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