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 50 /* --------------------------------------------------------------------- */ 51 #include <sys/malloc.h> 52 #ifdef _KERNEL 53 #include <sys/systm.h> 54 #endif 55 #include <sys/vmmeter.h> 56 #include <vm/swap_pager.h> 57 58 #ifdef MALLOC_DECLARE 59 MALLOC_DECLARE(M_TMPFSMNT); 60 #endif 61 62 /* --------------------------------------------------------------------- */ 63 64 /* 65 * Internal representation of a tmpfs directory entry. 66 */ 67 struct tmpfs_dirent { 68 RB_ENTRY(tmpfs_dirent) rb_node; 69 RB_ENTRY(tmpfs_dirent) rb_cookienode; 70 71 /* Length of the name stored in this directory entry. This avoids 72 * the need to recalculate it every time the name is used. */ 73 uint16_t td_namelen; 74 75 /* The name of the entry, allocated from a string pool. This 76 * string is not required to be zero-terminated; therefore, the 77 * td_namelen field must always be used when accessing its value. */ 78 char *td_name; 79 80 /* Pointer to the node this entry refers to. */ 81 struct tmpfs_node *td_node; 82 }; 83 84 struct tmpfs_dirtree; 85 RB_HEAD(tmpfs_dirtree, tmpfs_dirent); 86 RB_PROTOTYPE(tmpfs_dirtree, tmpfs_dirent, rb_node, 87 tmpfs_dirtree_compare); 88 89 RB_HEAD(tmpfs_dirtree_cookie, tmpfs_dirent); 90 RB_PROTOTYPE(tmpfs_dirtree_cookie, tmpfs_dirent, rb_cookienode, 91 tmpfs_dirtree_cookie_compare); 92 93 94 /* 95 * A directory in tmpfs holds a set of directory entries, which in 96 * turn point to other files (which can be directories themselves). 97 * 98 * In tmpfs, this set is managed by a red-black tree, whose root is defined 99 * by the struct tmpfs_dirtree type. 100 * 101 * It is important to notice that directories do not have entries for . and 102 * .. as other file systems do. These can be generated when requested 103 * based on information available by other means, such as the pointer to 104 * the node itself in the former case or the pointer to the parent directory 105 * in the latter case. This is done to simplify tmpfs's code and, more 106 * importantly, to remove redundancy. 107 * 108 * Each entry in a directory has a cookie that identifies it. Cookies 109 * supersede offsets within directories because, given how tmpfs stores 110 * directories in memory, there is no such thing as an offset. (Emulating 111 * a real offset could be very difficult.) 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 memory address on which 116 * stores their information is stored. 117 * 118 * DragonFly binaries use 64-bit cookies. We mask-off the signed bit to 119 * ensure that cookie 'offsets' are positive. 120 */ 121 #ifdef _KERNEL 122 123 #define TMPFS_ROOTINO ((ino_t)2) 124 125 #define TMPFS_DIRCOOKIE_DOT 0 126 #define TMPFS_DIRCOOKIE_DOTDOT 1 127 #define TMPFS_DIRCOOKIE_EOF 2 128 129 static __inline 130 off_t 131 tmpfs_dircookie(struct tmpfs_dirent *de) 132 { 133 return (((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFFFFFFFFFFLLU); 134 } 135 136 #endif /* _KERNEL */ 137 138 /* --------------------------------------------------------------------- */ 139 140 /* 141 * Internal representation of a tmpfs file system node. 142 * 143 * This structure is splitted in two parts: one holds attributes common 144 * to all file types and the other holds data that is only applicable to 145 * a particular type. The code must be careful to only access those 146 * attributes that are actually allowed by the node's type. 147 */ 148 struct tmpfs_node { 149 /* Doubly-linked list entry which links all existing nodes for a 150 * single file system. This is provided to ease the removal of 151 * all nodes during the unmount operation. */ 152 LIST_ENTRY(tmpfs_node) tn_entries; 153 154 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', 155 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode 156 * types instead of a custom enumeration is to make things simpler 157 * and faster, as we do not need to convert between two types. */ 158 enum vtype tn_type; 159 160 /* Node identifier. */ 161 ino_t tn_id; 162 163 /* Node's internal status. This is used by several file system 164 * operations to do modifications to the node in a delayed 165 * fashion. */ 166 int tn_blksize; /* small file optimization */ 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; /* atomic ops req */ 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 structure */ 208 struct lock tn_interlock; 209 210 /* 211 * tmpfs vnode state, may specify an allocation in-progress. 