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 MALLOC_DECLARE(M_TMPFSMNT); 60 61 /* --------------------------------------------------------------------- */ 62 63 /* 64 * Internal representation of a tmpfs directory entry. 65 */ 66 struct tmpfs_dirent { 67 RB_ENTRY(tmpfs_dirent) rb_node; 68 69 /* Length of the name stored in this directory entry. This avoids 70 * the need to recalculate it every time the name is used. */ 71 uint16_t td_namelen; 72 73 /* The name of the entry, allocated from a string pool. This 74 * string is not required to be zero-terminated; therefore, the 75 * td_namelen field must always be used when accessing its value. */ 76 char *td_name; 77 78 /* Pointer to the node this entry refers to. */ 79 struct tmpfs_node *td_node; 80 }; 81 82 struct tmpfs_dirtree; 83 RB_HEAD(tmpfs_dirtree, tmpfs_dirent); 84 RB_PROTOTYPE(tmpfs_dirtree, tmpfs_dirent, rb_node, 85 tmpfs_dirtree_compare); 86 87 88 /* A directory in tmpfs holds a set of directory entries, which in 89 * turn point to other files (which can be directories themselves). 90 * 91 * In tmpfs, this set is managed by a red-black tree, whose root is defined 92 * by the struct tmpfs_dirtree type. 93 * 94 * It is important to notice that directories do not have entries for . and 95 * .. as other file systems do. These can be generated when requested 96 * based on information available by other means, such as the pointer to 97 * the node itself in the former case or the pointer to the parent directory 98 * in the latter case. This is done to simplify tmpfs's code and, more 99 * importantly, to remove redundancy. */ 100 101 /* Each entry in a directory has a cookie that identifies it. Cookies 102 * supersede offsets within directories because, given how tmpfs stores 103 * directories in memory, there is no such thing as an offset. (Emulating 104 * a real offset could be very difficult.) 105 * 106 * The '.', '..' and the end of directory markers have fixed cookies which 107 * cannot collide with the cookies generated by other entries. The cookies 108 * for the other entries are generated based on the memory address on which 109 * stores their information is stored. 110 * 111 * Ideally, using the entry's memory pointer as the cookie would be enough 112 * to represent it and it wouldn't cause collisions in any system. 113 * Unfortunately, this results in "offsets" with very large values which 114 * later raise problems in the Linux compatibility layer (and maybe in other 115 * places) as described in PR kern/32034. Hence we need to workaround this 116 * with a rather ugly hack. 117 * 118 * Linux 32-bit binaries, unless built with _FILE_OFFSET_BITS=64, have off_t 119 * set to 'long', which is a 32-bit *signed* long integer. Regardless of 120 * the macro value, GLIBC (2.3 at least) always uses the getdents64 121 * system call (when calling readdir) which internally returns off64_t 122 * offsets. In order to make 32-bit binaries work, *GLIBC* converts the 123 * 64-bit values returned by the kernel to 32-bit ones and aborts with 124 * EOVERFLOW if the conversion results in values that won't fit in 32-bit 125 * integers (which it assumes is because the directory is extremely large). 126 * This wouldn't cause problems if we were dealing with unsigned integers, 127 * but as we have signed integers, this check fails due to sign expansion. 128 * 129 * For example, consider that the kernel returns the 0xc1234567 cookie to 130 * userspace in a off64_t integer. Later on, GLIBC casts this value to 131 * off_t (remember, signed) with code similar to: 132 * system call returns the offset in kernel_value; 133 * off_t casted_value = kernel_value; 134 * if (sizeof(off_t) != sizeof(off64_t) && 135 * kernel_value != casted_value) 136 * error! 