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/conf.h>
  52 #include <sys/fcntl.h>
  53 #include <sys/file.h>
  54 #include <sys/filedesc.h>
  55 #include <sys/fnv_hash.h>
  56 #include <sys/kernel.h>
  57 #include <sys/lock.h>
  58 #include <sys/malloc.h>
  59 #include <sys/mman.h>
  60 #include <sys/mutex.h>
  61 #include <sys/priv.h>
  62 #include <sys/proc.h>
  63 #include <sys/refcount.h>
  64 #include <sys/resourcevar.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 static struct unrhdr *shm_ino_unr;
 100 static dev_t shm_dev_ino;
 101
 102 #define SHM_HASH(fnv)   (&shm_dictionary[(fnv) & shm_hash])
 103
 104 static int      shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
 105 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
 106 static void     shm_init(void *arg);
 107 static void     shm_drop(struct shmfd *shmfd);
 108 static struct shmfd *shm_hold(struct shmfd *shmfd);
 109 static void     shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
 110 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
 111 static int      shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
 112 static int      shm_dotruncate(struct shmfd *shmfd, off_t length);
 113
 114 static fo_rdwr_t        shm_read;
 115 static fo_rdwr_t        shm_write;
 116 static fo_truncate_t    shm_truncate;
 117 static fo_ioctl_t       shm_ioctl;
 118 static fo_poll_t        shm_poll;
 119 static fo_kqfilter_t    shm_kqfilter;
 120 static fo_stat_t        shm_stat;
 121 static fo_close_t       shm_close;
 122 static fo_chmod_t       shm_chmod;
 123 static fo_chown_t       shm_chown;
 124
 125 /* File descriptor operations. */
 126 static struct fileops shm_ops = {
 127         .fo_read = shm_read,
 128         .fo_write = shm_write,
 129         .fo_truncate = shm_truncate,
 130         .fo_ioctl = shm_ioctl,
 131         .fo_poll = shm_poll,
 132         .fo_kqfilter = shm_kqfilter,
 133         .fo_stat = shm_stat,
 134         .fo_close = shm_close,
 135         .fo_chmod = shm_chmod,
 136         .fo_chown = shm_chown,
 137         .fo_flags = DFLAG_PASSABLE
 138 };
 139
 140 FEATURE(posix_shm, "POSIX shared memory");
 141
 142 static int
 143 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
 144     int flags, struct thread *td)
 145 {
 146
 147         return (EOPNOTSUPP);
 148 }
 149
 150 static int
 151 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
 152     int flags, struct thread *td)
 153 {
 154
 155         return (EOPNOTSUPP);
 156 }
 157
 158 static int
 159 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
 160     struct thread *td)
 161 {
 162         struct shmfd *shmfd;
 163 #ifdef MAC
 164         int error;
 165 #endif
 166
 167         shmfd = fp->f_data;
 168 #ifdef MAC
 169         error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
 170         if (error)
 171                 return (error);
 172 #endif
 173         return (shm_dotruncate(shmfd, length));
 174 }
 175
 176 static int
 177 shm_ioctl(struct file *fp, u_long com, void *data,
 178     struct ucred *active_cred, struct thread *td)
 179 {
 180
 181         return (EOPNOTSUPP);
 182 }
 183
 184 static int
 185 shm_poll(struct file *fp, int events, struct ucred *active_cred,
 186     struct thread *td)
 187 {
 188
 189         return (EOPNOTSUPP);
 190 }
 191
 192 static int
 193 shm_kqfilter(struct file *fp, struct knote *kn)
 194 {
 195
 196         return (EOPNOTSUPP);
 197 }
 198
 199 static int
 200 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
 201     struct thread *td)
 202 {
 203         struct shmfd *shmfd;
 204 #ifdef MAC
 205         int error;
 206 #endif
 207
 208         shmfd = fp->f_data;
 209
 210 #ifdef MAC
 211         error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
 212         if (error)
 213                 return (error);
 214 #endif
 215
 216         /*
 217          * Attempt to return sanish values for fstat() on a memory file
 218          * descriptor.
