1 /* $NetBSD: tmpfs_vfsops.c,v 1.77 2020/04/04 20:49:30 ad Exp $ */ 2 3 /* 4 * Copyright (c) 2005, 2006, 2007 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/cdefs.h> 45 __KERNEL_RCSID(0, "$NetBSD: tmpfs_vfsops.c,v 1.77 2020/04/04 20:49:30 ad Exp $"); 46 47 #include <sys/param.h> 48 #include <sys/atomic.h> 49 #include <sys/types.h> 50 #include <sys/kmem.h> 51 #include <sys/mount.h> 52 #include <sys/stat.h> 53 #include <sys/systm.h> 54 #include <sys/vnode.h> 55 #include <sys/kauth.h> 56 #include <sys/module.h> 57 58 #include <miscfs/genfs/genfs.h> 59 #include <fs/tmpfs/tmpfs.h> 60 #include <fs/tmpfs/tmpfs_args.h> 61 62 MODULE(MODULE_CLASS_VFS, tmpfs, NULL); 63 64 struct pool tmpfs_dirent_pool; 65 struct pool tmpfs_node_pool; 66 67 void 68 tmpfs_init(void) 69 { 70 71 pool_init(&tmpfs_dirent_pool, sizeof(tmpfs_dirent_t), 0, 0, 0, 72 "tmpfs_dirent", &pool_allocator_nointr, IPL_NONE); 73 pool_init(&tmpfs_node_pool, sizeof(tmpfs_node_t), 0, 0, 0, 74 "tmpfs_node", &pool_allocator_nointr, IPL_NONE); 75 } 76 77 void 78 tmpfs_done(void) 79 { 80 81 pool_destroy(&tmpfs_dirent_pool); 82 pool_destroy(&tmpfs_node_pool); 83 } 84 85 int 86 tmpfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 87 { 88 struct tmpfs_args *args = data; 89 tmpfs_mount_t *tmp; 90 tmpfs_node_t *root; 91 struct vattr va; 92 struct vnode *vp; 93 uint64_t memlimit; 94 ino_t nodes; 95 int error, flags; 96 bool set_memlimit; 97 bool set_nodes; 98 99 if (args == NULL) 100 return EINVAL; 101 102 /* Validate the version. */ 103 if (*data_len < sizeof(*args) || 104 args->ta_version != TMPFS_ARGS_VERSION) 105 return EINVAL; 106 107 /* Handle retrieval of mount point arguments. */ 108 if (mp->mnt_flag & MNT_GETARGS) { 109 if (mp->mnt_data == NULL) 110 return EIO; 111 tmp = VFS_TO_TMPFS(mp); 112 113 args->ta_version = TMPFS_ARGS_VERSION; 114 args->ta_nodes_max = tmp->tm_nodes_max; 115 args->ta_size_max = tmp->tm_mem_limit; 116 117 root = tmp->tm_root; 118 args->ta_root_uid = root->tn_uid; 119 args->ta_root_gid = root->tn_gid; 120 args->ta_root_mode = root->tn_mode; 121 122 *data_len = sizeof(*args); 123 return 0; 124 } 125 126 127 /* Prohibit mounts if there is not enough memory. */ 128 if (tmpfs_mem_info(true) < uvmexp.freetarg) 129 return EINVAL; 130 131 /* Check for invalid uid and gid arguments */ 132 if (args->ta_root_uid == VNOVAL || args->ta_root_gid == VNOVAL) 133 return EINVAL; 134 135 /* Get the memory usage limit for this file-system. */ 136 if (args->ta_size_max < PAGE_SIZE) { 137 memlimit = UINT64_MAX; 138 set_memlimit = false; 139 } else { 140 memlimit = args->ta_size_max; 141 set_memlimit = true; 142 } 143 KASSERT(memlimit > 0); 144 145 if (args->ta_nodes_max <= 3) { 146 nodes = 3 + (memlimit / 1024); 147 set_nodes = false; 148 } else { 149 nodes = args->ta_nodes_max; 150 set_nodes = true; 151 } 152 nodes = MIN(nodes, INT_MAX); 153 KASSERT(nodes >= 3); 