1 /* $NetBSD: tmpfs.h,v 1.26 2007/02/22 06:37:00 thorpej Exp $ */ 2 3 /*- 4 * Copyright (c) 2005, 2006 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 * $FreeBSD$ 33 */ 34 35 #ifndef _FS_TMPFS_TMPFS_H_ 36 #define _FS_TMPFS_TMPFS_H_ 37 38 #include <sys/dirent.h> 39 #include <sys/mount.h> 40 #include <sys/queue.h> 41 #include <sys/vnode.h> 42 #include <sys/file.h> 43 #include <sys/lock.h> 44 #include <sys/mutex.h> 45 46 #include <sys/malloc.h> 47 #include <sys/systm.h> 48 #include <sys/tree.h> 49 #include <sys/vmmeter.h> 50 #include <vm/swap_pager.h> 51 52 MALLOC_DECLARE(M_TMPFSMNT); 53 MALLOC_DECLARE(M_TMPFSNAME); 54 55 /* 56 * Internal representation of a tmpfs directory entry. 57 */ 58 59 LIST_HEAD(tmpfs_dir_duphead, tmpfs_dirent); 60 61 struct tmpfs_dirent { 62 /* 63 * Depending on td_cookie flag entry can be of 3 types: 64 * - regular -- no hash collisions, stored in RB-Tree 65 * - duphead -- synthetic linked list head for dup entries 66 * - dup -- stored in linked list instead of RB-Tree 67 */ 68 union { 69 /* regular and duphead entry types */ 70 RB_ENTRY(tmpfs_dirent) td_entries; 71 72 /* dup entry type */ 73 struct { 74 LIST_ENTRY(tmpfs_dirent) entries; 75 LIST_ENTRY(tmpfs_dirent) index_entries; 76 } td_dup; 77 } uh; 78 79 uint32_t td_cookie; 80 uint32_t td_hash; 81 u_int td_namelen; 82 83 /* 84 * Pointer to the node this entry refers to. In case this field 85 * is NULL, the node is a whiteout. 86 */ 87 struct tmpfs_node * td_node; 88 89 union { 90 /* 91 * The name of the entry, allocated from a string pool. This 92 * string is not required to be zero-terminated. 93 */ 94 char * td_name; /* regular, dup */ 95 struct tmpfs_dir_duphead td_duphead; /* duphead */ 96 } ud; 97 }; 98 99 /* 100 * A directory in tmpfs holds a collection of directory entries, which 101 * in turn point to other files (which can be directories themselves). 102 * 103 * In tmpfs, this collection is managed by a RB-Tree, whose head is 104 * defined by the struct tmpfs_dir type. 105 * 106 * It is important to notice that directories do not have entries for . and 107 * .. as other file systems do. These can be generated when requested 108 * based on information available by other means, such as the pointer to 109 * the node itself in the former case or the pointer to the parent directory 110 * in the latter case. This is done to simplify tmpfs's code and, more 111 * importantly, to remove redundancy. 112 */ 113 RB_HEAD(tmpfs_dir, tmpfs_dirent); 114 115 /* 116 * Each entry in a directory has a cookie that identifies it. Cookies 117 * supersede offsets within directories because, given how tmpfs stores 118 * directories in memory, there is no such thing as an offset. 119 * 120 * The '.', '..' and the end of directory markers have fixed cookies which 121 * cannot collide with the cookies generated by other entries. The cookies 122 * for the other entries are generated based on the file name hash value or 123 * unique number in case of name hash collision. 124 * 125 * To preserve compatibility cookies are limited to 31 bits. 126 */ 127 128 #define TMPFS_DIRCOOKIE_DOT 0 129 #define TMPFS_DIRCOOKIE_DOTDOT 1 130 #define TMPFS_DIRCOOKIE_EOF 2 131 #define TMPFS_DIRCOOKIE_MASK ((off_t)0x3fffffffU) 132 #define TMPFS_DIRCOOKIE_MIN ((off_t)0x00000004U) 133 #define TMPFS_DIRCOOKIE_DUP ((off_t)0x40000000U) 134 #define TMPFS_DIRCOOKIE_DUPHEAD ((off_t)0x80000000U) 135 #define TMPFS_DIRCOOKIE_DUP_MIN TMPFS_DIRCOOKIE_DUP 136 #define TMPFS_DIRCOOKIE_DUP_MAX \ 137 (TMPFS_DIRCOOKIE_DUP | TMPFS_DIRCOOKIE_MASK) 138 139 /* 140 * Internal representation of a tmpfs file system node. 141 * 142 * This structure is splitted in two parts: one holds attributes common 143 * to all file types and the other holds data that is only applicable to 144 * a particular type. The code must be careful to only access those 145 * attributes that are actually allowed by the node's type. 146 * 147 * Below is the key of locks used to protected the fields in the following 148 * structures. 