 219          */
 220         bzero(sb, sizeof(*sb));
 221         sb->st_blksize = PAGE_SIZE;
 222         sb->st_size = shmfd->shm_size;
 223         sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
 224         mtx_lock(&shm_timestamp_lock);
 225         sb->st_atim = shmfd->shm_atime;
 226         sb->st_ctim = shmfd->shm_ctime;
 227         sb->st_mtim = shmfd->shm_mtime;
 228         sb->st_birthtim = shmfd->shm_birthtime;
 229         sb->st_mode = S_IFREG | shmfd->shm_mode;                /* XXX */
 230         sb->st_uid = shmfd->shm_uid;
 231         sb->st_gid = shmfd->shm_gid;
 232         mtx_unlock(&shm_timestamp_lock);
 233         sb->st_dev = shm_dev_ino;
 234         sb->st_ino = shmfd->shm_ino;
 235
 236         return (0);
 237 }
 238
 239 static int
 240 shm_close(struct file *fp, struct thread *td)
 241 {
 242         struct shmfd *shmfd;
 243
 244         shmfd = fp->f_data;
 245         fp->f_data = NULL;
 246         shm_drop(shmfd);
 247
 248         return (0);
 249 }
 250
 251 static int
 252 shm_dotruncate(struct shmfd *shmfd, off_t length)
 253 {
 254         vm_object_t object;
 255         vm_page_t m, ma[1];
 256         vm_pindex_t idx, nobjsize;
 257         vm_ooffset_t delta;
 258         int base, rv;
 259
 260         object = shmfd->shm_object;
 261         VM_OBJECT_LOCK(object);
 262         if (length == shmfd->shm_size) {
 263                 VM_OBJECT_UNLOCK(object);
 264                 return (0);
 265         }
 266         nobjsize = OFF_TO_IDX(length + PAGE_MASK);
 267
 268         /* Are we shrinking?  If so, trim the end. */
 269         if (length < shmfd->shm_size) {
 270                 /*
 271                  * Disallow any requests to shrink the size if this
 272                  * object is mapped into the kernel.
 273                  */
 274                 if (shmfd->shm_kmappings > 0) {
 275                         VM_OBJECT_UNLOCK(object);
 276                         return (EBUSY);
 277                 }
 278
 279                 /*
 280                  * Zero the truncated part of the last page.
 281                  */
 282                 base = length & PAGE_MASK;
 283                 if (base != 0) {
 284                         idx = OFF_TO_IDX(length);
 285 retry:
 286                         m = vm_page_lookup(object, idx);
 287                         if (m != NULL) {
 288                                 if ((m->oflags & VPO_BUSY) != 0 ||
 289                                     m->busy != 0) {
 290                                         vm_page_sleep(m, "shmtrc");
 291                                         goto retry;
 292                                 }
 293                         } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
 294                                 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
 295                                 if (m == NULL) {
 296                                         VM_OBJECT_UNLOCK(object);
 297                                         VM_WAIT;
 298                                         VM_OBJECT_LOCK(object);
 299                                         goto retry;
 300                                 } else if (m->valid != VM_PAGE_BITS_ALL) {
 301                                         ma[0] = m;
 302                                         rv = vm_pager_get_pages(object, ma, 1,
 303                                             0);
 304                                         m = vm_page_lookup(object, idx);
 305                                 } else
 306                                         /* A cached page was reactivated. */
 307                                         rv = VM_PAGER_OK;
 308                                 vm_page_lock(m);
 309                                 if (rv == VM_PAGER_OK) {
 310                                         vm_page_deactivate(m);
 311                                         vm_page_unlock(m);
 312                                         vm_page_wakeup(m);
 313                                 } else {
 314                                         vm_page_free(m);
 315                                         vm_page_unlock(m);
 316                                         VM_OBJECT_UNLOCK(object);
 317                                         return (EIO);
 318                                 }
 319                         }
 320                         if (m != NULL) {
 321                                 pmap_zero_page_area(m, base, PAGE_SIZE - base);
 322                                 KASSERT(m->valid == VM_PAGE_BITS_ALL,
 323                                     ("shm_dotruncate: page %p is invalid", m));
 324                                 vm_page_dirty(m);
 325                                 vm_pager_page_unswapped(m);
 326                         }
 327                 }
 328                 delta = ptoa(object->size - nobjsize);
 329
 330                 /* Toss in memory pages. */
 331                 if (nobjsize < object->size)
 332                         vm_object_page_remove(object, nobjsize, object->size,
 333                             0);
 334
 335                 /* Toss pages from swap. */
 336                 if (object->type == OBJT_SWAP)
 337                         swap_pager_freespace(object, nobjsize, delta);
 338
 339                 /* Free the swap accounted for shm */
 340                 swap_release_by_cred(delta, object->cred);
 341                 object->charge -= delta;
 342         } else {
 343                 /* Attempt to reserve the swap */
 344                 delta = ptoa(nobjsize - object->size);
 345                 if (!swap_reserve_by_cred(delta, object->cred)) {
 346                         VM_OBJECT_UNLOCK(object);
 347                         return (ENOMEM);
 348                 }
 349                 object->charge += delta;
 350         }
 351         shmfd->shm_size = length;
 352         mtx_lock(&shm_timestamp_lock);
 353         vfs_timestamp(&shmfd->shm_ctime);
 354         shmfd->shm_mtime = shmfd->shm_ctime;
 355         mtx_unlock(&shm_timestamp_lock);
 356         object->size = nobjsize;
 357         VM_OBJECT_UNLOCK(object);
 358         return (0);
 359 }
 360
 361 /*
 362  * shmfd object management including creation and reference counting
 363  * routines.
