we shrinking?  If so, trim the end. */
 262         if (length < shmfd->shm_size) {
 263
 264                 /*
 265                  * Zero the truncated part of the last page.
 266                  */
 267                 base = length & PAGE_MASK;
 268                 if (base != 0) {
 269                         idx = OFF_TO_IDX(length);
 270 retry:
 271                         m = vm_page_lookup(object, idx);
 272                         if (m != NULL) {
 273                                 if ((m->oflags & VPO_BUSY) != 0 ||
 274                                     m->busy != 0) {
 275                                         vm_page_sleep(m, "shmtrc");
 276                                         goto retry;
 277                                 }
 278                         } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
 279                                 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
 280                                 if (m == NULL) {
 281                                         VM_OBJECT_UNLOCK(object);
 282                                         VM_WAIT;
 283                                         VM_OBJECT_LOCK(object);
 284                                         goto retry;
 285                                 } else if (m->valid != VM_PAGE_BITS_ALL) {
 286                                         ma[0] = m;
 287                                         rv = vm_pager_get_pages(object, ma, 1,
 288                                             0);
 289                                         m = vm_page_lookup(object, idx);
 290                                 } else
 291                                         /* A cached page was reactivated. */
 292                                         rv = VM_PAGER_OK;
 293                                 vm_page_lock(m);
 294                                 if (rv == VM_PAGER_OK) {
 295                                         vm_page_deactivate(m);
 296                                         vm_page_unlock(m);
 297                                         vm_page_wakeup(m);
 298                                 } else {
 299                                         vm_page_free(m);
 300                                         vm_page_unlock(m);
 301                                         VM_OBJECT_UNLOCK(object);
 302                                         return (EIO);
 303                                 }
 304                         }
 305                         if (m != NULL) {
 306                                 pmap_zero_page_area(m, base, PAGE_SIZE - base);
 307                                 KASSERT(m->valid == VM_PAGE_BITS_ALL,
 308                                     ("shm_dotruncate: page %p is invalid", m));
 309                                 vm_page_dirty(m);
 310                                 vm_pager_page_unswapped(m);
 311                         }
 312                 }
 313                 delta = ptoa(object->size - nobjsize);
 314
 315                 /* Toss in memory pages. */
 316                 if (nobjsize < object->size)
 317                         vm_object_page_remove(object, nobjsize, object->size,
 318                             0);
 319
 320                 /* Toss pages from swap. */
 321                 if (object->type == OBJT_SWAP)
 322                         swap_pager_freespace(object, nobjsize, delta);
 323
 324                 /* Free the swap accounted for shm */
 325                 swap_release_by_cred(delta, object->cred);
 326                 object->charge -= delta;
 327         } else {
 328                 /* Attempt to reserve the swap */
 329                 delta = ptoa(nobjsize - object->size);
 330                 if (!swap_reserve_by_cred(delta, object->cred)) {
 331                         VM_OBJECT_UNLOCK(object);
 332                         return (ENOMEM);
 333                 }
 334                 object->charge += delta;
 335         }
 336         shmfd->shm_size = length;
 337         mtx_lock(&shm_timestamp_lock);
 338         vfs_timestamp(&shmfd->shm_ctime);
 339         shmfd->shm_mtime = shmfd->shm_ctime;
 340         mtx_unlock(&shm_timestamp_lock);
 341         object->size = nobjsize;
 342         VM_OBJECT_UNLOCK(object);
 343         return (0);
 344 }
 345
 346 /*
 347  * shmfd object management including creation and reference counting
 348  * routines.
 349  */
 350 static struct shmfd *
 351 shm_alloc(struct ucred *ucred, mode_t mode)
 352 {
 353         struct shmfd *shmfd;
 354
 355         shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
 356         shmfd->shm_size = 0;
 357         shmfd->shm_uid = ucred->cr_uid;
 358         shmfd->shm_gid = ucred->cr_gid;
 359         shmfd->shm_mode = mode;
 360         shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
 361             shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
 362         KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
 363         VM_OBJECT_LOCK(shmfd->shm_object);
 364         vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
 365         vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
 366         VM_OBJECT_UNLOCK(shmfd->shm_object);
 367         vfs_timestamp(&shmfd->shm_birthtime);
 368         shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
 369             shmfd->shm_birthtime;
 370         refcount_init(&shmfd->shm_refs, 1);
 371 #ifdef MAC
 372         mac_posixshm_init(shmfd);
 373         mac_posixshm_create(ucred, shmfd);
 374 #endif
 375
 376         return (shmfd);
 377 }
 378
 379 static struct shmfd *
 380 shm_hold(struct shmfd *shmfd)
 381 {
 382
 383         refcount_acquire(&shmfd->shm_refs);
 384         return (shmfd);
 385 }
 386
 387 static void
 388 shm_drop(struct shmfd *shmfd)
 389 {
 390
 391         if (refcount_release(&shmfd->shm_refs)) {
 392 #ifdef MAC
 393                 mac_posixshm_destroy(shmfd);
 394 #endif
 395                 vm_object_deallocate(shmfd->shm_object);
 396                 free(shmfd, M_SHMFD);
 397         }
 398 }
 399
 400 /*
 401  * Determine if the credentials have sufficient permissions for a
 402  * specified combination of FREAD and FWRITE.