137 * In this case, casted_value still has 0xc1234567, but when it is compared 138 * for equality against kernel_value, it is promoted to a 64-bit integer and 139 * becomes 0xffffffffc1234567, which is different than 0x00000000c1234567. 140 * Then, GLIBC assumes this is because the directory is very large. 141 * 142 * Given that all the above happens in user-space, we have no control over 143 * it; therefore we must workaround the issue here. We do this by 144 * truncating the pointer value to a 32-bit integer and hope that there 145 * won't be collisions. In fact, this will not cause any problems in 146 * 32-bit platforms but some might arise in 64-bit machines (I'm not sure 147 * if they can happen at all in practice). 148 * 149 * XXX A nicer solution shall be attempted. */ 150 #ifdef _KERNEL 151 #define TMPFS_DIRCOOKIE_DOT 0 152 #define TMPFS_DIRCOOKIE_DOTDOT 1 153 #define TMPFS_DIRCOOKIE_EOF 2 154 static __inline 155 off_t 156 tmpfs_dircookie(struct tmpfs_dirent *de) 157 { 158 off_t cookie; 159 160 cookie = ((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFF; 161 KKASSERT(cookie != TMPFS_DIRCOOKIE_DOT); 162 KKASSERT(cookie != TMPFS_DIRCOOKIE_DOTDOT); 163 KKASSERT(cookie != TMPFS_DIRCOOKIE_EOF); 164 165 return cookie; 166 } 167 #endif 168 169 /* --------------------------------------------------------------------- */ 170 171 /* 172 * Internal representation of a tmpfs file system node. 173 * 174 * This structure is splitted in two parts: one holds attributes common 175 * to all file types and the other holds data that is only applicable to 176 * a particular type. The code must be careful to only access those 177 * attributes that are actually allowed by the node's type. 178 */ 179 struct tmpfs_node { 180 /* Doubly-linked list entry which links all existing nodes for a 181 * single file system. This is provided to ease the removal of 182 * all nodes during the unmount operation. */ 183 LIST_ENTRY(tmpfs_node) tn_entries; 184 185 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', 186 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode 187 * types instead of a custom enumeration is to make things simpler 188 * and faster, as we do not need to convert between two types. */ 189 enum vtype tn_type; 190 191 /* Node identifier. */ 192 ino_t tn_id; 193 194 /* Node's internal status. This is used by several file system 195 * operations to do modifications to the node in a delayed 196 * fashion. */ 197 int tn_status; 198 #define TMPFS_NODE_ACCESSED (1 << 1) 199 #define TMPFS_NODE_MODIFIED (1 << 2) 200 #define TMPFS_NODE_CHANGED (1 << 3) 201 202 /* The node size. It does not necessarily match the real amount 203 * of memory consumed by it. */ 204 off_t tn_size; 205 206 /* Generic node attributes. */ 207 uid_t tn_uid; 208 gid_t tn_gid; 209 mode_t tn_mode; 210 int tn_flags; 211 nlink_t tn_links; 212 int32_t tn_atime; 213 int32_t tn_atimensec; 214 int32_t tn_mtime; 215 int32_t tn_mtimensec; 216 int32_t tn_ctime; 217 int32_t tn_ctimensec; 218 unsigned long tn_gen; 219 struct lockf tn_advlock; 220 221 /* As there is a single vnode for each active file within the 222 * system, care has to be taken to avoid allocating more than one 223 * vnode per file. In order to do this, a bidirectional association 224 * is kept between vnodes and nodes. 225 * 226 * Whenever a vnode is allocated, its v_data field is updated to 227 * point to the node it references. At the same time, the node's 228 * tn_vnode field is modified to point to the new vnode representing 229 * it. Further attempts to allocate a vnode for this same node will 230 * result in returning a new reference to the value stored in 231 * tn_vnode. 