1/* $NetBSD: tmpfs.h,v 1.26 2007/02/22 06:37:00 thorpej Exp $ */ 2 3/*- 4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD 5 * 6 * Copyright (c) 2005, 2006 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 * $FreeBSD$ 35 */ 36 37#ifndef _FS_TMPFS_TMPFS_H_ 38#define _FS_TMPFS_TMPFS_H_ 39 40#include <sys/cdefs.h> 41#include <sys/queue.h> 42#include <sys/tree.h> 43 44#ifdef _SYS_MALLOC_H_ 45MALLOC_DECLARE(M_TMPFSNAME); 46#endif 47 48/* 49 * Internal representation of a tmpfs directory entry. 50 */ 51 52LIST_HEAD(tmpfs_dir_duphead, tmpfs_dirent); 53 54struct tmpfs_dirent { 55 /* 56 * Depending on td_cookie flag entry can be of 3 types: 57 * - regular -- no hash collisions, stored in RB-Tree 58 * - duphead -- synthetic linked list head for dup entries 59 * - dup -- stored in linked list instead of RB-Tree 60 */ 61 union { 62 /* regular and duphead entry types */ 63 RB_ENTRY(tmpfs_dirent) td_entries; 64 65 /* dup entry type */ 66 struct { 67 LIST_ENTRY(tmpfs_dirent) entries; 68 LIST_ENTRY(tmpfs_dirent) index_entries; 69 } td_dup; 70 } uh; 71 72 uint32_t td_cookie; 73 uint32_t td_hash; 74 u_int td_namelen; 75 76 /* 77 * Pointer to the node this entry refers to. In case this field 78 * is NULL, the node is a whiteout. 79 */ 80 struct tmpfs_node * td_node; 81 82 union { 83 /* 84 * The name of the entry, allocated from a string pool. This 85 * string is not required to be zero-terminated. 86 */ 87 char * td_name; /* regular, dup */ 88 struct tmpfs_dir_duphead td_duphead; /* duphead */ 89 } ud; 90}; 91 92/* 93 * A directory in tmpfs holds a collection of directory entries, which 94 * in turn point to other files (which can be directories themselves). 95 * 96 * In tmpfs, this collection is managed by a RB-Tree, whose head is 97 * defined by the struct tmpfs_dir type. 98 * 99 * It is important to notice that directories do not have entries for . and 100 * .. as other file systems do. These can be generated when requested 101 * based on information available by other means, such as the pointer to 102 * the node itself in the former case or the pointer to the parent directory 103 * in the latter case. This is done to simplify tmpfs's code and, more 104 * importantly, to remove redundancy. 105 */ 106RB_HEAD(tmpfs_dir, tmpfs_dirent); 107 108/* 109 * Each entry in a directory has a cookie that identifies it. Cookies 110 * supersede offsets within directories because, given how tmpfs stores 111 * directories in memory, there is no such thing as an offset. 112 * 113 * The '.', '..' and the end of directory markers have fixed cookies which 114 * cannot collide with the cookies generated by other entries. The cookies 115 * for the other entries are generated based on the file name hash value or 116 * unique number in case of name hash collision. 117 * 118 * To preserve compatibility cookies are limited to 31 bits. 119 */ 120 121#define TMPFS_DIRCOOKIE_DOT 0 122#define TMPFS_DIRCOOKIE_DOTDOT 1 123#define TMPFS_DIRCOOKIE_EOF 2 124#define TMPFS_DIRCOOKIE_MASK ((off_t)0x3fffffffU) 125#define TMPFS_DIRCOOKIE_MIN ((off_t)0x00000004U) 126#define TMPFS_DIRCOOKIE_DUP ((off_t)0x40000000U) 127#define TMPFS_DIRCOOKIE_DUPHEAD ((off_t)0x80000000U) 128#define TMPFS_DIRCOOKIE_DUP_MIN TMPFS_DIRCOOKIE_DUP 129#define TMPFS_DIRCOOKIE_DUP_MAX \ 130 (TMPFS_DIRCOOKIE_DUP | TMPFS_DIRCOOKIE_MASK) 131 132/* 133 * Internal representation of a tmpfs file system node. 134 * 135 * This structure is splitted in two parts: one holds attributes common 136 * to all file types and the other holds data that is only applicable to 137 * a particular type. The code must be careful to only access those 138 * attributes that are actually allowed by the node's type. 139 * 140 * Below is the key of locks used to protected the fields in the following 141 * structures. 142 * (v) vnode lock in exclusive mode 143 * (vi) vnode lock in exclusive mode, or vnode lock in shared vnode and 144 * tn_interlock 145 * (i) tn_interlock 146 * (m) tmpfs_mount tm_allnode_lock 147 * (c) stable after creation 148 */ 149struct tmpfs_node { 150 /* 151 * Doubly-linked list entry which links all existing nodes for 152 * a single file system. This is provided to ease the removal 153 * of all nodes during the unmount operation, and to support 154 * the implementation of VOP_VNTOCNP(). tn_attached is false 155 * when the node is removed from list and unlocked. 