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_blksize; /* small file optimization */ 168 int tn_status; 169 #define TMPFS_NODE_ACCESSED (1 << 1) 170 #define TMPFS_NODE_MODIFIED (1 << 2) 171 #define TMPFS_NODE_CHANGED (1 << 3) 172 173 /* The node size. It does not necessarily match the real amount 174 * of memory consumed by it. */ 175 off_t tn_size; 176 177 /* Generic node attributes. */ 178 uid_t tn_uid; 179 gid_t tn_gid; 180 mode_t tn_mode; 181 u_int tn_flags; 182 nlink_t tn_links; /* atomic ops req */ 183 long tn_atime; 184 long tn_atimensec; 185 long tn_mtime; 186 long tn_mtimensec; 187 long tn_ctime; 188 long tn_ctimensec; 189 unsigned long tn_gen; 190 struct lockf tn_advlock; 191 192 /* As there is a single vnode for each active file within the 193 * system, care has to be taken to avoid allocating more than one 194 * vnode per file. In order to do this, a bidirectional association 195 * is kept between vnodes and nodes. 196 * 197 * Whenever a vnode is allocated, its v_data field is updated to 198 * point to the node it references. At the same time, the node's 199 * tn_vnode field is modified to point to the new vnode representing 200 * it. Further attempts to allocate a vnode for this same node will 201 * result in returning a new reference to the value stored in 202 * tn_vnode. 203 * 204 * May be NULL when the node is unused (that is, no vnode has been 205 * allocated for it or it has been reclaimed). */ 206 struct vnode * tn_vnode; 207 208 /* interlock to protect structure */ 209 struct lock tn_interlock; 210 211 /* 212 * tmpfs vnode state, may specify an allocation in-progress. 213 */ 214 int tn_vpstate; 215 216 /* misc data field for different tn_type node */ 217 union { 218 /* Valid when tn_type == VBLK || tn_type == VCHR. */ 219 dev_t tn_rdev; /*int32_t ?*/ 220 221 /* Valid when tn_type == VDIR. */ 222 struct tn_dir { 223 /* 224 * Pointer to the parent directory. The root 225 * directory has a pointer to itself in this field; 226 * this property identifies the root node. 227 */ 228 struct tmpfs_node * tn_parent; 229 230 /* 231 * Directory entries are indexed by name and also 232 * indexed by cookie. 233 */ 234 struct tmpfs_dirtree tn_dirtree; 235 struct tmpfs_dirtree_cookie tn_cookietree; 236 } tn_dir; 237 238 /* Valid when tn_type == VLNK. */ 239 /* The link's target, allocated from a string pool. */ 240 char * tn_link; 241 242 /* 243 * Valid when tn_type == VREG. 244 * 245 * aobj is used as backing store for the vnode object. It 246 * typically only contains swap assignments, but we also use 247 * it to save the vnode object's vm_page's when the vnode 248 * becomes inactive. 249 */ 250 struct tn_reg { 251 vm_object_t tn_aobj; 252 size_t tn_aobj_pages; 253 int tn_pages_in_aobj; 254 } tn_reg; 255 256 /* Valid when tn_type = VFIFO */ 257 struct tn_fifo { 258 int (*tn_fo_read) (struct file *fp, struct uio *uio, 259 struct ucred *cred, int flags); 260 int (*tn_fo_write) (struct file *fp, struct uio *uio, 261 struct ucred *cred, int flags); 262 } tn_fifo; 263 } tn_spec; 264 }; 265 266 /* Only userspace needs this */ 267 #define VTOI(vp) ((struct tmpfs_node *)(vp)->v_data) 268 269 #ifdef _KERNEL 270 LIST_HEAD(tmpfs_node_list, tmpfs_node); 271 272 #define tn_rdev tn_spec.tn_rdev 273 #define tn_dir tn_spec.tn_dir 274 #define tn_link tn_spec.tn_link 275 #define tn_reg tn_spec.tn_reg 276 #define tn_fifo tn_spec.tn_fifo 277 278 #define TMPFS_NODE_LOCK(node) lockmgr(&(node)->tn_interlock, LK_EXCLUSIVE|LK_RETRY) 279 #define TMPFS_NODE_LOCK_SH(node) lockmgr(&(node)->tn_interlock, LK_SHARED|LK_RETRY) 280 #define TMPFS_NODE_UNLOCK(node) lockmgr(&(node)->tn_interlock, LK_RELEASE) 281 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock) 282 283 #ifdef INVARIANTS 284 #define TMPFS_ASSERT_LOCKED(node) do { \ 285 KKASSERT(node != NULL); \ 286 KKASSERT(node->tn_vnode != NULL); \ 287 if (!vn_islocked(node->tn_vnode) && \ 288 (lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE )) \ 289 panic("tmpfs: node is not locked: %p", node); \ 290 } while (0) 291 #define TMPFS_ASSERT_ELOCKED(node) do { \ 292 KKASSERT((node) != NULL); \ 293 KKASSERT(lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE); \ 294 } while (0) 295 #else 296 #define TMPFS_ASSERT_LOCKED(node) (void)0 297 #define TMPFS_ASSERT_ELOCKED(node) (void)0 298 #endif /* INVARIANTS */ 299 300 #define TMPFS_VNODE_DOOMED 0x0001 301 /* --------------------------------------------------------------------- */ 302 303 /* 304 * Internal representation of a tmpfs mount point. 