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 #include "opt_ktrace.h"
  49
  50 #include <sys/param.h>
  51 #include <sys/capsicum.h>
  52 #include <sys/conf.h>
  53 #include <sys/fcntl.h>
  54 #include <sys/file.h>
  55 #include <sys/filedesc.h>
  56 #include <sys/fnv_hash.h>
  57 #include <sys/kernel.h>
  58 #include <sys/uio.h>
  59 #include <sys/signal.h>
  60 #include <sys/ktrace.h>
  61 #include <sys/lock.h>
  62 #include <sys/malloc.h>
  63 #include <sys/mman.h>
  64 #include <sys/mutex.h>
  65 #include <sys/priv.h>
  66 #include <sys/proc.h>
  67 #include <sys/refcount.h>
  68 #include <sys/resourcevar.h>
  69 #include <sys/rwlock.h>
  70 #include <sys/stat.h>
  71 #include <sys/sysctl.h>
  72 #include <sys/sysproto.h>
  73 #include <sys/systm.h>
  74 #include <sys/sx.h>
  75 #include <sys/time.h>
  76 #include <sys/vnode.h>
  77 #include <sys/unistd.h>
  78 #include <sys/user.h>
  79
  80 #include <security/mac/mac_framework.h>
  81
  82 #include <vm/vm.h>
  83 #include <vm/vm_param.h>
  84 #include <vm/pmap.h>
  85 #include <vm/vm_extern.h>
  86 #include <vm/vm_map.h>
  87 #include <vm/vm_kern.h>
  88 #include <vm/vm_object.h>
  89 #include <vm/vm_page.h>
  90 #include <vm/vm_pageout.h>
  91 #include <vm/vm_pager.h>
  92 #include <vm/swap_pager.h>
  93
  94 struct shm_mapping {
  95         char            *sm_path;
  96         Fnv32_t         sm_fnv;
  97         struct shmfd    *sm_shmfd;
  98         LIST_ENTRY(shm_mapping) sm_link;
  99 };
 100
 101 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor");
 102 static LIST_HEAD(, shm_mapping) *shm_dictionary;
 103 static struct sx shm_dict_lock;
 104 static struct mtx shm_timestamp_lock;
 105 static u_long shm_hash;
 106 static struct unrhdr *shm_ino_unr;
 107 static dev_t shm_dev_ino;
 108
 109 #define SHM_HASH(fnv)   (&shm_dictionary[(fnv) & shm_hash])
 110
 111 static int      shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags);
 112 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode);
 113 static void     shm_init(void *arg);
 114 static void     shm_drop(struct shmfd *shmfd);
 115 static struct shmfd *shm_hold(struct shmfd *shmfd);
 116 static void     shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd);
 117 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv);
 118 static int      shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred);
 119 static int      shm_dotruncate(struct shmfd *shmfd, off_t length);
 120
 121 static fo_rdwr_t        shm_read;
 122 static fo_rdwr_t        shm_write;
 123 static fo_truncate_t    shm_truncate;
 124 static fo_stat_t        shm_stat;
 125 static fo_close_t       shm_close;
 126 static fo_chmod_t       shm_chmod;
 127 static fo_chown_t       shm_chown;
 128 static fo_seek_t        shm_seek;
 129 static fo_fill_kinfo_t  shm_fill_kinfo;
 130
 131 /* File descriptor operations. */
 132 static struct fileops shm_ops = {
 133         .fo_read = shm_read,
 134         .fo_write = shm_write,
 135         .fo_truncate = shm_truncate,
 136         .fo_ioctl = invfo_ioctl,
 137         .fo_poll = invfo_poll,
 138         .fo_kqfilter = invfo_kqfilter,
 139         .fo_stat = shm_stat,
 140         .fo_close = shm_close,
 141         .fo_chmod = shm_chmod,
 142         .fo_chown = shm_chown,
 143         .fo_sendfile = vn_sendfile,
 144         .fo_seek = shm_seek,
 145         .fo_fill_kinfo = shm_fill_kinfo,
 146         .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
 147 };
 148
 149 FEATURE(posix_shm, "POSIX shared memory");
 150
 151 static int
 152 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
 153 {
 154         vm_page_t m;
 155         vm_pindex_t idx;
 156         size_t tlen;
 157         int error, offset, rv;
 158
 159         idx = OFF_TO_IDX(uio->uio_offset);
 160         offset = uio->uio_offset & PAGE_MASK;
 161         tlen = MIN(PAGE_SIZE - offset, len);
 162
 163         VM_OBJECT_WLOCK(obj);
 164
 165         /*
 166          * Read I/O without either a corresponding resident page or swap
 167          * page: use zero_region.  This is intended to avoid instantiating
 168          * pages on read from a sparse region.