156 */ 157 LIST_ENTRY(tmpfs_node) tn_entries; /* (m) */ 158 bool tn_attached; /* (m) */ 159 160 /* 161 * The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', 162 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode 163 * types instead of a custom enumeration is to make things simpler 164 * and faster, as we do not need to convert between two types. 165 */ 166 enum vtype tn_type; /* (c) */ 167 168 /* Node identifier. */ 169 ino_t tn_id; /* (c) */ 170 171 /* 172 * Node's internal status. This is used by several file system 173 * operations to do modifications to the node in a delayed 174 * fashion. 175 */ 176 int tn_status; /* (vi) */ 177#define TMPFS_NODE_ACCESSED (1 << 1) 178#define TMPFS_NODE_MODIFIED (1 << 2) 179#define TMPFS_NODE_CHANGED (1 << 3) 180 181 /* 182 * The node size. It does not necessarily match the real amount 183 * of memory consumed by it. 184 */ 185 off_t tn_size; /* (v) */ 186 187 /* Generic node attributes. */ 188 uid_t tn_uid; /* (v) */ 189 gid_t tn_gid; /* (v) */ 190 mode_t tn_mode; /* (v) */ 191 int tn_links; /* (v) */ 192 u_long tn_flags; /* (v) */ 193 struct timespec tn_atime; /* (vi) */ 194 struct timespec tn_mtime; /* (vi) */ 195 struct timespec tn_ctime; /* (vi) */ 196 struct timespec tn_birthtime; /* (v) */ 197 unsigned long tn_gen; /* (c) */ 198 199 /* 200 * As there is a single vnode for each active file within the 201 * system, care has to be taken to avoid allocating more than one 202 * vnode per file. In order to do this, a bidirectional association 203 * is kept between vnodes and nodes. 204 * 205 * Whenever a vnode is allocated, its v_data field is updated to 206 * point to the node it references. At the same time, the node's 207 * tn_vnode field is modified to point to the new vnode representing 208 * it. Further attempts to allocate a vnode for this same node will 209 * result in returning a new reference to the value stored in 210 * tn_vnode. 211 * 212 * May be NULL when the node is unused (that is, no vnode has been 213 * allocated for it or it has been reclaimed). 214 */ 215 struct vnode * tn_vnode; /* (i) */ 216 217 /* 218 * Interlock to protect tn_vpstate, and tn_status under shared 219 * vnode lock. 220 */ 221 struct mtx tn_interlock; 222 223 /* 224 * Identify if current node has vnode assiocate with 225 * or allocating vnode. 226 */ 227 int tn_vpstate; /* (i) */ 228 229 /* Transient refcounter on this node. */ 230 u_int tn_refcount; /* (m) + (i) */ 231 232 /* misc data field for different tn_type node */ 233 union { 234 /* Valid when tn_type == VBLK || tn_type == VCHR. */ 235 dev_t tn_rdev; /* (c) */ 236 237 /* Valid when tn_type == VDIR. */ 238 struct tn_dir { 239 /* 240 * Pointer to the parent directory. The root 241 * directory has a pointer to itself in this field; 242 * this property identifies the root node. 243 */ 244 struct tmpfs_node * tn_parent; 245 246 /* 247 * Head of a tree that links the contents of 248 * the directory together. 249 */ 250 struct tmpfs_dir tn_dirhead; 251 252 /* 253 * Head of a list the contains fake directory entries 254 * heads, i.e. entries with TMPFS_DIRCOOKIE_DUPHEAD 255 * flag. 256 */ 257 struct tmpfs_dir_duphead tn_dupindex; 258 259 /* 260 * Number and pointer of the first directory entry 261 * returned by the readdir operation if it were 262 * called again to continue reading data from the 263 * same directory as before. This is used to speed 264 * up reads of long directories, assuming that no 265 * more than one read is in progress at a given time. 266 * Otherwise, these values are discarded. 267 */ 268 off_t tn_readdir_lastn; 269 struct tmpfs_dirent * tn_readdir_lastp; 270 } tn_dir; 271 272 /* Valid when tn_type == VLNK. */ 273 /* The link's target, allocated from a string pool. */ 274 char * tn_link; /* (c) */ 275 276 /* Valid when tn_type == VREG. */ 277 struct tn_reg { 278 /* 279 * The contents of regular files stored in a 280 * tmpfs file system are represented by a 281 * single anonymous memory object (aobj, for 282 * short). The aobj provides direct access to 283 * any position within the file. It is a task 284 * of the memory management subsystem to issue 285 * the required page ins or page outs whenever 286 * a position within the file is accessed. 