149 * (v) vnode lock in exclusive mode 150 * (vi) vnode lock in exclusive mode, or vnode lock in shared vnode and 151 * tn_interlock 152 * (i) tn_interlock 153 * (m) tmpfs_mount tm_allnode_lock 154 * (c) stable after creation 155 */ 156 struct tmpfs_node { 157 /* 158 * Doubly-linked list entry which links all existing nodes for 159 * a single file system. This is provided to ease the removal 160 * of all nodes during the unmount operation, and to support 161 * the implementation of VOP_VNTOCNP(). tn_attached is false 162 * when the node is removed from list and unlocked. 163 */ 164 LIST_ENTRY(tmpfs_node) tn_entries; /* (m) */ 165 bool tn_attached; /* (m) */ 166 167 /* 168 * The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', 169 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode 170 * types instead of a custom enumeration is to make things simpler 171 * and faster, as we do not need to convert between two types. 172 */ 173 enum vtype tn_type; /* (c) */ 174 175 /* Node identifier. */ 176 ino_t tn_id; /* (c) */ 177 178 /* 179 * Node's internal status. This is used by several file system 180 * operations to do modifications to the node in a delayed 181 * fashion. 182 */ 183 int tn_status; /* (vi) */ 184 #define TMPFS_NODE_ACCESSED (1 << 1) 185 #define TMPFS_NODE_MODIFIED (1 << 2) 186 #define TMPFS_NODE_CHANGED (1 << 3) 187 188 /* 189 * The node size. It does not necessarily match the real amount 190 * of memory consumed by it. 191 */ 192 off_t tn_size; /* (v) */ 193 194 /* Generic node attributes. */ 195 uid_t tn_uid; /* (v) */ 196 gid_t tn_gid; /* (v) */ 197 mode_t tn_mode; /* (v) */ 198 u_long tn_flags; /* (v) */ 199 nlink_t tn_links; /* (v) */ 200 struct timespec tn_atime; /* (vi) */ 201 struct timespec tn_mtime; /* (vi) */ 202 struct timespec tn_ctime; /* (vi) */ 203 struct timespec tn_birthtime; /* (v) */ 204 unsigned long tn_gen; /* (c) */ 205 206 /* 207 * As there is a single vnode for each active file within the 208 * system, care has to be taken to avoid allocating more than one 209 * vnode per file. In order to do this, a bidirectional association 210 * is kept between vnodes and nodes. 211 * 212 * Whenever a vnode is allocated, its v_data field is updated to 213 * point to the node it references. At the same time, the node's 214 * tn_vnode field is modified to point to the new vnode representing 215 * it. Further attempts to allocate a vnode for this same node will 216 * result in returning a new reference to the value stored in 217 * tn_vnode. 218 * 219 * May be NULL when the node is unused (that is, no vnode has been 220 * allocated for it or it has been reclaimed). 221 */ 222 struct vnode * tn_vnode; /* (i) */ 223 224 /* 225 * Interlock to protect tn_vpstate, and tn_status under shared 226 * vnode lock. 227 */ 228 struct mtx tn_interlock; 229 230 /* 231 * Identify if current node has vnode assiocate with 232 * or allocating vnode. 233 */ 234 int tn_vpstate; /* (i) */ 235 236 /* Transient refcounter on this node. */ 237 u_int tn_refcount; /* (m) + (i) */ 238 239 /* misc data field for different tn_type node */ 240 union { 241 /* Valid when tn_type == VBLK || tn_type == VCHR. */ 242 dev_t tn_rdev; /* (c) */ 243 244 /* Valid when tn_type == VDIR. */ 245 struct tn_dir { 246 /* 247 * Pointer to the parent directory. The root 248 * directory has a pointer to itself in this field; 249 * this property identifies the root node. 250 */ 251 struct tmpfs_node * tn_parent; 252 253 /* 254 * Head of a tree that links the contents of 255 * the directory together. 