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/rwlock.h>
  65 #include <sys/stat.h>
  66 #include <sys/sysctl.h>
  67 #include <sys/sysproto.h>
  68 #include <sys/systm.h>
  69 #include <sys/sx.h>
  70 #include <sys/time.h>
  71 #include <sys/vnode.h>
  72
  73 #include <security/mac/mac_framework.h>
  74
  75 #include <vm/vm.h>
  76 #include <vm/vm_param.h>
  77 #include <vm/pmap.h>
  78 #include <vm/vm_extern.h>
  79 #include <vm/vm_map.h>
  80 #include <vm/vm_kern.h>
  81 #include <vm/vm_object.h>
  82 #include <vm/vm_page.h>
  83 #include <vm/vm_pageout.h>
  84 #include <vm/vm_pager.h>
  85 #include <vm/swap_pager.h>
  86
  87 struct shm_mapping {
  88         char            *sm_path;
  89         Fnv32_t         sm_fnv;
  90         struct shmfd    *sm_shmfd;
  91         LIST_ENTRY(shm_mapping) sm_link;
  92 };
  93
  94 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
  95 static LIST_HEAD(, shm_mapping) *shm_dictionary;
  96 static struct sx shm_dict_lock;
  97 static struct mtx shm_timestamp_lock;
  98 static u_long shm_hash;
  99
 100 #define SHM_HASH(fnv)   (&shm_dictionary[(fnv) & shm_hash])
 101
 102 static int      shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
 103 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
 104 static void     shm_dict_init(void *arg);
 105 static void     shm_drop(struct shmfd *shmfd);
 106 static struct shmfd *shm_hold(struct shmfd *shmfd);
 107 static void     shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
 108 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
 109 static int      shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
 110 static int      shm_dotruncate(struct shmfd *shmfd, off_t length);
 111
 112 static fo_rdwr_t        shm_read;
 113 static fo_rdwr_t        shm_write;
 114 static fo_truncate_t    shm_truncate;
 115 static fo_ioctl_t       shm_ioctl;
 116 static fo_poll_t        shm_poll;
 117 static fo_kqfilter_t    shm_kqfilter;
 118 static fo_stat_t        shm_stat;
 119 static fo_close_t       shm_close;
 120 static fo_chmod_t       shm_chmod;
 121 static fo_chown_t       shm_chown;
 122
 123 /* File descriptor operations. */
 124 static struct fileops shm_ops = {
 125         .fo_read = shm_read,
 126         .fo_write = shm_write,
 127         .fo_truncate = shm_truncate,
 128         .fo_ioctl = shm_ioctl,
 129         .fo_poll = shm_poll,
 130         .fo_kqfilter = shm_kqfilter,
 131         .fo_stat = shm_stat,
 132         .fo_close = shm_close,
 133         .fo_chmod = shm_chmod,
 134         .fo_chown = shm_chown,
 135         .fo_sendfile = invfo_sendfile,
 136         .fo_flags = DFLAG_PASSABLE
 137 };
 138
 139 FEATURE(posix_shm, "POSIX shared memory");
 140
 141 static int
 142 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
 143 {
 144         vm_page_t m;
 145         vm_pindex_t idx;
 146         size_t tlen;
 147         int error, offset, rv;
 148
 149         idx = OFF_TO_IDX(uio->uio_offset);
 150         offset = uio->uio_offset & PAGE_MASK;
 151         tlen = MIN(PAGE_SIZE - offset, len);
 152
 153         VM_OBJECT_WLOCK(obj);
 154
 155         /*
 156          * Parallel reads of the page content from disk are prevented
 157          * by exclusive busy.
 158          *
 159          * Although the tmpfs vnode lock is held here, it is
 160          * nonetheless safe to sleep waiting for a free page.  The
 161          * pageout daemon does not need to acquire the tmpfs vnode
 162          * lock to page out tobj's pages because tobj is a OBJT_SWAP
 163          * type object.
