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: src/sys/fs/tmpfs/tmpfs.h,v 1.18 2009/10/11 07:03:56 delphij Exp $ 33 */ 34 35 #ifndef _VFS_TMPFS_TMPFS_H_ 36 #define _VFS_TMPFS_TMPFS_H_ 37 38 /* --------------------------------------------------------------------- 39 * KERNEL-SPECIFIC DEFINITIONS 40 * --------------------------------------------------------------------- */ 41 #include <sys/dirent.h> 42 #include <sys/mount.h> 43 #include <sys/tree.h> 44 #include <sys/vnode.h> 45 #include <sys/file.h> 46 #include <sys/lock.h> 47 #include <sys/lockf.h> 48 #include <sys/mutex.h> 49 #include <sys/objcache.h> 50 51 /* --------------------------------------------------------------------- */ 52 #include <sys/malloc.h> 53 #ifdef _KERNEL 54 #include <sys/systm.h> 55 #endif 56 #include <sys/vmmeter.h> 57 #include <vm/swap_pager.h> 58 59 #ifdef MALLOC_DECLARE 60 MALLOC_DECLARE(M_TMPFSMNT); 61 #endif 62 63 /* --------------------------------------------------------------------- */ 64 65 /* 66 * Internal representation of a tmpfs directory entry. 67 */ 68 struct tmpfs_dirent { 69 RB_ENTRY(tmpfs_dirent) rb_node; 70 RB_ENTRY(tmpfs_dirent) rb_cookienode; 71 72 /* Length of the name stored in this directory entry. This avoids 73 * the need to recalculate it every time the name is used. */ 74 uint16_t td_namelen; 75 76 /* The name of the entry, allocated from a string pool. This 77 * string is not required to be zero-terminated; therefore, the 78 * td_namelen field must always be used when accessing its value. */ 79 char *td_name; 80 81 /* Pointer to the node this entry refers to. */ 82 struct tmpfs_node *td_node; 83 }; 84 85 struct tmpfs_dirtree; 86 RB_HEAD(tmpfs_dirtree, tmpfs_dirent); 87 RB_PROTOTYPE(tmpfs_dirtree, tmpfs_dirent, rb_node, 88 tmpfs_dirtree_compare); 89 90 RB_HEAD(tmpfs_dirtree_cookie, tmpfs_dirent); 91 RB_PROTOTYPE(tmpfs_dirtree_cookie, tmpfs_dirent, rb_cookienode, 92 tmpfs_dirtree_cookie_compare); 93 94 95 /* 96 * A directory in tmpfs holds a set of directory entries, which in 97 * turn point to other files (which can be directories themselves). 98 * 99 * In tmpfs, this set is managed by a red-black tree, whose root is defined 100 * by the struct tmpfs_dirtree type. 101 * 102 * It is important to notice that directories do not have entries for . and 103 * .. as other file systems do. These can be generated when requested 104 * based on information available by other means, such as the pointer to 105 * the node itself in the former case or the pointer to the parent directory 106 * in the latter case. This is done to simplify tmpfs's code and, more 107 * importantly, to remove redundancy. 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. (Emulating 112 * a real offset could be very difficult.) 113 * 114 * The '.', '..' and the end of directory markers have fixed cookies which 115 * cannot collide with the cookies generated by other entries. The cookies 116 * for the other entries are generated based on the memory address on which 117 * stores their information is stored. 118 * 119 * DragonFly binaries use 64-bit cookies. We mask-off the signed bit to 120 * ensure that cookie 'offsets' are positive. 121 */ 122 #ifdef _KERNEL 123 124 #define TMPFS_ROOTINO ((ino_t)2) 125 126 #define TMPFS_DIRCOOKIE_DOT 0 127 #define TMPFS_DIRCOOKIE_DOTDOT 1 128 #define TMPFS_DIRCOOKIE_EOF 2 129 130 static __inline 131 off_t 132 tmpfs_dircookie(struct tmpfs_dirent *de) 133 { 134 return (((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFFFFFFFFFFLLU); 135 } 136 137 #endif /* _KERNEL */ 138 139 /* --------------------------------------------------------------------- */ 140 141 /* 142 * Internal representation of a tmpfs file system node. 143 * 144 * This structure is splitted in two parts: one holds attributes common 145 * to all file types and the other holds data that is only applicable to 146 * a particular type. The code must be careful to only access those 147 * attributes that are actually allowed by the node's type. 148 */ 149 struct tmpfs_node { 150 /* Doubly-linked list entry which links all existing nodes for a 151 * single file system. This is