 169          */
 170         if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL &&
 171             !vm_pager_has_page(obj, idx, NULL, NULL)) {
 172                 VM_OBJECT_WUNLOCK(obj);
 173                 return (uiomove(__DECONST(void *, zero_region), tlen, uio));
 174         }
 175
 176         /*
 177          * Parallel reads of the page content from disk are prevented
 178          * by exclusive busy.
 179          *
 180          * Although the tmpfs vnode lock is held here, it is
 181          * nonetheless safe to sleep waiting for a free page.  The
 182          * pageout daemon does not need to acquire the tmpfs vnode
 183          * lock to page out tobj's pages because tobj is a OBJT_SWAP
 184          * type object.
 185          */
 186         m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL);
 187         if (m->valid != VM_PAGE_BITS_ALL) {
 188                 if (vm_pager_has_page(obj, idx, NULL, NULL)) {
 189                         rv = vm_pager_get_pages(obj, &m, 1, 0);
 190                         m = vm_page_lookup(obj, idx);
 191                         if (m == NULL) {
 192                                 printf(
 193                     "uiomove_object: vm_obj %p idx %jd null lookup rv %d\n",
 194                                     obj, idx, rv);
 195                                 VM_OBJECT_WUNLOCK(obj);
 196                                 return (EIO);
 197                         }
 198                         if (rv != VM_PAGER_OK) {
 199                                 printf(
 200             "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
 201                                     obj, idx, m->valid, rv);
 202                                 vm_page_lock(m);
 203                                 vm_page_free(m);
 204                                 vm_page_unlock(m);
 205                                 VM_OBJECT_WUNLOCK(obj);
 206                                 return (EIO);
 207                         }
 208                 } else
 209                         vm_page_zero_invalid(m, TRUE);
 210         }
 211         vm_page_xunbusy(m);
 212         vm_page_lock(m);
 213         vm_page_hold(m);
 214         if (m->queue == PQ_NONE) {
 215                 vm_page_deactivate(m);
 216         } else {
 217                 /* Requeue to maintain LRU ordering. */
 218                 vm_page_requeue(m);
 219         }
 220         vm_page_unlock(m);
 221         VM_OBJECT_WUNLOCK(obj);
 222         error = uiomove_fromphys(&m, offset, tlen, uio);
 223         if (uio->uio_rw == UIO_WRITE && error == 0) {
 224                 VM_OBJECT_WLOCK(obj);
 225                 vm_page_dirty(m);
 226                 vm_pager_page_unswapped(m);
 227                 VM_OBJECT_WUNLOCK(obj);
 228         }
 229         vm_page_lock(m);
 230         vm_page_unhold(m);
 231         vm_page_unlock(m);
 232
 233         return (error);
 234 }
 235
 236 int
 237 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio)
 238 {
 239         ssize_t resid;
 240         size_t len;
 241         int error;
 242
 243         error = 0;
 244         while ((resid = uio->uio_resid) > 0) {
 245                 if (obj_size <= uio->uio_offset)
 246                         break;
 247                 len = MIN(obj_size - uio->uio_offset, resid);
 248                 if (len == 0)
 249                         break;
 250                 error = uiomove_object_page(obj, len, uio);
 251                 if (error != 0 || resid == uio->uio_resid)
 252                         break;
 253         }
 254         return (error);
 255 }
 256
 257 static int
 258 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td)
 259 {
 260         struct shmfd *shmfd;
 261         off_t foffset;
 262         int error;
 263
 264         shmfd = fp->f_data;
 265         foffset = foffset_lock(fp, 0);
 266         error = 0;
 267         switch (whence) {
 268         case L_INCR:
 269                 if (foffset < 0 ||
 270                     (offset > 0 && foffset > OFF_MAX - offset)) {
 271                         error = EOVERFLOW;
 272                         break;
 273                 }
 274                 offset += foffset;
 275                 break;
 276         case L_XTND:
 277                 if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) {
 278                         error = EOVERFLOW;
 279                         break;
 280                 }
 281                 offset += shmfd->shm_size;
 282                 break;
 283         case L_SET:
 284                 break;
 285         default:
 286                 error = EINVAL;
 287         }
 288         if (error == 0) {
 289                 if (offset < 0 || offset > shmfd->shm_size)
 290                         error = EINVAL;
 291                 else
 292                         td->td_uretoff.tdu_off = offset;
