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/queue.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 #include <sys/systm.h> 54 #include <sys/vmmeter.h> 55 #include <vm/swap_pager.h> 56 57 MALLOC_DECLARE(M_TMPFSMNT); 58 59 /* --------------------------------------------------------------------- */ 60 61 /* 62 * Internal representation of a tmpfs directory entry. 63 */ 64 struct tmpfs_dirent { 65 TAILQ_ENTRY(tmpfs_dirent) td_entries; 66 67 /* Length of the name stored in this directory entry. This avoids 68 * the need to recalculate it every time the name is used. */ 69 uint16_t td_namelen; 70 71 /* The name of the entry, allocated from a string pool. This 72 * string is not required to be zero-terminated; therefore, the 73 * td_namelen field must always be used when accessing its value. */ 74 char * td_name; 75 76 /* Pointer to the node this entry refers to. */ 77 struct tmpfs_node * td_node; 78 }; 79 80 /* A directory in tmpfs holds a sorted list of directory entries, which in 81 * turn point to other files (which can be directories themselves). 82 * 83 * In tmpfs, this list is managed by a tail queue, whose head is defined by 84 * the struct tmpfs_dir type. 85 * 86 * It is imporant to notice that directories do not have entries for . and 87 * .. as other file systems do. These can be generated when requested 88 * based on information available by other means, such as the pointer to 89 * the node itself in the former case or the pointer to the parent directory 90 * in the latter case. This is done to simplify tmpfs's code and, more 91 * importantly, to remove redundancy. */ 92 TAILQ_HEAD(tmpfs_dir, tmpfs_dirent); 93 94 /* Each entry in a directory has a cookie that identifies it. Cookies 95 * supersede offsets within directories because, given how tmpfs stores 96 * directories in memory, there is no such thing as an offset. (Emulating 97 * a real offset could be very difficult.) 98 * 99 * The '.', '..' and the end of directory markers have fixed cookies which 100 * cannot collide with the cookies generated by other entries. The cookies 101 * for the other entries are generated based on the memory address on which 102 * stores their information is stored. 103 * 104 * Ideally, using the entry's memory pointer as the cookie would be enough 105 * to represent it and it wouldn't cause collisions in any system. 106 * Unfortunately, this results in "offsets" with very large values which 107 * later raise problems in the Linux compatibility layer (and maybe in other 108 * places) as described in PR kern/32034. Hence we need to workaround this 109 * with a rather ugly hack. 110 * 111 * Linux 32-bit binaries, unless built with _FILE_OFFSET_BITS=64, have off_t 112 * set to 'long', which is a 32-bit *signed* long integer. Regardless of 113 * the macro value, GLIBC (2.3 at least) always uses the getdents64 114 * system call (when calling readdir) which internally returns off64_t 115 * offsets. In order to make 32-bit binaries work, *GLIBC* converts the 116 * 64-bit values returned by the kernel to 32-bit ones and aborts with 117 * EOVERFLOW if the conversion results in values that won't fit in 32-bit 118 * integers (which it assumes is because the directory is extremely large). 119 * This wouldn't cause problems if we were dealing with unsigned integers, 120 * but as we have signed integers, this check fails due to sign expansion. 121 * 122 * For example, consider that the kernel returns the 0xc1234567 cookie to 123 * userspace in a off64_t integer. Later on, GLIBC casts this value to 124 * off_t (remember, signed) with code similar to: 125 * system call returns the offset in kernel_value; 126 * off_t casted_value = kernel_value; 127 * if (sizeof(off_t) != sizeof(off64_t) && 128 * kernel_value != casted_value) 129 * error! 130 * In this case, casted_value still has 0xc1234567, but when it is compared 131 * for equality against kernel_value, it is promoted to a 64-bit integer and 132 * becomes 0xffffffffc1234567, which is different than 0x00000000c1234567. 133 * Then, GLIBC assumes this is because the directory is very large. 134 * 135 * Given that all the above happens in user-space, we have no control over 136 * it; therefore we must workaround the issue here. We do this by 137 * truncating the pointer value to a 32-bit integer and hope that there 138 * won't be collisions. In fact, this will not cause any problems in 139 * 32-bit platforms but some might arise in 64-bit machines (I'm not sure 140 * if they can happen at all in practice). 