 164          */
 165         m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
 166         if (m->valid != VM_PAGE_BITS_ALL) {
 167                 if (vm_pager_has_page(obj, idx, NULL, NULL)) {
 168                         rv = vm_pager_get_pages(obj, &m, 1, 0);
 169                         m = vm_page_lookup(obj, idx);
 170                         if (m == NULL) {
 171                                 printf(
 172                     "uiomove_object: vm_obj %p idx %jd null lookup rv %d\n",
 173                                     obj, idx, rv);
 174                                 VM_OBJECT_WUNLOCK(obj);
 175                                 return (EIO);
 176                         }
 177                         if (rv != VM_PAGER_OK) {
 178                                 printf(
 179             "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
 180                                     obj, idx, m->valid, rv);
 181                                 vm_page_lock(m);
 182                                 vm_page_free(m);
 183                                 vm_page_unlock(m);
 184                                 VM_OBJECT_WUNLOCK(obj);
 185                                 return (EIO);
 186                         }
 187                 } else
 188                         vm_page_zero_invalid(m, TRUE);
 189         }
 190         vm_page_xunbusy(m);
 191         vm_page_lock(m);
 192         vm_page_hold(m);
 193         vm_page_unlock(m);
 194         VM_OBJECT_WUNLOCK(obj);
 195         error = uiomove_fromphys(&m, offset, tlen, uio);
 196         if (uio->uio_rw == UIO_WRITE && error == 0) {
 197                 VM_OBJECT_WLOCK(obj);
 198                 vm_page_dirty(m);
 199                 VM_OBJECT_WUNLOCK(obj);
 200         }
 201         vm_page_lock(m);
 202         vm_page_unhold(m);
 203         if (m->queue == PQ_NONE) {
 204                 vm_page_deactivate(m);
 205         } else {
 206                 /* Requeue to maintain LRU ordering. */
 207                 vm_page_requeue(m);
 208         }
 209         vm_page_unlock(m);
 210
 211         return (error);
 212 }
 213
 214 int
 215 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio)
 216 {
 217         ssize_t resid;
 218         size_t len;
 219         int error;
 220
 221         error = 0;
 222         while ((resid = uio->uio_resid) > 0) {
 223                 if (obj_size <= uio->uio_offset)
 224                         break;
 225                 len = MIN(obj_size - uio->uio_offset, resid);
 226                 if (len == 0)
 227                         break;
 228                 error = uiomove_object_page(obj, len, uio);
 229                 if (error != 0 || resid == uio->uio_resid)
 230                         break;
 231         }
 232         return (error);
 233 }
 234
 235 static int
 236 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
 237     int flags, struct thread *td)
 238 {
 239
 240         return (EOPNOTSUPP);
 241 }
 242
 243 static int
 244 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
 245     int flags, struct thread *td)
 246 {
 247
 248         return (EOPNOTSUPP);
 249 }
 250
 251 static int
 252 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
 253     struct thread *td)
 254 {
 255         struct shmfd *shmfd;
 256 #ifdef MAC
 257         int error;
 258 #endif
 259
 260         shmfd = fp->f_data;
 261 #ifdef MAC
 262         error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
 263         if (error)
 264                 return (error);
 265 #endif
 266         return (shm_dotruncate(shmfd, length));
 267 }
 268
 269 static int
 270 shm_ioctl(struct file *fp, u_long com, void *data,
 271     struct ucred *active_cred, struct thread *td)
 272 {
 273
 274         return (EOPNOTSUPP);
 275 }
 276
 277 static int
 278 shm_poll(struct file *fp, int events, struct ucred *active_cred,
 279     struct thread *td)
 280 {
 281
 282         return (EOPNOTSUPP);
 283 }
 284
 285 static int
 286 shm_kqfilter(struct file *fp, struct knote *kn)
 287 {
 288
 289         return (EOPNOTSUPP);
 290 }
 291
 292 static int
 293 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
 294     struct thread *td)
 295 {
 296         struct shmfd *shmfd;
 297 #ifdef MAC
 298         int error;
 299 #endif
 300
 301         shmfd = fp->f_data;
 302
 303 #ifdef MAC
 304         error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
 305         if (error)
 306                 return (error);
 307 #endif
 308
 309         /*
 310          * Attempt to return sanish values for fstat() on a memory file
 311          * descriptor.