 293         }
 294         foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
 295         return (error);
 296 }
 297
 298 static int
 299 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
 300     int flags, struct thread *td)
 301 {
 302         struct shmfd *shmfd;
 303         void *rl_cookie;
 304         int error;
 305
 306         shmfd = fp->f_data;
 307         foffset_lock_uio(fp, uio, flags);
 308         rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset,
 309             uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
 310 #ifdef MAC
 311         error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd);
 312         if (error)
 313                 return (error);
 314 #endif
 315         error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
 316         rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
 317         foffset_unlock_uio(fp, uio, flags);
 318         return (error);
 319 }
 320
 321 static int
 322 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
 323     int flags, struct thread *td)
 324 {
 325         struct shmfd *shmfd;
 326         void *rl_cookie;
 327         int error;
 328
 329         shmfd = fp->f_data;
 330 #ifdef MAC
 331         error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd);
 332         if (error)
 333                 return (error);
 334 #endif
 335         foffset_lock_uio(fp, uio, flags);
 336         if ((flags & FOF_OFFSET) == 0) {
 337                 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX,
 338                     &shmfd->shm_mtx);
 339         } else {
 340                 rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset,
 341                     uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx);
 342         }
 343
 344         error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio);
 345         rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx);
 346         foffset_unlock_uio(fp, uio, flags);
 347         return (error);
 348 }
 349
 350 static int
 351 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred,
 352     struct thread *td)
 353 {
 354         struct shmfd *shmfd;
 355 #ifdef MAC
 356         int error;
 357 #endif
 358
 359         shmfd = fp->f_data;
 360 #ifdef MAC
 361         error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd);
 362         if (error)
 363                 return (error);
 364 #endif
 365         return (shm_dotruncate(shmfd, length));
 366 }
 367
 368 static int
 369 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
 370     struct thread *td)
 371 {
 372         struct shmfd *shmfd;
 373 #ifdef MAC
 374         int error;
 375 #endif
 376
 377         shmfd = fp->f_data;
 378
 379 #ifdef MAC
 380         error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd);
 381         if (error)
 382                 return (error);
 383 #endif
 384
 385         /*
 386          * Attempt to return sanish values for fstat() on a memory file
 387          * descriptor.
 388          */
 389         bzero(sb, sizeof(*sb));
 390         sb->st_blksize = PAGE_SIZE;
 391         sb->st_size = shmfd->shm_size;
 392         sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize;
 393         mtx_lock(&shm_timestamp_lock);
 394         sb->st_atim = shmfd->shm_atime;
 395         sb->st_ctim = shmfd->shm_ctime;
 396         sb->st_mtim = shmfd->shm_mtime;
 397         sb->st_birthtim = shmfd->shm_birthtime;
 398         sb->st_mode = S_IFREG | shmfd->shm_mode;                /* XXX */
 399         sb->st_uid = shmfd->shm_uid;
 400         sb->st_gid = shmfd->shm_gid;
 401         mtx_unlock(&shm_timestamp_lock);
 402         sb->st_dev = shm_dev_ino;
 403         sb->st_ino = shmfd->shm_ino;
 404
 405         return (0);
 406 }
 407
 408 static int
 409 shm_close(struct file *fp, struct thread *td)
 410 {
 411         struct shmfd *shmfd;
 412
 413         shmfd = fp->f_data;
 414         fp->f_data = NULL;
 415         shm_drop(shmfd);
 416
 417         return (0);
 418 }
 419
 420 static int
 421 shm_dotruncate(struct shmfd *shmfd, off_t length)
 422 {
 423         vm_object_t object;
 424         vm_page_t m, ma[1];
 425         vm_pindex_t idx, nobjsize;
 426         vm_ooffset_t delta;
 427         int base, rv;
 428
 429         object = shmfd->shm_object;
 430         VM_OBJECT_WLOCK(object);
 431         if (length == shmfd->shm_size) {
 432                 VM_OBJECT_WUNLOCK(object);
 433                 return (0);
 434         }
 435         nobjsize = OFF_TO_IDX(length + PAGE_MASK);
 436
 437         /* Are we shrinking?  If so, trim the end. */
 438         if (length < shmfd->shm_size) {
 439                 /*
 440                  * Disallow any requests to shrink the size if this
 441                  * object is mapped into the kernel.