305 */ 306 struct tmpfs_mount { 307 struct mount *tm_mount; 308 309 /* Maximum number of memory pages available for use by the file 310 * system, set during mount time. This variable must never be 311 * used directly as it may be bigger than the current amount of 312 * free memory; in the extreme case, it will hold the SIZE_MAX 313 * value. Instead, use the TMPFS_PAGES_MAX macro. */ 314 long tm_pages_max; 315 316 /* Number of pages in use by the file system. Cannot be bigger 317 * than the value returned by TMPFS_PAGES_MAX in any case. */ 318 long tm_pages_used; 319 320 /* Pointer to the node representing the root directory of this 321 * file system. */ 322 struct tmpfs_node * tm_root; 323 324 /* Maximum number of possible nodes for this file system; set 325 * during mount time. We need a hard limit on the maximum number 326 * of nodes to avoid allocating too much of them; their objects 327 * cannot be released until the file system is unmounted. 328 * Otherwise, we could easily run out of memory by creating lots 329 * of empty files and then simply removing them. */ 330 ino_t tm_nodes_max; 331 332 /* Number of nodes currently that are in use. */ 333 ino_t tm_nodes_inuse; 334 335 /* maximum representable file size */ 336 u_int64_t tm_maxfilesize; 337 338 /* Nodes are organized in two different lists. The used list 339 * contains all nodes that are currently used by the file system; 340 * i.e., they refer to existing files. The available list contains 341 * all nodes that are currently available for use by new files. 342 * Nodes must be kept in this list (instead of deleting them) 343 * because we need to keep track of their generation number (tn_gen 344 * field). 345 * 346 * Note that nodes are lazily allocated: if the available list is 347 * empty and we have enough space to create more nodes, they will be 348 * created and inserted in the used list. Once these are released, 349 * they will go into the available list, remaining alive until the 350 * file system is unmounted. */ 351 struct tmpfs_node_list tm_nodes_used; 352 353 /* Per-mount malloc zones for tmpfs nodes, names, and dirents */ 354 struct malloc_type *tm_node_zone; 355 struct malloc_type *tm_dirent_zone; 356 struct malloc_type *tm_name_zone; 357 358 struct objcache_malloc_args tm_node_zone_malloc_args; 359 struct objcache_malloc_args tm_dirent_zone_malloc_args; 360 361 /* Pools used to store file system meta data. */ 362 struct objcache *tm_dirent_pool; 363 struct objcache *tm_node_pool; 364 365 ino_t tm_ino; 366 int tm_flags; 367 368 struct netexport tm_export; 369 370 struct mount *tm_mnt; 371 }; 372 373 #define TMPFS_LOCK(tm) lwkt_gettoken(&(tm)->tm_mount->mnt_token) 374 #define TMPFS_UNLOCK(tm) lwkt_reltoken(&(tm)->tm_mount->mnt_token) 375 376 /* --------------------------------------------------------------------- */ 377 378 /* 379 * This structure maps a file identifier to a tmpfs node. Used by the 380 * NFS code. 381 */ 382 struct tmpfs_fid { 383 uint16_t tf_len; 384 uint16_t tf_pad; 385 ino_t tf_id; 386 unsigned long tf_gen; 387 } __packed; 388 389 /* --------------------------------------------------------------------- */ 390 391 /* 392 * Prototypes for tmpfs_subr.c. 393 */ 394 395 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype, 396 uid_t uid, gid_t gid, mode_t mode, char *, int, int, 397 struct tmpfs_node **); 398 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 399 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 400 const char *, uint16_t, struct tmpfs_dirent **); 