 403  */
 404 static int
 405 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
 406 {
 407         accmode_t accmode;
 408         int error;
 409
 410         accmode = 0;
 411         if (flags & FREAD)
 412                 accmode |= VREAD;
 413         if (flags & FWRITE)
 414                 accmode |= VWRITE;
 415         mtx_lock(&shm_timestamp_lock);
 416         error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
 417             accmode, ucred, NULL);
 418         mtx_unlock(&shm_timestamp_lock);
 419         return (error);
 420 }
 421
 422 /*
 423  * Dictionary management.  We maintain an in-kernel dictionary to map
 424  * paths to shmfd objects.  We use the FNV hash on the path to store
 425  * the mappings in a hash table.
 426  */
 427 static void
 428 shm_dict_init(void *arg)
 429 {
 430
 431         mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
 432         sx_init(&shm_dict_lock, "shm dictionary");
 433         shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
 434 }
 435 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
 436
 437 static struct shmfd *
 438 shm_lookup(char *path, Fnv32_t fnv)
 439 {
 440         struct shm_mapping *map;
 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                         return (map->sm_shmfd);
 447         }
 448
 449         return (NULL);
 450 }
 451
 452 static void
 453 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
 454 {
 455         struct shm_mapping *map;
 456
 457         map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
 458         map->sm_path = path;
 459         map->sm_fnv = fnv;
 460         map->sm_shmfd = shm_hold(shmfd);
 461         LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
 462 }
 463
 464 static int
 465 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
 466 {
 467         struct shm_mapping *map;
 468         int error;
 469
 470         LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
 471                 if (map->sm_fnv != fnv)
 472                         continue;
 473                 if (strcmp(map->sm_path, path) == 0) {
 474 #ifdef MAC
 475                         error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
 476                         if (error)
 477                                 return (error);
 478 #endif
 479                         error = shm_access(map->sm_shmfd, ucred,
 480                             FREAD | FWRITE);
 481                         if (error)
 482                                 return (error);
 483                         LIST_REMOVE(map, sm_link);
 484                         shm_drop(map->sm_shmfd);
 485                         free(map->sm_path, M_SHMFD);
 486                         free(map, M_SHMFD);
 487                         return (0);
 488                 }
 489         }
 490
 491         return (ENOENT);
 492 }
 493
 494 /* System calls. */
 495 int
 496 sys_shm_open(struct thread *td, struct shm_open_args *uap)
 497 {
 498         struct filedesc *fdp;
 499         struct shmfd *shmfd;
 500         struct file *fp;
 501         char *path;
 502         Fnv32_t fnv;
 503         mode_t cmode;
 504         int fd, error;
 505
 506 #ifdef CAPABILITY_MODE
 507         /*
 508          * shm_open(2) is only allowed for anonymous objects.
 509          */
 510         if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
 511                 return (ECAPMODE);
 512 #endif
 513
 514         if ((uap->flags & O_ACCMODE) != O_RDONLY &&
 515             (uap->flags & O_ACCMODE) != O_RDWR)
 516                 return (EINVAL);
 517
 518         if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
 519                 return (EINVAL);
 520
 521         fdp = td->td_proc->p_fd;
 522         cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
 523
 524         error = falloc(td, &fp, &fd, 0);
 525         if (error)
 526                 return (error);
 527
 528         /* A SHM_ANON path pointer creates an anonymous object. */
 529         if (uap->path == SHM_ANON) {
 530                 /* A read-only anonymous object is pointless. */
 531                 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
 532                         fdclose(fdp, fp, fd, td);
 533                         fdrop(fp, td);
 534                         return (EINVAL);
 535                 }
 536                 shmfd = shm_alloc(td->td_ucred, cmode);
 537         } else {
 538                 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
 539                 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 540
 541                 /* Require paths to start with a '/' character. */
 542                 if (error == 0 && path[0] != '/')
 543                         error = EINVAL;
 544                 if (error) {
 545                         fdclose(fdp, fp, fd, td);
 546                         fdrop(fp, td);
 547                         free(path, M_SHMFD);
 548                         return (error);
 549                 }
 550
 551                 fnv = fnv_32_str(path, FNV1_32_INIT);
 552                 sx_xlock(&shm_dict_lock);
 553                 shmfd = shm_lookup(path, fnv);
 554                 if (shmfd == NULL) {
 555                         /* Object does not yet exist, create it if requested. */
 556                         if (uap->flags & O_CREAT) {
 557 #ifdef MAC
 558                                 error = mac_posixshm_check_create(td->td_ucred,
 559                                     path);
 560                                 if (error == 0) {
 561 #endif
 562                                         shmfd = shm_alloc(td->td_ucred, cmode);
 563                                         shm_insert(path, fnv, shmfd);
 564 #ifdef MAC
 565                                 }
 566 #endif
 567                         } else {
 568                                 free(path, M_SHMFD);
 569                                 error = ENOENT;
 570                         }
 571                 } else {
 572                         /*
 573                          * Object already exists, obtain a new
 574                          * reference if requested and permitted.
