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 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40 /* 41 * Efficient memory file system. 42 * 43 * tmpfs is a file system that uses NetBSD's virtual memory sub-system 44 * (the well-known UVM) to store file data and metadata in an efficient 45 * way. This means that it does not follow the structure of an on-disk 46 * file system because it simply does not need to. Instead, it uses 47 * memory-specific data structures and algorithms to automatically 48 * allocate and release resources. 49 */ 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include <sys/param.h> 54 #include <sys/limits.h> 55 #include <sys/lock.h> 56 #include <sys/mutex.h> 57 #include <sys/kernel.h> 58 #include <sys/stat.h> 59 #include <sys/systm.h> 60 #include <sys/sysctl.h> 61 62 #include <vm/vm.h> 63 #include <vm/vm_object.h> 64 #include <vm/vm_param.h> 65 66 #include <fs/tmpfs/tmpfs.h> 67 68 /* 69 * Default permission for root node 70 */ 71 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) 72 73 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures"); 74 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names"); 75 76 /* --------------------------------------------------------------------- */ 77 78 static int tmpfs_mount(struct mount *, struct thread *); 79 static int tmpfs_unmount(struct mount *, int, struct thread *); 80 static int tmpfs_root(struct mount *, int flags, struct vnode **, 81 struct thread *); 82 static int tmpfs_fhtovp(struct mount *, struct fid *, struct vnode **); 83 static int tmpfs_statfs(struct mount *, struct statfs *, struct thread *); 84 85 /* --------------------------------------------------------------------- */ 86 87 static const char *tmpfs_opts[] = { 88 "from", "size", "inodes", "uid", "gid", "mode", "export", 89 NULL 90 }; 91 92 /* --------------------------------------------------------------------- */ 93 94 #define SWI_MAXMIB 3 95 96 static u_int 97 get_swpgtotal(void) 98 { 99 struct xswdev xsd; 100 char *sname = "vm.swap_info"; 101 int soid[SWI_MAXMIB], oid[2]; 102 u_int unswdev, total, dmmax, nswapdev; 103 size_t mibi, len; 104 105 total = 0; 106 107 len = sizeof(dmmax); 108 if (kernel_sysctlbyname(curthread, "vm.dmmax", &dmmax, &len, 109 NULL, 0, NULL, 0) != 0) 110 return total; 111 112 len = sizeof(nswapdev); 113 if (kernel_sysctlbyname(curthread, "vm.nswapdev", 114 &nswapdev, &len, 115 NULL, 0, NULL, 0) != 0) 116 return total; 117 118 mibi = (SWI_MAXMIB - 1) * sizeof(int); 119 oid[0] = 0; 120 oid[1] = 3; 121 122 if (kernel_sysctl(curthread, oid, 2, 123 soid, &mibi, (void *)sname, strlen(sname), 124 NULL, 0) != 0) 125 return total; 126 127 mibi = (SWI_MAXMIB - 1); 128 for (unswdev = 0; unswdev < nswapdev; ++unswdev) { 129 soid[mibi] = unswdev; 130 len = sizeof(struct xswdev); 131 if (kernel_sysctl(curthread, 132 soid, mibi + 1, &xsd, &len, NULL, 0, 133 NULL, 0) != 0) 134 return total; 135 if (len == sizeof(struct xswdev)) 136 total += (xsd.xsw_nblks - dmmax); 137 } 138 139 /* Not Reached */ 140 return total; 141 } 142 143 /* --------------------------------------------------------------------- */ 144 static int 145 tmpfs_node_ctor(void *mem, int size, void *arg, int flags) 146 { 147 struct tmpfs_node *node = (struct tmpfs_node *)mem; 148 149 node->tn_gen++; 150 node->tn_size = 0; 151 node->tn_status = 0; 152 node->tn_flags = 0; 153 node->tn_links = 0; 154 node->tn_lockf = NULL; 155 node->tn_vnode = NULL; 156 node->tn_vpstate = 0; 157 158 return (0); 159 } 160 161 static void 162 tmpfs_node_dtor(void *mem, int size, void *arg) 163 { 164 struct tmpfs_node *node = (struct tmpfs_node *)mem; 165 node->tn_type = VNON; 166 } 167 168 static int 169 tmpfs_node_init(void *mem, int size, int flags) 170 { 171 struct tmpfs_node *node = (struct tmpfs_node *)mem; 172 node->tn_id = 0; 173 174 mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF); 175 node->tn_gen = arc4random(); 176 177 return (0); 178 } 179 180 static void 181 tmpfs_node_fini(void *mem, int size) 182 { 183 struct tmpfs_node *node = (struct tmpfs_node *)mem; 184 185 