154 155 if (mp->mnt_flag & MNT_UPDATE) { 156 tmp = VFS_TO_TMPFS(mp); 157 if (set_nodes && nodes < tmp->tm_nodes_cnt) 158 return EBUSY; 159 if ((mp->mnt_iflag & IMNT_WANTRDONLY)) { 160 /* Changing from read/write to read-only. */ 161 flags = WRITECLOSE; 162 if ((mp->mnt_flag & MNT_FORCE)) 163 flags |= FORCECLOSE; 164 error = vflush(mp, NULL, flags); 165 if (error) 166 return error; 167 } 168 if (set_memlimit) { 169 if ((error = tmpfs_mntmem_set(tmp, memlimit)) != 0) 170 return error; 171 } 172 if (set_nodes) 173 tmp->tm_nodes_max = nodes; 174 root = tmp->tm_root; 175 root->tn_uid = args->ta_root_uid; 176 root->tn_gid = args->ta_root_gid; 177 root->tn_mode = args->ta_root_mode; 178 return 0; 179 } 180 181 mp->mnt_flag |= MNT_LOCAL; 182 mp->mnt_stat.f_namemax = TMPFS_MAXNAMLEN; 183 mp->mnt_fs_bshift = PAGE_SHIFT; 184 mp->mnt_dev_bshift = DEV_BSHIFT; 185 mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO | IMNT_SHRLOOKUP | 186 IMNT_NCLOOKUP; 187 vfs_getnewfsid(mp); 188 189 /* Allocate the tmpfs mount structure and fill it. */ 190 tmp = kmem_zalloc(sizeof(tmpfs_mount_t), KM_SLEEP); 191 tmp->tm_nodes_max = nodes; 192 tmp->tm_nodes_cnt = 0; 193 LIST_INIT(&tmp->tm_nodes); 194 195 mutex_init(&tmp->tm_lock, MUTEX_DEFAULT, IPL_NONE); 196 tmpfs_mntmem_init(tmp, memlimit); 197 mp->mnt_data = tmp; 198 199 /* Allocate the root node. */ 200 vattr_null(&va); 201 va.va_type = VDIR; 202 va.va_mode = args->ta_root_mode & ALLPERMS; 203 va.va_uid = args->ta_root_uid; 204 va.va_gid = args->ta_root_gid; 205 error = vcache_new(mp, NULL, &va, NOCRED, NULL, &vp); 206 if (error) { 207 mp->mnt_data = NULL; 208 tmpfs_mntmem_destroy(tmp); 209 mutex_destroy(&tmp->tm_lock); 210 kmem_free(tmp, sizeof(*tmp)); 211 return error; 212 } 213 KASSERT(vp != NULL); 214 root = VP_TO_TMPFS_NODE(vp); 215 KASSERT(root != NULL); 216 217 /* 218 * Parent of the root inode is itself. Also, root inode has no 219 * directory entry (i.e. is never attached), thus hold an extra 220 * reference (link) for it. 221 */ 222 root->tn_links++; 223 root->tn_spec.tn_dir.tn_parent = root; 224 tmp->tm_root = root; 225 vrele(vp); 226 227 error = set_statvfs_info(path, UIO_USERSPACE, "tmpfs", UIO_SYSSPACE, 228 mp->mnt_op->vfs_name, mp, curlwp); 229 if (error) { 230 (void)tmpfs_unmount(mp, MNT_FORCE); 231 } 232 return error; 233 } 234 235 int 236 tmpfs_start(struct mount *mp, int flags) 237 { 238 239 return 0; 240 } 241 242 int 243 tmpfs_unmount(struct mount *mp, int mntflags) 244 { 245 tmpfs_mount_t *tmp = VFS_TO_TMPFS(mp); 246 tmpfs_node_t *node, *cnode; 247 int error, flags = 0; 248 249 /* Handle forced unmounts. */ 250 if (mntflags & MNT_FORCE) 251 flags |= FORCECLOSE; 252 253 /* Finalize all pending I/O. */ 254 error = vflush(mp, NULL, flags); 255 if (error != 0) 256 return error; 257 258 /* 259 * First round, detach and destroy all directory entries. 