1 /* $NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $ */ 2 3 /*- 4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD 5 * 6 * Copyright (c) 2005 The NetBSD Foundation, Inc. 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to The NetBSD Foundation 10 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 11 * 2005 program. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 32 * POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 /* 36 * Efficient memory file system. 37 * 38 * tmpfs is a file system that uses FreeBSD's virtual memory 39 * sub-system to store file data and metadata in an efficient way. 40 * This means that it does not follow the structure of an on-disk file 41 * system because it simply does not need to. Instead, it uses 42 * memory-specific data structures and algorithms to automatically 43 * allocate and release resources. 44 */ 45 46 #include "opt_tmpfs.h" 47 48 #include <sys/cdefs.h> 49 __FBSDID("$FreeBSD$"); 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/dirent.h> 54 #include <sys/limits.h> 55 #include <sys/lock.h> 56 #include <sys/mount.h> 57 #include <sys/mutex.h> 58 #include <sys/proc.h> 59 #include <sys/jail.h> 60 #include <sys/kernel.h> 61 #include <sys/rwlock.h> 62 #include <sys/stat.h> 63 #include <sys/sysctl.h> 64 #include <sys/vnode.h> 65 66 #include <vm/vm.h> 67 #include <vm/vm_object.h> 68 #include <vm/vm_param.h> 69 70 #include <fs/tmpfs/tmpfs.h> 71 72 /* 73 * Default permission for root node 74 */ 75 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) 76 77 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures"); 78 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names"); 79 80 static int tmpfs_mount(struct mount *); 81 static int tmpfs_unmount(struct mount *, int); 82 static int tmpfs_root(struct mount *, int flags, struct vnode **); 83 static int tmpfs_fhtovp(struct mount *, struct fid *, int, 84 struct vnode **); 85 static int tmpfs_statfs(struct mount *, struct statfs *); 86 static void tmpfs_susp_clean(struct mount *); 87 88 static const char *tmpfs_opts[] = { 89 "from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export", 90 "union", "nonc", NULL 91 }; 92 93 static const char *tmpfs_updateopts[] = { 94 "from", "export", "size", NULL 95 }; 96 97 static int 98 tmpfs_node_ctor(void *mem, int size, void *arg, int flags) 99 { 100 struct tmpfs_node *node = (struct tmpfs_node *)mem; 101 102 node->tn_gen++; 103 node->tn_size = 0; 104 node->tn_status = 0; 105 node->tn_flags = 0; 106 node->tn_links = 0; 107 node->tn_vnode = NULL; 108 node->tn_vpstate = 0; 109 110 return (0); 111 } 112 113 static void 114 tmpfs_node_dtor(void *mem, int size, void *arg) 115 { 116 struct tmpfs_node *node = (struct tmpfs_node *)mem; 117 node->tn_type = VNON; 118 } 119 120 static int 121 tmpfs_node_init(void *mem, int size, int flags) 122 { 123 struct tmpfs_node *node = (struct tmpfs_node *)mem; 124 node->tn_id = 0; 125 126 mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF); 127 node->tn_gen = arc4random(); 128 129 return (0); 130 } 131 132 static void 133 tmpfs_node_fini(void *mem, int size) 134 { 135 struct tmpfs_node *node = (struct tmpfs_node *)mem; 136 137 mtx_destroy(&node->tn_interlock); 138 } 139 140 static int 141 tmpfs_mount(struct mount *mp) 142 { 143 const size_t nodes_per_page = howmany(PAGE_SIZE, 144 sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node)); 145 struct tmpfs_mount *tmp; 146 struct tmpfs_node *root; 147 int error, flags; 148 bool nonc; 149 /* Size counters. */ 150 u_quad_t pages; 151 off_t