mtx_destroy(&node->tn_interlock); 186 } 187 188 static int 189 tmpfs_mount(struct mount *mp, struct thread *td) 190 { 191 struct tmpfs_mount *tmp; 192 struct tmpfs_node *root; 193 size_t pages, mem_size; 194 ino_t nodes; 195 int error; 196 /* Size counters. */ 197 ino_t nodes_max; 198 size_t size_max; 199 200 /* Root node attributes. */ 201 uid_t root_uid; 202 gid_t root_gid; 203 mode_t root_mode; 204 205 struct vattr va; 206 207 if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts)) 208 return (EINVAL); 209 210 if (mp->mnt_flag & MNT_UPDATE) { 211 /* XXX: There is no support yet to update file system 212 * settings. Should be added. */ 213 214 return EOPNOTSUPP; 215 } 216 217 printf("WARNING: TMPFS is considered to be a highly experimental " 218 "feature in FreeBSD.\n"); 219 220 vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY); 221 error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred, td); 222 VOP_UNLOCK(mp->mnt_vnodecovered, 0); 223 if (error) 224 return (error); 225 226 if (mp->mnt_cred->cr_ruid != 0 || 227 vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1) 228 root_gid = va.va_gid; 229 if (mp->mnt_cred->cr_ruid != 0 || 230 vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1) 231 root_uid = va.va_uid; 232 if (mp->mnt_cred->cr_ruid != 0 || 233 vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1) 234 root_mode = va.va_mode; 235 if (vfs_scanopt(mp->mnt_optnew, "inodes", "%d", &nodes_max) != 1) 236 nodes_max = 0; 237 if (vfs_scanopt(mp->mnt_optnew, "size", "%qu", &size_max) != 1) 238 size_max = 0; 239 240 /* Do not allow mounts if we do not have enough memory to preserve 241 * the minimum reserved pages. */ 242 mem_size = cnt.v_free_count + cnt.v_inactive_count + get_swpgtotal(); 243 mem_size -= mem_size > cnt.v_wire_count ? cnt.v_wire_count : mem_size; 244 if (mem_size < TMPFS_PAGES_RESERVED) 245 return ENOSPC; 246 247 /* Get the maximum number of memory pages this file system is 248 * allowed to use, based on the maximum size the user passed in 249 * the mount structure. A value of zero is treated as if the 250 * maximum available space was requested. */ 251 if (size_max < PAGE_SIZE || size_max >= SIZE_MAX) 252 pages = SIZE_MAX; 253 else 254 pages = howmany(size_max, PAGE_SIZE); 255 MPASS(pages > 0); 256 257 if (nodes_max <= 3) 258 nodes = 3 + pages * PAGE_SIZE / 1024; 259 else 260 nodes = nodes_max; 261 MPASS(nodes >= 3); 262 263 /* Allocate the tmpfs mount structure and fill it. */ 264 tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount), 265 M_TMPFSMNT, M_WAITOK | M_ZERO); 266 267 mtx_init(&tmp->allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF); 268 tmp->tm_nodes_max = nodes; 269 tmp->tm_nodes_inuse = 0; 270 tmp->tm_maxfilesize = (u_int64_t)(cnt.v_page_count + get_swpgtotal()) * PAGE_SIZE; 271 LIST_INIT(&tmp->tm_nodes_used); 272 273 tmp->tm_pages_max = pages; 274 tmp->tm_pages_used = 0; 275 tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->allnode_lock); 276 tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent", 277 sizeof(struct tmpfs_dirent), 278 NULL, NULL, NULL, NULL, 279 UMA_ALIGN_PTR, 0); 280 tmp->tm_node_pool = uma_zcreate("TMPFS node", 281 sizeof(struct tmpfs_node), 282 tmpfs_node_ctor, tmpfs_node_dtor, 283 tmpfs_node_init, tmpfs_node_fini, 284 UMA_ALIGN_PTR, 0); 285 286 /* Allocate the root node. */ 287 error = tmpfs_alloc_node(tmp, VDIR, root_uid, 288 root_gid, root_mode & ALLPERMS, NULL, NULL, 289 VNOVAL, td, &root); 290 291 if (error != 0 || root == NULL) { 292 uma_zdestroy(tmp->tm_node_pool); 293 uma_zdestroy(tmp->tm_dirent_pool); 294 delete_unrhdr(tmp->tm_ino_unr); 295 free(tmp, M_TMPFSMNT); 296 return error; 297 } 298 KASSERT(root->tn_id == 2, ("tmpfs root with invalid ino: %d", root->tn_id)); 299 tmp->tm_root = root; 300 301 MNT_ILOCK(mp); 302 mp->mnt_flag |= MNT_LOCAL; 303 mp->mnt_kern_flag |= MNTK_MPSAFE; 304 MNT_IUNLOCK(mp); 305 306 mp->mnt_data = tmp; 307 mp->mnt_stat.f_namemax = MAXNAMLEN; 308 vfs_getnewfsid(mp); 309 vfs_mountedfrom(mp, "tmpfs"); 310 311 return 0; 312 } 313 314 /* --------------------------------------------------------------------- */ 