141 * 142 * XXX A nicer solution shall be attempted. */ 143 #ifdef _KERNEL 144 #define TMPFS_DIRCOOKIE_DOT 0 145 #define TMPFS_DIRCOOKIE_DOTDOT 1 146 #define TMPFS_DIRCOOKIE_EOF 2 147 static __inline 148 off_t 149 tmpfs_dircookie(struct tmpfs_dirent *de) 150 { 151 off_t cookie; 152 153 cookie = ((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFF; 154 KKASSERT(cookie != TMPFS_DIRCOOKIE_DOT); 155 KKASSERT(cookie != TMPFS_DIRCOOKIE_DOTDOT); 156 KKASSERT(cookie != TMPFS_DIRCOOKIE_EOF); 157 158 return cookie; 159 } 160 #endif 161 162 /* --------------------------------------------------------------------- */ 163 164 /* 165 * Internal representation of a tmpfs file system node. 166 * 167 * This structure is splitted in two parts: one holds attributes common 168 * to all file types and the other holds data that is only applicable to 169 * a particular type. The code must be careful to only access those 170 * attributes that are actually allowed by the node's type. 171 * 172 * 173 * Below is the key of locks used to protected the fields in the following 174 * structures. 175 * 176 */ 177 struct tmpfs_node { 178 /* Doubly-linked list entry which links all existing nodes for a 179 * single file system. This is provided to ease the removal of 180 * all nodes during the unmount operation. */ 181 LIST_ENTRY(tmpfs_node) tn_entries; 182 183 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', 184 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode 185 * types instead of a custom enumeration is to make things simpler 186 * and faster, as we do not need to convert between two types. */ 187 enum vtype tn_type; 188 189 /* Node identifier. */ 190 ino_t tn_id; 191 192 /* Node's internal status. This is used by several file system 193 * operations to do modifications to the node in a delayed 194 * fashion. */ 195 int tn_status; 196 #define TMPFS_NODE_ACCESSED (1 << 1) 197 #define TMPFS_NODE_MODIFIED (1 << 2) 198 #define TMPFS_NODE_CHANGED (1 << 3) 199 200 /* The node size. It does not necessarily match the real amount 201 * of memory consumed by it. */ 202 off_t tn_size; 203 204 /* Generic node attributes. */ 205 uid_t tn_uid; 206 gid_t tn_gid; 207 mode_t tn_mode; 208 int tn_flags; 209 nlink_t tn_links; 210 int32_t tn_atime; 211 int32_t tn_atimensec; 212 int32_t tn_mtime; 213 int32_t tn_mtimensec; 214 int32_t tn_ctime; 215 int32_t tn_ctimensec; 216 unsigned long tn_gen; 217 struct lockf tn_advlock; 218 219 /* As there is a single vnode for each active file within the 220 * system, care has to be taken to avoid allocating more than one 221 * vnode per file. In order to do this, a bidirectional association 222 * is kept between vnodes and nodes. 223 * 224 * Whenever a vnode is allocated, its v_data field is updated to 225 * point to the node it references. At the same time, the node's 226 * tn_vnode field is modified to point to the new vnode representing 227 * it. Further attempts to allocate a vnode for this same node will 228 * result in returning a new reference to the value stored in 229 * tn_vnode. 230 * 231 * May be NULL when the node is unused (that is, no vnode has been 232 * allocated for it or it has been reclaimed). */ 233 struct vnode * tn_vnode; 234 235 /* interlock to protect tn_vpstate */ 236 struct lock tn_interlock; 237 238 /* Identify if current node has vnode assiocate with 239 * or allocating vnode. 240 */ 241 int tn_vpstate; 242 243 /* misc data field for different tn_type node */ 244 union { 245 /* Valid when tn_type == VBLK || tn_type == VCHR. */ 246 dev_t tn_rdev; /*int32_t ?