 575                          */
 576                         free(path, M_SHMFD);
 577                         if ((uap->flags & (O_CREAT | O_EXCL)) ==
 578                             (O_CREAT | O_EXCL))
 579                                 error = EEXIST;
 580                         else {
 581 #ifdef MAC
 582                                 error = mac_posixshm_check_open(td->td_ucred,
 583                                     shmfd, FFLAGS(uap->flags & O_ACCMODE));
 584                                 if (error == 0)
 585 #endif
 586                                 error = shm_access(shmfd, td->td_ucred,
 587                                     FFLAGS(uap->flags & O_ACCMODE));
 588                         }
 589
 590                         /*
 591                          * Truncate the file back to zero length if
 592                          * O_TRUNC was specified and the object was
 593                          * opened with read/write.
 594                          */
 595                         if (error == 0 &&
 596                             (uap->flags & (O_ACCMODE | O_TRUNC)) ==
 597                             (O_RDWR | O_TRUNC)) {
 598 #ifdef MAC
 599                                 error = mac_posixshm_check_truncate(
 600                                         td->td_ucred, fp->f_cred, shmfd);
 601                                 if (error == 0)
 602 #endif
 603                                         shm_dotruncate(shmfd, 0);
 604                         }
 605                         if (error == 0)
 606                                 shm_hold(shmfd);
 607                 }
 608                 sx_xunlock(&shm_dict_lock);
 609
 610                 if (error) {
 611                         fdclose(fdp, fp, fd, td);
 612                         fdrop(fp, td);
 613                         return (error);
 614                 }
 615         }
 616
 617         finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
 618
 619         FILEDESC_XLOCK(fdp);
 620         if (fdp->fd_ofiles[fd] == fp)
 621                 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
 622         FILEDESC_XUNLOCK(fdp);
 623         td->td_retval[0] = fd;
 624         fdrop(fp, td);
 625
 626         return (0);
 627 }
 628
 629 int
 630 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
 631 {
 632         char *path;
 633         Fnv32_t fnv;
 634         int error;
 635
 636         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 637         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 638         if (error) {
 639                 free(path, M_TEMP);
 640                 return (error);
 641         }
 642
 643         fnv = fnv_32_str(path, FNV1_32_INIT);
 644         sx_xlock(&shm_dict_lock);
 645         error = shm_remove(path, fnv, td->td_ucred);
 646         sx_xunlock(&shm_dict_lock);
 647         free(path, M_TEMP);
 648
 649         return (error);
 650 }
 651
 652 /*
 653  * mmap() helper to validate mmap() requests against shm object state
 654  * and give mmap() the vm_object to use for the mapping.
 655  */
 656 int
 657 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
 658     vm_object_t *obj)
 659 {
 660
 661         /*
 662          * XXXRW: This validation is probably insufficient, and subject to
 663          * sign errors.  It should be fixed.
 664          */
 665         if (foff >= shmfd->shm_size ||
 666             foff + objsize > round_page(shmfd->shm_size))
 667                 return (EINVAL);
 668
 669         mtx_lock(&shm_timestamp_lock);
 670         vfs_timestamp(&shmfd->shm_atime);
 671         mtx_unlock(&shm_timestamp_lock);
 672         vm_object_reference(shmfd->shm_object);
 673         *obj = shmfd->shm_object;
 674         return (0);
 675 }
 676
 677 static int
 678 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
 679     struct thread *td)
 680 {
 681         struct shmfd *shmfd;
 682         int error;
 683
 684         error = 0;
 685         shmfd = fp->f_data;
 686         mtx_lock(&shm_timestamp_lock);
 687         /*
 688          * SUSv4 says that x bits of permission need not be affected.
 689          * Be consistent with our shm_open there.
 690          */
 691 #ifdef MAC
 692         error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
 693         if (error != 0)
 694                 goto out;
 695 #endif
 696         error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
 697             shmfd->shm_gid, VADMIN, active_cred, NULL);
 698         if (error != 0)
 699                 goto out;
 700         shmfd->shm_mode = mode & ACCESSPERMS;
 701 out:
 702         mtx_unlock(&shm_timestamp_lock);
 703         return (error);
 704 }
 705
 706 static int
 707 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 708     struct thread *td)
 709 {
 710         struct shmfd *shmfd;
 711         int error;
 712
 713         error = 0;
 714         shmfd = fp->f_data;
 715         mtx_lock(&shm_timestamp_lock);
 716 #ifdef MAC
 717         error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
 718         if (error != 0)
 719                 goto out;
 720 #endif
 721         if (uid == (uid_t)-1)
 722                 uid = shmfd->shm_uid;
 723         if (gid == (gid_t)-1)
 724                  gid = shmfd->shm_gid;
 725         if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
 726             (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
 727             (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
 728                 goto out;
 729         shmfd->shm_uid = uid;
 730         shmfd->shm_gid = gid;
 731 out:
 732         mtx_unlock(&shm_timestamp_lock);
 733         return (error);
 734 }