1 /* $NetBSD: tmpfs_vfsops.c,v 1.71 2017/03/01 10:44:47 hannken 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 1.71 2017/03/01 10:44:47 hannken 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 /* This can never happen? */ 136 if ((args->ta_root_mode & ALLPERMS) == VNOVAL) 137 return EINVAL; 138 139 /* Get the memory usage limit for this file-system. */ 140 if (args->ta_size_max < PAGE_SIZE) { 141 memlimit = UINT64_MAX; 142 set_memlimit = false; 143 } else { 144 memlimit = args->ta_size_max; 145 set_memlimit = true; 146 } 147 KASSERT(memlimit > 0); 148 149 if (args->ta_nodes_max <= 3) { 150 nodes = 3 + (memlimit / 1024); 151 set_nodes = false; 152 } else { 153 nodes = args->ta_nodes_max; 154 set_nodes = true; 155 } 156 nodes = MIN(nodes, INT_MAX); 157 KASSERT(nodes >= 3); 158 159 if (mp->mnt_flag & MNT_UPDATE) { 160 tmp = VFS_TO_TMPFS(mp); 161 if (set_nodes && nodes < tmp->tm_nodes_cnt) 162 return EBUSY; 163 if ((mp->mnt_iflag & IMNT_WANTRDONLY)) { 164 /* Changing from read/write to read-only. */ 165 flags = WRITECLOSE; 166 if ((mp->mnt_flag & MNT_FORCE)) 167 flags |= FORCECLOSE; 168 error = vflush(mp, NULL, flags); 169 if (error) 170 return error; 171 } 172 if (set_memlimit) { 173 if ((error = tmpfs_mntmem_set(tmp, memlimit)) != 0) 174 return error; 175 } 176 if (set_nodes) 177 tmp->tm_nodes_max = nodes; 178 root = tmp->tm_root; 179 root->tn_uid = args->ta_root_uid; 180 root->tn_gid = args->ta_root_gid; 181 root->tn_mode = args->ta_root_mode; 182 return 0; 183 } 184 185 /* Allocate the tmpfs mount structure and fill it. */ 186 tmp = kmem_zalloc(sizeof(tmpfs_mount_t), KM_SLEEP); 187 if (tmp == NULL) 188 return ENOMEM; 189 190 tmp->tm_nodes_max = nodes; 191 tmp->tm_nodes_cnt = 0; 192 LIST_INIT(&tmp->tm_nodes); 193 194 mutex_init(&tmp->tm_lock, MUTEX_DEFAULT, IPL_NONE); 195 tmpfs_mntmem_init(tmp, memlimit); 196 mp->mnt_data = tmp; 197 198 /* Allocate the root node. */ 199 vattr_null(&va); 200 va.va_type = VDIR; 201 va.va_mode = args->ta_root_mode & ALLPERMS; 202 va.va_uid = args->ta_root_uid; 203 va.va_gid = args->ta_root_gid; 204 error = vcache_new(mp, NULL, &va, NOCRED, &vp); 205 if (error) { 206 mp->mnt_data = NULL; 207 tmpfs_mntmem_destroy(tmp); 208 mutex_destroy(&tmp->tm_lock); 209 kmem_free(tmp, sizeof(*tmp)); 210 return error; 211 } 212 KASSERT(vp != NULL); 213 root = VP_TO_TMPFS_NODE(vp); 214 KASSERT(root != NULL); 215 216 /* 217 * Parent of the root inode is itself. Also, root inode has no 218 * directory entry (i.e. is never attached), thus hold an extra 219 * reference (link) for it. 220 */ 221 root->tn_links++; 222 root->tn_spec.tn_dir.tn_parent = root; 223 tmp->tm_root = root; 224 vrele(vp); 225 226 mp->mnt_flag |= MNT_LOCAL; 227 mp->mnt_stat.f_namemax = TMPFS_MAXNAMLEN; 228 mp->mnt_fs_bshift = PAGE_SHIFT; 229 mp->mnt_dev_bshift = DEV_BSHIFT; 230 mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO; 231 vfs_getnewfsid(mp); 232 233 error = set_statvfs_info(path, UIO_USERSPACE, "tmpfs", UIO_SYSSPACE, 234 mp->mnt_op->vfs_name, mp, curlwp); 235 if (error) { 236 (void)tmpfs_unmount(mp, MNT_FORCE); 237 } 238 return error; 239 } 240 241 int 