provided to ease the removal of 152 * all nodes during the unmount operation. */ 153 LIST_ENTRY(tmpfs_node) tn_entries; 154 155 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', 156 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode 157 * types instead of a custom enumeration is to make things simpler 158 * and faster, as we do not need to convert between two types. */ 159 enum vtype tn_type; 160 161 /* Node identifier. */ 162 ino_t tn_id; 163 164 /* Node's internal status. This is used by several file system 165 * operations to do modifications to the node in a delayed 166 * fashion. */ 167 int tn_status; 168 #define TMPFS_NODE_ACCESSED (1 << 1) 169 #define TMPFS_NODE_MODIFIED (1 << 2) 170 #define TMPFS_NODE_CHANGED (1 << 3) 171 172 /* The node size. It does not necessarily match the real amount 173 * of memory consumed by it. */ 174 off_t tn_size; 175 176 /* Generic node attributes. */ 177 uid_t tn_uid; 178 gid_t tn_gid; 179 mode_t tn_mode; 180 u_int tn_flags; 181 nlink_t tn_links; 182 long tn_atime; 183 long tn_atimensec; 184 long tn_mtime; 185 long tn_mtimensec; 186 long tn_ctime; 187 long tn_ctimensec; 188 unsigned long tn_gen; 189 struct lockf tn_advlock; 190 191 /* As there is a single vnode for each active file within the 192 * system, care has to be taken to avoid allocating more than one 193 * vnode per file. In order to do this, a bidirectional association 194 * is kept between vnodes and nodes. 195 * 196 * Whenever a vnode is allocated, its v_data field is updated to 197 * point to the node it references. At the same time, the node's 198 * tn_vnode field is modified to point to the new vnode representing 199 * it. Further attempts to allocate a vnode for this same node will 200 * result in returning a new reference to the value stored in 201 * tn_vnode. 202 * 203 * May be NULL when the node is unused (that is, no vnode has been 204 * allocated for it or it has been reclaimed). */ 205 struct vnode * tn_vnode; 206 207 /* interlock to protect tn_vpstate */ 208 struct lock tn_interlock; 209 210 /* Identify if current node has vnode assiocate with 211 * or allocating vnode. 212 */ 213 int tn_vpstate; 214 215 /* misc data field for different tn_type node */ 216 union { 217 /* Valid when tn_type == VBLK || tn_type == VCHR. */ 218 dev_t tn_rdev; /*int32_t ?*/ 219 220 /* Valid when tn_type == VDIR. */ 221 struct tn_dir { 222 /* 223 * Pointer to the parent directory. The root 224 * directory has a pointer to itself in this field; 225 * this property identifies the root node. 226 */ 227 struct tmpfs_node * tn_parent; 228 229 /* 230 * Directory entries are indexed by name and also 231 * indexed by cookie. 232 */ 233 struct tmpfs_dirtree tn_dirtree; 234 struct tmpfs_dirtree_cookie tn_cookietree; 235 } tn_dir; 236 237 /* Valid when tn_type == VLNK. */ 238 /* The link's target, allocated from a string pool. */ 239 char * tn_link; 240 241 /* 242 * Valid when tn_type == VREG. 243 * 244 * aobj is used as backing store for the vnode object. It 245 * typically only contains swap assignments, but we also use 246 * it to save the vnode object's vm_page's when the vnode 247 * becomes inactive. 248 */ 249 struct tn_reg { 250 vm_object_t tn_aobj; 251 size_t tn_aobj_pages; 252 int tn_pages_in_aobj; 253 } tn_reg; 254 255 /* Valid when tn_type = VFIFO */ 256 struct tn_fifo { 257 int (*tn_fo_read) (struct file *fp, struct uio *uio, 258 struct ucred *cred, int flags); 259 int (*tn_fo_write) (struct file *fp, struct uio *uio, 260 struct ucred *cred, int flags); 261 } tn_fifo; 262 } tn_spec; 263 }; 264 265 /* Only userspace needs this */ 266 #define VTOI(vp) ((struct tmpfs_node *)(vp)->v_data) 267 268 #ifdef _KERNEL 269 LIST_HEAD(tmpfs_node_list, tmpfs_node); 270 271 #define tn_rdev tn_spec.tn_rdev 272 #define tn_dir tn_spec.tn_dir 273 #define tn_link tn_spec.tn_link 274 #define tn_reg tn_spec.tn_reg 275 #define tn_fifo tn_spec.tn_fifo 276 277 #define TMPFS_NODE_LOCK(node) lockmgr(&(node)->tn_interlock, LK_EXCLUSIVE|LK_RETRY) 278 #define TMPFS_NODE_LOCK_SH(node) lockmgr(&(node)->tn_interlock, LK_SHARED|LK_RETRY) 279 #define