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 FreeBSD's virtual memory 37 * sub-system to store file data and metadata in an efficient way. 38 * This means that it does not follow the structure of an on-disk file 39 * 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 #include <sys/cdefs.h> 44 __FBSDID("$FreeBSD$"); 45 46 #include <sys/param.h> 47 #include <sys/limits.h> 48 #include <sys/lock.h> 49 #include <sys/mutex.h> 50 #include <sys/proc.h> 51 #include <sys/jail.h> 52 #include <sys/kernel.h> 53 #include <sys/rwlock.h> 54 #include <sys/stat.h> 55 #include <sys/systm.h> 56 #include <sys/sysctl.h> 57 58 #include <vm/vm.h> 59 #include <vm/vm_object.h> 60 #include <vm/vm_param.h> 61 62 #include <fs/tmpfs/tmpfs.h> 63 64 /* 65 * Default permission for root node 66 */ 67 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) 68 69 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures"); 70 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names"); 71 72 static int tmpfs_mount(struct mount *); 73 static int tmpfs_unmount(struct mount *, int); 74 static int tmpfs_root(struct mount *, int flags, struct vnode **); 75 static int tmpfs_fhtovp(struct mount *, struct fid *, int, 76 struct vnode **); 77 static int tmpfs_statfs(struct mount *, struct statfs *); 78 static void tmpfs_susp_clean(struct mount *); 79 80 static const char *tmpfs_opts[] = { 81 "from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export", 82 "union", NULL 83 }; 84 85 static const char *tmpfs_updateopts[] = { 86 "from", "export", NULL 87 }; 88 89 static int 90 tmpfs_node_ctor(void *mem, int size, void *arg, int flags) 91 { 92 struct tmpfs_node *node = (struct tmpfs_node *)mem; 93 94 node->tn_gen++; 95 node->tn_size = 0; 96 node->tn_status = 0; 97 node->tn_flags = 0; 98 node->tn_links = 0; 99 node->tn_vnode = NULL; 100 node->tn_vpstate = 0; 101 102 return (0); 103 } 104 105 static void 106 tmpfs_node_dtor(void *mem, int size, void *arg) 107 { 108 struct tmpfs_node *node = (struct tmpfs_node *)mem; 109 node->tn_type = VNON; 110 } 111 112 static int 113 tmpfs_node_init(void *mem, int size, int flags) 114 { 115 struct tmpfs_node *node = (struct tmpfs_node *)mem; 116 node->tn_id = 0; 117 118 mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF); 119 node->tn_gen = arc4random(); 120 121 return (0); 122 } 123 124 static void 125 tmpfs_node_fini(void *mem, int size) 126 { 127 struct tmpfs_node *node = (struct tmpfs_node *)mem; 128 129 mtx_destroy(&node->tn_interlock); 130 } 131 132 static int 133 tmpfs_mount(struct mount *mp) 134 { 135 const size_t nodes_per_page = howmany(PAGE_SIZE, 136 sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node)); 137 struct tmpfs_mount *tmp; 138 struct tmpfs_node *root; 139 struct thread *td = curthread; 140 int error; 141 /* Size counters. */ 142 u_quad_t pages; 143 off_t nodes_max, size_max, maxfilesize; 144 145 /* Root node attributes. */ 146 uid_t root_uid; 147 gid_t root_gid; 148 mode_t root_mode; 149 150 struct vattr va; 151 152 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_TMPFS)) 153 return (EPERM); 154 155 if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts)) 156 return (EINVAL); 157 158 if (mp->mnt_flag & MNT_UPDATE) { 159 /* Only support update mounts for certain options. */ 160 if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0) 161 return (EOPNOTSUPP); 162 if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) != 163 ((struct tmpfs_mount *)mp->mnt_data)->tm_ronly) 164 return (EOPNOTSUPP); 165 return (0); 166 } 167 168 vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY); 169 error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred); 170 VOP_UNLOCK(mp->mnt_vnodecovered, 0); 171 if (error) 172 return (error); 173 174 if (mp->mnt_cred->cr_ruid != 0 || 175 vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1) 176 root_gid = va.va_gid; 177 if (mp->mnt_cred->cr_ruid != 0 || 178 vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1) 179 root_uid = va.va_uid; 180 if (mp->mnt_cred->cr_ruid != 0 || 181 vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1) 