401 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *); 402 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, 403 struct tmpfs_node *, int, struct vnode **); 404 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 405 struct namecache *, struct ucred *, char *); 406 void tmpfs_dir_attach(struct tmpfs_node *, struct tmpfs_dirent *); 407 void tmpfs_dir_detach(struct tmpfs_node *, struct tmpfs_dirent *); 408 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 409 struct tmpfs_node *f, 410 struct namecache *ncp); 411 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *); 412 int tmpfs_dir_getdotdotdent(struct tmpfs_mount *, 413 struct tmpfs_node *, struct uio *); 414 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t); 415 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *); 416 int tmpfs_reg_resize(struct vnode *, off_t, int); 417 int tmpfs_chflags(struct vnode *, u_long, struct ucred *); 418 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *); 419 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *); 420 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *); 421 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *, 422 int, struct ucred *); 423 void tmpfs_itimes(struct vnode *, const struct timespec *, 424 const struct timespec *); 425 426 void tmpfs_update(struct vnode *); 427 int tmpfs_truncate(struct vnode *, off_t); 428 boolean_t tmpfs_node_ctor(void *obj, void *privdata, int flags); 429 void tmpfs_lock4(struct tmpfs_node *node1, struct tmpfs_node *node2, 430 struct tmpfs_node *node3, struct tmpfs_node *node4); 431 void tmpfs_unlock4(struct tmpfs_node *node1, struct tmpfs_node *node2, 432 struct tmpfs_node *node3, struct tmpfs_node *node4); 433 434 /* --------------------------------------------------------------------- */ 435 436 /* 437 * Convenience macros to simplify some logical expressions. 438 */ 439 #define IMPLIES(a, b) (!(a) || (b)) 440 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 441 442 /* --------------------------------------------------------------------- */ 443 444 /* 445 * Checks that the directory entry pointed by 'de' matches the name 'name' 446 * with a length of 'len'. 447 */ 448 #define TMPFS_DIRENT_MATCHES(de, name, len) \ 449 (de->td_namelen == (uint16_t)len && \ 450 bcmp((de)->td_name, (name), (de)->td_namelen) == 0) 451 452 /* --------------------------------------------------------------------- */ 453 454 /* 455 * Ensures that the node pointed by 'node' is a directory and that its 456 * contents are consistent with respect to directories. 457 */ 458 #define TMPFS_VALIDATE_DIR(node) do { \ 459 KKASSERT((node)->tn_type == VDIR); \ 460 KKASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 461 } while(0) 462 463 /* --------------------------------------------------------------------- */ 464 465 /* 466 * Macros/functions to convert from generic data structures to tmpfs 467 * specific ones. Kernel code use VP_TO_TMPFS_NODE() instead of VTOI(). 468 */ 469 470 static inline 471 struct tmpfs_mount * 472 VFS_TO_TMPFS(struct mount *mp) 473 { 474 struct tmpfs_mount *tmp; 475 476 KKASSERT((mp) != NULL && (mp)->mnt_data != NULL); 477 tmp = (struct tmpfs_mount *)(mp)->mnt_data; 478 return tmp; 479 } 480 481 static inline 482 struct tmpfs_node * 483 VP_TO_TMPFS_NODE(struct vnode *vp) 484 { 485 struct tmpfs_node *node; 486 487 KKASSERT((vp) != NULL && (vp)->v_data != NULL); 488 node = (struct tmpfs_node *)vp->v_data; 489 return node; 490 } 491 492 static inline 493 struct tmpfs_node * 494 VP_TO_TMPFS_DIR(struct vnode *vp) 495 { 496 struct tmpfs_node *node; 497 498 node = VP_TO_TMPFS_NODE(vp); 499 TMPFS_VALIDATE_DIR(node); 500 return node; 501 } 502 503 /* --------------------------------------------------------------------- */ 504 /* 505 * buffer cache size 506 */ 507 #define TMPFS_BLKSIZE 16384 /* buffer cache size*/ 508 #define TMPFS_BLKMASK (TMPFS_BLKSIZE - 1) 509 #define TMPFS_BLKMASK64 ((off_t)(TMPFS_BLKSIZE - 1)) 510 #endif /* _KERNEL */ 511 512 #endif /* _VFS_TMPFS_TMPFS_H_ */ 513