 312          */
 313         bzero(sb, sizeof(*sb));
 314         sb->st_blksize = PAGE_SIZE;
 315         sb->st_size = shmfd->shm_size;
 316         sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
 317         mtx_lock(&shm_timestamp_lock);
 318         sb->st_atim = shmfd->shm_atime;
 319         sb->st_ctim = shmfd->shm_ctime;
 320         sb->st_mtim = shmfd->shm_mtime;
 321         sb->st_birthtim = shmfd->shm_birthtime;
 322         sb->st_mode = S_IFREG | shmfd->shm_mode;                /* XXX */
 323         sb->st_uid = shmfd->shm_uid;
 324         sb->st_gid = shmfd->shm_gid;
 325         mtx_unlock(&shm_timestamp_lock);
 326
 327         return (0);
 328 }
 329
 330 static int
 331 shm_close(struct file *fp, struct thread *td)
 332 {
 333         struct shmfd *shmfd;
 334
 335         shmfd = fp->f_data;
 336         fp->f_data = NULL;
 337         shm_drop(shmfd);
 338
 339         return (0);
 340 }
 341
 342 static int
 343 shm_dotruncate(struct shmfd *shmfd, off_t length)
 344 {
 345         vm_object_t object;
 346         vm_page_t m, ma[1];
 347         vm_pindex_t idx, nobjsize;
 348         vm_ooffset_t delta;
 349         int base, rv;
 350
 351         object = shmfd->shm_object;
 352         VM_OBJECT_WLOCK(object);
 353         if (length == shmfd->shm_size) {
 354                 VM_OBJECT_WUNLOCK(object);
 355                 return (0);
 356         }
 357         nobjsize = OFF_TO_IDX(length + PAGE_MASK);
 358
 359         /* Are we shrinking?  If so, trim the end. */
 360         if (length < shmfd->shm_size) {
 361                 /*
 362                  * Disallow any requests to shrink the size if this
 363                  * object is mapped into the kernel.
 364                  */
 365                 if (shmfd->shm_kmappings > 0) {
 366                         VM_OBJECT_WUNLOCK(object);
 367                         return (EBUSY);
 368                 }
 369
 370                 /*
 371                  * Zero the truncated part of the last page.
 372                  */
 373                 base = length & PAGE_MASK;
 374                 if (base != 0) {
 375                         idx = OFF_TO_IDX(length);
 376 retry:
 377                         m = vm_page_lookup(object, idx);
 378                         if (m != NULL) {
 379                                 if (vm_page_sleep_if_busy(m, "shmtrc"))
 380                                         goto retry;
 381                         } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
 382                                 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
 383                                 if (m == NULL) {
 384                                         VM_OBJECT_WUNLOCK(object);
 385                                         VM_WAIT;
 386                                         VM_OBJECT_WLOCK(object);
 387                                         goto retry;
 388                                 } else if (m->valid != VM_PAGE_BITS_ALL) {
 389                                         ma[0] = m;
 390                                         rv = vm_pager_get_pages(object, ma, 1,
 391                                             0);
 392                                         m = vm_page_lookup(object, idx);
 393                                 } else
 394                                         /* A cached page was reactivated. */
 395                                         rv = VM_PAGER_OK;
 396                                 vm_page_lock(m);
 397                                 if (rv == VM_PAGER_OK) {
 398                                         vm_page_deactivate(m);
 399                                         vm_page_unlock(m);
 400                                         vm_page_xunbusy(m);
 401                                 } else {
 402                                         vm_page_free(m);
 403                                         vm_page_unlock(m);
 404                                         VM_OBJECT_WUNLOCK(object);
 405                                         return (EIO);
 406                                 }
 407                         }
 408                         if (m != NULL) {
 409                                 pmap_zero_page_area(m, base, PAGE_SIZE - base);
 410                                 KASSERT(m->valid == VM_PAGE_BITS_ALL,
 411                                     ("shm_dotruncate: page %p is invalid", m));
 412                                 vm_page_dirty(m);
 413                                 vm_pager_page_unswapped(m);
 414                         }
 415                 }
 416                 delta = ptoa(object->size - nobjsize);
 417
 418                 /* Toss in memory pages. */
 419                 if (nobjsize < object->size)
 420                         vm_object_page_remove(object, nobjsize, object->size,
 421                             0);
 422
 423                 /* Toss pages from swap. */
 424                 if (object->type == OBJT_SWAP)
 425                         swap_pager_freespace(object, nobjsize, delta);
 426
 427                 /* Free the swap accounted for shm */
 428                 swap_release_by_cred(delta, object->cred);
 429                 object->charge -= delta;
 430         } else {
 431                 /* Attempt to reserve the swap */
 432                 delta = ptoa(nobjsize - object->size);
 433                 if (!swap_reserve_by_cred(delta, object->cred)) {
 434                         VM_OBJECT_WUNLOCK(object);
 435                         return (ENOMEM);
 436                 }
 437                 object->charge += delta;
 438         }
 439         shmfd->shm_size = length;
 440         mtx_lock(&shm_timestamp_lock);
 441         vfs_timestamp(&shmfd->shm_ctime);
 442         shmfd->shm_mtime = shmfd->shm_ctime;
 443         mtx_unlock(&shm_timestamp_lock);
 444         object->size = nobjsize;
 445         VM_OBJECT_WUNLOCK(object);
 446         return (0);
 447 }
 448
 449 /*
 450  * shmfd object management including creation and reference counting
 451  * routines.
