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_extern.h>
  78 #include <vm/vm_map.h>
  79 #include <vm/vm_kern.h>
  80 #include <vm/vm_object.h>
  81 #include <vm/vm_page.h>
  82 #include <vm/vm_pageout.h>
  83 #include <vm/vm_pager.h>
  84 #include <vm/swap_pager.h>
  85
  86 struct shm_mapping {
  87         char            *sm_path;
  88         Fnv32_t         sm_fnv;
  89         struct shmfd    *sm_shmfd;
  90         LIST_ENTRY(shm_mapping) sm_link;
  91 };
  92
  93 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
  94 static LIST_HEAD(, shm_mapping) *shm_dictionary;
  95 static struct sx shm_dict_lock;
  96 static struct mtx shm_timestamp_lock;
  97 static u_long shm_hash;
  98
  99 #define SHM_HASH(fnv)   (&shm_dictionary[(fnv) & shm_hash])
 100
 101 static int      shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
 102 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
 103 static void     shm_dict_init(void *arg);
 104 static void     shm_drop(struct shmfd *shmfd);
 105 static struct shmfd *shm_hold(struct shmfd *shmfd);
 106 static void     shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
 107 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
 108 static int      shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
 109 static int      shm_dotruncate(struct shmfd *shmfd, off_t length);
 110
 111 static fo_rdwr_t        shm_read;
 112 static fo_rdwr_t        shm_write;
 113 static fo_truncate_t    shm_truncate;
 114 static fo_ioctl_t       shm_ioctl;
 115 static fo_poll_t        shm_poll;
 116 static fo_kqfilter_t    shm_kqfilter;
 117 static fo_stat_t        shm_stat;
 118 static fo_close_t       shm_close;
 119 static fo_chmod_t       shm_chmod;
 120 static fo_chown_t       shm_chown;
 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_chmod = shm_chmod,
 133         .fo_chown = shm_chown,
 134         .fo_flags = DFLAG_PASSABLE
 135 };
 136
 137 FEATURE(posix_shm, "POSIX shared memory");
 138
 139 static int
 140 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
 141     int flags, struct thread *td)
 142 {
 143
 144         return (EOPNOTSUPP);
 145 }
 146
 147 static int
 148 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
 149     int flags, struct thread *td)
 150 {
 151
 152         return (EOPNOTSUPP);
 153 }
 154
 155 static int
 156 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
 157     struct thread *td)
 158 {
 159         struct shmfd *shmfd;
 160 #ifdef MAC
 161         int error;
 162 #endif
 163
 164         shmfd = fp->f_data;
 165 #ifdef MAC
 166         error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
 167         if (error)
 168                 return (error);
 169 #endif
 170         return (shm_dotruncate(shmfd, length));
 171 }
 172
 173 static int
 174 shm_ioctl(struct file *fp, u_long com, void *data,
 175     struct ucred *active_cred, struct thread *td)
 176 {
 177
 178         return (EOPNOTSUPP);
 179 }
 180
 181 static int
 182 shm_poll(struct file *fp, int events, struct ucred *active_cred,
 183     struct thread *td)
 184 {
 185
 186         return (EOPNOTSUPP);
 187 }
 188
 189 static int
 190 shm_kqfilter(struct file *fp, struct knote *kn)
 191 {
 192
 193         return (EOPNOTSUPP);
 194 }
 195
 196 static int
 197 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
 198     struct thread *td)
 199 {
 200         struct shmfd *shmfd;
 201 #ifdef MAC
 202         int error;
 203 #endif
 204
 205         shmfd = fp->f_data;
 206
 207 #ifdef MAC
 208         error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
 209         if (error)
 210                 return (error);
 211 #endif
 212
 213         /*
 214          * Attempt to return sanish values for fstat() on a memory file
 215          * descriptor.
