1 /* $NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $ */ 2 3 /*- 4 * Copyright (c) 2005 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 9 * 2005 program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Efficient memory file system. 35 * 36 * tmpfs is a file system that uses NetBSD's virtual memory sub-system 37 * (the well-known UVM) to store file data and metadata in an efficient 38 * way. This means that it does not follow the structure of an on-disk 39 * file system because it simply does not need to. Instead, it uses 40 * memory-specific data structures and algorithms to automatically 41 * allocate and release resources. 42 */ 43 44 #include <sys/conf.h> 45 #include <sys/param.h> 46 #include <sys/limits.h> 47 #include <sys/lock.h> 48 #include <sys/mutex.h> 49 #include <sys/kernel.h> 50 #include <sys/stat.h> 51 #include <sys/systm.h> 52 #include <sys/sysctl.h> 53 #include <sys/objcache.h> 54 55 #include <vm/vm.h> 56 #include <vm/vm_object.h> 57 #include <vm/vm_param.h> 58 59 #include <vfs/tmpfs/tmpfs.h> 60 #include <vfs/tmpfs/tmpfs_vnops.h> 61 #include <vfs/tmpfs/tmpfs_args.h> 62 63 /* 64 * Default permission for root node 65 */ 66 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) 67 68 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures"); 69 70 /* --------------------------------------------------------------------- */ 71 72 static int tmpfs_mount(struct mount *, char *, caddr_t, struct ucred *); 73 static int tmpfs_unmount(struct mount *, int); 74 static int tmpfs_root(struct mount *, struct vnode **); 75 static int tmpfs_fhtovp(struct mount *, struct vnode *, struct fid *, struct vnode **); 76 static int tmpfs_statfs(struct mount *, struct statfs *, struct ucred *cred); 77 78 /* --------------------------------------------------------------------- */ 79 int 80 tmpfs_node_ctor(void *obj, void *privdata, int flags) 81 { 82 struct tmpfs_node *node = (struct tmpfs_node *)obj; 83 84 node->tn_gen++; 85 node->tn_size = 0; 86 node->tn_status = 0; 87 node->tn_flags = 0; 88 node->tn_links = 0; 89 node->tn_vnode = NULL; 90 node->tn_vpstate = TMPFS_VNODE_WANT; 91 bzero(&node->tn_spec, sizeof(node->tn_spec)); 92 93 return (1); 94 } 95 96 static void 97 tmpfs_node_dtor(void *obj, void *privdata) 98 { 99 struct tmpfs_node *node = (struct tmpfs_node *)obj; 100 node->tn_type = VNON; 101 node->tn_vpstate = TMPFS_VNODE_DOOMED; 102 } 103 104 static void* 105 tmpfs_node_init(void *args, int flags) 106 { 107 struct tmpfs_node *node = (struct tmpfs_node *)objcache_malloc_alloc(args, flags); 108 if (node == NULL) 109 return (NULL); 110 node->tn_id = 0; 111 112 lockinit(&node->tn_interlock, "tmpfs node interlock", 0, LK_CANRECURSE); 113 node->tn_gen = karc4random(); 114 115 return node; 116 } 117 118 static void 119 tmpfs_node_fini(void *obj, void *args) 120 { 121 struct tmpfs_node *node = (struct tmpfs_node *)obj; 122 lockuninit(&node->tn_interlock); 123 objcache_malloc_free(obj, args); 124 } 125 126 static int 127 tmpfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred) 128 { 129 struct tmpfs_mount *tmp; 130 struct tmpfs_node *root; 131 struct tmpfs_args args; 132 vm_pindex_t pages; 133 vm_pindex_t pages_limit; 134 ino_t nodes; 135 u_int64_t maxfsize; 136 int error; 137 /* Size counters. */ 138 ino_t nodes_max; 139 off_t size_max; 140 size_t maxfsize_max; 141 size_t size; 142 143 /* Root node attributes. */ 144 uid_t root_uid = cred->cr_uid; 145 gid_t root_gid = cred->cr_gid; 146 mode_t root_mode = (VREAD | VWRITE); 147 148 if (mp->mnt_flag & MNT_UPDATE) { 149 /* XXX: There is no support yet to update file system 150 * settings. Should be added. */ 151 152 return EOPNOTSUPP; 153 } 154 155 /* 156 * mount info 157 */ 158 bzero(&args, sizeof(args)); 159 size_max = 0; 160 nodes_max = 0; 161 maxfsize_max = 0; 162 163 if (path) { 164 if (data) { 165 error = copyin(data, &args, sizeof(args)); 166 if (error) 167 return (error); 168 } 169 size_max = args.ta_size_max; 170 nodes_max = args.ta_nodes_max; 171 maxfsize_max = args.ta_maxfsize_max; 172 root_uid = args.ta_root_uid; 173 root_gid = args.ta_root_gid; 174 root_mode = args.ta_root_mode; 175 } 176 177 /* 178 * If mount by non-root, then verify that user has necessary 179 * permissions on the device. 180 */ 181 if (cred->cr_uid != 0) { 182 root_mode = VREAD; 183 if ((mp->mnt_flag & MNT_RDONLY) == 0) 184 root_mode |= VWRITE; 185 } 186 187 pages_limit = vm_swap_max + vmstats.v_page_count / 2; 188 189 if (size_max == 0) 190 pages = pages_limit / 2; 191 else if (size_max < PAGE_SIZE) 192 pages = 1; 193 else if (OFF_TO_IDX(size_max) > pages_limit) 194 pages = pages_limit; 195 else 196 pages = OFF_TO_IDX(size_max); 197 198 if (nodes_max == 0) 199 nodes = 3 + pages * PAGE_SIZE / 1024; 200 else if (nodes_max < 3) 201 nodes = 3; 202 else if (nodes_max > pages) 203 nodes = pages; 204 else 205 nodes = nodes_max; 206 207 maxfsize = IDX_TO_OFF(pages_limit); 208 if (maxfsize_max != 0 && maxfsize > maxfsize_max) 209 maxfsize = maxfsize_max; 210 211 /* Allocate the tmpfs mount structure and fill it. */ 212 tmp = kmalloc(sizeof(*tmp), M_TMPFSMNT, M_WAITOK | M_ZERO); 213 214 lockinit(&(tmp->allnode_lock), "tmpfs allnode lock", 0, LK_CANRECURSE); 215 tmp->tm_nodes_max = nodes; 216 tmp->tm_nodes_inuse = 0; 217 tmp->tm_maxfilesize = maxfsize; 218 LIST_INIT(&tmp->tm_nodes_used); 219 220 tmp->tm_pages_max = pages; 221 tmp->tm_pages_used = 0; 222 223 kmalloc_create(&tmp->tm_node_zone, "tmpfs node"); 224 kmalloc_create(&tmp->tm_dirent_zone, "tmpfs dirent"); 225 kmalloc_create(&tmp->tm_name_zone, "tmpfs name zone"); 226 227 kmalloc_raise_limit(tmp->tm_node_zone, sizeof(struct tmpfs_node) * 228 tmp->tm_nodes_max); 229 230 tmp->tm_node_zone_malloc_args.objsize = sizeof(struct tmpfs_node); 231 tmp->tm_node_zone_malloc_args.mtype = tmp->tm_node_zone; 232 233 tmp->tm_dirent_zone_malloc_args.objsize = sizeof(struct tmpfs_dirent); 234 tmp->tm_dirent_zone_malloc_args.mtype = tmp->tm_dirent_zone; 235 236 tmp->tm_dirent_pool = objcache_create( "tmpfs dirent cache", 237 0, 0, 238 NULL, NULL, NULL, 239 objcache_malloc_alloc, objcache_malloc_free, 240 &tmp->tm_dirent_zone_malloc_args); 241 tmp->tm_node_pool = objcache_create( "tmpfs node cache", 242 0, 0, 243 tmpfs_node_ctor, tmpfs_node_dtor, NULL, 244 tmpfs_node_init, tmpfs_node_fini, 245 &tmp->tm_node_zone_malloc_args); 246 247 /* Allocate the root node. */ 248 error = tmpfs_alloc_node(tmp, VDIR, root_uid, root_gid, 249 root_mode & ALLPERMS, NULL, NULL, 250 VNOVAL, VNOVAL, &root); 251 252 /* 253 * We are backed by swap, set snocache chflags flag so we 254 * don't trip over swapcache. 255 */ 256 root->tn_flags = SF_NOCACHE; 257 258 if (error != 0 || root == NULL) { 259 objcache_destroy(tmp->tm_node_pool); 260 objcache_destroy(tmp->tm_dirent_pool); 261 kfree(tmp, M_TMPFSMNT); 262 return error; 263 } 264 KASSERT(root->tn_id >= 0, ("tmpfs root with invalid ino: %d", (int)root->tn_id)); 265 tmp->tm_root = root; 266 267 mp->mnt_flag |= MNT_LOCAL; 268 #if 0 269 mp->mnt_kern_flag |= MNTK_RD_MPSAFE | MNTK_WR_MPSAFE | MNTK_GA_MPSAFE | 270 MNTK_IN_MPSAFE | MNTK_SG_MPSAFE; 271 #endif 272 mp->mnt_kern_flag |= MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_SG_MPSAFE; 273 mp->mnt_kern_flag |= MNTK_WR_MPSAFE; 274 mp->mnt_kern_flag |= MNTK_NOMSYNC; 275 mp->mnt_data = (qaddr_t)tmp; 276 vfs_getnewfsid(mp); 277 278 279 vfs_add_vnodeops(mp, &tmpfs_vnode_vops, &mp->mnt_vn_norm_ops); 280 vfs_add_vnodeops(mp, &tmpfs_fifo_vops, &mp->mnt_vn_fifo_ops); 281 282 copystr("tmpfs", mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size); 283 bzero(mp->mnt_stat.f_mntfromname +size, MNAMELEN - size); 284 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname)); 285 copyinstr(path, mp->mnt_stat.f_mntonname, 286 sizeof(mp->mnt_stat.f_mntonname) -1, 287 &size); 288 289 tmpfs_statfs(mp, &mp->mnt_stat, cred); 290 291 return 0; 292 } 293 294 /* --------------------------------------------------------------------- */ 295 296 /* ARGSUSED2 */ 297 static int 298 tmpfs_unmount(struct mount *mp, int mntflags) 299 { 300 int error; 301 int flags = 0; 302 int found; 303 struct tmpfs_mount *tmp; 304 struct tmpfs_node *node; 305 306 /* Handle forced unmounts. */ 307 if (mntflags & MNT_FORCE) 308 flags |= FORCECLOSE; 309 310 tmp = VFS_TO_TMPFS(mp); 311 312 /* 313 * Finalize all pending I/O. In the case of tmpfs we want 314 * to throw all the data away so clean out the buffer cache 315 * and vm objects before calling vflush(). 316 */ 317 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) { 318 if (node->tn_type == VREG && node->tn_vnode) { 319 ++node->tn_links; 320 TMPFS_NODE_LOCK(node); 321 vx_get(node->tn_vnode); 322 tmpfs_truncate(node->tn_vnode, 0); 323 vx_put(node->tn_vnode); 324 TMPFS_NODE_UNLOCK(node); 325 --node->tn_links; 326 } 327 } 328 error = vflush(mp, 0, flags); 329 if (error != 0) 330 return error; 331 332 /* 333 * First pass get rid of all the directory entries and 334 * vnode associations. The directory structure will 335 * remain via the extra link count representing tn_dir.tn_parent. 