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_DOOMED 0x0001 300 /* --------------------------------------------------------------------- */ 301 302 /* 303 * Internal representation of a tmpfs mount point. 304 */ 305 struct tmpfs_mount { 306 struct mount *tm_mount; 307 308 /* Maximum number of memory pages available for use by the file 309 * system, set during mount time. This variable must never be 310 * used directly as it may be bigger than the current amount of 311 * free memory; in the extreme case, it will hold the SIZE_MAX 312 * value. Instead, use the TMPFS_PAGES_MAX macro. */ 313 long tm_pages_max; 314 315 /* Number of pages in use by the file system. Cannot be bigger 316 * than the value returned by TMPFS_PAGES_MAX in any case. */ 317 long tm_pages_used; 318 319 /* Pointer to the node representing the root directory of this 320 * file system. */ 321 struct tmpfs_node * tm_root; 322 323 /* Maximum number of possible nodes for this file system; set 324 * during mount time. We need a hard limit on the maximum number 325 * of nodes to avoid allocating too much of them; their objects 326 * cannot be released until the file system is unmounted. 327 * Otherwise, we could easily run out of memory by creating lots 328 * of empty files and then simply removing them. */ 329 ino_t tm_nodes_max; 330 331 /* Number of nodes currently that are in use. */ 332 ino_t tm_nodes_inuse; 333 334 /* maximum representable file size */ 335 u_int64_t tm_maxfilesize; 336 337 /* Nodes are organized in two different lists. The used list 338 * contains all nodes that are currently used by the file system; 339 * i.e., they refer to existing files. The available list contains 340 * all nodes that are currently available for use by new files. 341 * Nodes must be kept in this list (instead of deleting them) 342 * because we need to keep track of their generation number (tn_gen 343 * field). 344 * 345 * Note that nodes are lazily allocated: if the available list is 346 * empty and we have enough space to create more nodes, they will be 347 * created and inserted in the used list. Once these are released, 348 * they will go into the available list, remaining alive until the 349 * file system is unmounted. */ 350 struct tmpfs_node_list tm_nodes_used; 351 352 /* Per-mount malloc zones for tmpfs nodes, names, and dirents */ 353 struct malloc_type *tm_node_zone_obj; 354 struct malloc_type *tm_dirent_zone_obj; 355 struct malloc_type *tm_name_zone; 356 357 ino_t tm_ino; 358 int tm_flags; 359 360 struct netexport tm_export; 361 362 struct mount *tm_mnt; 363 }; 364 365 #define TMPFS_LOCK(tm) lwkt_gettoken(&(tm)->tm_mount->mnt_token) 366 #define TMPFS_UNLOCK(tm) lwkt_reltoken(&(tm)->tm_mount->mnt_token) 367 368 /* --------------------------------------------------------------------- */ 369 370 /* 371 * This structure maps a file identifier to a tmpfs node. Used by the 372 * NFS code. 373 */ 374 struct tmpfs_fid { 375 uint16_t tf_len; 376 uint16_t tf_pad; 377 ino_t tf_id; 378 unsigned long tf_gen; 379 } __packed; 380 381 /* --------------------------------------------------------------------- */ 382 383 /* 384 * Prototypes for tmpfs_subr.c. 385 */ 386 387 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype, 388 uid_t uid, gid_t gid, mode_t mode, char *, int, int, 389 struct tmpfs_node **); 390 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 391 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 392 const char *, uint16_t, struct tmpfs_dirent **); 393 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *); 