 442                  */
 443                 if (shmfd->shm_kmappings > 0) {
 444                         VM_OBJECT_WUNLOCK(object);
 445                         return (EBUSY);
 446                 }
 447
 448                 /*
 449                  * Zero the truncated part of the last page.
 450                  */
 451                 base = length & PAGE_MASK;
 452                 if (base != 0) {
 453                         idx = OFF_TO_IDX(length);
 454 retry:
 455                         m = vm_page_lookup(object, idx);
 456                         if (m != NULL) {
 457                                 if (vm_page_sleep_if_busy(m, "shmtrc"))
 458                                         goto retry;
 459                         } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
 460                                 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
 461                                 if (m == NULL) {
 462                                         VM_OBJECT_WUNLOCK(object);
 463                                         VM_WAIT;
 464                                         VM_OBJECT_WLOCK(object);
 465                                         goto retry;
 466                                 } else if (m->valid != VM_PAGE_BITS_ALL) {
 467                                         ma[0] = m;
 468                                         rv = vm_pager_get_pages(object, ma, 1,
 469                                             0);
 470                                         m = vm_page_lookup(object, idx);
 471                                 } else
 472                                         /* A cached page was reactivated. */
 473                                         rv = VM_PAGER_OK;
 474                                 vm_page_lock(m);
 475                                 if (rv == VM_PAGER_OK) {
 476                                         vm_page_deactivate(m);
 477                                         vm_page_unlock(m);
 478                                         vm_page_xunbusy(m);
 479                                 } else {
 480                                         vm_page_free(m);
 481                                         vm_page_unlock(m);
 482                                         VM_OBJECT_WUNLOCK(object);
 483                                         return (EIO);
 484                                 }
 485                         }
 486                         if (m != NULL) {
 487                                 pmap_zero_page_area(m, base, PAGE_SIZE - base);
 488                                 KASSERT(m->valid == VM_PAGE_BITS_ALL,
 489                                     ("shm_dotruncate: page %p is invalid", m));
 490                                 vm_page_dirty(m);
 491                                 vm_pager_page_unswapped(m);
 492                         }
 493                 }
 494                 delta = ptoa(object->size - nobjsize);
 495
 496                 /* Toss in memory pages. */
 497                 if (nobjsize < object->size)
 498                         vm_object_page_remove(object, nobjsize, object->size,
 499                             0);
 500
 501                 /* Toss pages from swap. */
 502                 if (object->type == OBJT_SWAP)
 503                         swap_pager_freespace(object, nobjsize, delta);
 504
 505                 /* Free the swap accounted for shm */
 506                 swap_release_by_cred(delta, object->cred);
 507                 object->charge -= delta;
 508         } else {
 509                 /* Attempt to reserve the swap */
 510                 delta = ptoa(nobjsize - object->size);
 511                 if (!swap_reserve_by_cred(delta, object->cred)) {
 512                         VM_OBJECT_WUNLOCK(object);
 513                         return (ENOMEM);
 514                 }
 515                 object->charge += delta;
 516         }
 517         shmfd->shm_size = length;
 518         mtx_lock(&shm_timestamp_lock);
 519         vfs_timestamp(&shmfd->shm_ctime);
 520         shmfd->shm_mtime = shmfd->shm_ctime;
 521         mtx_unlock(&shm_timestamp_lock);
 522         object->size = nobjsize;
 523         VM_OBJECT_WUNLOCK(object);
 524         return (0);
 525 }
 526
 527 /*
 528  * shmfd object management including creation and reference counting
 529  * routines.
