uipc_shm.c revision 228509
1175164Sjhb/*- 2225344Srwatson * Copyright (c) 2006, 2011 Robert N. M. Watson 3175164Sjhb * All rights reserved. 4175164Sjhb * 5175164Sjhb * Redistribution and use in source and binary forms, with or without 6175164Sjhb * modification, are permitted provided that the following conditions 7175164Sjhb * are met: 8175164Sjhb * 1. Redistributions of source code must retain the above copyright 9175164Sjhb * notice, this list of conditions and the following disclaimer. 10175164Sjhb * 2. Redistributions in binary form must reproduce the above copyright 11175164Sjhb * notice, this list of conditions and the following disclaimer in the 12175164Sjhb * documentation and/or other materials provided with the distribution. 13175164Sjhb * 14175164Sjhb * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15175164Sjhb * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16175164Sjhb * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17175164Sjhb * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18175164Sjhb * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19175164Sjhb * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20175164Sjhb * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21175164Sjhb * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22175164Sjhb * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23175164Sjhb * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24175164Sjhb * SUCH DAMAGE. 25175164Sjhb */ 26175164Sjhb 27175164Sjhb/* 28175164Sjhb * Support for shared swap-backed anonymous memory objects via 29175164Sjhb * shm_open(2) and shm_unlink(2). While most of the implementation is 30175164Sjhb * here, vm_mmap.c contains mapping logic changes. 31175164Sjhb * 32175164Sjhb * TODO: 33175164Sjhb * 34225344Srwatson * (1) Need to export data to a userland tool via a sysctl. Should ipcs(1) 35175164Sjhb * and ipcrm(1) be expanded or should new tools to manage both POSIX 36175164Sjhb * kernel semaphores and POSIX shared memory be written? 37175164Sjhb * 38225344Srwatson * (2) Add support for this file type to fstat(1). 39175164Sjhb * 40225344Srwatson * (3) Resource limits? Does this need its own resource limits or are the 41175164Sjhb * existing limits in mmap(2) sufficient? 42175164Sjhb * 43225344Srwatson * (4) Partial page truncation. vnode_pager_setsize() will zero any parts 44175164Sjhb * of a partially mapped page as a result of ftruncate(2)/truncate(2). 45175164Sjhb * We can do the same (with the same pmap evil), but do we need to 46175164Sjhb * worry about the bits on disk if the page is swapped out or will the 47175164Sjhb * swapper zero the parts of a page that are invalid if the page is 48175164Sjhb * swapped back in for us? 49175164Sjhb */ 50175164Sjhb 51175164Sjhb#include <sys/cdefs.h> 52175164Sjhb__FBSDID("$FreeBSD: head/sys/kern/uipc_shm.c 228509 2011-12-14 22:22:19Z jhb $"); 53175164Sjhb 54223692Sjonathan#include "opt_capsicum.h" 55223692Sjonathan 56175164Sjhb#include <sys/param.h> 57223692Sjonathan#include <sys/capability.h> 58175164Sjhb#include <sys/fcntl.h> 59175164Sjhb#include <sys/file.h> 60175164Sjhb#include <sys/filedesc.h> 61175164Sjhb#include <sys/fnv_hash.h> 62175164Sjhb#include <sys/kernel.h> 63175164Sjhb#include <sys/lock.h> 64175164Sjhb#include <sys/malloc.h> 65175164Sjhb#include <sys/mman.h> 66175164Sjhb#include <sys/mutex.h> 67224914Skib#include <sys/priv.h> 68175164Sjhb#include <sys/proc.h> 69175164Sjhb#include <sys/refcount.h> 70175164Sjhb#include <sys/resourcevar.h> 71175164Sjhb#include <sys/stat.h> 72175164Sjhb#include <sys/sysctl.h> 73175164Sjhb#include <sys/sysproto.h> 74175164Sjhb#include <sys/systm.h> 75175164Sjhb#include <sys/sx.h> 76175164Sjhb#include <sys/time.h> 77175164Sjhb#include <sys/vnode.h> 78175164Sjhb 79175164Sjhb#include <security/mac/mac_framework.h> 80175164Sjhb 81175164Sjhb#include <vm/vm.h> 82175164Sjhb#include <vm/vm_param.h> 83175164Sjhb#include <vm/pmap.h> 