182 root_mode = va.va_mode; 183 if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0) 184 nodes_max = 0; 185 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0) 186 size_max = 0; 187 if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0) 188 maxfilesize = 0; 189 190 /* Do not allow mounts if we do not have enough memory to preserve 191 * the minimum reserved pages. */ 192 if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED) 193 return ENOSPC; 194 195 /* Get the maximum number of memory pages this file system is 196 * allowed to use, based on the maximum size the user passed in 197 * the mount structure. A value of zero is treated as if the 198 * maximum available space was requested. */ 199 if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE || 200 (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX)) 201 pages = SIZE_MAX; 202 else { 203 size_max = roundup(size_max, PAGE_SIZE); 204 pages = howmany(size_max, PAGE_SIZE); 205 } 206 MPASS(pages > 0); 207 208 if (nodes_max <= 3) { 209 if (pages < INT_MAX / nodes_per_page) 210 nodes_max = pages * nodes_per_page; 211 else 212 nodes_max = INT_MAX; 213 } 214 if (nodes_max > INT_MAX) 215 nodes_max = INT_MAX; 216 MPASS(nodes_max >= 3); 217 218 /* Allocate the tmpfs mount structure and fill it. */ 219 tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount), 220 M_TMPFSMNT, M_WAITOK | M_ZERO); 221 222 mtx_init(&tmp->allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF); 223 tmp->tm_nodes_max = nodes_max; 224 tmp->tm_nodes_inuse = 0; 225 tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX; 226 LIST_INIT(&tmp->tm_nodes_used); 227 228 tmp->tm_pages_max = pages; 229 tmp->tm_pages_used = 0; 230 tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->allnode_lock); 231 tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent", 232 sizeof(struct tmpfs_dirent), 233 NULL, NULL, NULL, NULL, 234 UMA_ALIGN_PTR, 0); 235 tmp->tm_node_pool = uma_zcreate("TMPFS node", 236 sizeof(struct tmpfs_node), 237 tmpfs_node_ctor, tmpfs_node_dtor, 238 tmpfs_node_init, tmpfs_node_fini, 239 UMA_ALIGN_PTR, 0); 240 tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 241 242 /* Allocate the root node. */ 243 error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, 244 root_gid, root_mode & ALLPERMS, NULL, NULL, 245 VNOVAL, &root); 246 247 if (error != 0 || root == NULL) { 248 uma_zdestroy(tmp->tm_node_pool); 249 uma_zdestroy(tmp->tm_dirent_pool); 250 delete_unrhdr(tmp->tm_ino_unr); 251 free(tmp, M_TMPFSMNT); 252 return error; 253 } 254 KASSERT(root->tn_id == 2, 255 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id)); 256 tmp->tm_root = root; 257 258 MNT_ILOCK(mp); 259 mp->mnt_flag |= MNT_LOCAL; 260 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 261 MNT_IUNLOCK(mp); 262 263 mp->mnt_data = tmp; 264 mp->mnt_stat.f_namemax = MAXNAMLEN; 265 vfs_getnewfsid(mp); 266 vfs_mountedfrom(mp, "tmpfs"); 267 268 return 0; 269 } 270 271 /* ARGSUSED2 */ 272 static int 273 tmpfs_unmount(struct mount *mp, int mntflags) 274 { 275 struct tmpfs_mount *tmp; 276 struct tmpfs_node *node; 277 int error, flags; 278 279 flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0; 280 tmp = VFS_TO_TMPFS(mp); 281 282 /* Stop writers */ 283 error = vfs_write_suspend_umnt(mp); 284 if (error != 0) 285 return (error); 286 /* 287 * At this point, nodes cannot be destroyed by any other 288 * thread because write suspension is started. 289 */ 290 291 for (;;) { 292 error = vflush(mp, 0, flags, curthread); 293 if (error != 0) { 294 vfs_write_resume(mp, VR_START_WRITE); 295 return (error); 296 } 297 MNT_ILOCK(mp); 298 if (mp->mnt_nvnodelistsize == 0) { 299 MNT_IUNLOCK(mp); 300 break; 301 } 302 MNT_IUNLOCK(mp); 303 if ((mntflags & MNT_FORCE) == 0) { 304 vfs_write_resume(mp, VR_START_WRITE); 305 return (EBUSY); 306 } 307 } 308 309 TMPFS_LOCK(tmp); 310 while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) { 311 TMPFS_UNLOCK(tmp); 312 if (node->tn_type == VDIR) 313 tmpfs_dir_destroy(tmp, node); 314 tmpfs_free_node(tmp, node); 315 TMPFS_LOCK(tmp); 316 } 317 TMPFS_UNLOCK(tmp); 318 319 