 530  */
 531 static struct shmfd *
 532 shm_alloc(struct ucred *ucred, mode_t mode)
 533 {
 534         struct shmfd *shmfd;
 535         int ino;
 536
 537         shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
 538         shmfd->shm_size = 0;
 539         shmfd->shm_uid = ucred->cr_uid;
 540         shmfd->shm_gid = ucred->cr_gid;
 541         shmfd->shm_mode = mode;
 542         shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
 543             shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
 544         KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
 545         shmfd->shm_object->pg_color = 0;
 546         VM_OBJECT_WLOCK(shmfd->shm_object);
 547         vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
 548         vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT);
 549         VM_OBJECT_WUNLOCK(shmfd->shm_object);
 550         vfs_timestamp(&shmfd->shm_birthtime);
 551         shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
 552             shmfd->shm_birthtime;
 553         ino = alloc_unr(shm_ino_unr);
 554         if (ino == -1)
 555                 shmfd->shm_ino = 0;
 556         else
 557                 shmfd->shm_ino = ino;
 558         refcount_init(&shmfd->shm_refs, 1);
 559         mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF);
 560         rangelock_init(&shmfd->shm_rl);
 561 #ifdef MAC
 562         mac_posixshm_init(shmfd);
 563         mac_posixshm_create(ucred, shmfd);
 564 #endif
 565
 566         return (shmfd);
 567 }
 568
 569 static struct shmfd *
 570 shm_hold(struct shmfd *shmfd)
 571 {
 572
 573         refcount_acquire(&shmfd->shm_refs);
 574         return (shmfd);
 575 }
 576
 577 static void
 578 shm_drop(struct shmfd *shmfd)
 579 {
 580
 581         if (refcount_release(&shmfd->shm_refs)) {
 582 #ifdef MAC
 583                 mac_posixshm_destroy(shmfd);
 584 #endif
 585                 rangelock_destroy(&shmfd->shm_rl);
 586                 mtx_destroy(&shmfd->shm_mtx);
 587                 vm_object_deallocate(shmfd->shm_object);
 588                 if (shmfd->shm_ino != 0)
 589                         free_unr(shm_ino_unr, shmfd->shm_ino);
 590                 free(shmfd, M_SHMFD);
 591         }
 592 }
 593
 594 /*
 595  * Determine if the credentials have sufficient permissions for a
 596  * specified combination of FREAD and FWRITE.
 597  */
 598 static int
 599 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags)
 600 {
 601         accmode_t accmode;
 602         int error;
 603
 604         accmode = 0;
 605         if (flags & FREAD)
 606                 accmode |= VREAD;
 607         if (flags & FWRITE)
 608                 accmode |= VWRITE;
 609         mtx_lock(&shm_timestamp_lock);
 610         error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid,
 611             accmode, ucred, NULL);
 612         mtx_unlock(&shm_timestamp_lock);
 613         return (error);
 614 }
 615
 616 /*
 617  * Dictionary management.  We maintain an in-kernel dictionary to map
 618  * paths to shmfd objects.  We use the FNV hash on the path to store
 619  * the mappings in a hash table.
 620  */
 621 static void
 622 shm_init(void *arg)
 623 {
 624
 625         mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF);
 626         sx_init(&shm_dict_lock, "shm dictionary");
 627         shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash);
 628         shm_ino_unr = new_unrhdr(1, INT32_MAX, NULL);
 629         KASSERT(shm_ino_unr != NULL, ("shm fake inodes not initialized"));
 630         shm_dev_ino = devfs_alloc_cdp_inode();
 631         KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized"));
 632 }
 633 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL);
 634
 635 static struct shmfd *
 636 shm_lookup(char *path, Fnv32_t fnv)
 637 {
 638         struct shm_mapping *map;
 639
 640         LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
 641                 if (map->sm_fnv != fnv)
 642                         continue;
 643                 if (strcmp(map->sm_path, path) == 0)
 644                         return (map->sm_shmfd);
 645         }
 646
 647         return (NULL);
 648 }
 649
 650 static void
 651 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd)
 652 {
 653         struct shm_mapping *map;
 654
 655         map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK);
 656         map->sm_path = path;
 657         map->sm_fnv = fnv;
 658         map->sm_shmfd = shm_hold(shmfd);
 659         shmfd->shm_path = path;
 660         LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link);
 661 }
 662
 663 static int
 664 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred)
 665 {
 666         struct shm_mapping *map;
 667         int error;
 668
 669         LIST_FOREACH(map, SHM_HASH(fnv), sm_link) {
 670                 if (map->sm_fnv != fnv)
 671                         continue;
 672                 if (strcmp(map->sm_path, path) == 0) {
 673 #ifdef MAC
 674                         error = mac_posixshm_check_unlink(ucred, map->sm_shmfd);
 675                         if (error)
 676                                 return (error);
 677 #endif
 678                         error = shm_access(map->sm_shmfd, ucred,
 679                             FREAD | FWRITE);
 680                         if (error)
 681                                 return (error);
 682                         map->sm_shmfd->shm_path = NULL;
 683                         LIST_REMOVE(map, sm_link);
 684                         shm_drop(map->sm_shmfd);
 685                         free(map->sm_path, M_SHMFD);
 686                         free(map, M_SHMFD);
 687                         return (0);
 688                 }
 689         }
 690
 691         return (ENOENT);
 692 }
 693
 694 /* System calls. */
 695 int
 696 sys_shm_open(struct thread *td, struct shm_open_args *uap)
 697 {
 698         struct filedesc *fdp;
 699         struct shmfd *shmfd;
 700         struct file *fp;
 701         char *path;
 702         Fnv32_t fnv;
 703         mode_t cmode;
 704         int fd, error;
 705
 706 #ifdef CAPABILITY_MODE
 707         /*
 708          * shm_open(2) is only allowed for anonymous objects.