256 */ 257 struct tmpfs_dir tn_dirhead; 258 259 /* 260 * Head of a list the contains fake directory entries 261 * heads, i.e. entries with TMPFS_DIRCOOKIE_DUPHEAD 262 * flag. 263 */ 264 struct tmpfs_dir_duphead tn_dupindex; 265 266 /* 267 * Number and pointer of the first directory entry 268 * returned by the readdir operation if it were 269 * called again to continue reading data from the 270 * same directory as before. This is used to speed 271 * up reads of long directories, assuming that no 272 * more than one read is in progress at a given time. 273 * Otherwise, these values are discarded. 274 */ 275 off_t tn_readdir_lastn; 276 struct tmpfs_dirent * tn_readdir_lastp; 277 } tn_dir; 278 279 /* Valid when tn_type == VLNK. */ 280 /* The link's target, allocated from a string pool. */ 281 char * tn_link; /* (c) */ 282 283 /* Valid when tn_type == VREG. */ 284 struct tn_reg { 285 /* 286 * The contents of regular files stored in a 287 * tmpfs file system are represented by a 288 * single anonymous memory object (aobj, for 289 * short). The aobj provides direct access to 290 * any position within the file. It is a task 291 * of the memory management subsystem to issue 292 * the required page ins or page outs whenever 293 * a position within the file is accessed. 294 */ 295 vm_object_t tn_aobj; /* (c) */ 296 } tn_reg; 297 } tn_spec; /* (v) */ 298 }; 299 LIST_HEAD(tmpfs_node_list, tmpfs_node); 300 301 #define tn_rdev tn_spec.tn_rdev 302 #define tn_dir tn_spec.tn_dir 303 #define tn_link tn_spec.tn_link 304 #define tn_reg tn_spec.tn_reg 305 #define tn_fifo tn_spec.tn_fifo 306 307 #define TMPFS_NODE_LOCK(node) mtx_lock(&(node)->tn_interlock) 308 #define TMPFS_NODE_UNLOCK(node) mtx_unlock(&(node)->tn_interlock) 309 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock) 310 #define TMPFS_NODE_ASSERT_LOCKED(node) mtx_assert(TMPFS_NODE_MTX(node), \ 311 MA_OWNED) 312 313 #ifdef INVARIANTS 314 #define TMPFS_ASSERT_LOCKED(node) do { \ 315 MPASS((node) != NULL); \ 316 MPASS((node)->tn_vnode != NULL); \ 317 ASSERT_VOP_LOCKED((node)->tn_vnode, "tmpfs assert"); \ 318 } while (0) 319 #else 320 #define TMPFS_ASSERT_LOCKED(node) (void)0 321 #endif 322 323 #define TMPFS_VNODE_ALLOCATING 1 324 #define TMPFS_VNODE_WANT 2 325 #define TMPFS_VNODE_DOOMED 4 326 #define TMPFS_VNODE_WRECLAIM 8 327 328 /* 329 * Internal representation of a tmpfs mount point. 330 */ 331 struct tmpfs_mount { 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 unrhdr * 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 /* Zones used to store file system meta data, per tmpfs mount. */ 382 uma_zone_t tm_dirent_pool; 383 uma_zone_t tm_node_pool; 384 385 /* Read-only status. */ 386 bool tm_ronly; 387 /* Do not use namecache. */ 388 bool tm_nonc; 389 }; 390 #define TMPFS_LOCK(tm) mtx_lock(&(tm)->tm_allnode_lock) 391 #define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->tm_allnode_lock) 392 #define TMPFS_MP_ASSERT_LOCKED(tm) mtx_assert(&(tm)->tm_allnode_lock, MA_OWNED) 393 394 /* 395 * This structure maps a file identifier to a tmpfs node. Used by the 396 * NFS code. 397 */ 398 struct tmpfs_fid { 399 uint16_t tf_len; 400 uint16_t tf_pad; 401 ino_t tf_id; 402 unsigned long tf_gen; 403 }; 404 405 struct tmpfs_dir_cursor { 406 struct tmpfs_dirent *tdc_current; 407 struct tmpfs_dirent *tdc_tree; 408 }; 409 410 #ifdef _KERNEL 411 /* 412 * Prototypes for tmpfs_subr.c. 413 */ 414 415 void tmpfs_ref_node(struct tmpfs_node *node); 416 void tmpfs_ref_node_locked(struct tmpfs_node *node); 417 int tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *, enum vtype, 418 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *, 419 char *, dev_t, struct tmpfs_node **); 420 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 421 bool tmpfs_free_node_locked(struct tmpfs_mount *, struct tmpfs_node *, bool); 422 void tmpfs_free_tmp(struct tmpfs_mount *); 423 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 424 const char *, u_int, struct tmpfs_dirent **); 425 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *); 426 void