232 * 233 * May be NULL when the node is unused (that is, no vnode has been 234 * allocated for it or it has been reclaimed). */ 235 struct vnode * tn_vnode; 236 237 /* interlock to protect tn_vpstate */ 238 struct lock tn_interlock; 239 240 /* Identify if current node has vnode assiocate with 241 * or allocating vnode. 242 */ 243 int tn_vpstate; 244 245 /* misc data field for different tn_type node */ 246 union { 247 /* Valid when tn_type == VBLK || tn_type == VCHR. */ 248 dev_t tn_rdev; /*int32_t ?*/ 249 250 /* Valid when tn_type == VDIR. */ 251 struct tn_dir { 252 /* Pointer to the parent directory. The root 253 * directory has a pointer to itself in this field; 254 * this property identifies the root node. */ 255 struct tmpfs_node * tn_parent; 256 257 /* Root of a red-black tree that links the contents of 258 * the directory together. See above for a 259 * description of its contents. */ 260 struct tmpfs_dirtree tn_dirtree; 261 262 /* Number and pointer of the first directory entry 263 * returned by the readdir operation if it were 264 * called again to continue reading data from the 265 * same directory as before. This is used to speed 266 * up reads of long directories, assuming that no 267 * more than one read is in progress at a given time. 268 * Otherwise, these values are discarded and a linear 269 * scan is performed from the beginning up to the 270 * point where readdir starts returning values. */ 271 off_t tn_readdir_lastn; 272 struct tmpfs_dirent * tn_readdir_lastp; 273 } tn_dir; 274 275 /* Valid when tn_type == VLNK. */ 276 /* The link's target, allocated from a string pool. */ 277 char * tn_link; 278 279 /* Valid when tn_type == VREG. */ 280 struct tn_reg { 281 /* The contents of regular files stored in a tmpfs 282 * file system are represented by a single anonymous 283 * memory object (aobj, for short). The aobj provides 284 * direct access to any position within the file, 285 * because its contents are always mapped in a 286 * contiguous region of virtual memory. It is a task 287 * of the memory management subsystem (see uvm(9)) to 288 * issue the required page ins or page outs whenever 289 * a position within the file is accessed. */ 290 vm_object_t tn_aobj; 291 size_t tn_aobj_pages; 292 293 } tn_reg; 294 295 /* Valid when tn_type = VFIFO */ 296 struct tn_fifo { 297 int (*tn_fo_read) (struct file *fp, struct uio *uio, 298 struct ucred *cred, int flags); 299 int (*tn_fo_write) (struct file *fp, struct uio *uio, 300 struct ucred *cred, int flags); 301 } tn_fifo; 302 } tn_spec; 303 }; 304 305 #define VTOI(vp) ((struct tmpfs_node *)(vp)->v_data) 306 307 #ifdef _KERNEL 308 LIST_HEAD(tmpfs_node_list, tmpfs_node); 309 310 #define tn_rdev tn_spec.tn_rdev 311 #define tn_dir tn_spec.tn_dir 312 #define tn_link tn_spec.tn_link 313 #define tn_reg tn_spec.tn_reg 314 #define tn_fifo tn_spec.tn_fifo 315 316 #define TMPFS_NODE_LOCK(node) lockmgr(&(node)->tn_interlock, LK_EXCLUSIVE|LK_RETRY) 317 #define TMPFS_NODE_LOCK_SH(node) lockmgr(&(node)->tn_interlock, LK_SHARED|LK_RETRY) 318 #define TMPFS_NODE_UNLOCK(node) lockmgr(&(node)->tn_interlock, LK_RELEASE) 319 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock) 320 321 #ifdef INVARIANTS 322 #define TMPFS_ASSERT_LOCKED(node) do { \ 323 KKASSERT(node != NULL); \ 324 KKASSERT(node->tn_vnode != NULL); \ 325 if (!vn_islocked(node->tn_vnode) && \ 326 (lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE )) \ 327 panic("tmpfs: node is not locked: %p", node); \ 328 } while (0) 329 #define TMPFS_ASSERT_ELOCKED(node) do { \ 330 KKASSERT((node) != NULL); \ 331 KKASSERT(lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE); \ 332 } while (0) 333 #else 334 #define TMPFS_ASSERT_LOCKED(node) (void)0 335 #define TMPFS_ASSERT_ELOCKED(node) (void)0 336 #endif 337 338 #define TMPFS_VNODE_ALLOCATING 1 339 #define TMPFS_VNODE_WANT 2 340 #define TMPFS_VNODE_DOOMED 4 341 /* --------------------------------------------------------------------- */ 342 343 /* 344 * Internal representation of a tmpfs mount point. 