 709          */
 710         if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON))
 711                 return (ECAPMODE);
 712 #endif
 713
 714         if ((uap->flags & O_ACCMODE) != O_RDONLY &&
 715             (uap->flags & O_ACCMODE) != O_RDWR)
 716                 return (EINVAL);
 717
 718         if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0)
 719                 return (EINVAL);
 720
 721         fdp = td->td_proc->p_fd;
 722         cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS;
 723
 724         error = falloc(td, &fp, &fd, O_CLOEXEC);
 725         if (error)
 726                 return (error);
 727
 728         /* A SHM_ANON path pointer creates an anonymous object. */
 729         if (uap->path == SHM_ANON) {
 730                 /* A read-only anonymous object is pointless. */
 731                 if ((uap->flags & O_ACCMODE) == O_RDONLY) {
 732                         fdclose(td, fp, fd);
 733                         fdrop(fp, td);
 734                         return (EINVAL);
 735                 }
 736                 shmfd = shm_alloc(td->td_ucred, cmode);
 737         } else {
 738                 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK);
 739                 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 740 #ifdef KTRACE
 741                 if (error == 0 && KTRPOINT(curthread, KTR_NAMEI))
 742                         ktrnamei(path);
 743 #endif
 744                 /* Require paths to start with a '/' character. */
 745                 if (error == 0 && path[0] != '/')
 746                         error = EINVAL;
 747                 if (error) {
 748                         fdclose(td, fp, fd);
 749                         fdrop(fp, td);
 750                         free(path, M_SHMFD);
 751                         return (error);
 752                 }
 753
 754                 fnv = fnv_32_str(path, FNV1_32_INIT);
 755                 sx_xlock(&shm_dict_lock);
 756                 shmfd = shm_lookup(path, fnv);
 757                 if (shmfd == NULL) {
 758                         /* Object does not yet exist, create it if requested. */
 759                         if (uap->flags & O_CREAT) {
 760 #ifdef MAC
 761                                 error = mac_posixshm_check_create(td->td_ucred,
 762                                     path);
 763                                 if (error == 0) {
 764 #endif
 765                                         shmfd = shm_alloc(td->td_ucred, cmode);
 766                                         shm_insert(path, fnv, shmfd);
 767 #ifdef MAC
 768                                 }
 769 #endif
 770                         } else {
 771                                 free(path, M_SHMFD);
 772                                 error = ENOENT;
 773                         }
 774                 } else {
 775                         /*
 776                          * Object already exists, obtain a new
 777                          * reference if requested and permitted.
 778                          */
 779                         free(path, M_SHMFD);
 780                         if ((uap->flags & (O_CREAT | O_EXCL)) ==
 781                             (O_CREAT | O_EXCL))
 782                                 error = EEXIST;
 783                         else {
 784 #ifdef MAC
 785                                 error = mac_posixshm_check_open(td->td_ucred,
 786                                     shmfd, FFLAGS(uap->flags & O_ACCMODE));
 787                                 if (error == 0)
 788 #endif
 789                                 error = shm_access(shmfd, td->td_ucred,
 790                                     FFLAGS(uap->flags & O_ACCMODE));
 791                         }
 792
 793                         /*
 794                          * Truncate the file back to zero length if
 795                          * O_TRUNC was specified and the object was
 796                          * opened with read/write.