 452  */
 453 static struct shmfd *
 454 shm_alloc(struct ucred *ucred, mode_t mode)
 455 {
 456         struct shmfd *shmfd;
 457
 458         shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
 459         shmfd->shm_size = 0;
 460         shmfd->shm_uid = ucred->cr_uid;
 461         shmfd->shm_gid = ucred->cr_gid;
 462         shmfd->shm_mode = mode;
 463         shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
 464             shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
 465         KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
 466         VM_OBJECT_WLOCK(shmfd->shm_object);
 467         vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
 468         vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
 469         VM_OBJECT_WUNLOCK(shmfd->shm_object);
 470         vfs_timestamp(&shmfd->shm_birthtime);
 471         shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
 472             shmfd->shm_birthtime;
 473         refcount_init(&shmfd->shm_refs, 1);
 474 #ifdef MAC
 475         mac_posixshm_init(shmfd);
 476         mac_posixshm_create(ucred, shmfd);
 477 #endif
 478
 479         return (shmfd);
 480 }
 481
 482 static struct shmfd *
 483 shm_hold(struct shmfd *shmfd)
 484 {
 485
 486         refcount_acquire(&shmfd->shm_refs);
 487         return (shmfd);
 488 }
 489
 490 static void
 491 shm_drop(struct shmfd *shmfd)
 492 {
 493
 494         if (refcount_release(&shmfd->shm_refs)) {
 495 #ifdef MAC
 496                 mac_posixshm_destroy(shmfd);
 497 #endif
 498                 vm_object_deallocate(shmfd->shm_object);
 499                 free(shmfd, M_SHMFD);
 500         }
 501 }
 502
 503 /*
 504  * Determine if the credentials have sufficient permissions for a
 505  * specified combination of FREAD and FWRITE.
 506  */
 507 static int
 508 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
 509 {
 510         accmode_t accmode;
 511         int error;
 512
 513         accmode = 0;
 514         if (flags & FREAD)
 515                 accmode |= VREAD;
 516         if (flags & FWRITE)
 517                 accmode |= VWRITE;
 518         mtx_lock(&shm_timestamp_lock);
 519         error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
 520             accmode, ucred, NULL);
 521         mtx_unlock(&shm_timestamp_lock);
 522         return (error);
 523 }
 524
 525 /*
 526  * Dictionary management.  We maintain an in-kernel dictionary to map
 527  * paths to shmfd objects.  We use the FNV hash on the path to store
 528  * the mappings in a hash table.