84175164Sjhb#include <vm/vm_map.h> 85228509Sjhb#include <vm/vm_kern.h> 86175164Sjhb#include <vm/vm_object.h> 87175164Sjhb#include <vm/vm_page.h> 88175164Sjhb#include <vm/vm_pager.h> 89175164Sjhb#include <vm/swap_pager.h> 90175164Sjhb 91175164Sjhbstruct shm_mapping { 92175164Sjhb char *sm_path; 93175164Sjhb Fnv32_t sm_fnv; 94175164Sjhb struct shmfd *sm_shmfd; 95175164Sjhb LIST_ENTRY(shm_mapping) sm_link; 96175164Sjhb}; 97175164Sjhb 98175164Sjhbstatic MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor"); 99175164Sjhbstatic LIST_HEAD(, shm_mapping) *shm_dictionary; 100175164Sjhbstatic struct sx shm_dict_lock; 101175164Sjhbstatic struct mtx shm_timestamp_lock; 102175164Sjhbstatic u_long shm_hash; 103175164Sjhb 104175164Sjhb#define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash]) 105175164Sjhb 106175164Sjhbstatic int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags); 107175164Sjhbstatic struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode); 108175164Sjhbstatic void shm_dict_init(void *arg); 109175164Sjhbstatic void shm_drop(struct shmfd *shmfd); 110175164Sjhbstatic struct shmfd *shm_hold(struct shmfd *shmfd); 111175164Sjhbstatic void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd); 112175164Sjhbstatic struct shmfd *shm_lookup(char *path, Fnv32_t fnv); 113175164Sjhbstatic int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred); 114194766Skibstatic int shm_dotruncate(struct shmfd *shmfd, off_t length); 115175164Sjhb 116175164Sjhbstatic fo_rdwr_t shm_read; 117175164Sjhbstatic fo_rdwr_t shm_write; 118175164Sjhbstatic fo_truncate_t shm_truncate; 119175164Sjhbstatic fo_ioctl_t shm_ioctl; 120175164Sjhbstatic fo_poll_t shm_poll; 121175164Sjhbstatic fo_kqfilter_t shm_kqfilter; 122175164Sjhbstatic fo_stat_t shm_stat; 123175164Sjhbstatic fo_close_t shm_close; 124224914Skibstatic fo_chmod_t shm_chmod; 125224914Skibstatic fo_chown_t shm_chown; 126175164Sjhb 127175164Sjhb/* File descriptor operations. */ 128175164Sjhbstatic struct fileops shm_ops = { 129175164Sjhb .fo_read = shm_read, 130175164Sjhb .fo_write = shm_write, 131175164Sjhb .fo_truncate = shm_truncate, 132175164Sjhb .fo_ioctl = shm_ioctl, 133175164Sjhb .fo_poll = shm_poll, 134175164Sjhb .fo_kqfilter = shm_kqfilter, 135175164Sjhb .fo_stat = shm_stat, 136175164Sjhb .fo_close = shm_close, 137224914Skib .fo_chmod = shm_chmod, 138224914Skib .fo_chown = shm_chown, 139175164Sjhb .fo_flags = DFLAG_PASSABLE 140175164Sjhb}; 141175164Sjhb 142175164SjhbFEATURE(posix_shm, "POSIX shared memory"); 143175164Sjhb 144175164Sjhbstatic int 145175164Sjhbshm_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 146175164Sjhb int flags, struct thread *td) 147175164Sjhb{ 148175164Sjhb 149175164Sjhb return (EOPNOTSUPP); 150175164Sjhb} 151175164Sjhb 152175164Sjhbstatic int 153175164Sjhbshm_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 154175164Sjhb int flags, struct thread *td) 155175164Sjhb{ 156175164Sjhb 157175164Sjhb return (EOPNOTSUPP); 158175164Sjhb} 159175164Sjhb 160175164Sjhbstatic int 161175164Sjhbshm_truncate(struct file *fp, off_t length, struct ucred *active_cred, 162175164Sjhb struct thread *td) 163175164Sjhb{ 164175164Sjhb struct shmfd *shmfd; 165175164Sjhb#ifdef MAC 166175164Sjhb int error; 167175164Sjhb#endif 168175164Sjhb 169175164Sjhb shmfd = fp->f_data; 170175164Sjhb#ifdef MAC 171175164Sjhb error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd); 172175164Sjhb if (error) 173175164Sjhb return (error); 174175164Sjhb#endif 175194766Skib return (shm_dotruncate(shmfd, length)); 176175164Sjhb} 177175164Sjhb 178175164Sjhbstatic int 179175164Sjhbshm_ioctl(struct file *fp, u_long com, void *data, 180175164Sjhb struct ucred *active_cred, struct thread *td) 181175164Sjhb{ 182175164Sjhb 183175164Sjhb return (EOPNOTSUPP); 184175164Sjhb} 185175164Sjhb 186175164Sjhbstatic int 187175164Sjhbshm_poll(struct file *fp, int events, struct ucred *active_cred, 188175164Sjhb struct thread *td) 