 797                          */
 798                         if (error == 0 &&
 799                             (uap->flags & (O_ACCMODE | O_TRUNC)) ==
 800                             (O_RDWR | O_TRUNC)) {
 801 #ifdef MAC
 802                                 error = mac_posixshm_check_truncate(
 803                                         td->td_ucred, fp->f_cred, shmfd);
 804                                 if (error == 0)
 805 #endif
 806                                         shm_dotruncate(shmfd, 0);
 807                         }
 808                         if (error == 0)
 809                                 shm_hold(shmfd);
 810                 }
 811                 sx_xunlock(&shm_dict_lock);
 812
 813                 if (error) {
 814                         fdclose(td, fp, fd);
 815                         fdrop(fp, td);
 816                         return (error);
 817                 }
 818         }
 819
 820         finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops);
 821
 822         td->td_retval[0] = fd;
 823         fdrop(fp, td);
 824
 825         return (0);
 826 }
 827
 828 int
 829 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap)
 830 {
 831         char *path;
 832         Fnv32_t fnv;
 833         int error;
 834
 835         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 836         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 837         if (error) {
 838                 free(path, M_TEMP);
 839                 return (error);
 840         }
 841 #ifdef KTRACE
 842         if (KTRPOINT(curthread, KTR_NAMEI))
 843                 ktrnamei(path);
 844 #endif
 845         fnv = fnv_32_str(path, FNV1_32_INIT);
 846         sx_xlock(&shm_dict_lock);
 847         error = shm_remove(path, fnv, td->td_ucred);
 848         sx_xunlock(&shm_dict_lock);
 849         free(path, M_TEMP);
 850
 851         return (error);
 852 }
 853
 854 /*
 855  * mmap() helper to validate mmap() requests against shm object state
 856  * and give mmap() the vm_object to use for the mapping.
 857  */
 858 int
 859 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff,
 860     vm_object_t *obj)
 861 {
 862
 863         /*
 864          * XXXRW: This validation is probably insufficient, and subject to
 865          * sign errors.  It should be fixed.
 866          */
 867         if (foff >= shmfd->shm_size ||
 868             foff + objsize > round_page(shmfd->shm_size))
 869                 return (EINVAL);
 870
 871         mtx_lock(&shm_timestamp_lock);
 872         vfs_timestamp(&shmfd->shm_atime);
 873         mtx_unlock(&shm_timestamp_lock);
 874         vm_object_reference(shmfd->shm_object);
 875         *obj = shmfd->shm_object;
 876         return (0);
 877 }
 878
 879 static int
 880 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
 881     struct thread *td)
 882 {
 883         struct shmfd *shmfd;
 884         int error;
 885
 886         error = 0;
 887         shmfd = fp->f_data;
 888         mtx_lock(&shm_timestamp_lock);
 889         /*
 890          * SUSv4 says that x bits of permission need not be affected.
 891          * Be consistent with our shm_open there.
 892          */
 893 #ifdef MAC
 894         error = mac_posixshm_check_setmode(active_cred, shmfd, mode);
 895         if (error != 0)
 896                 goto out;
 897 #endif
 898         error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid,
 899             shmfd->shm_gid, VADMIN, active_cred, NULL);
 900         if (error != 0)
 901                 goto out;
 902         shmfd->shm_mode = mode & ACCESSPERMS;
 903 out:
 904         mtx_unlock(&shm_timestamp_lock);
 905         return (error);
 906 }
 907
 908 static int
 909 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 910     struct thread *td)
 911 {
 912         struct shmfd *shmfd;
 913         int error;
 914
 915         error = 0;
 916         shmfd = fp->f_data;
 917         mtx_lock(&shm_timestamp_lock);
 918 #ifdef MAC
 919         error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid);
 920         if (error != 0)
 921                 goto out;
 922 #endif
 923         if (uid == (uid_t)-1)
 924                 uid = shmfd->shm_uid;
 925         if (gid == (gid_t)-1)
 926                  gid = shmfd->shm_gid;
 927         if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) ||
 928             (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) &&
 929             (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
 930                 goto out;
 931         shmfd->shm_uid = uid;
 932         shmfd->shm_gid = gid;
 933 out:
 934         mtx_unlock(&shm_timestamp_lock);
 935         return (error);
 936 }
 937
 938 /*
 939  * Helper routines to allow the backing object of a shared memory file
 940  * descriptor to be mapped in the kernel.