nodes_max, size_max, maxfilesize; 152 153 /* Root node attributes. */ 154 uid_t root_uid; 155 gid_t root_gid; 156 mode_t root_mode; 157 158 struct vattr va; 159 160 if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts)) 161 return (EINVAL); 162 163 if (mp->mnt_flag & MNT_UPDATE) { 164 /* Only support update mounts for certain options. */ 165 if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0) 166 return (EOPNOTSUPP); 167 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) == 0) { 168 /* 169 * On-the-fly resizing is not supported (yet). We still 170 * need to have "size" listed as "supported", otherwise 171 * trying to update fs that is listed in fstab with size 172 * parameter, say trying to change rw to ro or vice 173 * versa, would cause vfs_filteropt() to bail. 174 */ 175 if (size_max != VFS_TO_TMPFS(mp)->tm_size_max) 176 return (EOPNOTSUPP); 177 } 178 if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) && 179 !(VFS_TO_TMPFS(mp)->tm_ronly)) { 180 /* RW -> RO */ 181 error = VFS_SYNC(mp, MNT_WAIT); 182 if (error) 183 return (error); 184 flags = WRITECLOSE; 185 if (mp->mnt_flag & MNT_FORCE) 186 flags |= FORCECLOSE; 187 error = vflush(mp, 0, flags, curthread); 188 if (error) 189 return (error); 190 VFS_TO_TMPFS(mp)->tm_ronly = 1; 191 MNT_ILOCK(mp); 192 mp->mnt_flag |= MNT_RDONLY; 193 MNT_IUNLOCK(mp); 194 } else if (!vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) && 195 VFS_TO_TMPFS(mp)->tm_ronly) { 196 /* RO -> RW */ 197 VFS_TO_TMPFS(mp)->tm_ronly = 0; 198 MNT_ILOCK(mp); 199 mp->mnt_flag &= ~MNT_RDONLY; 200 MNT_IUNLOCK(mp); 201 } 202 return (0); 203 } 204 205 vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY); 206 error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred); 207 VOP_UNLOCK(mp->mnt_vnodecovered, 0); 208 if (error) 209 return (error); 210 211 if (mp->mnt_cred->cr_ruid != 0 || 212 vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1) 213 root_gid = va.va_gid; 214 if (mp->mnt_cred->cr_ruid != 0 || 215 vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1) 216 root_uid = va.va_uid; 217 if (mp->mnt_cred->cr_ruid != 0 || 218 vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1) 219 root_mode = va.va_mode; 220 if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0) 221 nodes_max = 0; 222 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0) 223 size_max = 0; 224 if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0) 225 maxfilesize = 0; 226 nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0; 227 228 /* Do not allow mounts if we do not have enough memory to preserve 229 * the minimum reserved pages. */ 230 if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED) 231 return (ENOSPC); 232 233 /* Get the maximum number of memory pages this file system is 234 * allowed to use, based on the maximum size the user passed in 235 * the mount structure. A value of zero is treated as if the 236 * maximum available space was requested. */ 237 if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE || 238 (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX)) 239 pages = SIZE_MAX; 240 else { 241 size_max = roundup(size_max, PAGE_SIZE); 242 pages = howmany(size_max, PAGE_SIZE); 243 } 244 MPASS(pages > 0); 245 246 if (nodes_max <= 3) { 247 if (pages < INT_MAX / nodes_per_page) 248 nodes_max = pages * nodes_per_page; 249 else 250 nodes_max = INT_MAX; 251 } 252 if (nodes_max > INT_MAX) 253 nodes_max = INT_MAX; 254 MPASS(nodes_max >= 3); 255 256 /* Allocate the tmpfs mount structure and fill it. */ 257 tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount), 