189175164Sjhb{ 190175164Sjhb 191175164Sjhb return (EOPNOTSUPP); 192175164Sjhb} 193175164Sjhb 194175164Sjhbstatic int 195175164Sjhbshm_kqfilter(struct file *fp, struct knote *kn) 196175164Sjhb{ 197175164Sjhb 198175164Sjhb return (EOPNOTSUPP); 199175164Sjhb} 200175164Sjhb 201175164Sjhbstatic int 202175164Sjhbshm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 203175164Sjhb struct thread *td) 204175164Sjhb{ 205175164Sjhb struct shmfd *shmfd; 206175164Sjhb#ifdef MAC 207175164Sjhb int error; 208175164Sjhb#endif 209175164Sjhb 210175164Sjhb shmfd = fp->f_data; 211175164Sjhb 212175164Sjhb#ifdef MAC 213175164Sjhb error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd); 214175164Sjhb if (error) 215175164Sjhb return (error); 216175164Sjhb#endif 217175164Sjhb 218175164Sjhb /* 219175164Sjhb * Attempt to return sanish values for fstat() on a memory file 220175164Sjhb * descriptor. 221175164Sjhb */ 222175164Sjhb bzero(sb, sizeof(*sb)); 223175164Sjhb sb->st_blksize = PAGE_SIZE; 224175164Sjhb sb->st_size = shmfd->shm_size; 225175164Sjhb sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize; 226224914Skib mtx_lock(&shm_timestamp_lock); 227205792Sed sb->st_atim = shmfd->shm_atime; 228205792Sed sb->st_ctim = shmfd->shm_ctime; 229205792Sed sb->st_mtim = shmfd->shm_mtime; 230224914Skib sb->st_birthtim = shmfd->shm_birthtime; 231224914Skib sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 232175164Sjhb sb->st_uid = shmfd->shm_uid; 233175164Sjhb sb->st_gid = shmfd->shm_gid; 234224914Skib mtx_unlock(&shm_timestamp_lock); 235175164Sjhb 236175164Sjhb return (0); 237175164Sjhb} 238175164Sjhb 239175164Sjhbstatic int 240175164Sjhbshm_close(struct file *fp, struct thread *td) 241175164Sjhb{ 242175164Sjhb struct shmfd *shmfd; 243175164Sjhb 244175164Sjhb shmfd = fp->f_data; 245175164Sjhb fp->f_data = NULL; 246175164Sjhb shm_drop(shmfd); 247175164Sjhb 248175164Sjhb return (0); 249175164Sjhb} 250175164Sjhb 251194766Skibstatic int 252175164Sjhbshm_dotruncate(struct shmfd *shmfd, off_t length) 253175164Sjhb{ 254175164Sjhb vm_object_t object; 255175164Sjhb vm_page_t m; 256175164Sjhb vm_pindex_t nobjsize; 257194766Skib vm_ooffset_t delta; 258175164Sjhb 259175164Sjhb object = shmfd->shm_object; 260175164Sjhb VM_OBJECT_LOCK(object); 261175164Sjhb if (length == shmfd->shm_size) { 262175164Sjhb VM_OBJECT_UNLOCK(object); 263194766Skib return (0); 264175164Sjhb } 265175164Sjhb nobjsize = OFF_TO_IDX(length + PAGE_MASK); 266175164Sjhb 267175164Sjhb /* Are we shrinking? If so, trim the end. */ 268175164Sjhb if (length < shmfd->shm_size) { 269228509Sjhb /* 270228509Sjhb * Disallow any requests to shrink the size if this 271228509Sjhb * object is mapped into the kernel. 272228509Sjhb */ 273228509Sjhb if (shmfd->shm_kmappings > 0) { 274228509Sjhb VM_OBJECT_UNLOCK(object); 275228509Sjhb return (EBUSY); 276228509Sjhb } 277194766Skib delta = ptoa(object->size - nobjsize); 278194766Skib 279175164Sjhb /* Toss in memory pages. */ 280175164Sjhb if (nobjsize < object->size) 281175164Sjhb vm_object_page_remove(object, nobjsize, object->size, 282223677Salc 0); 283175164Sjhb 284175164Sjhb /* Toss pages from swap. */ 285175164Sjhb if (object->type == OBJT_SWAP) 286194766Skib swap_pager_freespace(object, nobjsize, delta); 287175164Sjhb 288194766Skib /* Free the swap accounted for shm */ 289216128Strasz swap_release_by_cred(delta, object->cred); 290194766Skib object->charge -= delta; 291194766Skib 292175164Sjhb /* 293175164Sjhb * If the last page is partially mapped, then zero out 294175164Sjhb * the garbage at the end of the page. See comments 295193303Salc * in vnode_pager_setsize() for more details. 