 364  */
 365 static struct shmfd *
 366 shm_alloc(struct ucred *ucred, mode_t mode)
 367 {
 368         struct shmfd *shmfd;
 369         int ino;
 370
 371         shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
 372         shmfd->shm_size = 0;
 373         shmfd->shm_uid = ucred->cr_uid;
 374         shmfd->shm_gid = ucred->cr_gid;
 375         shmfd->shm_mode = mode;
 376         shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
 377             shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
 378         KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
 379         VM_OBJECT_LOCK(shmfd->shm_object);
 380         vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
 381         vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
 382         VM_OBJECT_UNLOCK(shmfd->shm_object);
 383         vfs_timestamp(&shmfd->shm_birthtime);
 384         shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
 385             shmfd->shm_birthtime;
 386         ino = alloc_unr(shm_ino_unr);
 387         if (ino == -1)
 388                 shmfd->shm_ino = 0;
 389         else
 390                 shmfd->shm_ino = ino;
 391         refcount_init(&shmfd->shm_refs, 1);
 392 #ifdef MAC
 393         mac_posixshm_init(shmfd);
 394         mac_posixshm_create(ucred, shmfd);
 395 #endif
 396
 397         return (shmfd);
 398 }
 399
 400 static struct shmfd *
 401 shm_hold(struct shmfd *shmfd)
 402 {
 403
 404         refcount_acquire(&shmfd->shm_refs);
 405         return (shmfd);
 406 }
 407
 408 static void
 409 shm_drop(struct shmfd *shmfd)
 410 {
 411
 412         if (refcount_release(&shmfd->shm_refs)) {
 413 #ifdef MAC
 414                 mac_posixshm_destroy(shmfd);
 415 #endif
 416                 vm_object_deallocate(shmfd->shm_object);
 417                 if (shmfd->shm_ino != 0)
 418                         free_unr(shm_ino_unr, shmfd->shm_ino);
 419                 free(shmfd, M_SHMFD);
 420         }
 421 }
 422
 423 /*
 424  * Determine if the credentials have sufficient permissions for a
 425  * specified combination of FREAD and FWRITE.
 426  */
 427 static int
 428 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
 429 {
 430         accmode_t accmode;
 431         int error;
 432
 433         accmode = 0;
 434         if (flags & FREAD)
 435                 accmode |= VREAD;
 436         if (flags & FWRITE)
 437                 accmode |= VWRITE;
 438         mtx_lock(&shm_timestamp_lock);
 439         error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
 440             accmode, ucred, NULL);
 441         mtx_unlock(&shm_timestamp_lock);
 442         return (error);
 443 }
 444
 445 /*
 446  * Dictionary management.  We maintain an in-kernel dictionary to map
 447  * paths to shmfd objects.  We use the FNV hash on the path to store
 448  * the mappings in a hash table.