 216          */
 217         bzero(sb, sizeof(*sb));
 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         mtx_lock(&shm_timestamp_lock);
 222         sb->st_atim = shmfd->shm_atime;
 223         sb->st_ctim = shmfd->shm_ctime;
 224         sb->st_mtim = shmfd->shm_mtime;
 225         sb->st_birthtim = shmfd->shm_birthtime;
 226         sb->st_mode = S_IFREG | shmfd->shm_mode;                /* XXX */
 227         sb->st_uid = shmfd->shm_uid;
 228         sb->st_gid = shmfd->shm_gid;
 229         mtx_unlock(&shm_timestamp_lock);
 230
 231         return (0);
 232 }
 233
 234 static int
 235 shm_close(struct file *fp, struct thread *td)
 236 {
 237         struct shmfd *shmfd;
 238
 239         shmfd = fp->f_data;
 240         fp->f_data = NULL;
 241         shm_drop(shmfd);
 242
 243         return (0);
 244 }
 245
 246 static int
 247 shm_dotruncate(struct shmfd *shmfd, off_t length)
 248 {
 249         vm_object_t object;
 250         vm_page_t m, ma[1];
 251         vm_pindex_t idx, nobjsize;
 252         vm_ooffset_t delta;
 253         int base, rv;
 254
 255         object = shmfd->shm_object;
 256         VM_OBJECT_LOCK(object);
 257         if (length == shmfd->shm_size) {
 258                 VM_OBJECT_UNLOCK(object);
 259                 return (0);
 260         }
 261         nobjsize = OFF_TO_IDX(length + PAGE_MASK);
 262
 263         /* Are we shrinking?  If so, trim the end. */
 264         if (length < shmfd->shm_size) {
 265                 /*
 266                  * Disallow any requests to shrink the size if this
 267                  * object is mapped into the kernel.
 268                  */
 269                 if (shmfd->shm_kmappings > 0) {
 270                         VM_OBJECT_UNLOCK(object);
 271                         return (EBUSY);
 272                 }
 273
 274                 /*
 275                  * Zero the truncated part of the last page.
 276                  */
 277                 base = length & PAGE_MASK;
 278                 if (base != 0) {
 279                         idx = OFF_TO_IDX(length);
 280 retry:
 281                         m = vm_page_lookup(object, idx);
 282                         if (m != NULL) {
 283                                 if ((m->oflags & VPO_BUSY) != 0 ||
 284                                     m->busy != 0) {
 285                                         vm_page_sleep(m, "shmtrc");
 286                                         goto retry;
 287                                 }
 288                         } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
 289                                 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
 290                                 if (m == NULL) {
 291                                         VM_OBJECT_UNLOCK(object);
 292                                         VM_WAIT;
 293                                         VM_OBJECT_LOCK(object);
 294                                         goto retry;
 295                                 } else if (m->valid != VM_PAGE_BITS_ALL) {
 296                                         ma[0] = m;
 297                                         rv = vm_pager_get_pages(object, ma, 1,
 298                                             0);
 299                                         m = vm_page_lookup(object, idx);
 300                                 } else
 301                                         /* A cached page was reactivated. */
 302                                         rv = VM_PAGER_OK;
 303                                 vm_page_lock(m);
 304                                 if (rv == VM_PAGER_OK) {
 305                                         vm_page_deactivate(m);
 306                                         vm_page_unlock(m);
 307                                         vm_page_wakeup(m);
 308                                 } else {
 309                                         vm_page_free(m);
 310                                         vm_page_unlock(m);
 311                                         VM_OBJECT_UNLOCK(object);
 312                                         return (EIO);
 313                                 }
 314                         }
 315                         if (m != NULL) {
 316                                 pmap_zero_page_area(m, base, PAGE_SIZE - base);
 317                                 KASSERT(m->valid == VM_PAGE_BITS_ALL,
 318                                     ("shm_dotruncate: page %p is invalid", m));
 319                                 vm_page_dirty(m);
 320                                 vm_pager_page_unswapped(m);
 321                         }
 322                 }
 323                 delta = ptoa(object->size - nobjsize);
 324
 325                 /* Toss in memory pages. */
 326                 if (nobjsize < object->size)
 327                         vm_object_page_remove(object, nobjsize, object->size,
 328                             0);
 329
 330                 /* Toss pages from swap. */
 331                 if (object->type == OBJT_SWAP)
 332                         swap_pager_freespace(object, nobjsize, delta);
 333
 334                 /* Free the swap accounted for shm */
 335                 swap_release_by_cred(delta, object->cred);
 336                 object->charge -= delta;
 337         } else {
 338                 /* Attempt to reserve the swap */
 339                 delta = ptoa(nobjsize - object->size);
 340                 if (!swap_reserve_by_cred(delta, object->cred)) {
 341                         VM_OBJECT_UNLOCK(object);
 342                         return (ENOMEM);
 343                 }
 344                 object->charge += delta;
 345         }
 346         shmfd->shm_size = length;
 347         mtx_lock(&shm_timestamp_lock);
 348         vfs_timestamp(&shmfd->shm_ctime);
 349         shmfd->shm_mtime = shmfd->shm_ctime;
 350         mtx_unlock(&shm_timestamp_lock);
 351         object->size = nobjsize;
 352         VM_OBJECT_UNLOCK(object);
 353         return (0);
 354 }
 355
 356 /*
 357  * shmfd object management including creation and reference counting
 358  * routines.