260 * Also, clear the pointers to the vnodes - they are gone. 261 */ 262 LIST_FOREACH(node, &tmp->tm_nodes, tn_entries) { 263 tmpfs_dirent_t *de; 264 265 node->tn_vnode = NULL; 266 if (node->tn_type != VDIR) { 267 continue; 268 } 269 while ((de = TAILQ_FIRST(&node->tn_spec.tn_dir.tn_dir)) != NULL) { 270 cnode = de->td_node; 271 if (cnode && cnode != TMPFS_NODE_WHITEOUT) { 272 cnode->tn_vnode = NULL; 273 } 274 tmpfs_dir_detach(node, de); 275 tmpfs_free_dirent(tmp, de); 276 } 277 /* Extra virtual entry (itself for the root). */ 278 node->tn_links--; 279 } 280 281 /* Release the reference on root (diagnostic). */ 282 node = tmp->tm_root; 283 node->tn_links--; 284 285 /* Second round, destroy all inodes. */ 286 while ((node = LIST_FIRST(&tmp->tm_nodes)) != NULL) { 287 tmpfs_free_node(tmp, node); 288 } 289 290 /* Throw away the tmpfs_mount structure. */ 291 tmpfs_mntmem_destroy(tmp); 292 mutex_destroy(&tmp->tm_lock); 293 kmem_free(tmp, sizeof(*tmp)); 294 mp->mnt_data = NULL; 295 296 return 0; 297 } 298 299 int 300 tmpfs_root(struct mount *mp, int lktype, vnode_t **vpp) 301 { 302 tmpfs_node_t *node = VFS_TO_TMPFS(mp)->tm_root; 303 int error; 304 305 error = vcache_get(mp, &node, sizeof(node), vpp); 306 if (error) 307 return error; 308 error = vn_lock(*vpp, lktype); 309 if (error) { 310 vrele(*vpp); 311 *vpp = NULL; 312 return error; 313 } 314 315 return 0; 316 } 317 318 int 319 tmpfs_vget(struct mount *mp, ino_t ino, int lktype, vnode_t **vpp) 320 { 321 322 return EOPNOTSUPP; 323 } 324 325 int 326 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int lktype, vnode_t **vpp) 327 { 328 tmpfs_mount_t *tmp = VFS_TO_TMPFS(mp); 329 tmpfs_node_t *node; 330 tmpfs_fid_t tfh; 331 int error; 332 333 if (fhp->fid_len != sizeof(tmpfs_fid_t)) { 334 return EINVAL; 335 } 336 memcpy(&tfh, fhp, sizeof(tmpfs_fid_t)); 337 338 mutex_enter(&tmp->tm_lock); 339 /* XXX big oof .. use a better data structure */ 340 LIST_FOREACH(node, &tmp->tm_nodes, tn_entries) { 341 if (node->tn_id == tfh.tf_id) { 342 /* Prevent this node from disappearing. */ 343 atomic_inc_32(&node->tn_holdcount); 344 break; 345 } 346 } 347 mutex_exit(&tmp->tm_lock); 348 if (node == NULL) 349 return ESTALE; 350 351 error = vcache_get(mp, &node, sizeof(node), vpp); 352 /* If this node has been reclaimed free it now. */ 353 if (atomic_dec_32_nv(&node->tn_holdcount) == TMPFS_NODE_RECLAIMED) { 354 KASSERT(error != 0); 355 tmpfs_free_node(tmp, node); 356 } 357 if (error) 358 return (error == ENOENT ? ESTALE : error); 359 error = vn_lock(*vpp, lktype); 360 if (error) { 361 vrele(*vpp); 362 *vpp = NULL; 363 return error; 364 } 365 if (TMPFS_NODE_GEN(node) != tfh.tf_gen) { 366 vput(*vpp); 367 *vpp = NULL; 368 return ESTALE; 369 } 370 371 return 0; 372 } 373 374 int 375 tmpfs_vptofh(vnode_t *vp, struct fid *fhp, size_t *fh_size) 376 { 377 tmpfs_fid_t tfh; 378 tmpfs_node_t *node; 379 380 if (*fh_size < sizeof(tmpfs_fid_t)) { 