242 tmpfs_start(struct mount *mp, int flags) 243 { 244 245 return 0; 246 } 247 248 int 249 tmpfs_unmount(struct mount *mp, int mntflags) 250 { 251 tmpfs_mount_t *tmp = VFS_TO_TMPFS(mp); 252 tmpfs_node_t *node, *cnode; 253 int error, flags = 0; 254 255 /* Handle forced unmounts. */ 256 if (mntflags & MNT_FORCE) 257 flags |= FORCECLOSE; 258 259 /* Finalize all pending I/O. */ 260 error = vflush(mp, NULL, flags); 261 if (error != 0) 262 return error; 263 264 /* 265 * First round, detach and destroy all directory entries. 266 * Also, clear the pointers to the vnodes - they are gone. 267 */ 268 LIST_FOREACH(node, &tmp->tm_nodes, tn_entries) { 269 tmpfs_dirent_t *de; 270 271 node->tn_vnode = NULL; 272 if (node->tn_type != VDIR) { 273 continue; 274 } 275 while ((de = TAILQ_FIRST(&node->tn_spec.tn_dir.tn_dir)) != NULL) { 276 cnode = de->td_node; 277 if (cnode && cnode != TMPFS_NODE_WHITEOUT) { 278 cnode->tn_vnode = NULL; 279 } 280 tmpfs_dir_detach(node, de); 281 tmpfs_free_dirent(tmp, de); 282 } 283 /* Extra virtual entry (itself for the root). */ 284 node->tn_links--; 285 } 286 287 /* Release the reference on root (diagnostic). */ 288 node = tmp->tm_root; 289 node->tn_links--; 290 291 /* Second round, destroy all inodes. */ 292 while ((node = LIST_FIRST(&tmp->tm_nodes)) != NULL) { 293 tmpfs_free_node(tmp, node); 294 } 295 296 /* Throw away the tmpfs_mount structure. */ 297 tmpfs_mntmem_destroy(tmp); 298 mutex_destroy(&tmp->tm_lock); 299 kmem_free(tmp, sizeof(*tmp)); 300 mp->mnt_data = NULL; 301 302 return 0; 303 } 304 305 int 306 tmpfs_root(struct mount *mp, vnode_t **vpp) 307 { 308 tmpfs_node_t *node = VFS_TO_TMPFS(mp)->tm_root; 309 int error; 310 311 error = vcache_get(mp, &node, sizeof(node), vpp); 312 if (error) 313 return error; 314 error = vn_lock(*vpp, LK_EXCLUSIVE); 315 if (error) { 316 vrele(*vpp); 317 *vpp = NULL; 318 return error; 319 } 320 321 return 0; 322 } 323 324 int 325 tmpfs_vget(struct mount *mp, ino_t ino, vnode_t **vpp) 326 { 327 328 return EOPNOTSUPP; 329 } 330 331 int 332 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, vnode_t **vpp) 333 { 334 tmpfs_mount_t *tmp = VFS_TO_TMPFS(mp); 335 tmpfs_node_t *node; 336 tmpfs_fid_t tfh; 337 int error; 338 339 if (fhp->fid_len != sizeof(tmpfs_fid_t)) { 340 return EINVAL; 341 } 342 memcpy(&tfh, fhp, sizeof(tmpfs_fid_t)); 343 344 mutex_enter(&tmp->tm_lock); 345 LIST_FOREACH(node, &tmp->tm_nodes, tn_entries) { 346 if (node->tn_id == tfh.tf_id) { 347 /* Prevent this node from disappearing. */ 348 atomic_inc_32(&node->tn_holdcount); 349 break; 350 } 351 } 352 mutex_exit(&tmp->tm_lock); 353 if (node == NULL) 354 return ESTALE; 355 356 error = vcache_get(mp, &node, sizeof(node), vpp); 357 /* If this node has been reclaimed free it now. */ 358 if (atomic_dec_32_nv(&node->tn_holdcount) == TMPFS_NODE_RECLAIMED) { 359 KASSERT(error != 0); 360 tmpfs_free_node(tmp, node); 361 } 362 if (error) 363 return (error == ENOENT ? ESTALE : error); 364 error = vn_lock(*vpp, LK_EXCLUSIVE); 365 if (error) { 366 vrele(*vpp); 367 *vpp = NULL; 368 return error; 369 } 370 if (TMPFS_NODE_GEN(node) != tfh.tf_gen) { 371 vput(*vpp); 