336 * 337 * No vnodes should remain after the vflush above. 338 */ 339 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) { 340 ++node->tn_links; 341 TMPFS_NODE_LOCK(node); 342 if (node->tn_type == VDIR) { 343 struct tmpfs_dirent *de; 344 345 while (!TAILQ_EMPTY(&node->tn_dir.tn_dirhead)) { 346 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead); 347 tmpfs_dir_detach(node, de); 348 tmpfs_free_dirent(tmp, de); 349 node->tn_size -= sizeof(struct tmpfs_dirent); 350 } 351 } 352 KKASSERT(node->tn_vnode == NULL); 353 #if 0 354 vp = node->tn_vnode; 355 if (vp != NULL) { 356 tmpfs_free_vp(vp); 357 vrecycle(vp); 358 node->tn_vnode = NULL; 359 } 360 #endif 361 TMPFS_NODE_UNLOCK(node); 362 --node->tn_links; 363 } 364 365 /* 366 * Now get rid of all nodes. We can remove any node with a 367 * link count of 0 or any directory node with a link count of 368 * 1. The parents will not be destroyed until all their children 369 * have been destroyed. 370 * 371 * Recursion in tmpfs_free_node() can further modify the list so 372 * we cannot use a next pointer here. 373 * 374 * The root node will be destroyed by this loop (it will be last). 375 */ 376 while (!LIST_EMPTY(&tmp->tm_nodes_used)) { 377 found = 0; 378 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) { 379 if (node->tn_links == 0 || 380 (node->tn_links == 1 && node->tn_type == VDIR)) { 381 TMPFS_NODE_LOCK(node); 382 tmpfs_free_node(tmp, node); 383 /* eats lock */ 384 found = 1; 385 break; 386 } 387 } 388 if (found == 0) { 389 kprintf("tmpfs: Cannot free entire node tree!"); 390 break; 391 } 392 } 393 394 KKASSERT(tmp->tm_root == NULL); 395 396 objcache_destroy(tmp->tm_dirent_pool); 397 objcache_destroy(tmp->tm_node_pool); 398 399 kmalloc_destroy(&tmp->tm_name_zone); 400 kmalloc_destroy(&tmp->tm_dirent_zone); 401 kmalloc_destroy(&tmp->tm_node_zone); 402 403 tmp->tm_node_zone = tmp->tm_dirent_zone = NULL; 404 405 lockuninit(&tmp->allnode_lock); 406 KKASSERT(tmp->tm_pages_used == 0); 407 KKASSERT(tmp->tm_nodes_inuse == 0); 408 409 /* Throw away the tmpfs_mount structure. */ 410 kfree(tmp, M_TMPFSMNT); 411 mp->mnt_data = NULL; 412 413 mp->mnt_flag &= ~MNT_LOCAL; 414 return 0; 415 } 416 417 /* --------------------------------------------------------------------- */ 418 419 static int 420 tmpfs_root(struct mount *mp, struct vnode **vpp) 421 { 422 struct tmpfs_mount *tmp; 423 int error; 424 425 tmp = VFS_TO_TMPFS(mp); 426 if (tmp->tm_root == NULL) { 427 kprintf("tmpfs_root: called without root node %p\n", mp); 428 print_backtrace(-1); 429 *vpp = NULL; 430 error = EINVAL; 431 } else { 432 error = tmpfs_alloc_vp(mp, tmp->tm_root, LK_EXCLUSIVE, vpp); 433 (*vpp)->v_flag |= VROOT; 434 (*vpp)->v_type = VDIR; 435 } 436 return error; 437 } 438 439 /* --------------------------------------------------------------------- */ 