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_ 45 MALLOC_DECLARE(M_TMPFSNAME); 46 #endif 47 48 /* 49 * Internal representation of a tmpfs directory entry. 50 */ 51 52 LIST_HEAD(tmpfs_dir_duphead, tmpfs_dirent); 53 54 struct 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 */ 106 RB_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 */ 149 struct 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) */ 158 159 /* Node identifier. */ 160 ino_t tn_id; /* (c) */ 161 162 /* 163 * The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', 164 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode 165 * types instead of a custom enumeration is to make things simpler 166 * and faster, as we do not need to convert between two types. 167 */ 168 enum vtype tn_type; /* (c) */ 169 170 /* 171 * See the top comment. Reordered here to fill LP64 hole. 172 */ 173 bool tn_attached; /* (m) */ 174 175 /* 176 * Node's internal status. This is used by several file system 177 * operations to do modifications to the node in a delayed 178 * fashion. 179 * 180 * tn_accessed has a dedicated byte to allow update by store without 181 * using atomics. This provides a micro-optimization to e.g. 182 * tmpfs_read_pgcache(). 183 */ 184 uint8_t tn_status; /* (vi) */ 185 uint8_t tn_accessed; /* unlocked */ 186 187 /* 188 * The node size. It does not necessarily match the real amount 189 * of memory consumed by it. 190 */ 191 off_t tn_size; /* (v) */ 192 193 /* Generic node attributes. */ 194 uid_t tn_uid; /* (v) */ 195 gid_t tn_gid; /* (v) */ 196 mode_t tn_mode; /* (v) */ 197 int tn_links; /* (v) */ 198 u_long tn_flags; /* (v) */ 199 struct timespec tn_atime; /* (vi) */ 200 struct timespec tn_mtime; /* (vi) */ 201 struct timespec tn_ctime; /* (vi) */ 202 struct timespec tn_birthtime; /* (v) */ 203 unsigned long tn_gen; /* (c) */ 204 205 /* 206 * As there is a single vnode for each active file within the 207 * system, care has to be taken to avoid allocating more than one 208 * vnode per file. In order to do this, a bidirectional association 209 * is kept between vnodes and nodes. 210 * 211 * Whenever a vnode is allocated, its v_data field is updated to 212 * point to the node it references. At the same time, the node's 213 * tn_vnode field is modified to point to the new vnode representing 214 * it. Further attempts to allocate a vnode for this same node will 215 * result in returning a new reference to the value stored in 216 * tn_vnode. 217 * 218 * May be NULL when the node is unused (that is, no vnode has been 219 * allocated for it or it has been reclaimed). 220 */ 221 struct vnode * tn_vnode; /* (i) */ 222 223 /* 224 * Interlock to protect tn_vpstate, and tn_status under shared 225 * vnode lock. 226 */ 227 struct mtx tn_interlock; 228 229 /* 230 * Identify if current node has vnode assiocate with 231 * or allocating vnode. 232 */ 233 int tn_vpstate; /* (i) */ 234 235 /* Transient refcounter on this node. */ 236 u_int tn_refcount; /* 0<->1 (m) + (i) */ 237 238 /* misc data field for different tn_type node */ 239 union { 240 /* Valid when tn_type == VBLK || tn_type == VCHR. */ 241 dev_t tn_rdev; /* (c) */ 242 243 /* Valid when tn_type == VDIR. */ 244 struct tn_dir { 245 /* 246 * Pointer to the parent directory. The root 247 * directory has a pointer to itself in this field; 248 * this property identifies the root node. 249 */ 250 struct tmpfs_node * tn_parent; 251 252 /* 253 * Head of a tree that links the contents of 254 * the directory together. 