345 */ 346 struct tmpfs_mount { 347 struct mount *tm_mount; 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 SIZE_MAX 353 * value. Instead, use the TMPFS_PAGES_MAX macro. */ 354 vm_pindex_t tm_pages_max; 355 356 /* Number of pages in use by the file system. Cannot be bigger 357 * than the value returned by TMPFS_PAGES_MAX in any case. */ 358 vm_pindex_t tm_pages_used; 359 360 /* Pointer to the node representing the root directory of this 361 * file system. */ 362 struct tmpfs_node * tm_root; 363 364 /* Maximum number of possible nodes for this file system; set 365 * during mount time. We need a hard limit on the maximum number 366 * of nodes to avoid allocating too much of them; their objects 367 * cannot be released until the file system is unmounted. 368 * Otherwise, we could easily run out of memory by creating lots 369 * of empty files and then simply removing them. */ 370 ino_t tm_nodes_max; 371 372 /* Number of nodes currently that are in use. */ 373 ino_t tm_nodes_inuse; 374 375 /* maximum representable file size */ 376 u_int64_t tm_maxfilesize; 377 378 /* Nodes are organized in two different lists. The used list 379 * contains all nodes that are currently used by the file system; 380 * i.e., they refer to existing files. The available list contains 381 * all nodes that are currently available for use by new files. 382 * Nodes must be kept in this list (instead of deleting them) 383 * because we need to keep track of their generation number (tn_gen 384 * field). 385 * 386 * Note that nodes are lazily allocated: if the available list is 387 * empty and we have enough space to create more nodes, they will be 388 * created and inserted in the used list. Once these are released, 389 * they will go into the available list, remaining alive until the 390 * file system is unmounted. */ 391 struct tmpfs_node_list tm_nodes_used; 392 393 /* Per-mount malloc zones for tmpfs nodes, names, and dirents */ 394 struct malloc_type *tm_node_zone; 395 struct malloc_type *tm_dirent_zone; 396 struct malloc_type *tm_name_zone; 397 398 struct objcache_malloc_args tm_node_zone_malloc_args; 399 struct objcache_malloc_args tm_dirent_zone_malloc_args; 400 401 /* Pools used to store file system meta data. */ 402 struct objcache *tm_dirent_pool; 403 struct objcache *tm_node_pool; 404 405 int tm_ino; 406 int tm_flags; 407 408 struct netexport tm_export; 409 410 struct mount *tm_mnt; 411 }; 412 413 #define TMPFS_LOCK(tm) lwkt_gettoken(&(tm)->tm_mount->mnt_token) 414 #define TMPFS_UNLOCK(tm) lwkt_reltoken(&(tm)->tm_mount->mnt_token) 415 416 /* --------------------------------------------------------------------- */ 417 418 /* 419 * This structure maps a file identifier to a tmpfs node. Used by the 420 * NFS code. 421 */ 422 struct tmpfs_fid { 423 uint16_t tf_len; 424 uint16_t tf_pad; 425 ino_t tf_id; 426 unsigned long tf_gen; 427 }; 428 429 /* --------------------------------------------------------------------- */ 430 431 #ifdef _KERNEL 432 /* 433 * Prototypes for tmpfs_subr.c. 434 */ 435 436 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype, 437 uid_t uid, gid_t gid, mode_t mode, char *, int, int, 438 struct tmpfs_node **); 439 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 440 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 441 const char *, uint16_t, struct tmpfs_dirent **); 442 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *); 443 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int, 444 