 529  */
 530 static void
 531 shm_dict_init(void *arg)
 532 {
 533
 534         mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
 535         sx_init(&shm_dict_lock, "shm dictionary");
 536         shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
 537 }
 538 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL);
 539
 540 static struct shmfd *
 541 shm_lookup(char *path, Fnv32_t fnv)
 542 {
 543         struct shm_mapping *map;
 544
 545         LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
 546                 if (map->sm_fnv != fnv)
 547                         continue;
 548                 if (strcmp(map->sm_path, path) == 0)
 549                         return (map->sm_shmfd);
 550         }
 551
 552         return (NULL);
 553 }
 554
 555 static void
 556 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
 557 {
 558         struct shm_mapping *map;
 559
 560         map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
 561         map->sm_path = path;
 562         map->sm_fnv = fnv;
 563         map->sm_shmfd = shm_hold(shmfd);
 564         shmfd->shm_path = path;
 565         LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
 566 }
 567
 568 static int
 569 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
 570 {
 571         struct shm_mapping *map;
 572         int error;
 573
 574         LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
 575                 if (map->sm_fnv != fnv)
 576                         continue;
 577                 if (strcmp(map->sm_path, path) == 0) {
 578 #ifdef MAC
 579                         error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
 580                         if (error)
 581                                 return (error);
 582 #endif
 583                         error = shm_access(map->sm_shmfd, ucred,
 584                             FREAD | FWRITE);
 585                         if (error)
 586                                 return (error);
 587                         map->sm_shmfd->shm_path = NULL;
 588                         LIST_REMOVE(map, sm_link);
 589                         shm_drop(map->sm_shmfd);
 590                         free(map->sm_path, M_SHMFD);
 591                         free(map, M_SHMFD);
 592                         return (0);
 593                 }
 594         }
 595
 596         return (ENOENT);
 597 }
 598
 599 /* System calls. */
 600 int
 601 sys_shm_open(struct thread *td, struct shm_open_args *uap)
 602 {
 603         struct filedesc *fdp;
 604         struct shmfd *shmfd;
 605         struct file *fp;
 606         char *path;
 607         Fnv32_t fnv;
 608         mode_t cmode;
 609         int fd, error;
 610
 611 #ifdef CAPABILITY_MODE
 612         /*
 613          * shm_open(2) is only allowed for anonymous objects.
 614          */
 615         if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
 616                 return (ECAPMODE);
 617 #endif
 618
 619         if ((uap->flags & O_ACCMODE) != O_RDONLY &&
 620             (uap->flags & O_ACCMODE) != O_RDWR)
 621                 return (EINVAL);
 622
 623         if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
 624                 return (EINVAL);
 625
 626         fdp = td->td_proc->p_fd;
 627         cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
 628
 629         error = falloc(td, &fp, &fd, O_CLOEXEC);
 630         if (error)
 631                 return (error);
 632
 633         /* A SHM_ANON path pointer creates an anonymous object. */
 634         if (uap->path == SHM_ANON) {
 635                 /* A read-only anonymous object is pointless. */
 636                 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
 637                         fdclose(fdp, fp, fd, td);
 638                         fdrop(fp, td);
 639                         return (EINVAL);
 640                 }
 641                 shmfd = shm_alloc(td->td_ucred, cmode);
 642         } else {
 643                 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
 644                 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 645
 646                 /* Require paths to start with a '/' character. */
 647                 if (error == 0 && path[0] != '/')
 648                         error = EINVAL;
 649                 if (error) {
 650                         fdclose(fdp, fp, fd, td);
 651                         fdrop(fp, td);
 652                         free(path, M_SHMFD);
 653                         return (error);
 654                 }
 655
 656                 fnv = fnv_32_str(path, FNV1_32_INIT);
 657                 sx_xlock(&shm_dict_lock);
 658                 shmfd = shm_lookup(path, fnv);
 659                 if (shmfd == NULL) {
 660                         /* Object does not yet exist, create it if requested. */
 661                         if (uap->flags & O_CREAT) {
 662 #ifdef MAC
 663                                 error = mac_posixshm_check_create(td->td_ucred,
 664                                     path);
 665                                 if (error == 0) {
 666 #endif
 667                                         shmfd = shm_alloc(td->td_ucred, cmode);
 668                                         shm_insert(path, fnv, shmfd);
 669 #ifdef MAC
 670                                 }
 671 #endif
 672                         } else {
 673                                 free(path, M_SHMFD);
 674                                 error = ENOENT;
 675                         }
 676                 } else {
 677                         /*
 678                          * Object already exists, obtain a new
 679                          * reference if requested and permitted.
 680                          */
 681                         free(path, M_SHMFD);
 682                         if ((uap->flags & (O_CREAT | O_EXCL)) ==
 683                             (O_CREAT | O_EXCL))
 684                                 error = EEXIST;
 685                         else {
 686 #ifdef MAC
 687                                 error = mac_posixshm_check_open(td->td_ucred,
 688                                     shmfd, FFLAGS(uap->flags & O_ACCMODE));
 689                                 if (error == 0)
 690 #endif
 691                                 error = shm_access(shmfd, td->td_ucred,
 692                                     FFLAGS(uap->flags & O_ACCMODE));
 693                         }
 694
 695                         /*
 696                          * Truncate the file back to zero length if
 697                          * O_TRUNC was specified and the object was
 698                          * opened with read/write.