296175164Sjhb * 297175164Sjhb * XXXJHB: This handles in memory pages, but what about 298175164Sjhb * a page swapped out to disk? 299175164Sjhb */ 300175164Sjhb if ((length & PAGE_MASK) && 301175164Sjhb (m = vm_page_lookup(object, OFF_TO_IDX(length))) != NULL && 302175164Sjhb m->valid != 0) { 303175164Sjhb int base = (int)length & PAGE_MASK; 304175164Sjhb int size = PAGE_SIZE - base; 305175164Sjhb 306175164Sjhb pmap_zero_page_area(m, base, size); 307193303Salc 308193303Salc /* 309193303Salc * Update the valid bits to reflect the blocks that 310193303Salc * have been zeroed. Some of these valid bits may 311193303Salc * have already been set. 312193303Salc */ 313228156Skib vm_page_set_valid_range(m, base, size); 314193303Salc 315193303Salc /* 316193303Salc * Round "base" to the next block boundary so that the 317193303Salc * dirty bit for a partially zeroed block is not 318193303Salc * cleared. 319193303Salc */ 320193303Salc base = roundup2(base, DEV_BSIZE); 321193303Salc 322193303Salc vm_page_clear_dirty(m, base, PAGE_SIZE - base); 323176075Salc } else if ((length & PAGE_MASK) && 324176075Salc __predict_false(object->cache != NULL)) { 325176075Salc vm_page_cache_free(object, OFF_TO_IDX(length), 326176075Salc nobjsize); 327175164Sjhb } 328194766Skib } else { 329194766Skib 330194766Skib /* Attempt to reserve the swap */ 331194766Skib delta = ptoa(nobjsize - object->size); 332216128Strasz if (!swap_reserve_by_cred(delta, object->cred)) { 333194766Skib VM_OBJECT_UNLOCK(object); 334194766Skib return (ENOMEM); 335194766Skib } 336194766Skib object->charge += delta; 337175164Sjhb } 338175164Sjhb shmfd->shm_size = length; 339175164Sjhb mtx_lock(&shm_timestamp_lock); 340175164Sjhb vfs_timestamp(&shmfd->shm_ctime); 341175164Sjhb shmfd->shm_mtime = shmfd->shm_ctime; 342175164Sjhb mtx_unlock(&shm_timestamp_lock); 343175164Sjhb object->size = nobjsize; 344175164Sjhb VM_OBJECT_UNLOCK(object); 345194766Skib return (0); 346175164Sjhb} 347175164Sjhb 348175164Sjhb/* 349175164Sjhb * shmfd object management including creation and reference counting 350175164Sjhb * routines. 351175164Sjhb */ 352175164Sjhbstatic struct shmfd * 353175164Sjhbshm_alloc(struct ucred *ucred, mode_t mode) 354175164Sjhb{ 355175164Sjhb struct shmfd *shmfd; 356175164Sjhb 357175164Sjhb shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO); 358175164Sjhb shmfd->shm_size = 0; 359175164Sjhb shmfd->shm_uid = ucred->cr_uid; 360175164Sjhb shmfd->shm_gid = ucred->cr_gid; 361175164Sjhb shmfd->shm_mode = mode; 362175164Sjhb shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL, 363194766Skib shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred); 364175164Sjhb KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate")); 365178181Salc VM_OBJECT_LOCK(shmfd->shm_object); 366178181Salc vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING); 367178181Salc vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT); 368178181Salc VM_OBJECT_UNLOCK(shmfd->shm_object); 369175164Sjhb vfs_timestamp(&shmfd->shm_birthtime); 370175164Sjhb shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime = 371175164Sjhb shmfd->shm_birthtime; 372175164Sjhb refcount_init(&shmfd->shm_refs, 1); 373175164Sjhb#ifdef MAC 374175164Sjhb mac_posixshm_init(shmfd); 375175164Sjhb mac_posixshm_create(ucred, shmfd); 376175164Sjhb#endif 377175164Sjhb 378175164Sjhb return (shmfd); 379175164Sjhb} 380175164Sjhb 381175164Sjhbstatic struct shmfd * 382175164Sjhbshm_hold(struct shmfd *shmfd) 383175164Sjhb{ 384175164Sjhb 385175164Sjhb refcount_acquire(&shmfd->shm_refs); 386175164Sjhb return (shmfd); 387175164Sjhb} 388175164Sjhb 389175164Sjhbstatic void 390175164Sjhbshm_drop(struct shmfd *shmfd) 391175164Sjhb{ 392175164Sjhb 393175164Sjhb if (refcount_release(&shmfd->shm_refs)) { 394175164Sjhb#ifdef MAC 395175164Sjhb mac_posixshm_destroy(shmfd); 396175164Sjhb#endif 397175164Sjhb vm_object_deallocate(shmfd->shm_object); 398175164Sjhb free(shmfd, M_SHMFD); 399175164Sjhb } 400175164Sjhb} 401175164Sjhb 402175164Sjhb/* 403175164Sjhb * Determine if the credentials have sufficient permissions for a 404175164Sjhb * specified combination of FREAD and FWRITE. 