 359  */
 360 static struct shmfd *
 361 shm_alloc(struct ucred *ucred, mode_t mode)
 362 {
 363         struct shmfd *shmfd;
 364
 365         shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
 366         shmfd->shm_size = 0;
 367         shmfd->shm_uid = ucred->cr_uid;
 368         shmfd->shm_gid = ucred->cr_gid;
 369         shmfd->shm_mode = mode;
 370         shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
 371             shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
 372         KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
 373         VM_OBJECT_LOCK(shmfd->shm_object);
 374         vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
 375         vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
 376         VM_OBJECT_UNLOCK(shmfd->shm_object);
 377         vfs_timestamp(&shmfd->shm_birthtime);
 378         shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
 379             shmfd->shm_birthtime;
 380         refcount_init(&shmfd->shm_refs, 1);
 381 #ifdef MAC
 382         mac_posixshm_init(shmfd);
 383         mac_posixshm_create(ucred, shmfd);
 384 #endif
 385
 386         return (shmfd);
 387 }
 388
 389 static struct shmfd *
 390 shm_hold(struct shmfd *shmfd)
 391 {
 392
 393         refcount_acquire(&shmfd->shm_refs);
 394         return (shmfd);
 395 }
 396
 397 static void
 398 shm_drop(struct shmfd *shmfd)
 399 {
 400
 401         if (refcount_release(&shmfd->shm_refs)) {
 402 #ifdef MAC
 403                 mac_posixshm_destroy(shmfd);
 404 #endif
 405                 vm_object_deallocate(shmfd->shm_object);
 406                 free(shmfd, M_SHMFD);
 407         }
 408 }
 409
 410 /*
 411  * Determine if the credentials have sufficient permissions for a
 412  * specified combination of FREAD and FWRITE.
 413  */
 414 static int
 415 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
 416 {
 417         accmode_t accmode;
 418         int error;
 419
 420         accmode = 0;
 421         if (flags & FREAD)
 422                 accmode |= VREAD;
 423         if (flags & FWRITE)
 424                 accmode |= VWRITE;
 425         mtx_lock(&shm_timestamp_lock);
 426         error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
 427             accmode, ucred, NULL);
 428         mtx_unlock(&shm_timestamp_lock);
 429         return (error);
 430 }
 431
 432 /*
 433  * Dictionary management.  We maintain an in-kernel dictionary to map
 434  * paths to shmfd objects.  We use the FNV hash on the path to store
 435  * the mappings in a hash table.