 699                          */
 700                         if (error == 0 &&
 701                             (uap->flags & (O_ACCMODE | O_TRUNC)) ==
 702                             (O_RDWR | O_TRUNC)) {
 703 #ifdef MAC
 704                                 error = mac_posixshm_check_truncate(
 705                                         td->td_ucred, fp->f_cred, shmfd);
 706                                 if (error == 0)
 707 #endif
 708                                         shm_dotruncate(shmfd, 0);
 709                         }
 710                         if (error == 0)
 711                                 shm_hold(shmfd);
 712                 }
 713                 sx_xunlock(&shm_dict_lock);
 714
 715                 if (error) {
 716                         fdclose(fdp, fp, fd, td);
 717                         fdrop(fp, td);
 718                         return (error);
 719                 }
 720         }
 721
 722         finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
 723
 724         td->td_retval[0] = fd;
 725         fdrop(fp, td);
 726
 727         return (0);
 728 }
 729
 730 int
 731 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
 732 {
 733         char *path;
 734         Fnv32_t fnv;
 735         int error;
 736
 737         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 738         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 739         if (error) {
 740                 free(path, M_TEMP);
 741                 return (error);
 742         }
 743
 744         fnv = fnv_32_str(path, FNV1_32_INIT);
 745         sx_xlock(&shm_dict_lock);
 746         error = shm_remove(path, fnv, td->td_ucred);
 747         sx_xunlock(&shm_dict_lock);
 748         free(path, M_TEMP);
 749
 750         return (error);
 751 }
 752
 753 /*
 754  * mmap() helper to validate mmap() requests against shm object state
 755  * and give mmap() the vm_object to use for the mapping.
 756  */
 757 int
 758 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
 759     vm_object_t *obj)
 760 {
 761
 762         /*
 763          * XXXRW: This validation is probably insufficient, and subject to
 764          * sign errors.  It should be fixed.
 765          */
 766         if (foff >= shmfd->shm_size ||
 767             foff + objsize > round_page(shmfd->shm_size))
 768                 return (EINVAL);
 769
 770         mtx_lock(&shm_timestamp_lock);
 771         vfs_timestamp(&shmfd->shm_atime);
 772         mtx_unlock(&shm_timestamp_lock);
 773         vm_object_reference(shmfd->shm_object);
 774         *obj = shmfd->shm_object;
 775         return (0);
 776 }
 777
 778 static int
 779 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
 780     struct thread *td)
 781 {
 782         struct shmfd *shmfd;
 783         int error;
 784
 785         error = 0;
 786         shmfd = fp->f_data;
 787         mtx_lock(&shm_timestamp_lock);
 788         /*
 789          * SUSv4 says that x bits of permission need not be affected.
 790          * Be consistent with our shm_open there.
 791          */
 792 #ifdef MAC
 793         error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
 794         if (error != 0)
 795                 goto out;
 796 #endif
 797         error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
 798             shmfd->shm_gid, VADMIN, active_cred, NULL);
 799         if (error != 0)
 800                 goto out;
 801         shmfd->shm_mode = mode & ACCESSPERMS;
 802 out:
 803         mtx_unlock(&shm_timestamp_lock);
 804         return (error);
 805 }
 806
 807 static int
 808 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 809     struct thread *td)
 810 {
 811         struct shmfd *shmfd;
 812         int error;
 813
 814         error = 0;
 815         shmfd = fp->f_data;
 816         mtx_lock(&shm_timestamp_lock);
 817 #ifdef MAC
 818         error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
 819         if (error != 0)
 820                 goto out;
 821 #endif
 822         if (uid == (uid_t)-1)
 823                 uid = shmfd->shm_uid;
 824         if (gid == (gid_t)-1)
 825                  gid = shmfd->shm_gid;
 826         if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
 827             (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
 828             (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
 829                 goto out;
 830         shmfd->shm_uid = uid;
 831         shmfd->shm_gid = gid;
 832 out:
 833         mtx_unlock(&shm_timestamp_lock);
 834         return (error);
 835 }
 836
 837 /*
 838  * Helper routines to allow the backing object of a shared memory file
 839  * descriptor to be mapped in the kernel.