255 */ 256 struct tmpfs_dir tn_dirhead; 257 258 /* 259 * Head of a list the contains fake directory entries 260 * heads, i.e. entries with TMPFS_DIRCOOKIE_DUPHEAD 261 * flag. 262 */ 263 struct tmpfs_dir_duphead tn_dupindex; 264 265 /* 266 * Number and pointer of the first directory entry 267 * returned by the readdir operation if it were 268 * called again to continue reading data from the 269 * same directory as before. This is used to speed 270 * up reads of long directories, assuming that no 271 * more than one read is in progress at a given time. 272 * Otherwise, these values are discarded. 273 */ 274 off_t tn_readdir_lastn; 275 struct tmpfs_dirent * tn_readdir_lastp; 276 } tn_dir; 277 278 /* Valid when tn_type == VLNK. */ 279 /* The link's target, allocated from a string pool. */ 280 struct tn_link { 281 char * tn_link_target; /* (c) */ 282 char tn_link_smr; /* (c) */ 283 } tn_link; 284 285 /* Valid when tn_type == VREG. */ 286 struct tn_reg { 287 /* 288 * The contents of regular files stored in a 289 * tmpfs file system are represented by a 290 * single anonymous memory object (aobj, for 291 * short). The aobj provides direct access to 292 * any position within the file. It is a task 293 * of the memory management subsystem to issue 294 * the required page ins or page outs whenever 295 * a position within the file is accessed. 296 */ 297 vm_object_t tn_aobj; /* (c) */ 298 struct tmpfs_mount *tn_tmp; /* (c) */ 299 } tn_reg; 300 } tn_spec; /* (v) */ 301 }; 302 LIST_HEAD(tmpfs_node_list, tmpfs_node); 303 304 #define tn_rdev tn_spec.tn_rdev 305 #define tn_dir tn_spec.tn_dir 306 #define tn_link_target tn_spec.tn_link.tn_link_target 307 #define tn_link_smr tn_spec.tn_link.tn_link_smr 308 #define tn_reg tn_spec.tn_reg 309 #define tn_fifo tn_spec.tn_fifo 310 311 #define TMPFS_LINK_MAX INT_MAX 312 313 #define TMPFS_NODE_LOCK(node) mtx_lock(&(node)->tn_interlock) 314 #define TMPFS_NODE_UNLOCK(node) mtx_unlock(&(node)->tn_interlock) 315 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock) 316 #define TMPFS_NODE_ASSERT_LOCKED(node) mtx_assert(TMPFS_NODE_MTX(node), \ 317 MA_OWNED) 318 319 #ifdef INVARIANTS 320 #define TMPFS_ASSERT_LOCKED(node) do { \ 321 MPASS((node) != NULL); \ 322 MPASS((node)->tn_vnode != NULL); \ 323 ASSERT_VOP_LOCKED((node)->tn_vnode, "tmpfs assert"); \ 324 } while (0) 325 #else 326 #define TMPFS_ASSERT_LOCKED(node) (void)0 327 #endif 328 329 /* tn_vpstate */ 330 #define TMPFS_VNODE_ALLOCATING 1 331 #define TMPFS_VNODE_WANT 2 332 #define TMPFS_VNODE_DOOMED 4 333 #define TMPFS_VNODE_WRECLAIM 8 334 335 /* tn_status */ 336 #define TMPFS_NODE_MODIFIED 0x01 337 #define TMPFS_NODE_CHANGED 0x02 338 339 /* 340 * Internal representation of a tmpfs mount point. 341 */ 342 struct tmpfs_mount { 343 /* 344 * Original value of the "size" parameter, for reference purposes, 345 * mostly. 346 */ 347 off_t tm_size_max; 348 /* 349 * Maximum number of memory pages available for use by the file 350 * system, set during mount time. This variable must never be 351 * used directly as it may be bigger than the current amount of 352 * free memory; in the extreme case, it will hold the ULONG_MAX 353 * value. 354 */ 355 u_long tm_pages_max; 356 357 /* Number of pages in use by the file system. */ 358 u_long tm_pages_used; 359 360 /* 361 * Pointer to the node representing the root directory of this 362 * file system. 363 */ 364 struct tmpfs_node * tm_root; 365 366 /* 367 * Maximum number of possible nodes for this file system; set 368 * during mount time. We need a hard limit on the maximum number 369 * of nodes to avoid allocating too much of them; their objects 370 * cannot be released until the file system is unmounted. 