 436  */
 437 static void
 438 shm_dict_init(void *arg)
 439 {
 440
 441         mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
 442         sx_init(&shm_dict_lock, "shm dictionary");
 443         shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
 444 }
 445 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
 446
 447 static struct shmfd *
 448 shm_lookup(char *path, Fnv32_t fnv)
 449 {
 450         struct shm_mapping *map;
 451
 452         LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
 453                 if (map->sm_fnv != fnv)
 454                         continue;
 455                 if (strcmp(map->sm_path, path) == 0)
 456                         return (map->sm_shmfd);
 457         }
 458
 459         return (NULL);
 460 }
 461
 462 static void
 463 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
 464 {
 465         struct shm_mapping *map;
 466
 467         map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
 468         map->sm_path = path;
 469         map->sm_fnv = fnv;
 470         map->sm_shmfd = shm_hold(shmfd);
 471         LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
 472 }
 473
 474 static int
 475 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
 476 {
 477         struct shm_mapping *map;
 478         int error;
 479
 480         LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
 481                 if (map->sm_fnv != fnv)
 482                         continue;
 483                 if (strcmp(map->sm_path, path) == 0) {
 484 #ifdef MAC
 485                         error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
 486                         if (error)
 487                                 return (error);
 488 #endif
 489                         error = shm_access(map->sm_shmfd, ucred,
 490                             FREAD | FWRITE);
 491                         if (error)
 492                                 return (error);
 493                         LIST_REMOVE(map, sm_link);
 494                         shm_drop(map->sm_shmfd);
 495                         free(map->sm_path, M_SHMFD);
 496                         free(map, M_SHMFD);
 497                         return (0);
 498                 }
 499         }
 500
 501         return (ENOENT);
 502 }
 503
 504 /* System calls. */
 505 int
 506 sys_shm_open(struct thread *td, struct shm_open_args *uap)
 507 {
 508         struct filedesc *fdp;
 509         struct shmfd *shmfd;
 510         struct file *fp;
 511         char *path;
 512         Fnv32_t fnv;
 513         mode_t cmode;
 514         int fd, error;
 515
 516 #ifdef CAPABILITY_MODE
 517         /*
 518          * shm_open(2) is only allowed for anonymous objects.
 519          */
 520         if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
 521                 return (ECAPMODE);
 522 #endif
 523
 524         if ((uap->flags & O_ACCMODE) != O_RDONLY &&
 525             (uap->flags & O_ACCMODE) != O_RDWR)
 526                 return (EINVAL);
 527
 528         if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
 529                 return (EINVAL);
 530
 531         fdp = td->td_proc->p_fd;
 532         cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
 533
 534         error = falloc(td, &fp, &fd, 0);
 535         if (error)
 536                 return (error);
 537
 538         /* A SHM_ANON path pointer creates an anonymous object. */
 539         if (uap->path == SHM_ANON) {
 540                 /* A read-only anonymous object is pointless. */
 541                 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
 542                         fdclose(fdp, fp, fd, td);
 543                         fdrop(fp, td);
 544                         return (EINVAL);
 545                 }
 546                 shmfd = shm_alloc(td->td_ucred, cmode);
 547         } else {
 548                 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
 549                 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 550
 551                 /* Require paths to start with a '/' character. */
 552                 if (error == 0 && path[0] != '/')
 553                         error = EINVAL;
 554                 if (error) {
 555                         fdclose(fdp, fp, fd, td);
 556                         fdrop(fp, td);
 557                         free(path, M_SHMFD);
 558                         return (error);
 559                 }
 560
 561                 fnv = fnv_32_str(path, FNV1_32_INIT);
 562                 sx_xlock(&shm_dict_lock);
 563                 shmfd = shm_lookup(path, fnv);
 564                 if (shmfd == NULL) {
 565                         /* Object does not yet exist, create it if requested. */
 566                         if (uap->flags & O_CREAT) {
 567 #ifdef MAC
 568                                 error = mac_posixshm_check_create(td->td_ucred,
 569                                     path);
 570                                 if (error == 0) {
 571 #endif
 572                                         shmfd = shm_alloc(td->td_ucred, cmode);
 573                                         shm_insert(path, fnv, shmfd);
 574 #ifdef MAC
 575                                 }
 576 #endif
 577                         } else {
 578                                 free(path, M_SHMFD);
 579                                 error = ENOENT;
 580                         }
 581                 } else {
 582                         /*
 583                          * Object already exists, obtain a new
 584                          * reference if requested and permitted.
 585                          */
 586                         free(path, M_SHMFD);
 587                         if ((uap->flags & (O_CREAT | O_EXCL)) ==
 588                             (O_CREAT | O_EXCL))
 589                                 error = EEXIST;
 590                         else {
 591 #ifdef MAC
 592                                 error = mac_posixshm_check_open(td->td_ucred,
 593                                     shmfd, FFLAGS(uap->flags & O_ACCMODE));
 594                                 if (error == 0)
 595 #endif
 596                                 error = shm_access(shmfd, td->td_ucred,
 597                                     FFLAGS(uap->flags & O_ACCMODE));
 598                         }
 599
 600                         /*
 601                          * Truncate the file back to zero length if
 602                          * O_TRUNC was specified and the object was
 603                          * opened with read/write.