315 316 /* ARGSUSED2 */ 317 static int 318 tmpfs_unmount(struct mount *mp, int mntflags, struct thread *l) 319 { 320 int error; 321 int flags = 0; 322 struct tmpfs_mount *tmp; 323 struct tmpfs_node *node; 324 325 /* Handle forced unmounts. */ 326 if (mntflags & MNT_FORCE) 327 flags |= FORCECLOSE; 328 329 /* Finalize all pending I/O. */ 330 error = vflush(mp, 0, flags, l); 331 if (error != 0) 332 return error; 333 334 tmp = VFS_TO_TMPFS(mp); 335 336 /* Free all associated data. The loop iterates over the linked list 337 * we have containing all used nodes. For each of them that is 338 * a directory, we free all its directory entries. Note that after 339 * freeing a node, it will automatically go to the available list, 340 * so we will later have to iterate over it to release its items. */ 341 node = LIST_FIRST(&tmp->tm_nodes_used); 342 while (node != NULL) { 343 struct tmpfs_node *next; 344 345 if (node->tn_type == VDIR) { 346 struct tmpfs_dirent *de; 347 348 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead); 349 while (de != NULL) { 350 struct tmpfs_dirent *nde; 351 352 nde = TAILQ_NEXT(de, td_entries); 353 tmpfs_free_dirent(tmp, de, FALSE); 354 de = nde; 355 node->tn_size -= sizeof(struct tmpfs_dirent); 356 } 357 } 358 359 next = LIST_NEXT(node, tn_entries); 360 tmpfs_free_node(tmp, node); 361 node = next; 362 } 363 364 uma_zdestroy(tmp->tm_dirent_pool); 365 uma_zdestroy(tmp->tm_node_pool); 366 delete_unrhdr(tmp->tm_ino_unr); 367 368 mtx_destroy(&tmp->allnode_lock); 369 MPASS(tmp->tm_pages_used == 0); 370 MPASS(tmp->tm_nodes_inuse == 0); 371 372 /* Throw away the tmpfs_mount structure. */ 373 free(mp->mnt_data, M_TMPFSMNT); 374 mp->mnt_data = NULL; 375 376 MNT_ILOCK(mp); 377 mp->mnt_flag &= ~MNT_LOCAL; 378 MNT_IUNLOCK(mp); 379 return 0; 380 } 381 382 /* --------------------------------------------------------------------- */ 383 384 static int 385 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td) 386 { 387 int error; 388 error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp, td); 389 390 if (!error) 391 (*vpp)->v_vflag |= VV_ROOT; 392 393 return error; 394 } 395 396 /* --------------------------------------------------------------------- */ 397 398 static int 399 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 400 { 401 boolean_t found; 402 struct tmpfs_fid *tfhp; 403 struct tmpfs_mount *tmp; 404 struct tmpfs_node *node; 405 406 tmp = VFS_TO_TMPFS(mp); 407 408 tfhp = (struct tmpfs_fid *)fhp; 409 if (tfhp->tf_len != sizeof(struct tmpfs_fid)) 410 return EINVAL; 411 412 if (tfhp->tf_id >= tmp->tm_nodes_max) 413 return EINVAL; 414 415 found = FALSE; 416 417 TMPFS_LOCK(tmp); 418 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) { 419 if (node->tn_id == tfhp->tf_id && 420 node->tn_gen == tfhp->tf_gen) { 421 found = TRUE; 422 break; 423 } 424 } 425 TMPFS_UNLOCK(tmp); 426 427 if (found) 428 return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp, curthread)); 429 430 return (EINVAL); 431 } 432 433 /* --------------------------------------------------------------------- */ 434 435 /* ARGSUSED2 */ 436 static int 437 tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *l) 438 { 439 fsfilcnt_t freenodes; 440 struct tmpfs_mount *tmp; 441 442 tmp = VFS_TO_TMPFS(mp); 443 444 sbp->f_iosize = PAGE_SIZE; 445 sbp->f_bsize = PAGE_SIZE; 446 447 sbp->f_blocks = TMPFS_PAGES_MAX(tmp); 448 sbp->f_bavail = sbp->f_bfree = TMPFS_PAGES_AVAIL(tmp); 449 450 freenodes = MIN(tmp->tm_nodes_max - tmp->tm_nodes_inuse, 451 TMPFS_PAGES_AVAIL(tmp) * PAGE_SIZE / sizeof(struct tmpfs_node)); 452 453 sbp->f_files = freenodes + tmp->tm_nodes_inuse; 454 sbp->f_ffree = freenodes; 455 /* sbp->f_owner = tmp->tn_uid; */ 456 457 return 0; 458 } 459 460 /* --------------------------------------------------------------------- */ 461 462 /* 463 * tmpfs vfs operations. 464 */ 465 466 struct vfsops tmpfs_vfsops = { 467 .vfs_mount = tmpfs_mount, 468 .vfs_unmount = tmpfs_unmount, 469 .vfs_root = tmpfs_root, 470 .vfs_statfs = tmpfs_statfs, 471 .vfs_fhtovp = tmpfs_fhtovp, 472 }; 473 VFS_SET(tmpfs_vfsops, tmpfs, 0); 474