 449  */
 450 static void
 451 shm_init(void *arg)
 452 {
 453
 454         mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
 455         sx_init(&shm_dict_lock, "shm dictionary");
 456         shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
 457         shm_ino_unr = new_unrhdr(1, INT32_MAX, NULL);
 458         KASSERT(shm_ino_unr != NULL, ("shm fake inodes not initialized"));
 459         shm_dev_ino = devfs_alloc_cdp_inode();
 460         KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized"));
 461 }
 462 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL);
 463
 464 static struct shmfd *
 465 shm_lookup(char *path, Fnv32_t fnv)
 466 {
 467         struct shm_mapping *map;
 468
 469         LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
 470                 if (map->sm_fnv != fnv)
 471                         continue;
 472                 if (strcmp(map->sm_path, path) == 0)
 473                         return (map->sm_shmfd);
 474         }
 475
 476         return (NULL);
 477 }
 478
 479 static void
 480 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
 481 {
 482         struct shm_mapping *map;
 483
 484         map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
 485         map->sm_path = path;
 486         map->sm_fnv = fnv;
 487         map->sm_shmfd = shm_hold(shmfd);
 488         shmfd->shm_path = path;
 489         LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
 490 }
 491
 492 static int
 493 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
 494 {
 495         struct shm_mapping *map;
 496         int error;
 497
 498         LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
 499                 if (map->sm_fnv != fnv)
 500                         continue;
 501                 if (strcmp(map->sm_path, path) == 0) {
 502 #ifdef MAC
 503                         error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
 504                         if (error)
 505                                 return (error);
 506 #endif
 507                         error = shm_access(map->sm_shmfd, ucred,
 508                             FREAD | FWRITE);
 509                         if (error)
 510                                 return (error);
 511                         map->sm_shmfd->shm_path = NULL;
 512                         LIST_REMOVE(map, sm_link);
 513                         shm_drop(map->sm_shmfd);
 514                         free(map->sm_path, M_SHMFD);
 515                         free(map, M_SHMFD);
 516                         return (0);
 517                 }
 518         }
 519
 520         return (ENOENT);
 521 }
 522
 523 /* System calls. */
 524 int
 525 sys_shm_open(struct thread *td, struct shm_open_args *uap)
 526 {
 527         struct filedesc *fdp;
 528         struct shmfd *shmfd;
 529         struct file *fp;
 530         char *path;
 531         Fnv32_t fnv;
 532         mode_t cmode;
 533         int fd, error;
 534
 535 #ifdef CAPABILITY_MODE
 536         /*
 537          * shm_open(2) is only allowed for anonymous objects.
 538          */
 539         if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
 540                 return (ECAPMODE);
 541 #endif
 542
 543         if ((uap->flags & O_ACCMODE) != O_RDONLY &&
 544             (uap->flags & O_ACCMODE) != O_RDWR)
 545                 return (EINVAL);
 546
 547         if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0)
 548                 return (EINVAL);
 549
 550         fdp = td->td_proc->p_fd;
 551         cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
 552
 553         error = falloc(td, &fp, &fd, O_CLOEXEC);
 554         if (error)
 555                 return (error);
 556
 557         /* A SHM_ANON path pointer creates an anonymous object. */
 558         if (uap->path == SHM_ANON) {
 559                 /* A read-only anonymous object is pointless. */
 560                 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
 561                         fdclose(fdp, fp, fd, td);
 562                         fdrop(fp, td);
 563                         return (EINVAL);
 564                 }
 565                 shmfd = shm_alloc(td->td_ucred, cmode);
 566         } else {
 567                 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
 568                 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 569
 570                 /* Require paths to start with a '/' character. */
 571                 if (error == 0 && path[0] != '/')
 572                         error = EINVAL;
 573                 if (error) {
 574                         fdclose(fdp, fp, fd, td);
 575                         fdrop(fp, td);
 576                         free(path, M_SHMFD);
 577                         return (error);
 578                 }
 579
 580                 fnv = fnv_32_str(path, FNV1_32_INIT);
 581                 sx_xlock(&shm_dict_lock);
 582                 shmfd = shm_lookup(path, fnv);
 583                 if (shmfd == NULL) {
 584                         /* Object does not yet exist, create it if requested. */
 585                         if (uap->flags & O_CREAT) {
 586 #ifdef MAC
 587                                 error = mac_posixshm_check_create(td->td_ucred,
 588                                     path);
 589                                 if (error == 0) {
 590 #endif
 591                                         shmfd = shm_alloc(td->td_ucred, cmode);
 592                                         shm_insert(path, fnv, shmfd);
 593 #ifdef MAC
 594                                 }
 595 #endif
 596                         } else {
 597                                 free(path, M_SHMFD);
 598                                 error = ENOENT;
 599                         }
 600                 } else {
 601                         /*
 602                          * Object already exists, obtain a new
 603                          * reference if requested and permitted.