 604                          */
 605                         if (error == 0 &&
 606                             (uap->flags & (O_ACCMODE | O_TRUNC)) ==
 607                             (O_RDWR | O_TRUNC)) {
 608 #ifdef MAC
 609                                 error = mac_posixshm_check_truncate(
 610                                         td->td_ucred, fp->f_cred, shmfd);
 611                                 if (error == 0)
 612 #endif
 613                                         shm_dotruncate(shmfd, 0);
 614                         }
 615                         if (error == 0)
 616                                 shm_hold(shmfd);
 617                 }
 618                 sx_xunlock(&shm_dict_lock);
 619
 620                 if (error) {
 621                         fdclose(fdp, fp, fd, td);
 622                         fdrop(fp, td);
 623                         return (error);
 624                 }
 625         }
 626
 627         finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
 628
 629         FILEDESC_XLOCK(fdp);
 630         if (fdp->fd_ofiles[fd] == fp)
 631                 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
 632         FILEDESC_XUNLOCK(fdp);
 633         td->td_retval[0] = fd;
 634         fdrop(fp, td);
 635
 636         return (0);
 637 }
 638
 639 int
 640 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
 641 {
 642         char *path;
 643         Fnv32_t fnv;
 644         int error;
 645
 646         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 647         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 648         if (error) {
 649                 free(path, M_TEMP);
 650                 return (error);
 651         }
 652
 653         fnv = fnv_32_str(path, FNV1_32_INIT);
 654         sx_xlock(&shm_dict_lock);
 655         error = shm_remove(path, fnv, td->td_ucred);
 656         sx_xunlock(&shm_dict_lock);
 657         free(path, M_TEMP);
 658
 659         return (error);
 660 }
 661
 662 /*
 663  * mmap() helper to validate mmap() requests against shm object state
 664  * and give mmap() the vm_object to use for the mapping.
 665  */
 666 int
 667 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
 668     vm_object_t *obj)
 669 {
 670
 671         /*
 672          * XXXRW: This validation is probably insufficient, and subject to
 673          * sign errors.  It should be fixed.
 674          */
 675         if (foff >= shmfd->shm_size ||
 676             foff + objsize > round_page(shmfd->shm_size))
 677                 return (EINVAL);
 678
 679         mtx_lock(&shm_timestamp_lock);
 680         vfs_timestamp(&shmfd->shm_atime);
 681         mtx_unlock(&shm_timestamp_lock);
 682         vm_object_reference(shmfd->shm_object);
 683         *obj = shmfd->shm_object;
 684         return (0);
 685 }
 686
 687 static int
 688 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
 689     struct thread *td)
 690 {
 691         struct shmfd *shmfd;
 692         int error;
 693
 694         error = 0;
 695         shmfd = fp->f_data;
 696         mtx_lock(&shm_timestamp_lock);
 697         /*
 698          * SUSv4 says that x bits of permission need not be affected.
 699          * Be consistent with our shm_open there.
 700          */
 701 #ifdef MAC
 702         error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
 703         if (error != 0)
 704                 goto out;
 705 #endif
 706         error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
 707             shmfd->shm_gid, VADMIN, active_cred, NULL);
 708         if (error != 0)
 709                 goto out;
 710         shmfd->shm_mode = mode & ACCESSPERMS;
 711 out:
 712         mtx_unlock(&shm_timestamp_lock);
 713         return (error);
 714 }
 715
 716 static int
 717 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 718     struct thread *td)
 719 {
 720         struct shmfd *shmfd;
 721         int error;
 722
 723         error = 0;
 724         shmfd = fp->f_data;
 725         mtx_lock(&shm_timestamp_lock);
 726 #ifdef MAC
 727         error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
 728         if (error != 0)
 729                 goto out;
 730 #endif
 731         if (uid == (uid_t)-1)
 732                 uid = shmfd->shm_uid;
 733         if (gid == (gid_t)-1)
 734                  gid = shmfd->shm_gid;
 735         if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
 736             (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
 737             (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
 738                 goto out;
 739         shmfd->shm_uid = uid;
 740         shmfd->shm_gid = gid;
 741 out:
 742         mtx_unlock(&shm_timestamp_lock);
 743         return (error);
 744 }
 745
 746 /*
 747  * Helper routines to allow the backing object of a shared memory file
 748  * descriptor to be mapped in the kernel.