struct vnode **); 445 void tmpfs_free_vp(struct vnode *); 446 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 447 struct namecache *, struct ucred *, char *); 448 void tmpfs_dir_attach(struct tmpfs_node *, struct tmpfs_dirent *); 449 void tmpfs_dir_detach(struct tmpfs_node *, struct tmpfs_dirent *); 450 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 451 struct tmpfs_node *f, 452 struct namecache *ncp); 453 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *); 454 int tmpfs_dir_getdotdotdent(struct tmpfs_mount *, 455 struct tmpfs_node *, struct uio *); 456 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t); 457 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *); 458 int tmpfs_reg_resize(struct vnode *, off_t, int); 459 int tmpfs_chflags(struct vnode *, int, struct ucred *); 460 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *); 461 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *); 462 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *); 463 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *, 464 int, struct ucred *); 465 void tmpfs_itimes(struct vnode *, const struct timespec *, 466 const struct timespec *); 467 468 void tmpfs_update(struct vnode *); 469 int tmpfs_truncate(struct vnode *, off_t); 470 int tmpfs_node_ctor(void *obj, void *privdata, int flags); 471 472 /* --------------------------------------------------------------------- */ 473 474 /* 475 * Convenience macros to simplify some logical expressions. 476 */ 477 #define IMPLIES(a, b) (!(a) || (b)) 478 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 479 480 /* --------------------------------------------------------------------- */ 481 482 /* 483 * Checks that the directory entry pointed by 'de' matches the name 'name' 484 * with a length of 'len'. 485 */ 486 #define TMPFS_DIRENT_MATCHES(de, name, len) \ 487 (de->td_namelen == (uint16_t)len && \ 488 bcmp((de)->td_name, (name), (de)->td_namelen) == 0) 489 490 /* --------------------------------------------------------------------- */ 491 492 /* 493 * Ensures that the node pointed by 'node' is a directory and that its 494 * contents are consistent with respect to directories. 495 */ 496 #define TMPFS_VALIDATE_DIR(node) \ 497 KKASSERT((node)->tn_type == VDIR); \ 498 KKASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 499 KKASSERT((node)->tn_dir.tn_readdir_lastp == NULL || \ 500 tmpfs_dircookie((node)->tn_dir.tn_readdir_lastp) == (node)->tn_dir.tn_readdir_lastn); 501 502 #endif 503 504 /* --------------------------------------------------------------------- */ 505 506 /* 507 * Macros/functions to convert from generic data structures to tmpfs 508 * specific ones. 509 */ 510 511 static inline 512 struct tmpfs_mount * 513 VFS_TO_TMPFS(struct mount *mp) 514 { 515 struct tmpfs_mount *tmp; 516 517 KKASSERT((mp) != NULL && (mp)->mnt_data != NULL); 518 tmp = (struct tmpfs_mount *)(mp)->mnt_data; 519 return tmp; 520 } 521 522 static inline 523 struct tmpfs_node * 524 VP_TO_TMPFS_NODE(struct vnode *vp) 525 { 526 struct tmpfs_node *node; 527 528 KKASSERT((vp) != NULL && (vp)->v_data != NULL); 529 node = (struct tmpfs_node *)vp->v_data; 530 return node; 531 } 532 533 static inline 534 struct tmpfs_node * 535 VP_TO_TMPFS_DIR(struct vnode *vp) 536 { 537 struct tmpfs_node *node; 538 539 node = VP_TO_TMPFS_NODE(vp); 540 TMPFS_VALIDATE_DIR(node); 541 return node; 542 } 543 544 /* --------------------------------------------------------------------- */ 545 /* 546 * buffer cache size 547 */ 548 #define TMPFS_BLKSIZE 16384 /* buffer cache size*/ 549 #define TMPFS_BLKMASK (TMPFS_BLKSIZE - 1) 550 #define TMPFS_BLKMASK64 ((off_t)(TMPFS_BLKSIZE - 1)) 551 #endif /* _KERNEL */ 552 553 #endif /* _VFS_TMPFS_TMPFS_H_ */ 554