258 M_TMPFSMNT, M_WAITOK | M_ZERO); 259 260 mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF); 261 tmp->tm_nodes_max = nodes_max; 262 tmp->tm_nodes_inuse = 0; 263 tmp->tm_refcount = 1; 264 tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX; 265 LIST_INIT(&tmp->tm_nodes_used); 266 267 tmp->tm_size_max = size_max; 268 tmp->tm_pages_max = pages; 269 tmp->tm_pages_used = 0; 270 new_unrhdr64(&tmp->tm_ino_unr, 2); 271 tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent", 272 sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL, 273 UMA_ALIGN_PTR, 0); 274 tmp->tm_node_pool = uma_zcreate("TMPFS node", 275 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor, 276 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0); 277 tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 278 tmp->tm_nonc = nonc; 279 280 /* Allocate the root node. */ 281 error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid, 282 root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root); 283 284 if (error != 0 || root == NULL) { 285 uma_zdestroy(tmp->tm_node_pool); 286 uma_zdestroy(tmp->tm_dirent_pool); 287 free(tmp, M_TMPFSMNT); 288 return (error); 289 } 290 KASSERT(root->tn_id == 2, 291 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id)); 292 tmp->tm_root = root; 293 294 MNT_ILOCK(mp); 295 mp->mnt_flag |= MNT_LOCAL; 296 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED; 297 MNT_IUNLOCK(mp); 298 299 mp->mnt_data = tmp; 300 mp->mnt_stat.f_namemax = MAXNAMLEN; 301 vfs_getnewfsid(mp); 302 vfs_mountedfrom(mp, "tmpfs"); 303 304 return 0; 305 } 306 307 /* ARGSUSED2 */ 308 static int 309 tmpfs_unmount(struct mount *mp, int mntflags) 310 { 311 struct tmpfs_mount *tmp; 312 struct tmpfs_node *node; 313 int error, flags; 314 315 flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0; 316 tmp = VFS_TO_TMPFS(mp); 317 318 /* Stop writers */ 319 error = vfs_write_suspend_umnt(mp); 320 if (error != 0) 321 return (error); 322 /* 323 * At this point, nodes cannot be destroyed by any other 324 * thread because write suspension is started. 325 */ 326 327 for (;;) { 328 error = vflush(mp, 0, flags, curthread); 329 if (error != 0) { 330 vfs_write_resume(mp, VR_START_WRITE); 331 return (error); 332 } 333 MNT_ILOCK(mp); 334 if (mp->mnt_nvnodelistsize == 0) { 335 MNT_IUNLOCK(mp); 336 break; 337 } 338 MNT_IUNLOCK(mp); 339 if ((mntflags & MNT_FORCE) == 0) { 340 vfs_write_resume(mp, VR_START_WRITE); 341 return (EBUSY); 342 } 343 } 344 345 TMPFS_LOCK(tmp); 346 while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) { 347 TMPFS_NODE_LOCK(node); 348 if (node->tn_type == VDIR) 349 tmpfs_dir_destroy(tmp, node); 350 if (tmpfs_free_node_locked(tmp, node, true)) 351 TMPFS_LOCK(tmp); 352 else 353 TMPFS_NODE_UNLOCK(node); 354 } 355 356 mp->mnt_data = NULL; 357 tmpfs_free_tmp(tmp); 358 vfs_write_resume(mp, VR_START_WRITE); 359 360 MNT_ILOCK(mp); 361 mp->mnt_flag &= ~MNT_LOCAL; 362 MNT_IUNLOCK(mp); 363 364 return (0); 365 } 366 367 void 368 tmpfs_free_tmp(struct tmpfs_mount *tmp) 369 { 370 371 MPASS(tmp->tm_refcount > 0); 372 tmp->tm_refcount--; 373 if (tmp->tm_refcount > 0) { 374 TMPFS_UNLOCK(tmp); 375 return; 376 } 377 TMPFS_UNLOCK(tmp); 378 379 uma_zdestroy(tmp->tm_dirent_pool); 380 uma_zdestroy(tmp->tm_node_pool); 381 382 mtx_destroy(&tmp->tm_allnode_lock); 383 MPASS(tmp->tm_pages_used == 0); 384 MPASS(tmp->tm_nodes_inuse == 0); 385 386 free(tmp, M_TMPFSMNT); 387 } 388 389 static int 390 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp) 391 { 392 int error; 393 394 