405175164Sjhb */ 406175164Sjhbstatic int 407175164Sjhbshm_access(struct shmfd *shmfd, struct ucred *ucred, int flags) 408175164Sjhb{ 409184413Strasz accmode_t accmode; 410224914Skib int error; 411175164Sjhb 412184413Strasz accmode = 0; 413175164Sjhb if (flags & FREAD) 414184413Strasz accmode |= VREAD; 415175164Sjhb if (flags & FWRITE) 416184413Strasz accmode |= VWRITE; 417224914Skib mtx_lock(&shm_timestamp_lock); 418224914Skib error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid, 419224914Skib accmode, ucred, NULL); 420224914Skib mtx_unlock(&shm_timestamp_lock); 421224914Skib return (error); 422175164Sjhb} 423175164Sjhb 424175164Sjhb/* 425175164Sjhb * Dictionary management. We maintain an in-kernel dictionary to map 426175164Sjhb * paths to shmfd objects. We use the FNV hash on the path to store 427175164Sjhb * the mappings in a hash table. 428175164Sjhb */ 429175164Sjhbstatic void 430175164Sjhbshm_dict_init(void *arg) 431175164Sjhb{ 432175164Sjhb 433175164Sjhb mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF); 434175164Sjhb sx_init(&shm_dict_lock, "shm dictionary"); 435175164Sjhb shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash); 436175164Sjhb} 437175164SjhbSYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL); 438175164Sjhb 439175164Sjhbstatic struct shmfd * 440175164Sjhbshm_lookup(char *path, Fnv32_t fnv) 441175164Sjhb{ 442175164Sjhb struct shm_mapping *map; 443175164Sjhb 444175164Sjhb LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 445175164Sjhb if (map->sm_fnv != fnv) 446175164Sjhb continue; 447175164Sjhb if (strcmp(map->sm_path, path) == 0) 448175164Sjhb return (map->sm_shmfd); 449175164Sjhb } 450175164Sjhb 451175164Sjhb return (NULL); 452175164Sjhb} 453175164Sjhb 454175164Sjhbstatic void 455175164Sjhbshm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd) 456175164Sjhb{ 457175164Sjhb struct shm_mapping *map; 458175164Sjhb 459175164Sjhb map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK); 460175164Sjhb map->sm_path = path; 461175164Sjhb map->sm_fnv = fnv; 462175164Sjhb map->sm_shmfd = shm_hold(shmfd); 463175164Sjhb LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link); 464175164Sjhb} 465175164Sjhb 466175164Sjhbstatic int 467175164Sjhbshm_remove(char *path, Fnv32_t fnv, struct ucred *ucred) 468175164Sjhb{ 469175164Sjhb struct shm_mapping *map; 470175164Sjhb int error; 471175164Sjhb 472175164Sjhb LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 473175164Sjhb if (map->sm_fnv != fnv) 474175164Sjhb continue; 475175164Sjhb if (strcmp(map->sm_path, path) == 0) { 476175164Sjhb#ifdef MAC 477175164Sjhb error = mac_posixshm_check_unlink(ucred, map->sm_shmfd); 478175164Sjhb if (error) 479175164Sjhb return (error); 480175164Sjhb#endif 481175164Sjhb error = shm_access(map->sm_shmfd, ucred, 482175164Sjhb FREAD | FWRITE); 483175164Sjhb if (error) 484175164Sjhb return (error); 485175164Sjhb LIST_REMOVE(map, sm_link); 486175164Sjhb shm_drop(map->sm_shmfd); 487175164Sjhb free(map->sm_path, M_SHMFD); 488175164Sjhb free(map, M_SHMFD); 489175164Sjhb return (0); 490175164Sjhb } 491175164Sjhb } 492175164Sjhb 493175164Sjhb return (ENOENT); 494175164Sjhb} 495175164Sjhb 496175164Sjhb/* System calls. */ 497175164Sjhbint 498225617Skmacysys_shm_open(struct thread *td, struct shm_open_args *uap) 499175164Sjhb{ 500175164Sjhb struct filedesc *fdp; 501175164Sjhb struct shmfd *shmfd; 502175164Sjhb struct file *fp; 503175164Sjhb char *path; 504175164Sjhb Fnv32_t fnv; 505175164Sjhb mode_t cmode; 506175164Sjhb int fd, error; 507175164Sjhb 508223692Sjonathan#ifdef CAPABILITY_MODE 509223692Sjonathan /* 510223692Sjonathan * shm_open(2) is only allowed for anonymous objects. 