372 *vpp = NULL; 373 return ESTALE; 374 } 375 376 return 0; 377 } 378 379 int 380 tmpfs_vptofh(vnode_t *vp, struct fid *fhp, size_t *fh_size) 381 { 382 tmpfs_fid_t tfh; 383 tmpfs_node_t *node; 384 385 if (*fh_size < sizeof(tmpfs_fid_t)) { 386 *fh_size = sizeof(tmpfs_fid_t); 387 return E2BIG; 388 } 389 *fh_size = sizeof(tmpfs_fid_t); 390 node = VP_TO_TMPFS_NODE(vp); 391 392 memset(&tfh, 0, sizeof(tfh)); 393 tfh.tf_len = sizeof(tmpfs_fid_t); 394 tfh.tf_gen = TMPFS_NODE_GEN(node); 395 tfh.tf_id = node->tn_id; 396 memcpy(fhp, &tfh, sizeof(tfh)); 397 398 return 0; 399 } 400 401 int 402 tmpfs_statvfs(struct mount *mp, struct statvfs *sbp) 403 { 404 tmpfs_mount_t *tmp; 405 fsfilcnt_t freenodes; 406 size_t avail; 407 408 tmp = VFS_TO_TMPFS(mp); 409 410 sbp->f_iosize = sbp->f_frsize = sbp->f_bsize = PAGE_SIZE; 411 412 mutex_enter(&tmp->tm_acc_lock); 413 avail = tmpfs_pages_avail(tmp); 414 sbp->f_blocks = (tmpfs_bytes_max(tmp) >> PAGE_SHIFT); 415 sbp->f_bavail = sbp->f_bfree = avail; 416 sbp->f_bresvd = 0; 417 418 freenodes = MIN(tmp->tm_nodes_max - tmp->tm_nodes_cnt, 419 avail * PAGE_SIZE / sizeof(tmpfs_node_t)); 420 421 sbp->f_files = tmp->tm_nodes_cnt + freenodes; 422 sbp->f_favail = sbp->f_ffree = freenodes; 423 sbp->f_fresvd = 0; 424 mutex_exit(&tmp->tm_acc_lock); 425 426 copy_statvfs_info(sbp, mp); 427 428 return 0; 429 } 430 431 int 432 tmpfs_sync(struct mount *mp, int waitfor, kauth_cred_t uc) 433 { 434 435 return 0; 436 } 437 438 int 439 tmpfs_snapshot(struct mount *mp, vnode_t *vp, struct timespec *ctime) 440 { 441 442 return EOPNOTSUPP; 443 } 444 445 /* 446 * tmpfs vfs operations. 447 */ 448 449 extern const struct vnodeopv_desc tmpfs_fifoop_opv_desc; 450 extern const struct vnodeopv_desc tmpfs_specop_opv_desc; 451 extern const struct vnodeopv_desc tmpfs_vnodeop_opv_desc; 452 453 const struct vnodeopv_desc * const tmpfs_vnodeopv_descs[] = { 454 &tmpfs_fifoop_opv_desc, 455 &tmpfs_specop_opv_desc, 456 &tmpfs_vnodeop_opv_desc, 457 NULL, 458 }; 459 460 struct vfsops tmpfs_vfsops = { 461 .vfs_name = MOUNT_TMPFS, 462 .vfs_min_mount_data = sizeof (struct tmpfs_args), 463 .vfs_mount = tmpfs_mount, 464 .vfs_start = tmpfs_start, 465 .vfs_unmount = tmpfs_unmount, 466 .vfs_root = tmpfs_root, 467 .vfs_quotactl = (void *)eopnotsupp, 468 .vfs_statvfs = tmpfs_statvfs, 469 .vfs_sync = tmpfs_sync, 470 .vfs_vget = tmpfs_vget, 471 .vfs_loadvnode = tmpfs_loadvnode, 472 .vfs_newvnode = tmpfs_newvnode, 473 .vfs_fhtovp = tmpfs_fhtovp, 474 .vfs_vptofh = tmpfs_vptofh, 475 .vfs_init = tmpfs_init, 476 .vfs_done = tmpfs_done, 477 .vfs_snapshot = tmpfs_snapshot, 478 .vfs_extattrctl = vfs_stdextattrctl, 479 .vfs_suspendctl = genfs_suspendctl, 480 .vfs_renamelock_enter = genfs_renamelock_enter, 481 .vfs_renamelock_exit = genfs_renamelock_exit, 482 .vfs_fsync = (void *)eopnotsupp, 483 .vfs_opv_descs = tmpfs_vnodeopv_descs 484 }; 485 486 static int 487 tmpfs_modcmd(modcmd_t cmd, void *arg) 488 { 489 490 switch (cmd) { 491 case MODULE_CMD_INIT: 492 return vfs_attach(&tmpfs_vfsops); 493 case MODULE_CMD_FINI: 494 return vfs_detach(&tmpfs_vfsops); 495 default: 496 return ENOTTY; 497 } 498 } 499