371 * Otherwise, we could easily run out of memory by creating lots 372 * of empty files and then simply removing them. 373 */ 374 ino_t tm_nodes_max; 375 376 /* unrhdr used to allocate inode numbers */ 377 struct unrhdr64 tm_ino_unr; 378 379 /* Number of nodes currently that are in use. */ 380 ino_t tm_nodes_inuse; 381 382 /* Refcounter on this struct tmpfs_mount. */ 383 uint64_t tm_refcount; 384 385 /* maximum representable file size */ 386 u_int64_t tm_maxfilesize; 387 388 /* 389 * The used list contains all nodes that are currently used by 390 * the file system; i.e., they refer to existing files. 391 */ 392 struct tmpfs_node_list tm_nodes_used; 393 394 /* All node lock to protect the node list and tmp_pages_used. */ 395 struct mtx tm_allnode_lock; 396 397 /* Read-only status. */ 398 bool tm_ronly; 399 /* Do not use namecache. */ 400 bool tm_nonc; 401 /* Do not update mtime on writes through mmaped areas. */ 402 bool tm_nomtime; 403 }; 404 #define TMPFS_LOCK(tm) mtx_lock(&(tm)->tm_allnode_lock) 405 #define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->tm_allnode_lock) 406 #define TMPFS_MP_ASSERT_LOCKED(tm) mtx_assert(&(tm)->tm_allnode_lock, MA_OWNED) 407 408 /* 409 * This structure maps a file identifier to a tmpfs node. Used by the 410 * NFS code. 411 */ 412 struct tmpfs_fid_data { 413 ino_t tfd_id; 414 unsigned long tfd_gen; 415 }; 416 _Static_assert(sizeof(struct tmpfs_fid_data) <= MAXFIDSZ, 417 "(struct tmpfs_fid_data) is larger than (struct fid).fid_data"); 418 419 struct tmpfs_dir_cursor { 420 struct tmpfs_dirent *tdc_current; 421 struct tmpfs_dirent *tdc_tree; 422 }; 423 424 #ifdef _KERNEL 425 /* 426 * Prototypes for tmpfs_subr.c. 427 */ 428 429 void tmpfs_ref_node(struct tmpfs_node *node); 430 int tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *, enum vtype, 431 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *, 432 const char *, dev_t, struct tmpfs_node **); 433 int tmpfs_fo_close(struct file *fp, struct thread *td); 434 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 435 bool tmpfs_free_node_locked(struct tmpfs_mount *, struct tmpfs_node *, bool); 436 void tmpfs_free_tmp(struct tmpfs_mount *); 437 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 438 const char *, u_int, struct tmpfs_dirent **); 439 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *); 440 void tmpfs_dirent_init(struct tmpfs_dirent *, const char *, u_int); 441 void tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj); 442 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int, 443 struct vnode **); 444 void tmpfs_free_vp(struct vnode *); 445 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 446 struct componentname *, const char *); 447 void tmpfs_check_mtime(struct vnode *); 448 void tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *); 449 void tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *); 450 void tmpfs_dir_destroy(struct tmpfs_mount *, struct tmpfs_node *); 451 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 452 struct tmpfs_node *f, 453 struct componentname *cnp); 454 int tmpfs_dir_getdents(struct tmpfs_mount *, struct tmpfs_node *, 455 struct uio *, int, u_long *, int *); 456 int tmpfs_dir_whiteout_add(struct vnode *, struct componentname *); 457 void tmpfs_dir_whiteout_remove(struct vnode *, struct componentname *); 458 int tmpfs_reg_resize(struct vnode *, off_t, boolean_t); 459 int tmpfs_chflags(struct vnode *, u_long, struct