440 441 static int 442 tmpfs_fhtovp(struct mount *mp, struct vnode *rootvp, struct fid *fhp, struct vnode **vpp) 443 { 444 boolean_t found; 445 struct tmpfs_fid *tfhp; 446 struct tmpfs_mount *tmp; 447 struct tmpfs_node *node; 448 449 tmp = VFS_TO_TMPFS(mp); 450 451 tfhp = (struct tmpfs_fid *)fhp; 452 if (tfhp->tf_len != sizeof(struct tmpfs_fid)) 453 return EINVAL; 454 455 if (tfhp->tf_id >= tmp->tm_nodes_max) 456 return EINVAL; 457 458 found = FALSE; 459 460 TMPFS_LOCK(tmp); 461 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) { 462 if (node->tn_id == tfhp->tf_id && 463 node->tn_gen == tfhp->tf_gen) { 464 found = TRUE; 465 break; 466 } 467 } 468 TMPFS_UNLOCK(tmp); 469 470 if (found) 471 return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp)); 472 473 return (EINVAL); 474 } 475 476 /* --------------------------------------------------------------------- */ 477 478 /* ARGSUSED2 */ 479 static int 480 tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred) 481 { 482 fsfilcnt_t freenodes; 483 struct tmpfs_mount *tmp; 484 485 tmp = VFS_TO_TMPFS(mp); 486 487 sbp->f_iosize = PAGE_SIZE; 488 sbp->f_bsize = PAGE_SIZE; 489 490 sbp->f_blocks = tmp->tm_pages_max; 491 sbp->f_bavail = tmp->tm_pages_max - tmp->tm_pages_used; 492 sbp->f_bfree = sbp->f_bavail; 493 494 freenodes = tmp->tm_nodes_max - tmp->tm_nodes_inuse; 495 496 sbp->f_files = freenodes + tmp->tm_nodes_inuse; 497 sbp->f_ffree = freenodes; 498 sbp->f_owner = tmp->tm_root->tn_uid; 499 500 return 0; 501 } 502 503 /* --------------------------------------------------------------------- */ 504 505 static int 506 tmpfs_vptofh(struct vnode *vp, struct fid *fhp) 507 { 508 struct tmpfs_node *node; 509 struct tmpfs_fid tfh; 510 node = VP_TO_TMPFS_NODE(vp); 511 memset(&tfh, 0, sizeof(tfh)); 512 tfh.tf_len = sizeof(struct tmpfs_fid); 513 tfh.tf_gen = node->tn_gen; 514 tfh.tf_id = node->tn_id; 515 memcpy(fhp, &tfh, sizeof(tfh)); 516 return (0); 517 } 518 519 /* --------------------------------------------------------------------- */ 520 521 static int 522 tmpfs_checkexp(struct mount *mp, struct sockaddr *nam, int *exflagsp, 523 struct ucred **credanonp) 524 { 525 struct tmpfs_mount *tmp; 526 struct netcred *nc; 527 528 tmp = (struct tmpfs_mount *) mp->mnt_data; 529 nc = vfs_export_lookup(mp, &tmp->tm_export, nam); 530 if (nc == NULL) 531 return (EACCES); 532 533 *exflagsp = nc->netc_exflags; 534 *credanonp = &nc->netc_anon; 535 536 return (0); 537 } 538 539 /* --------------------------------------------------------------------- */ 540 541 /* 542 * tmpfs vfs operations. 543 */ 544 545 static struct vfsops tmpfs_vfsops = { 546 .vfs_mount = tmpfs_mount, 547 .vfs_unmount = tmpfs_unmount, 548 .vfs_root = tmpfs_root, 549 .vfs_statfs = tmpfs_statfs, 550 .vfs_fhtovp = tmpfs_fhtovp, 551 .vfs_vptofh = tmpfs_vptofh, 552 .vfs_sync = vfs_stdsync, 553 .vfs_checkexp = tmpfs_checkexp, 554 }; 555 556 VFS_SET(tmpfs_vfsops, tmpfs, 0); 557 MODULE_VERSION(tmpfs, 1); 558