 749  */
 750 int
 751 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
 752 {
 753         struct shmfd *shmfd;
 754         vm_offset_t kva, ofs;
 755         vm_object_t obj;
 756         int rv;
 757
 758         if (fp->f_type != DTYPE_SHM)
 759                 return (EINVAL);
 760         shmfd = fp->f_data;
 761         obj = shmfd->shm_object;
 762         VM_OBJECT_LOCK(obj);
 763         /*
 764          * XXXRW: This validation is probably insufficient, and subject to
 765          * sign errors.  It should be fixed.
 766          */
 767         if (offset >= shmfd->shm_size ||
 768             offset + size > round_page(shmfd->shm_size)) {
 769                 VM_OBJECT_UNLOCK(obj);
 770                 return (EINVAL);
 771         }
 772
 773         shmfd->shm_kmappings++;
 774         vm_object_reference_locked(obj);
 775         VM_OBJECT_UNLOCK(obj);
 776
 777         /* Map the object into the kernel_map and wire it. */
 778         kva = vm_map_min(kernel_map);
 779         ofs = offset & PAGE_MASK;
 780         offset = trunc_page(offset);
 781         size = round_page(size + ofs);
 782         rv = vm_map_find(kernel_map, obj, offset, &kva, size,
 783             VMFS_ALIGNED_SPACE, VM_PROT_READ | VM_PROT_WRITE,
 784             VM_PROT_READ | VM_PROT_WRITE, 0);
 785         if (rv == KERN_SUCCESS) {
 786                 rv = vm_map_wire(kernel_map, kva, kva + size,
 787                     VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
 788                 if (rv == KERN_SUCCESS) {
 789                         *memp = (void *)(kva + ofs);
 790                         return (0);
 791                 }
 792                 vm_map_remove(kernel_map, kva, kva + size);
 793         } else
 794                 vm_object_deallocate(obj);
 795
 796         /* On failure, drop our mapping reference. */
 797         VM_OBJECT_LOCK(obj);
 798         shmfd->shm_kmappings--;
 799         VM_OBJECT_UNLOCK(obj);
 800
 801         return (vm_mmap_to_errno(rv));
 802 }
 803
 804 /*
 805  * We require the caller to unmap the entire entry.  This allows us to
 806  * safely decrement shm_kmappings when a mapping is removed.
 807  */
 808 int
 809 shm_unmap(struct file *fp, void *mem, size_t size)
 810 {
 811         struct shmfd *shmfd;
 812         vm_map_entry_t entry;
 813         vm_offset_t kva, ofs;
 814         vm_object_t obj;
 815         vm_pindex_t pindex;
 816         vm_prot_t prot;
 817         boolean_t wired;
 818         vm_map_t map;
 819         int rv;
 820
 821         if (fp->f_type != DTYPE_SHM)
 822                 return (EINVAL);
 823         shmfd = fp->f_data;
 824         kva = (vm_offset_t)mem;
 825         ofs = kva & PAGE_MASK;
 826         kva = trunc_page(kva);
 827         size = round_page(size + ofs);
 828         map = kernel_map;
 829         rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
 830             &obj, &pindex, &prot, &wired);
 831         if (rv != KERN_SUCCESS)
 832                 return (EINVAL);
 833         if (entry->start != kva || entry->end != kva + size) {
 834                 vm_map_lookup_done(map, entry);
 835                 return (EINVAL);
 836         }
 837         vm_map_lookup_done(map, entry);
 838         if (obj != shmfd->shm_object)
 839                 return (EINVAL);
 840         vm_map_remove(map, kva, kva + size);
 841         VM_OBJECT_LOCK(obj);
 842         KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
 843         shmfd->shm_kmappings--;
 844         VM_OBJECT_UNLOCK(obj);
 845         return (0);
 846 }