 840  */
 841 int
 842 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
 843 {
 844         struct shmfd *shmfd;
 845         vm_offset_t kva, ofs;
 846         vm_object_t obj;
 847         int rv;
 848
 849         if (fp->f_type != DTYPE_SHM)
 850                 return (EINVAL);
 851         shmfd = fp->f_data;
 852         obj = shmfd->shm_object;
 853         VM_OBJECT_WLOCK(obj);
 854         /*
 855          * XXXRW: This validation is probably insufficient, and subject to
 856          * sign errors.  It should be fixed.
 857          */
 858         if (offset >= shmfd->shm_size ||
 859             offset + size > round_page(shmfd->shm_size)) {
 860                 VM_OBJECT_WUNLOCK(obj);
 861                 return (EINVAL);
 862         }
 863
 864         shmfd->shm_kmappings++;
 865         vm_object_reference_locked(obj);
 866         VM_OBJECT_WUNLOCK(obj);
 867
 868         /* Map the object into the kernel_map and wire it. */
 869         kva = vm_map_min(kernel_map);
 870         ofs = offset & PAGE_MASK;
 871         offset = trunc_page(offset);
 872         size = round_page(size + ofs);
 873         rv = vm_map_find(kernel_map, obj, offset, &kva, size,
 874             VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
 875             VM_PROT_READ | VM_PROT_WRITE, 0);
 876         if (rv == KERN_SUCCESS) {
 877                 rv = vm_map_wire(kernel_map, kva, kva + size,
 878                     VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
 879                 if (rv == KERN_SUCCESS) {
 880                         *memp = (void *)(kva + ofs);
 881                         return (0);
 882                 }
 883                 vm_map_remove(kernel_map, kva, kva + size);
 884         } else
 885                 vm_object_deallocate(obj);
 886
 887         /* On failure, drop our mapping reference. */
 888         VM_OBJECT_WLOCK(obj);
 889         shmfd->shm_kmappings--;
 890         VM_OBJECT_WUNLOCK(obj);
 891
 892         return (vm_mmap_to_errno(rv));
 893 }
 894
 895 /*
 896  * We require the caller to unmap the entire entry.  This allows us to
 897  * safely decrement shm_kmappings when a mapping is removed.
 898  */
 899 int
 900 shm_unmap(struct file *fp, void *mem, size_t size)
 901 {
 902         struct shmfd *shmfd;
 903         vm_map_entry_t entry;
 904         vm_offset_t kva, ofs;
 905         vm_object_t obj;
 906         vm_pindex_t pindex;
 907         vm_prot_t prot;
 908         boolean_t wired;
 909         vm_map_t map;
 910         int rv;
 911
 912         if (fp->f_type != DTYPE_SHM)
 913                 return (EINVAL);
 914         shmfd = fp->f_data;
 915         kva = (vm_offset_t)mem;
 916         ofs = kva & PAGE_MASK;
 917         kva = trunc_page(kva);
 918         size = round_page(size + ofs);
 919         map = kernel_map;
 920         rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
 921             &obj, &pindex, &prot, &wired);
 922         if (rv != KERN_SUCCESS)
 923                 return (EINVAL);
 924         if (entry->start != kva || entry->end != kva + size) {
 925                 vm_map_lookup_done(map, entry);
 926                 return (EINVAL);
 927         }
 928         vm_map_lookup_done(map, entry);
 929         if (obj != shmfd->shm_object)
 930                 return (EINVAL);
 931         vm_map_remove(map, kva, kva + size);
 932         VM_OBJECT_WLOCK(obj);
 933         KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
 934         shmfd->shm_kmappings--;
 935         VM_OBJECT_WUNLOCK(obj);
 936         return (0);
 937 }
 938
 939 void
 940 shm_path(struct shmfd *shmfd, char *path, size_t size)
 941 {
 942
 943         if (shmfd->shm_path == NULL)
 944                 return;
 945         sx_slock(&shm_dict_lock);
 946         if (shmfd->shm_path != NULL)
 947                 strlcpy(path, shmfd->shm_path, size);
 948         sx_sunlock(&shm_dict_lock);
 949 }