 604                          */
 605                         free(path, M_SHMFD);
 606                         if ((uap->flags & (O_CREAT | O_EXCL)) ==
 607                             (O_CREAT | O_EXCL))
 608                                 error = EEXIST;
 609                         else {
 610 #ifdef MAC
 611                                 error = mac_posixshm_check_open(td->td_ucred,
 612                                     shmfd, FFLAGS(uap->flags & O_ACCMODE));
 613                                 if (error == 0)
 614 #endif
 615                                 error = shm_access(shmfd, td->td_ucred,
 616                                     FFLAGS(uap->flags & O_ACCMODE));
 617                         }
 618
 619                         /*
 620                          * Truncate the file back to zero length if
 621                          * O_TRUNC was specified and the object was
 622                          * opened with read/write.
 623                          */
 624                         if (error == 0 &&
 625                             (uap->flags & (O_ACCMODE | O_TRUNC)) ==
 626                             (O_RDWR | O_TRUNC)) {
 627 #ifdef MAC
 628                                 error = mac_posixshm_check_truncate(
 629                                         td->td_ucred, fp->f_cred, shmfd);
 630                                 if (error == 0)
 631 #endif
 632                                         shm_dotruncate(shmfd, 0);
 633                         }
 634                         if (error == 0)
 635                                 shm_hold(shmfd);
 636                 }
 637                 sx_xunlock(&shm_dict_lock);
 638
 639                 if (error) {
 640                         fdclose(fdp, fp, fd, td);
 641                         fdrop(fp, td);
 642                         return (error);
 643                 }
 644         }
 645
 646         finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
 647
 648         td->td_retval[0] = fd;
 649         fdrop(fp, td);
 650
 651         return (0);
 652 }
 653
 654 int
 655 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
 656 {
 657         char *path;
 658         Fnv32_t fnv;
 659         int error;
 660
 661         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 662         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 663         if (error) {
 664                 free(path, M_TEMP);
 665                 return (error);
 666         }
 667
 668         fnv = fnv_32_str(path, FNV1_32_INIT);
 669         sx_xlock(&shm_dict_lock);
 670         error = shm_remove(path, fnv, td->td_ucred);
 671         sx_xunlock(&shm_dict_lock);
 672         free(path, M_TEMP);
 673
 674         return (error);
 675 }
 676
 677 /*
 678  * mmap() helper to validate mmap() requests against shm object state
 679  * and give mmap() the vm_object to use for the mapping.
 680  */
 681 int
 682 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
 683     vm_object_t *obj)
 684 {
 685
 686         /*
 687          * XXXRW: This validation is probably insufficient, and subject to
 688          * sign errors.  It should be fixed.
 689          */
 690         if (foff >= shmfd->shm_size ||
 691             foff + objsize > round_page(shmfd->shm_size))
 692                 return (EINVAL);
 693
 694         mtx_lock(&shm_timestamp_lock);
 695         vfs_timestamp(&shmfd->shm_atime);
 696         mtx_unlock(&shm_timestamp_lock);
 697         vm_object_reference(shmfd->shm_object);
 698         *obj = shmfd->shm_object;
 699         return (0);
 700 }
 701
 702 static int
 703 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
 704     struct thread *td)
 705 {
 706         struct shmfd *shmfd;
 707         int error;
 708
 709         error = 0;
 710         shmfd = fp->f_data;
 711         mtx_lock(&shm_timestamp_lock);
 712         /*
 713          * SUSv4 says that x bits of permission need not be affected.
 714          * Be consistent with our shm_open there.
 715          */
 716 #ifdef MAC
 717         error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
 718         if (error != 0)
 719                 goto out;
 720 #endif
 721         error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
 722             shmfd->shm_gid, VADMIN, active_cred, NULL);
 723         if (error != 0)
 724                 goto out;
 725         shmfd->shm_mode = mode & ACCESSPERMS;
 726 out:
 727         mtx_unlock(&shm_timestamp_lock);
 728         return (error);
 729 }
 730
 731 static int
 732 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 733     struct thread *td)
 734 {
 735         struct shmfd *shmfd;
 736         int error;
 737
 738         error = 0;
 739         shmfd = fp->f_data;
 740         mtx_lock(&shm_timestamp_lock);
 741 #ifdef MAC
 742         error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
 743         if (error != 0)
 744                 goto out;
 745 #endif
 746         if (uid == (uid_t)-1)
 747                 uid = shmfd->shm_uid;
 748         if (gid == (gid_t)-1)
 749                  gid = shmfd->shm_gid;
 750         if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
 751             (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
 752             (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
 753                 goto out;
 754         shmfd->shm_uid = uid;
 755         shmfd->shm_gid = gid;
 756 out:
 757         mtx_unlock(&shm_timestamp_lock);
 758         return (error);
 759 }
 760
 761 /*
 762  * Helper routines to allow the backing object of a shared memory file
 763  * descriptor to be mapped in the kernel.