error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp); 395 if (error == 0) 396 (*vpp)->v_vflag |= VV_ROOT; 397 return (error); 398 } 399 400 static int 401 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags, 402 struct vnode **vpp) 403 { 404 struct tmpfs_fid *tfhp; 405 struct tmpfs_mount *tmp; 406 struct tmpfs_node *node; 407 int error; 408 409 tmp = VFS_TO_TMPFS(mp); 410 411 tfhp = (struct tmpfs_fid *)fhp; 412 if (tfhp->tf_len != sizeof(struct tmpfs_fid)) 413 return (EINVAL); 414 415 if (tfhp->tf_id >= tmp->tm_nodes_max) 416 return (EINVAL); 417 418 TMPFS_LOCK(tmp); 419 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) { 420 if (node->tn_id == tfhp->tf_id && 421 node->tn_gen == tfhp->tf_gen) { 422 tmpfs_ref_node(node); 423 break; 424 } 425 } 426 TMPFS_UNLOCK(tmp); 427 428 if (node != NULL) { 429 error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp); 430 tmpfs_free_node(tmp, node); 431 } else 432 error = EINVAL; 433 return (error); 434 } 435 436 /* ARGSUSED2 */ 437 static int 438 tmpfs_statfs(struct mount *mp, struct statfs *sbp) 439 { 440 struct tmpfs_mount *tmp; 441 size_t used; 442 443 tmp = VFS_TO_TMPFS(mp); 444 445 sbp->f_iosize = PAGE_SIZE; 446 sbp->f_bsize = PAGE_SIZE; 447 448 used = tmpfs_pages_used(tmp); 449 if (tmp->tm_pages_max != ULONG_MAX) 450 sbp->f_blocks = tmp->tm_pages_max; 451 else 452 sbp->f_blocks = used + tmpfs_mem_avail(); 453 if (sbp->f_blocks <= used) 454 sbp->f_bavail = 0; 455 else 456 sbp->f_bavail = sbp->f_blocks - used; 457 sbp->f_bfree = sbp->f_bavail; 458 used = tmp->tm_nodes_inuse; 459 sbp->f_files = tmp->tm_nodes_max; 460 if (sbp->f_files <= used) 461 sbp->f_ffree = 0; 462 else 463 sbp->f_ffree = sbp->f_files - used; 464 /* sbp->f_owner = tmp->tn_uid; */ 465 466 return 0; 467 } 468 469 static int 470 tmpfs_sync(struct mount *mp, int waitfor) 471 { 472 struct vnode *vp, *mvp; 473 struct vm_object *obj; 474 475 if (waitfor == MNT_SUSPEND) { 476 MNT_ILOCK(mp); 477 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 478 MNT_IUNLOCK(mp); 479 } else if (waitfor == MNT_LAZY) { 480 /* 481 * Handle lazy updates of mtime from writes to mmaped 482 * regions. Use MNT_VNODE_FOREACH_ALL instead of 483 * MNT_VNODE_FOREACH_ACTIVE, since unmap of the 484 * tmpfs-backed vnode does not call vinactive(), due 485 * to vm object type is OBJT_SWAP. 486 */ 487 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 488 if (vp->v_type != VREG) { 489 VI_UNLOCK(vp); 490 continue; 491 } 492 obj = vp->v_object; 493 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) == 494 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj")); 495 496 /* 497 * Unlocked read, avoid taking vnode lock if 498 * not needed. Lost update will be handled on 499 * the next call. 500 */ 501 if ((obj->flags & OBJ_TMPFS_DIRTY) == 0) { 502 VI_UNLOCK(vp); 503 continue; 504 } 505 if (vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 506 curthread) != 0) 507 continue; 508 tmpfs_check_mtime(vp); 509 vput(vp); 510 } 511 } 512 return (0); 513 } 514 515 /* 516 * The presence of a susp_clean method tells the VFS to track writes. 517 */ 518 static void 519 tmpfs_susp_clean(struct mount *mp __unused) 520 { 521 } 522 523 /* 524 * tmpfs vfs operations. 525 */ 526 527 struct vfsops tmpfs_vfsops = { 528 .vfs_mount = tmpfs_mount, 529 .vfs_unmount = tmpfs_unmount, 530 .vfs_root = tmpfs_root, 531 .vfs_statfs = tmpfs_statfs, 532 .vfs_fhtovp = tmpfs_fhtovp, 533 .vfs_sync = tmpfs_sync, 534 .vfs_susp_clean = tmpfs_susp_clean, 535 }; 536 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL); 537