uma_zdestroy(tmp->tm_dirent_pool); 320 uma_zdestroy(tmp->tm_node_pool); 321 delete_unrhdr(tmp->tm_ino_unr); 322 323 mtx_destroy(&tmp->allnode_lock); 324 MPASS(tmp->tm_pages_used == 0); 325 MPASS(tmp->tm_nodes_inuse == 0); 326 327 /* Throw away the tmpfs_mount structure. */ 328 free(mp->mnt_data, M_TMPFSMNT); 329 mp->mnt_data = NULL; 330 vfs_write_resume(mp, VR_START_WRITE); 331 332 MNT_ILOCK(mp); 333 mp->mnt_flag &= ~MNT_LOCAL; 334 MNT_IUNLOCK(mp); 335 336 return (0); 337 } 338 339 static int 340 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp) 341 { 342 int error; 343 error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp); 344 345 if (!error) 346 (*vpp)->v_vflag |= VV_ROOT; 347 348 return error; 349 } 350 351 static int 352 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags, 353 struct vnode **vpp) 354 { 355 boolean_t found; 356 struct tmpfs_fid *tfhp; 357 struct tmpfs_mount *tmp; 358 struct tmpfs_node *node; 359 360 tmp = VFS_TO_TMPFS(mp); 361 362 tfhp = (struct tmpfs_fid *)fhp; 363 if (tfhp->tf_len != sizeof(struct tmpfs_fid)) 364 return EINVAL; 365 366 if (tfhp->tf_id >= tmp->tm_nodes_max) 367 return EINVAL; 368 369 found = FALSE; 370 371 TMPFS_LOCK(tmp); 372 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) { 373 if (node->tn_id == tfhp->tf_id && 374 node->tn_gen == tfhp->tf_gen) { 375 found = TRUE; 376 break; 377 } 378 } 379 TMPFS_UNLOCK(tmp); 380 381 if (found) 382 return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp)); 383 384 return (EINVAL); 385 } 386 387 /* ARGSUSED2 */ 388 static int 389 tmpfs_statfs(struct mount *mp, struct statfs *sbp) 390 { 391 struct tmpfs_mount *tmp; 392 size_t used; 393 394 tmp = VFS_TO_TMPFS(mp); 395 396 sbp->f_iosize = PAGE_SIZE; 397 sbp->f_bsize = PAGE_SIZE; 398 399 used = tmpfs_pages_used(tmp); 400 if (tmp->tm_pages_max != SIZE_MAX) 401 sbp->f_blocks = tmp->tm_pages_max; 402 else 403 sbp->f_blocks = used + tmpfs_mem_avail(); 404 if (sbp->f_blocks <= used) 405 sbp->f_bavail = 0; 406 else 407 sbp->f_bavail = sbp->f_blocks - used; 408 sbp->f_bfree = sbp->f_bavail; 409 used = tmp->tm_nodes_inuse; 410 sbp->f_files = tmp->tm_nodes_max; 411 if (sbp->f_files <= used) 412 sbp->f_ffree = 0; 413 else 414 sbp->f_ffree = sbp->f_files - used; 415 /* sbp->f_owner = tmp->tn_uid; */ 416 417 return 0; 418 } 419 420 static int 421 tmpfs_sync(struct mount *mp, int waitfor) 422 { 423 struct vnode *vp, *mvp; 424 struct vm_object *obj; 425 426 if (waitfor == MNT_SUSPEND) { 427 MNT_ILOCK(mp); 428 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 429 MNT_IUNLOCK(mp); 430 } else if (waitfor == MNT_LAZY) { 431 /* 432 * Handle lazy updates of mtime from writes to mmaped 433 * regions. Use MNT_VNODE_FOREACH_ALL instead of 434 * MNT_VNODE_FOREACH_ACTIVE, since unmap of the 435 * tmpfs-backed vnode does not call vinactive(), due 436 * to vm object type is OBJT_SWAP. 437 */ 438 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 439 if (vp->v_type != VREG) { 440 VI_UNLOCK(vp); 441 continue; 442 } 443 obj = vp->v_object; 444 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) == 445 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj")); 446 447 /* 448 * Unlocked read, avoid taking vnode lock if 449 * not needed. Lost update will be handled on 450 * the next call. 451 */ 452 if ((obj->flags & OBJ_TMPFS_DIRTY) == 0) { 453 VI_UNLOCK(vp); 454 continue; 455 } 456 if (vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 457 curthread) != 0) 458 continue; 459 tmpfs_check_mtime(vp); 460 vput(vp); 461 } 462 } 463 return (0); 464 } 465 466 /* 467 * The presence of a susp_clean method tells the VFS to track writes. 468 */ 469 static void 470 tmpfs_susp_clean(struct mount *mp __unused) 471 { 472 } 473 474 /* 475 * tmpfs vfs operations. 476 */ 477 478 struct vfsops tmpfs_vfsops = { 479 .vfs_mount = tmpfs_mount, 480 .vfs_unmount = tmpfs_unmount, 481 .vfs_root = tmpfs_root, 482 .vfs_statfs = tmpfs_statfs, 483 .vfs_fhtovp = tmpfs_fhtovp, 484 .vfs_sync = tmpfs_sync, 485 .vfs_susp_clean = tmpfs_susp_clean, 486 }; 487 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL); 488