tmpfs_dirent_init(struct tmpfs_dirent *, const char *, u_int); 427 void tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj); 428 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int, 429 struct vnode **); 430 void tmpfs_free_vp(struct vnode *); 431 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 432 struct componentname *, char *); 433 void tmpfs_check_mtime(struct vnode *); 434 void tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *); 435 void tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *); 436 void tmpfs_dir_destroy(struct tmpfs_mount *, struct tmpfs_node *); 437 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 438 struct tmpfs_node *f, 439 struct componentname *cnp); 440 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, int, 441 u_long *, int *); 442 int tmpfs_dir_whiteout_add(struct vnode *, struct componentname *); 443 void tmpfs_dir_whiteout_remove(struct vnode *, struct componentname *); 444 int tmpfs_reg_resize(struct vnode *, off_t, boolean_t); 445 int tmpfs_chflags(struct vnode *, u_long, struct ucred *, struct thread *); 446 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct thread *); 447 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *, 448 struct thread *); 449 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct thread *); 450 int tmpfs_chtimes(struct vnode *, struct vattr *, struct ucred *cred, 451 struct thread *); 452 void tmpfs_itimes(struct vnode *, const struct timespec *, 453 const struct timespec *); 454 455 void tmpfs_set_status(struct tmpfs_node *node, int status); 456 void tmpfs_update(struct vnode *); 457 int tmpfs_truncate(struct vnode *, off_t); 458 struct tmpfs_dirent *tmpfs_dir_first(struct tmpfs_node *dnode, 459 struct tmpfs_dir_cursor *dc); 460 struct tmpfs_dirent *tmpfs_dir_next(struct tmpfs_node *dnode, 461 struct tmpfs_dir_cursor *dc); 462 463 /* 464 * Convenience macros to simplify some logical expressions. 465 */ 466 #define IMPLIES(a, b) (!(a) || (b)) 467 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 468 469 /* 470 * Checks that the directory entry pointed by 'de' matches the name 'name' 471 * with a length of 'len'. 472 */ 473 #define TMPFS_DIRENT_MATCHES(de, name, len) \ 474 (de->td_namelen == len && \ 475 bcmp((de)->ud.td_name, (name), (de)->td_namelen) == 0) 476 477 /* 478 * Ensures that the node pointed by 'node' is a directory and that its 479 * contents are consistent with respect to directories. 480 */ 481 #define TMPFS_VALIDATE_DIR(node) do { \ 482 MPASS((node)->tn_type == VDIR); \ 483 MPASS((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 484 } while (0) 485 486 /* 487 * Amount of memory pages to reserve for the system (e.g., to not use by 488 * tmpfs). 489 */ 490 #define TMPFS_PAGES_MINRESERVED (4 * 1024 * 1024 / PAGE_SIZE) 491 492 size_t tmpfs_mem_avail(void); 493 494 size_t tmpfs_pages_used(struct tmpfs_mount *tmp); 495 496 #endif 497 498 /* 499 * Macros/functions to convert from generic data structures to tmpfs 500 * specific ones. 501 */ 502 503 static inline struct tmpfs_mount * 504 VFS_TO_TMPFS(struct mount *mp) 505 { 506 struct tmpfs_mount *tmp; 507 508 MPASS(mp != NULL && mp->mnt_data != NULL); 509 tmp = (struct tmpfs_mount *)mp->mnt_data; 510 return (tmp); 511 } 512 513 static inline struct tmpfs_node * 514 VP_TO_TMPFS_NODE(struct vnode *vp) 515 { 516 struct tmpfs_node *node; 517 518 MPASS(vp != NULL && vp->v_data != NULL); 519 node = (struct tmpfs_node *)vp->v_data; 520 return (node); 521 } 522 523 static inline struct tmpfs_node * 524 VP_TO_TMPFS_DIR(struct vnode *vp) 525 { 526 struct tmpfs_node *node; 527 528 node = VP_TO_TMPFS_NODE(vp); 529 TMPFS_VALIDATE_DIR(node); 530 return (node); 531 } 532 533 static inline bool 534 tmpfs_use_nc(struct vnode *vp) 535 { 536 537 return (!(VFS_TO_TMPFS(vp->v_mount)->tm_nonc)); 538 } 539 540 #endif /* _FS_TMPFS_TMPFS_H_ */ 541