TMPFS_NODE_UNLOCK(node) lockmgr(&(node)->tn_interlock, LK_RELEASE) 280 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock) 281 282 #ifdef INVARIANTS 283 #define TMPFS_ASSERT_LOCKED(node) do { \ 284 KKASSERT(node != NULL); \ 285 KKASSERT(node->tn_vnode != NULL); \ 286 if (!vn_islocked(node->tn_vnode) && \ 287 (lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE )) \ 288 panic("tmpfs: node is not locked: %p", node); \ 289 } while (0) 290 #define TMPFS_ASSERT_ELOCKED(node) do { \ 291 KKASSERT((node) != NULL); \ 292 KKASSERT(lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE); \ 293 } while (0) 294 #else 295 #define TMPFS_ASSERT_LOCKED(node) (void)0 296 #define TMPFS_ASSERT_ELOCKED(node) (void)0 297 #endif /* INVARIANTS */ 298 299 #define TMPFS_VNODE_ALLOCATING 1 300 #define TMPFS_VNODE_WANT 2 301 #define TMPFS_VNODE_DOOMED 4 302 /* --------------------------------------------------------------------- */ 303 304 /* 305 * Internal representation of a tmpfs mount point. 306 */ 307 struct tmpfs_mount { 308 struct mount *tm_mount; 309 310 /* Maximum number of memory pages available for use by the file 311 * system, set during mount time. This variable must never be 312 * used directly as it may be bigger than the current amount of 313 * free memory; in the extreme case, it will hold the SIZE_MAX 314 * value. Instead, use the TMPFS_PAGES_MAX macro. */ 315 long tm_pages_max; 316 317 /* Number of pages in use by the file system. Cannot be bigger 318 * than the value returned by TMPFS_PAGES_MAX in any case. */ 319 long tm_pages_used; 320 321 /* Pointer to the node representing the root directory of this 322 * file system. */ 323 struct tmpfs_node * tm_root; 324 325 /* Maximum number of possible nodes for this file system; set 326 * during mount time. We need a hard limit on the maximum number 327 * of nodes to avoid allocating too much of them; their objects 328 * cannot be released until the file system is unmounted. 329 * Otherwise, we could easily run out of memory by creating lots 330 * of empty files and then simply removing them. */ 331 ino_t tm_nodes_max; 332 333 /* Number of nodes currently that are in use. */ 334 ino_t tm_nodes_inuse; 335 336 /* maximum representable file size */ 337 u_int64_t tm_maxfilesize; 338 339 /* Nodes are organized in two different lists. The used list 340 * contains all nodes that are currently used by the file system; 341 * i.e., they refer to existing files. The available list contains 342 * all nodes that are currently available for use by new files. 343 * Nodes must be kept in this list (instead of deleting them) 344 * because we need to keep track of their generation number (tn_gen 345 * field). 346 * 347 * Note that nodes are lazily allocated: if the available list is 348 * empty and we have enough space to create more nodes, they will be 349 * created and inserted in the used list. Once these are released, 350 * they will go into the available list, remaining alive until the 351 * file system is unmounted. */ 352 struct tmpfs_node_list tm_nodes_used; 353 354 /* Per-mount malloc zones for tmpfs nodes, names, and dirents */ 355 struct malloc_type *tm_node_zone; 356 struct malloc_type *tm_dirent_zone; 357 struct malloc_type *tm_name_zone; 358 359 struct objcache_malloc_args tm_node_zone_malloc_args; 360 struct objcache_malloc_args tm_dirent_zone_malloc_args; 361 362 /* Pools used to store file system meta data. */ 363 struct objcache *tm_dirent_pool; 364 struct objcache *tm_node_pool; 365 366 ino_t tm_ino; 367 int tm_flags; 368 369 struct netexport tm_export; 370 371 struct mount *tm_mnt; 372 }; 373 374 #define TMPFS_LOCK(tm) lwkt_gettoken(&(tm)->tm_mount->mnt_token) 375 #define TMPFS_UNLOCK(tm) lwkt_reltoken(&(tm)->tm_mount->mnt_token) 376 377 /* --------------------------------------------------------------------- */ 378 379 /* 380 * This structure maps a file identifier to a tmpfs node. Used by the 381 * NFS code. 382 */ 383 struct tmpfs_fid { 384 uint16_t tf_len; 385 uint16_t tf_pad; 386 ino_t tf_id; 387 unsigned long tf_gen; 388 } __packed; 389 390 /* --------------------------------------------------------------------- */ 391 392 /* 393 * Prototypes for tmpfs_subr.c. 394 */ 395 396 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype, 397 uid_t uid, gid_t gid, mode_t mode, char *, int, int, 398 struct tmpfs_node **); 399 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 400 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 401 const char *, uint16_t, struct tmpfs_dirent **); 402 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *); 403 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, 404 struct tmpfs_node *, int, struct vnode **); 405 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 406 struct namecache *, struct ucred *, char *); 407 void tmpfs_dir_attach(struct tmpfs_node *, struct tmpfs_dirent *); 408 void tmpfs_dir_detach(struct tmpfs_node *, struct tmpfs_dirent *); 409 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 410 struct tmpfs_node *f, 411 struct namecache *ncp); 412 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *); 413 int tmpfs_dir_getdotdotdent(struct tmpfs_mount *, 414 struct tmpfs_node *, struct uio *); 415 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t); 416 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *); 417 int tmpfs_reg_resize(struct vnode *, off_t, int); 418 int tmpfs_chflags(struct vnode *, u_long, struct ucred *); 419 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *); 420 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *); 421 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *); 422 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *, 423 int, struct ucred *); 424 void tmpfs_itimes(struct vnode *, const struct timespec *, 425 const struct timespec *); 426 427 void tmpfs_update(struct vnode *); 428 int tmpfs_truncate(struct vnode *, off_t); 429 boolean_t tmpfs_node_ctor(void *obj, void *privdata, int flags); 430 431 /* --------------------------------------------------------------------- */ 432 433 /* 434 * Convenience macros to simplify some logical expressions. 435 */ 436 #define IMPLIES(a, b) (!(a) || (b)) 437 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 438 439 /* --------------------------------------------------------------------- */ 440 441 /* 442 * Checks that the directory entry pointed by 'de' matches the name 'name' 443 * with a length of 'len'. 444 */ 445 #define TMPFS_DIRENT_MATCHES(de, name, len) \ 446 (de->td_namelen == (uint16_t)len && \ 447 bcmp((de)->td_name, (name), (de)->td_namelen) == 0) 448 449 /* --------------------------------------------------------------------- */ 450 451 /* 452 * Ensures that the node pointed by 'node' is a directory and that its 453 * contents are consistent with respect to directories. 454 */ 455 #define TMPFS_VALIDATE_DIR(node) do { \ 456 KKASSERT((node)->tn_type == VDIR); \ 457 KKASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 458 } while(0) 459 460 /* --------------------------------------------------------------------- */ 461 462 /* 463 * Macros/functions to convert from generic data structures to tmpfs 464 * specific ones. Kernel code use VP_TO_TMPFS_NODE() instead of VTOI(). 465 */ 466 467 static inline 468 struct tmpfs_mount * 469 VFS_TO_TMPFS(struct mount *mp) 470 { 471 struct tmpfs_mount *tmp; 472 473 KKASSERT((mp) != NULL && (mp)->mnt_data != NULL); 474 tmp = (struct tmpfs_mount *)(mp)->mnt_data; 475 return tmp; 476 } 477 478 static inline 479 struct tmpfs_node * 480 VP_TO_TMPFS_NODE(struct vnode *vp) 481 { 482 struct tmpfs_node *node; 483 484 KKASSERT((vp) != NULL && (vp)->v_data != NULL); 485 node = (struct tmpfs_node *)vp->v_data; 486 return node; 487 } 488 489 static inline 490 struct tmpfs_node * 491 VP_TO_TMPFS_DIR(struct vnode *vp) 492 { 493 struct tmpfs_node *node; 494 495 node = VP_TO_TMPFS_NODE(vp); 496 TMPFS_VALIDATE_DIR(node); 497 return node; 498 } 499 500 /* --------------------------------------------------------------------- */ 501 /* 502 * buffer cache size 503 */ 504 #define TMPFS_BLKSIZE 16384 /* buffer cache size*/ 505 #define TMPFS_BLKMASK (TMPFS_BLKSIZE - 1) 506 #define TMPFS_BLKMASK64 ((off_t)(TMPFS_BLKSIZE - 1)) 507 #endif /* _KERNEL */ 508 509 #endif /* _VFS_TMPFS_TMPFS_H_ */ 510