 941  */
 942 int
 943 shm_map(struct file *fp, size_t size, off_t offset, void **memp)
 944 {
 945         struct shmfd *shmfd;
 946         vm_offset_t kva, ofs;
 947         vm_object_t obj;
 948         int rv;
 949
 950         if (fp->f_type != DTYPE_SHM)
 951                 return (EINVAL);
 952         shmfd = fp->f_data;
 953         obj = shmfd->shm_object;
 954         VM_OBJECT_WLOCK(obj);
 955         /*
 956          * XXXRW: This validation is probably insufficient, and subject to
 957          * sign errors.  It should be fixed.
 958          */
 959         if (offset >= shmfd->shm_size ||
 960             offset + size > round_page(shmfd->shm_size)) {
 961                 VM_OBJECT_WUNLOCK(obj);
 962                 return (EINVAL);
 963         }
 964
 965         shmfd->shm_kmappings++;
 966         vm_object_reference_locked(obj);
 967         VM_OBJECT_WUNLOCK(obj);
 968
 969         /* Map the object into the kernel_map and wire it. */
 970         kva = vm_map_min(kernel_map);
 971         ofs = offset & PAGE_MASK;
 972         offset = trunc_page(offset);
 973         size = round_page(size + ofs);
 974         rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0,
 975             VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
 976             VM_PROT_READ | VM_PROT_WRITE, 0);
 977         if (rv == KERN_SUCCESS) {
 978                 rv = vm_map_wire(kernel_map, kva, kva + size,
 979                     VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
 980                 if (rv == KERN_SUCCESS) {
 981                         *memp = (void *)(kva + ofs);
 982                         return (0);
 983                 }
 984                 vm_map_remove(kernel_map, kva, kva + size);
 985         } else
 986                 vm_object_deallocate(obj);
 987
 988         /* On failure, drop our mapping reference. */
 989         VM_OBJECT_WLOCK(obj);
 990         shmfd->shm_kmappings--;
 991         VM_OBJECT_WUNLOCK(obj);
 992
 993         return (vm_mmap_to_errno(rv));
 994 }
 995
 996 /*
 997  * We require the caller to unmap the entire entry.  This allows us to
 998  * safely decrement shm_kmappings when a mapping is removed.
 999  */
1000 int
1001 shm_unmap(struct file *fp, void *mem, size_t size)
1002 {
1003         struct shmfd *shmfd;
1004         vm_map_entry_t entry;
1005         vm_offset_t kva, ofs;
1006         vm_object_t obj;
1007         vm_pindex_t pindex;
1008         vm_prot_t prot;
1009         boolean_t wired;
1010         vm_map_t map;
1011         int rv;
1012
1013         if (fp->f_type != DTYPE_SHM)
1014                 return (EINVAL);
1015         shmfd = fp->f_data;
1016         kva = (vm_offset_t)mem;
1017         ofs = kva & PAGE_MASK;
1018         kva = trunc_page(kva);
1019         size = round_page(size + ofs);
1020         map = kernel_map;
1021         rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
1022             &obj, &pindex, &prot, &wired);
1023         if (rv != KERN_SUCCESS)
1024                 return (EINVAL);
1025         if (entry->start != kva || entry->end != kva + size) {
1026                 vm_map_lookup_done(map, entry);
1027                 return (EINVAL);
1028         }
1029         vm_map_lookup_done(map, entry);
1030         if (obj != shmfd->shm_object)
1031                 return (EINVAL);
1032         vm_map_remove(map, kva, kva + size);
1033         VM_OBJECT_WLOCK(obj);
1034         KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
1035         shmfd->shm_kmappings--;
1036         VM_OBJECT_WUNLOCK(obj);
1037         return (0);
1038 }
1039
1040 static int
1041 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
1042 {
1043         struct shmfd *shmfd;
1044
1045         kif->kf_type = KF_TYPE_SHM;
1046         shmfd = fp->f_data;
1047
1048         mtx_lock(&shm_timestamp_lock);
1049         kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode;    /* XXX */
1050         mtx_unlock(&shm_timestamp_lock);
1051         kif->kf_un.kf_file.kf_file_size = shmfd->shm_size;
1052         if (shmfd->shm_path != NULL) {
1053                 sx_slock(&shm_dict_lock);
1054                 if (shmfd->shm_path != NULL)
1055                         strlcpy(kif->kf_path, shmfd->shm_path,
1056                             sizeof(kif->kf_path));
1057                 sx_sunlock(&shm_dict_lock);
1058         }
1059         return (0);
1060 }