381 *fh_size = sizeof(tmpfs_fid_t); 382 return E2BIG; 383 } 384 *fh_size = sizeof(tmpfs_fid_t); 385 node = VP_TO_TMPFS_NODE(vp); 386 387 memset(&tfh, 0, sizeof(tfh)); 388 tfh.tf_len = sizeof(tmpfs_fid_t); 389 tfh.tf_gen = TMPFS_NODE_GEN(node); 390 tfh.tf_id = node->tn_id; 391 memcpy(fhp, &tfh, sizeof(tfh)); 392 393 return 0; 394 } 395 396 int 397 tmpfs_statvfs(struct mount *mp, struct statvfs *sbp) 398 { 399 tmpfs_mount_t *tmp; 400 fsfilcnt_t freenodes; 401 size_t avail; 402 403 tmp = VFS_TO_TMPFS(mp); 404 405 sbp->f_iosize = sbp->f_frsize = sbp->f_bsize = PAGE_SIZE; 406 407 mutex_enter(&tmp->tm_acc_lock); 408 avail = tmpfs_pages_avail(tmp); 409 sbp->f_blocks = (tmpfs_bytes_max(tmp) >> PAGE_SHIFT); 410 sbp->f_bavail = sbp->f_bfree = avail; 411 sbp->f_bresvd = 0; 412 413 freenodes = MIN(tmp->tm_nodes_max - tmp->tm_nodes_cnt, 414 avail * PAGE_SIZE / sizeof(tmpfs_node_t)); 415 416 sbp->f_files = tmp->tm_nodes_cnt + freenodes; 417 sbp->f_favail = sbp->f_ffree = freenodes; 418 sbp->f_fresvd = 0; 419 mutex_exit(&tmp->tm_acc_lock); 420 421 copy_statvfs_info(sbp, mp); 422 423 return 0; 424 } 425 426 int 427 tmpfs_sync(struct mount *mp, int waitfor, kauth_cred_t uc) 428 { 429 430 return 0; 431 } 432 433 int 434 tmpfs_snapshot(struct mount *mp, vnode_t *vp, struct timespec *ctime) 435 { 436 437 return EOPNOTSUPP; 438 } 439 440 /* 441 * tmpfs vfs operations. 442 */ 443 444 extern const struct vnodeopv_desc tmpfs_fifoop_opv_desc; 445 extern const struct vnodeopv_desc tmpfs_specop_opv_desc; 446 extern const struct vnodeopv_desc tmpfs_vnodeop_opv_desc; 447 448 const struct vnodeopv_desc * const tmpfs_vnodeopv_descs[] = { 449 &tmpfs_fifoop_opv_desc, 450 &tmpfs_specop_opv_desc, 451 &tmpfs_vnodeop_opv_desc, 452 NULL, 453 }; 454 455 struct vfsops tmpfs_vfsops = { 456 .vfs_name = MOUNT_TMPFS, 457 .vfs_min_mount_data = sizeof (struct tmpfs_args), 458 .vfs_mount = tmpfs_mount, 459 .vfs_start = tmpfs_start, 460 .vfs_unmount = tmpfs_unmount, 461 .vfs_root = tmpfs_root, 462 .vfs_quotactl = (void *)eopnotsupp, 463 .vfs_statvfs = tmpfs_statvfs, 464 .vfs_sync = tmpfs_sync, 465 .vfs_vget = tmpfs_vget, 466 .vfs_loadvnode = tmpfs_loadvnode, 467 .vfs_newvnode = tmpfs_newvnode, 468 .vfs_fhtovp = tmpfs_fhtovp, 469 .vfs_vptofh = tmpfs_vptofh, 470 .vfs_init = tmpfs_init, 471 .vfs_done = tmpfs_done, 472 .vfs_snapshot = tmpfs_snapshot, 473 .vfs_extattrctl = vfs_stdextattrctl, 474 .vfs_suspendctl = genfs_suspendctl, 475 .vfs_renamelock_enter = genfs_renamelock_enter, 476 .vfs_renamelock_exit = genfs_renamelock_exit, 477 .vfs_fsync = (void *)eopnotsupp, 478 .vfs_opv_descs = tmpfs_vnodeopv_descs 479 }; 480 481 static int 482 tmpfs_modcmd(modcmd_t cmd, void *arg) 483 { 484 485 switch (cmd) { 486 case MODULE_CMD_INIT: 487 return vfs_attach(&tmpfs_vfsops); 488 case MODULE_CMD_FINI: 489 return vfs_detach(&tmpfs_vfsops); 490 default: 491 return ENOTTY; 492 } 493 } 494