511223692Sjonathan */ 512223692Sjonathan if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON)) 513223692Sjonathan return (ECAPMODE); 514223692Sjonathan#endif 515223692Sjonathan 516175164Sjhb if ((uap->flags & O_ACCMODE) != O_RDONLY && 517175164Sjhb (uap->flags & O_ACCMODE) != O_RDWR) 518175164Sjhb return (EINVAL); 519175164Sjhb 520175164Sjhb if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0) 521175164Sjhb return (EINVAL); 522175164Sjhb 523175164Sjhb fdp = td->td_proc->p_fd; 524175164Sjhb cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS; 525175164Sjhb 526220245Skib error = falloc(td, &fp, &fd, 0); 527175164Sjhb if (error) 528175164Sjhb return (error); 529175164Sjhb 530175164Sjhb /* A SHM_ANON path pointer creates an anonymous object. */ 531175164Sjhb if (uap->path == SHM_ANON) { 532175164Sjhb /* A read-only anonymous object is pointless. */ 533175164Sjhb if ((uap->flags & O_ACCMODE) == O_RDONLY) { 534175164Sjhb fdclose(fdp, fp, fd, td); 535175164Sjhb fdrop(fp, td); 536175164Sjhb return (EINVAL); 537175164Sjhb } 538175164Sjhb shmfd = shm_alloc(td->td_ucred, cmode); 539175164Sjhb } else { 540175164Sjhb path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK); 541175164Sjhb error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 542175164Sjhb 543175164Sjhb /* Require paths to start with a '/' character. */ 544175164Sjhb if (error == 0 && path[0] != '/') 545175164Sjhb error = EINVAL; 546175164Sjhb if (error) { 547175164Sjhb fdclose(fdp, fp, fd, td); 548175164Sjhb fdrop(fp, td); 549175164Sjhb free(path, M_SHMFD); 550175164Sjhb return (error); 551175164Sjhb } 552175164Sjhb 553175164Sjhb fnv = fnv_32_str(path, FNV1_32_INIT); 554175164Sjhb sx_xlock(&shm_dict_lock); 555175164Sjhb shmfd = shm_lookup(path, fnv); 556175164Sjhb if (shmfd == NULL) { 557175164Sjhb /* Object does not yet exist, create it if requested. */ 558175164Sjhb if (uap->flags & O_CREAT) { 559225344Srwatson#ifdef MAC 560225344Srwatson error = mac_posixshm_check_create(td->td_ucred, 561225344Srwatson path); 562225344Srwatson if (error == 0) { 563225344Srwatson#endif 564225344Srwatson shmfd = shm_alloc(td->td_ucred, cmode); 565225344Srwatson shm_insert(path, fnv, shmfd); 566225344Srwatson#ifdef MAC 567225344Srwatson } 568225344Srwatson#endif 569175164Sjhb } else { 570175164Sjhb free(path, M_SHMFD); 571175164Sjhb error = ENOENT; 572175164Sjhb } 573175164Sjhb } else { 574175164Sjhb /* 575175164Sjhb * Object already exists, obtain a new 576175164Sjhb * reference if requested and permitted. 577175164Sjhb */ 578175164Sjhb free(path, M_SHMFD); 579175164Sjhb if ((uap->flags & (O_CREAT | O_EXCL)) == 580175164Sjhb (O_CREAT | O_EXCL)) 581175164Sjhb error = EEXIST; 582175164Sjhb else { 583175164Sjhb#ifdef MAC 584175164Sjhb error = mac_posixshm_check_open(td->td_ucred, 585225344Srwatson shmfd, FFLAGS(uap->flags & O_ACCMODE)); 586175164Sjhb if (error == 0) 587175164Sjhb#endif 588175164Sjhb error = shm_access(shmfd, td->td_ucred, 589175164Sjhb FFLAGS(uap->flags & O_ACCMODE)); 590175164Sjhb } 591175164Sjhb 592175164Sjhb /* 593175164Sjhb * Truncate the file back to zero length if 594175164Sjhb * O_TRUNC was specified and the object was 595175164Sjhb * opened with read/write. 596175164Sjhb */ 597175164Sjhb if (error == 0 && 598175164Sjhb (uap->flags & (O_ACCMODE | O_TRUNC)) == 599175164Sjhb (O_RDWR | O_TRUNC)) { 600175164Sjhb#ifdef MAC 601175164Sjhb error = mac_posixshm_check_truncate( 602175164Sjhb td->td_ucred, fp->f_cred, shmfd); 603175164Sjhb if (error == 0) 604175164Sjhb#endif 605175164Sjhb shm_dotruncate(shmfd, 0); 606175164Sjhb } 607175164Sjhb if (error == 0) 608175164Sjhb shm_hold(shmfd); 609175164Sjhb } 610175164Sjhb sx_xunlock(&shm_dict_lock); 611175164Sjhb 612175164Sjhb if (error) { 613175164Sjhb fdclose(fdp, fp, fd, td); 614175164Sjhb fdrop(fp, td); 615175164Sjhb return (error); 616175164Sjhb } 617175164Sjhb } 618175164Sjhb 619175164Sjhb finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops); 620175164Sjhb 621175164Sjhb FILEDESC_XLOCK(fdp); 622175164Sjhb if (fdp->fd_ofiles[fd] == fp) 623175164Sjhb fdp->fd_ofileflags[fd] |= UF_EXCLOSE; 624175164Sjhb FILEDESC_XUNLOCK(fdp); 625175164Sjhb td->td_retval[0] = fd; 626175164Sjhb fdrop(fp, td); 627175164Sjhb 628175164Sjhb return (0); 629175164Sjhb} 630175164Sjhb 631175164Sjhbint 