 764  */
 765 int
 766 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
 767 {
 768         struct shmfd *shmfd;
 769         vm_offset_t kva, ofs;
 770         vm_object_t obj;
 771         int rv;
 772
 773         if (fp->f_type != DTYPE_SHM)
 774                 return (EINVAL);
 775         shmfd = fp->f_data;
 776         obj = shmfd->shm_object;
 777         VM_OBJECT_LOCK(obj);
 778         /*
 779          * XXXRW: This validation is probably insufficient, and subject to
 780          * sign errors.  It should be fixed.
 781          */
 782         if (offset >= shmfd->shm_size ||
 783             offset + size > round_page(shmfd->shm_size)) {
 784                 VM_OBJECT_UNLOCK(obj);
 785                 return (EINVAL);
 786         }
 787
 788         shmfd->shm_kmappings++;
 789         vm_object_reference_locked(obj);
 790         VM_OBJECT_UNLOCK(obj);
 791
 792         /* Map the object into the kernel_map and wire it. */
 793         kva = vm_map_min(kernel_map);
 794         ofs = offset & PAGE_MASK;
 795         offset = trunc_page(offset);
 796         size = round_page(size + ofs);
 797         rv = vm_map_find(kernel_map, obj, offset, &kva, size,
 798             VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
 799             VM_PROT_READ | VM_PROT_WRITE, 0);
 800         if (rv == KERN_SUCCESS) {
 801                 rv = vm_map_wire(kernel_map, kva, kva + size,
 802                     VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
 803                 if (rv == KERN_SUCCESS) {
 804                         *memp = (void *)(kva + ofs);
 805                         return (0);
 806                 }
 807                 vm_map_remove(kernel_map, kva, kva + size);
 808         } else
 809                 vm_object_deallocate(obj);
 810
 811         /* On failure, drop our mapping reference. */
 812         VM_OBJECT_LOCK(obj);
 813         shmfd->shm_kmappings--;
 814         VM_OBJECT_UNLOCK(obj);
 815
 816         return (vm_mmap_to_errno(rv));
 817 }
 818
 819 /*
 820  * We require the caller to unmap the entire entry.  This allows us to
 821  * safely decrement shm_kmappings when a mapping is removed.
 822  */
 823 int
 824 shm_unmap(struct file *fp, void *mem, size_t size)
 825 {
 826         struct shmfd *shmfd;
 827         vm_map_entry_t entry;
 828         vm_offset_t kva, ofs;
 829         vm_object_t obj;
 830         vm_pindex_t pindex;
 831         vm_prot_t prot;
 832         boolean_t wired;
 833         vm_map_t map;
 834         int rv;
 835
 836         if (fp->f_type != DTYPE_SHM)
 837                 return (EINVAL);
 838         shmfd = fp->f_data;
 839         kva = (vm_offset_t)mem;
 840         ofs = kva & PAGE_MASK;
 841         kva = trunc_page(kva);
 842         size = round_page(size + ofs);
 843         map = kernel_map;
 844         rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
 845             &obj, &pindex, &prot, &wired);
 846         if (rv != KERN_SUCCESS)
 847                 return (EINVAL);
 848         if (entry->start != kva || entry->end != kva + size) {
 849                 vm_map_lookup_done(map, entry);
 850                 return (EINVAL);
 851         }
 852         vm_map_lookup_done(map, entry);
 853         if (obj != shmfd->shm_object)
 854                 return (EINVAL);
 855         vm_map_remove(map, kva, kva + size);
 856         VM_OBJECT_LOCK(obj);
 857         KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
 858         shmfd->shm_kmappings--;
 859         VM_OBJECT_UNLOCK(obj);
 860         return (0);
 861 }
 862
 863 void
 864 shm_path(struct shmfd *shmfd, char *path, size_t size)
 865 {
 866
 867         if (shmfd->shm_path == NULL)
 868                 return;
 869         sx_slock(&shm_dict_lock);
 870         if (shmfd->shm_path != NULL)
 871                 strlcpy(path, shmfd->shm_path, size);
 872         sx_sunlock(&shm_dict_lock);
 873 }