287 */ 288 vm_object_t tn_aobj; /* (c) */ 289 } tn_reg; 290 } tn_spec; /* (v) */ 291}; 292LIST_HEAD(tmpfs_node_list, tmpfs_node); 293 294#define tn_rdev tn_spec.tn_rdev 295#define tn_dir tn_spec.tn_dir 296#define tn_link tn_spec.tn_link 297#define tn_reg tn_spec.tn_reg 298#define tn_fifo tn_spec.tn_fifo 299 300#define TMPFS_LINK_MAX INT_MAX 301 302#define TMPFS_NODE_LOCK(node) mtx_lock(&(node)->tn_interlock) 303#define TMPFS_NODE_UNLOCK(node) mtx_unlock(&(node)->tn_interlock) 304#define TMPFS_NODE_MTX(node) (&(node)->tn_interlock) 305#define TMPFS_NODE_ASSERT_LOCKED(node) mtx_assert(TMPFS_NODE_MTX(node), \ 306 MA_OWNED) 307 308#ifdef INVARIANTS 309#define TMPFS_ASSERT_LOCKED(node) do { \ 310 MPASS((node) != NULL); \ 311 MPASS((node)->tn_vnode != NULL); \ 312 ASSERT_VOP_LOCKED((node)->tn_vnode, "tmpfs assert"); \ 313 } while (0) 314#else 315#define TMPFS_ASSERT_LOCKED(node) (void)0 316#endif 317 318#define TMPFS_VNODE_ALLOCATING 1 319#define TMPFS_VNODE_WANT 2 320#define TMPFS_VNODE_DOOMED 4 321#define TMPFS_VNODE_WRECLAIM 8 322 323/* 324 * Internal representation of a tmpfs mount point. 325 */ 326struct tmpfs_mount { 327 /* 328 * Original value of the "size" parameter, for reference purposes, 329 * mostly. 330 */ 331 off_t tm_size_max; 332 /* 333 * Maximum number of memory pages available for use by the file 334 * system, set during mount time. This variable must never be 335 * used directly as it may be bigger than the current amount of 336 * free memory; in the extreme case, it will hold the ULONG_MAX 337 * value. 338 */ 339 u_long tm_pages_max; 340 341 /* Number of pages in use by the file system. */ 342 u_long tm_pages_used; 343 344 /* 345 * Pointer to the node representing the root directory of this 346 * file system. 347 */ 348 struct tmpfs_node * tm_root; 349 350 /* 351 * Maximum number of possible nodes for this file system; set 352 * during mount time. We need a hard limit on the maximum number 353 * of nodes to avoid allocating too much of them; their objects 354 * cannot be released until the file system is unmounted. 355 * Otherwise, we could easily run out of memory by creating lots 356 * of empty files and then simply removing them. 357 */ 358 ino_t tm_nodes_max; 359 360 /* unrhdr used to allocate inode numbers */ 361 struct unrhdr64 tm_ino_unr; 362 363 /* Number of nodes currently that are in use. */ 364 ino_t tm_nodes_inuse; 365 366 /* Refcounter on this struct tmpfs_mount. */ 367 uint64_t tm_refcount; 368 369 /* maximum representable file size */ 370 u_int64_t tm_maxfilesize; 371 372 /* 373 * The used list contains all nodes that are currently used by 374 * the file system; i.e., they refer to existing files. 375 */ 376 struct tmpfs_node_list tm_nodes_used; 377 378 /* All node lock to protect the node list and tmp_pages_used. */ 379 struct mtx tm_allnode_lock; 380 381 /* Read-only status. */ 382 bool tm_ronly; 383 /* Do not use namecache. */ 384 bool tm_nonc; 385}; 386#define TMPFS_LOCK(tm) mtx_lock(&(tm)->tm_allnode_lock) 387#define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->tm_allnode_lock) 388#define TMPFS_MP_ASSERT_LOCKED(tm) mtx_assert(&(tm)->tm_allnode_lock, MA_OWNED) 389 390/* 391 * This structure maps a file identifier to a tmpfs node. Used by the 392 * NFS code. 393 */ 394struct tmpfs_fid_data { 395 ino_t tfd_id; 396 unsigned long tfd_gen; 397}; 398_Static_assert(sizeof(struct tmpfs_fid_data) <= MAXFIDSZ, 399 "(struct tmpfs_fid_data) is larger than (struct fid).fid_data"); 400 401struct tmpfs_dir_cursor { 402 struct tmpfs_dirent *tdc_current; 403 struct tmpfs_dirent *tdc_tree; 404}; 405 406#ifdef _KERNEL 407/* 408 * Prototypes for tmpfs_subr.c. 409 */ 410 411void tmpfs_ref_node(struct tmpfs_node *node); 412void tmpfs_ref_node_locked(struct tmpfs_node *node); 413int tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *, enum vtype, 414 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *, 415 char *, dev_t, struct tmpfs_node **); 416void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 417bool tmpfs_free_node_locked(struct tmpfs_mount *, struct tmpfs_node *, bool); 418void tmpfs_free_tmp(struct tmpfs_mount *); 419int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 420 const char *, u_int, struct tmpfs_dirent **); 421void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *); 422void tmpfs_dirent_init(struct tmpfs_dirent *, const char *, u_int); 423void tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj); 424int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int, 425 struct vnode **); 426void tmpfs_free_vp(struct vnode *); 427int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 428 struct componentname *, char *); 429void tmpfs_check_mtime(struct vnode *); 430void tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *); 431void tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *); 432void tmpfs_dir_destroy(struct tmpfs_mount *, struct tmpfs_node *); 433struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 434 struct tmpfs_node *f, 435 struct componentname *cnp); 436int tmpfs_dir_getdents(struct tmpfs_mount *, struct tmpfs_node *, 437 struct uio *, int, u_long *, int *); 438int tmpfs_dir_whiteout_add(struct vnode *, struct componentname *); 439void tmpfs_dir_whiteout_remove(struct vnode *, struct componentname *); 440int tmpfs_reg_resize(struct vnode *, off_t, boolean_t); 441int tmpfs_chflags(struct vnode *, u_long, struct ucred *, struct thread *); 442int tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct thread *); 443int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *, 444 struct thread *); 445int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct thread *); 446int tmpfs_chtimes(struct vnode *, struct vattr *, struct ucred *cred, 447 struct thread *); 448void tmpfs_itimes(struct vnode *, const struct timespec *, 449 const struct timespec *); 450 451void tmpfs_set_status(struct tmpfs_mount *tm, struct tmpfs_node *node, 452 int status); 453void tmpfs_update(struct vnode *); 454int tmpfs_truncate(struct vnode *, off_t); 455struct tmpfs_dirent *tmpfs_dir_first(struct tmpfs_node *dnode, 456 struct tmpfs_dir_cursor *dc); 457struct tmpfs_dirent *tmpfs_dir_next(struct tmpfs_node *dnode, 458 struct tmpfs_dir_cursor *dc); 459 460/* 461 * Convenience macros to simplify some logical expressions. 462 */ 463#define IMPLIES(a, b) (!(a) || (b)) 464#define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 465 466/* 467 * Checks that the directory entry pointed by 'de' matches the name 'name' 468 * with a length of 'len'. 469 */ 470#define TMPFS_DIRENT_MATCHES(de, name, len) \ 471 (de->td_namelen == len && \ 472 bcmp((de)->ud.td_name, (name), (de)->td_namelen) == 0) 473 474/* 475 * Ensures that the node pointed by 'node' is a directory and that its 476 * contents are consistent with respect to directories. 477 */ 478#define TMPFS_VALIDATE_DIR(node) do { \ 479 MPASS((node)->tn_type == VDIR); \ 480 MPASS((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 481} while (0) 482 483/* 484 * Amount of memory pages to reserve for the system (e.g., to not use by 485 * tmpfs). 486 */ 487#define TMPFS_PAGES_MINRESERVED (4 * 1024 * 1024 / PAGE_SIZE) 488 489size_t tmpfs_mem_avail(void); 490size_t tmpfs_pages_used(struct tmpfs_mount *tmp); 491void tmpfs_subr_init(void); 492void tmpfs_subr_uninit(void); 493 494/* 495 * Macros/functions to convert from generic data structures to tmpfs 496 * specific ones. 497 */ 498 499static inline struct tmpfs_mount * 500VFS_TO_TMPFS(struct mount *mp) 501{ 502 struct tmpfs_mount *tmp; 503 504 MPASS(mp != NULL && mp->mnt_data != NULL); 505 tmp = (struct tmpfs_mount *)mp->mnt_data; 506 return (tmp); 507} 508 509static inline struct tmpfs_node * 510VP_TO_TMPFS_NODE(struct vnode *vp) 511{ 512 struct tmpfs_node *node; 513 514 MPASS(vp != NULL && vp->v_data != NULL); 515 node = (struct tmpfs_node *)vp->v_data; 516 return (node); 517} 518 519static inline struct tmpfs_node * 520VP_TO_TMPFS_DIR(struct vnode *vp) 521{ 522 struct tmpfs_node *node; 523 524 node = VP_TO_TMPFS_NODE(vp); 525 TMPFS_VALIDATE_DIR(node); 526 return (node); 527} 528 529static inline bool 530tmpfs_use_nc(struct vnode *vp) 531{ 532 533 return (!(VFS_TO_TMPFS(vp->v_mount)->tm_nonc)); 534} 535#endif /* _KERNEL */ 536 537#endif /* _FS_TMPFS_TMPFS_H_ */ 538