ucred *, struct thread *); 460 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct thread *); 461 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *, 462 struct thread *); 463 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct thread *); 464 int tmpfs_chtimes(struct vnode *, struct vattr *, struct ucred *cred, 465 struct thread *); 466 void tmpfs_itimes(struct vnode *, const struct timespec *, 467 const struct timespec *); 468 469 void tmpfs_set_accessed(struct tmpfs_mount *tm, struct tmpfs_node *node); 470 void tmpfs_set_status(struct tmpfs_mount *tm, struct tmpfs_node *node, 471 int status); 472 int tmpfs_truncate(struct vnode *, off_t); 473 struct tmpfs_dirent *tmpfs_dir_first(struct tmpfs_node *dnode, 474 struct tmpfs_dir_cursor *dc); 475 struct tmpfs_dirent *tmpfs_dir_next(struct tmpfs_node *dnode, 476 struct tmpfs_dir_cursor *dc); 477 static __inline void 478 tmpfs_update(struct vnode *vp) 479 { 480 481 tmpfs_itimes(vp, NULL, NULL); 482 } 483 484 /* 485 * Convenience macros to simplify some logical expressions. 486 */ 487 #define IMPLIES(a, b) (!(a) || (b)) 488 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 489 490 /* 491 * Checks that the directory entry pointed by 'de' matches the name 'name' 492 * with a length of 'len'. 493 */ 494 #define TMPFS_DIRENT_MATCHES(de, name, len) \ 495 (de->td_namelen == len && \ 496 bcmp((de)->ud.td_name, (name), (de)->td_namelen) == 0) 497 498 /* 499 * Ensures that the node pointed by 'node' is a directory and that its 500 * contents are consistent with respect to directories. 501 */ 502 #define TMPFS_VALIDATE_DIR(node) do { \ 503 MPASS((node)->tn_type == VDIR); \ 504 MPASS((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 505 } while (0) 506 507 /* 508 * Amount of memory pages to reserve for the system (e.g., to not use by 509 * tmpfs). 510 */ 511 #if !defined(TMPFS_PAGES_MINRESERVED) 512 #define TMPFS_PAGES_MINRESERVED (4 * 1024 * 1024 / PAGE_SIZE) 513 #endif 514 515 size_t tmpfs_mem_avail(void); 516 size_t tmpfs_pages_used(struct tmpfs_mount *tmp); 517 void tmpfs_subr_init(void); 518 void tmpfs_subr_uninit(void); 519 520 /* 521 * Macros/functions to convert from generic data structures to tmpfs 522 * specific ones. 523 */ 524 525 static inline struct tmpfs_mount * 526 VFS_TO_TMPFS(struct mount *mp) 527 { 528 struct tmpfs_mount *tmp; 529 530 MPASS(mp != NULL && mp->mnt_data != NULL); 531 tmp = (struct tmpfs_mount *)mp->mnt_data; 532 return (tmp); 533 } 534 535 static inline struct tmpfs_node * 536 VP_TO_TMPFS_NODE(struct vnode *vp) 537 { 538 struct tmpfs_node *node; 539 540 MPASS(vp != NULL && vp->v_data != NULL); 541 node = (struct tmpfs_node *)vp->v_data; 542 return (node); 543 } 544 545 #define VP_TO_TMPFS_NODE_SMR(vp) \ 546 ((struct tmpfs_node *)vn_load_v_data_smr(vp)) 547 548 static inline struct tmpfs_node * 549 VP_TO_TMPFS_DIR(struct vnode *vp) 550 { 551 struct tmpfs_node *node; 552 553 node = VP_TO_TMPFS_NODE(vp); 554 TMPFS_VALIDATE_DIR(node); 555 return (node); 556 } 557 558 static inline bool 559 tmpfs_use_nc(struct vnode *vp) 560 { 561 562 return (!(VFS_TO_TMPFS(vp->v_mount)->tm_nonc)); 563 } 564 565 static inline void 566 tmpfs_update_getattr(struct vnode *vp) 567 { 568 struct tmpfs_node *node; 569 570 node = VP_TO_TMPFS_NODE(vp); 571 if (__predict_false((node->tn_status & (TMPFS_NODE_MODIFIED | 572 TMPFS_NODE_CHANGED)) != 0 || node->tn_accessed)) 573 tmpfs_update(vp); 574 } 575 576 extern struct fileops tmpfs_fnops; 577 578 #endif /* _KERNEL */ 579 580 #endif /* _FS_TMPFS_TMPFS_H_ */ 581