394 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, 395 struct tmpfs_node *, int, struct vnode **); 396 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 397 struct namecache *, struct ucred *, char *); 398 void tmpfs_dir_attach_locked(struct tmpfs_node *, struct tmpfs_dirent *); 399 void tmpfs_dir_detach_locked(struct tmpfs_node *, struct tmpfs_dirent *); 400 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 401 struct tmpfs_node *f, 402 struct namecache *ncp); 403 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *); 404 int tmpfs_dir_getdotdotdent(struct tmpfs_mount *, 405 struct tmpfs_node *, struct uio *); 406 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t); 407 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *); 408 int tmpfs_reg_resize(struct vnode *, off_t, int); 409 int tmpfs_chflags(struct vnode *, u_long, struct ucred *); 410 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *); 411 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *); 412 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *); 413 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *, 414 int, struct ucred *); 415 void tmpfs_itimes(struct vnode *, const struct timespec *, 416 const struct timespec *); 417 418 void tmpfs_node_init(struct tmpfs_node *node); 419 void tmpfs_node_uninit(struct tmpfs_node *node); 420 void tmpfs_update(struct vnode *); 421 int tmpfs_truncate(struct vnode *, off_t); 422 void tmpfs_lock4(struct tmpfs_node *node1, struct tmpfs_node *node2, 423 struct tmpfs_node *node3, struct tmpfs_node *node4); 424 void tmpfs_unlock4(struct tmpfs_node *node1, struct tmpfs_node *node2, 425 struct tmpfs_node *node3, struct tmpfs_node *node4); 426 427 /* --------------------------------------------------------------------- */ 428 429 /* 430 * Convenience macros to simplify some logical expressions. 431 */ 432 #define IMPLIES(a, b) (!(a) || (b)) 433 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 434 435 /* --------------------------------------------------------------------- */ 436 437 /* 438 * Checks that the directory entry pointed by 'de' matches the name 'name' 439 * with a length of 'len'. 440 */ 441 #define TMPFS_DIRENT_MATCHES(de, name, len) \ 442 (de->td_namelen == (uint16_t)len && \ 443 bcmp((de)->td_name, (name), (de)->td_namelen) == 0) 444 445 /* --------------------------------------------------------------------- */ 446 447 /* 448 * Ensures that the node pointed by 'node' is a directory and that its 449 * contents are consistent with respect to directories. 450 */ 451 #define TMPFS_VALIDATE_DIR(node) do { \ 452 KKASSERT((node)->tn_type == VDIR); \ 453 KKASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 454 } while(0) 455 456 /* --------------------------------------------------------------------- */ 457 458 /* 459 * Macros/functions to convert from generic data structures to tmpfs 460 * specific ones. Kernel code use VP_TO_TMPFS_NODE() instead of VTOI(). 461 */ 462 463 static inline 464 struct tmpfs_mount * 465 VFS_TO_TMPFS(struct mount *mp) 466 { 467 struct tmpfs_mount *tmp; 468 469 KKASSERT((mp) != NULL && (mp)->mnt_data != NULL); 470 tmp = (struct tmpfs_mount *)(mp)->mnt_data; 471 return tmp; 472 } 473 474 static inline 475 struct tmpfs_node * 476 VP_TO_TMPFS_NODE(struct vnode *vp) 477 { 478 struct tmpfs_node *node; 479 480 KKASSERT((vp) != NULL && (vp)->v_data != NULL); 481 node = (struct tmpfs_node *)vp->v_data; 482 return node; 483 } 484 485 static inline 486 struct tmpfs_node * 487 VP_TO_TMPFS_DIR(struct vnode *vp) 488 { 489 struct tmpfs_node *node; 490 491 node = VP_TO_TMPFS_NODE(vp); 492 TMPFS_VALIDATE_DIR(node); 493 return node; 494 } 495 496 /* --------------------------------------------------------------------- */ 497 /* 498 * buffer cache size 499 */ 500 #define TMPFS_BLKSIZE 16384 /* buffer cache size*/ 501 #define TMPFS_BLKMASK (TMPFS_BLKSIZE - 1) 502 #define TMPFS_BLKMASK64 ((off_t)(TMPFS_BLKSIZE - 1)) 503 #endif /* _KERNEL */ 504 505 #endif /* _VFS_TMPFS_TMPFS_H_ */ 506