632225617Skmacysys_shm_unlink(struct thread *td, struct shm_unlink_args *uap) 633175164Sjhb{ 634175164Sjhb char *path; 635175164Sjhb Fnv32_t fnv; 636175164Sjhb int error; 637175164Sjhb 638175164Sjhb path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 639175164Sjhb error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 640175164Sjhb if (error) { 641175164Sjhb free(path, M_TEMP); 642175164Sjhb return (error); 643175164Sjhb } 644175164Sjhb 645175164Sjhb fnv = fnv_32_str(path, FNV1_32_INIT); 646175164Sjhb sx_xlock(&shm_dict_lock); 647175164Sjhb error = shm_remove(path, fnv, td->td_ucred); 648175164Sjhb sx_xunlock(&shm_dict_lock); 649175164Sjhb free(path, M_TEMP); 650175164Sjhb 651175164Sjhb return (error); 652175164Sjhb} 653175164Sjhb 654175164Sjhb/* 655175164Sjhb * mmap() helper to validate mmap() requests against shm object state 656175164Sjhb * and give mmap() the vm_object to use for the mapping. 657175164Sjhb */ 658175164Sjhbint 659175164Sjhbshm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff, 660175164Sjhb vm_object_t *obj) 661175164Sjhb{ 662175164Sjhb 663175164Sjhb /* 664175164Sjhb * XXXRW: This validation is probably insufficient, and subject to 665175164Sjhb * sign errors. It should be fixed. 666175164Sjhb */ 667185533Skan if (foff >= shmfd->shm_size || 668185533Skan foff + objsize > round_page(shmfd->shm_size)) 669175164Sjhb return (EINVAL); 670175164Sjhb 671175164Sjhb mtx_lock(&shm_timestamp_lock); 672175164Sjhb vfs_timestamp(&shmfd->shm_atime); 673175164Sjhb mtx_unlock(&shm_timestamp_lock); 674175164Sjhb vm_object_reference(shmfd->shm_object); 675175164Sjhb *obj = shmfd->shm_object; 676175164Sjhb return (0); 677175164Sjhb} 678224914Skib 679224914Skibstatic int 680224914Skibshm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 681224914Skib struct thread *td) 682224914Skib{ 683224914Skib struct shmfd *shmfd; 684224914Skib int error; 685224914Skib 686224914Skib error = 0; 687224914Skib shmfd = fp->f_data; 688224914Skib mtx_lock(&shm_timestamp_lock); 689224914Skib /* 690224914Skib * SUSv4 says that x bits of permission need not be affected. 691224914Skib * Be consistent with our shm_open there. 692224914Skib */ 693224914Skib#ifdef MAC 694224914Skib error = mac_posixshm_check_setmode(active_cred, shmfd, mode); 695224914Skib if (error != 0) 696224914Skib goto out; 697224914Skib#endif 698224914Skib error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, 699224914Skib shmfd->shm_gid, VADMIN, active_cred, NULL); 700224914Skib if (error != 0) 701224914Skib goto out; 702224914Skib shmfd->shm_mode = mode & ACCESSPERMS; 703224914Skibout: 704224914Skib mtx_unlock(&shm_timestamp_lock); 705224914Skib return (error); 706224914Skib} 707224914Skib 708224914Skibstatic int 709224914Skibshm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 710224914Skib struct thread *td) 711224914Skib{ 712224914Skib struct shmfd *shmfd; 713224914Skib int error; 714224914Skib 715224935Skib error = 0; 716224914Skib shmfd = fp->f_data; 717224914Skib mtx_lock(&shm_timestamp_lock); 718224914Skib#ifdef MAC 719224914Skib error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid); 720224914Skib if (error != 0) 721224914Skib goto out; 722224914Skib#endif 723224914Skib if (uid == (uid_t)-1) 724224914Skib uid = shmfd->shm_uid; 725224914Skib if (gid == (gid_t)-1) 726224914Skib gid = shmfd->shm_gid; 727224914Skib if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) || 728224914Skib (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) && 729224914Skib (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0))) 730224914Skib goto out; 731224914Skib shmfd->shm_uid = uid; 732224914Skib shmfd->shm_gid = gid; 733224914Skibout: 734224914Skib mtx_unlock(&shm_timestamp_lock); 735224914Skib return (error); 736224914Skib} 737228509Sjhb 738228509Sjhb/* 739228509Sjhb * Helper routines to allow the backing object of a shared memory file 740228509Sjhb * descriptor to be mapped in the kernel. 741228509Sjhb */ 742228509Sjhbint 743228509Sjhbshm_map(struct file *fp, size_t size, off_t offset, void **memp) 744228509Sjhb{ 745228509Sjhb struct shmfd *shmfd; 746228509Sjhb vm_offset_t kva, ofs; 747228509Sjhb vm_object_t obj; 748228509Sjhb int rv; 749228509Sjhb 750228509Sjhb if (fp->f_type != DTYPE_SHM) 751228509Sjhb return (EINVAL); 752228509Sjhb shmfd = fp->f_data; 753228509Sjhb obj = shmfd->shm_object; 754228509Sjhb VM_OBJECT_LOCK(obj); 755228509Sjhb /* 756228509Sjhb * XXXRW: This validation is probably insufficient, and subject to 757228509Sjhb * sign errors. It should be fixed. 758228509Sjhb */ 759228509Sjhb if (offset >= shmfd->shm_size || 760228509Sjhb offset + size > round_page(shmfd->shm_size)) { 761228509Sjhb VM_OBJECT_UNLOCK(obj); 762228509Sjhb return (EINVAL); 763228509Sjhb } 764228509Sjhb 765228509Sjhb shmfd->shm_kmappings++; 766228509Sjhb vm_object_reference_locked(obj); 767228509Sjhb VM_OBJECT_UNLOCK(obj); 768228509Sjhb 769228509Sjhb /* Map the object into the kernel_map and wire it. */ 770228509Sjhb kva = vm_map_min(kernel_map); 771228509Sjhb ofs = offset & PAGE_MASK; 772228509Sjhb offset = trunc_page(offset); 773228509Sjhb size = round_page(size + ofs); 774228509Sjhb rv = vm_map_find(kernel_map, obj, offset, &kva, size, 775228509Sjhb VMFS_ALIGNED_SPACE, VM_PROT_READ | VM_PROT_WRITE, 776228509Sjhb VM_PROT_READ | VM_PROT_WRITE, 0); 777228509Sjhb if (rv == KERN_SUCCESS) { 778228509Sjhb rv = vm_map_wire(kernel_map, kva, kva + size, 779228509Sjhb VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 780228509Sjhb if (rv == KERN_SUCCESS) { 781228509Sjhb *memp = (void *)(kva + ofs); 782228509Sjhb return (0); 783228509Sjhb } 784228509Sjhb vm_map_remove(kernel_map, kva, kva + size); 785228509Sjhb } else 786228509Sjhb vm_object_deallocate(obj); 787228509Sjhb 788228509Sjhb /* On failure, drop our mapping reference. */ 789228509Sjhb VM_OBJECT_LOCK(obj); 790228509Sjhb shmfd->shm_kmappings--; 791228509Sjhb VM_OBJECT_UNLOCK(obj); 792228509Sjhb 793228509Sjhb switch (rv) { 794228509Sjhb case KERN_INVALID_ADDRESS: 795228509Sjhb case KERN_NO_SPACE: 796228509Sjhb return (ENOMEM); 797228509Sjhb case KERN_PROTECTION_FAILURE: 798228509Sjhb return (EACCES); 799228509Sjhb default: 800228509Sjhb return (EINVAL); 801228509Sjhb } 802228509Sjhb} 803228509Sjhb 804228509Sjhb/* 805228509Sjhb * We require the caller to unmap the entire entry. This allows us to 806228509Sjhb * safely decrement shm_kmappings when a mapping is removed. 807228509Sjhb */ 808228509Sjhbint 809228509Sjhbshm_unmap(struct file *fp, void *mem, size_t size) 810228509Sjhb{ 811228509Sjhb struct shmfd *shmfd; 812228509Sjhb vm_map_entry_t entry; 813228509Sjhb vm_offset_t kva, ofs; 814228509Sjhb vm_object_t obj; 815228509Sjhb vm_pindex_t pindex; 816228509Sjhb vm_prot_t prot; 817228509Sjhb boolean_t wired; 818228509Sjhb vm_map_t map; 819228509Sjhb int rv; 820228509Sjhb 821228509Sjhb if (fp->f_type != DTYPE_SHM) 822228509Sjhb return (EINVAL); 823228509Sjhb shmfd = fp->f_data; 824228509Sjhb kva = (vm_offset_t)mem; 825228509Sjhb ofs = kva & PAGE_MASK; 826228509Sjhb kva = trunc_page(kva); 827228509Sjhb size = round_page(size + ofs); 828228509Sjhb map = kernel_map; 829228509Sjhb rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry, 830228509Sjhb &obj, &pindex, &prot, &wired); 831228509Sjhb if (rv != KERN_SUCCESS) 832228509Sjhb return (EINVAL); 833228509Sjhb if (entry->start != kva || entry->end != kva + size) { 834228509Sjhb vm_map_lookup_done(map, entry); 835228509Sjhb return (EINVAL); 836228509Sjhb } 837228509Sjhb vm_map_lookup_done(map, entry); 838228509Sjhb if (obj != shmfd->shm_object) 839228509Sjhb return (EINVAL); 840228509Sjhb vm_map_remove(map, kva, kva + size); 841228509Sjhb VM_OBJECT_LOCK(obj); 842228509Sjhb KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped")); 843228509Sjhb shmfd->shm_kmappings--; 844228509Sjhb VM_OBJECT_UNLOCK(obj); 845228509Sjhb return (0); 846228509Sjhb} 847