*/ 247 248 /* Valid when tn_type == VDIR. */ 249 struct tn_dir{ 250 /* Pointer to the parent directory. The root 251 * directory has a pointer to itself in this field; 252 * this property identifies the root node. */ 253 struct tmpfs_node * tn_parent; 254 255 /* Head of a tail-queue that links the contents of 256 * the directory together. See above for a 257 * description of its contents. */ 258 struct tmpfs_dir tn_dirhead; 259 260 /* Number and pointer of the first directory entry 261 * returned by the readdir operation if it were 262 * called again to continue reading data from the 263 * same directory as before. This is used to speed 264 * up reads of long directories, assuming that no 265 * more than one read is in progress at a given time. 266 * Otherwise, these values are discarded and a linear 267 * scan is performed from the beginning up to the 268 * point where readdir starts returning values. */ 269 off_t tn_readdir_lastn; 270 struct tmpfs_dirent * tn_readdir_lastp; 271 }tn_dir; 272 273 /* Valid when tn_type == VLNK. */ 274 /* The link's target, allocated from a string pool. */ 275 char * tn_link; 276 277 /* Valid when tn_type == VREG. */ 278 struct tn_reg { 279 /* The contents of regular files stored in a tmpfs 280 * file system are represented by a single anonymous 281 * memory object (aobj, for short). The aobj provides 282 * direct access to any position within the file, 283 * because its contents are always mapped in a 284 * contiguous region of virtual memory. It is a task 285 * of the memory management subsystem (see uvm(9)) to 286 * issue the required page ins or page outs whenever 287 * a position within the file is accessed. */ 288 vm_object_t tn_aobj; 289 size_t tn_aobj_pages; 290 291 }tn_reg; 292 293 /* Valid when tn_type = VFIFO */ 294 struct tn_fifo { 295 int (*tn_fo_read) (struct file *fp, struct uio *uio, 296 struct ucred *cred, int flags); 297 int (*tn_fo_write) (struct file *fp, struct uio *uio, 298 struct ucred *cred, int flags); 299 }tn_fifo; 300 }tn_spec; 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 tn_spec.tn_link 307 #define tn_reg tn_spec.tn_reg 308 #define tn_fifo tn_spec.tn_fifo 309 310 #define TMPFS_NODE_LOCK(node) lockmgr(&(node)->tn_interlock, LK_EXCLUSIVE|LK_RETRY) 311 #define TMPFS_NODE_UNLOCK(node) lockmgr(&(node)->tn_interlock, LK_RELEASE) 312 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock) 313 314 #ifdef INVARIANTS 315 #define TMPFS_ASSERT_LOCKED(node) do { \ 316 KKASSERT(node != NULL); \ 317 KKASSERT(node->tn_vnode != NULL); \ 318 if (!vn_islocked(node->tn_vnode) && \ 319 (lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE )) \ 320 panic("tmpfs: node is not locked: %p", node); \ 321 } while (0) 322 #define TMPFS_ASSERT_ELOCKED(node) do { \ 323 KKASSERT((node) != NULL); \ 324 KKASSERT(lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE); \ 325 } while (0) 326 #else 327 #define TMPFS_ASSERT_LOCKED(node) (void)0 328 #define TMPFS_ASSERT_ELOCKED(node) (void)0 329 #endif 330 331 #define TMPFS_VNODE_ALLOCATING 1 332 #define TMPFS_VNODE_WANT 2 333 #define TMPFS_VNODE_DOOMED 4 334 /* --------------------------------------------------------------------- */ 335 336 /* 337 * Internal representation of a tmpfs mount point. 338 */ 339 struct tmpfs_mount { 340 /* Maximum number of memory pages available for use by the file 341 * system, set during mount time. This variable must never be 342 * used directly as it may be bigger than the current amount of 343 * free memory; in the extreme case, it will hold the SIZE_MAX 344 * value. Instead, use the TMPFS_PAGES_MAX macro. */ 345 vm_pindex_t tm_pages_max; 346 347 /* Number of pages in use by the file system. Cannot be bigger 348 * than the value returned by TMPFS_PAGES_MAX in any case. */ 349 vm_pindex_t tm_pages_used; 350 351 /* Pointer to the node representing the root directory of this 352 * file system. */ 353 struct tmpfs_node * tm_root; 354 355 /* Maximum number of possible nodes for this file system; set 356 * during mount time. We need a hard limit on the maximum number 357 * of nodes to avoid allocating too much of them; their objects 358 * cannot be released until the file system is unmounted. 359 * Otherwise, we could easily run out of memory by creating lots 360 * of empty files and then simply removing them. */ 361 ino_t tm_nodes_max; 362 363 /* Number of nodes currently that are in use. */ 364 ino_t tm_nodes_inuse; 365 366 /* maximum representable file size */ 367 u_int64_t tm_maxfilesize; 368 369 /* Nodes are organized in two different lists. The used list 370 * contains all nodes that are currently used by the file system; 371 * i.e., they refer to existing files. The available list contains 372 * all nodes that are currently available for use by new files. 373 * Nodes must be kept in this list (instead of deleting them) 374 * because we need to keep track of their generation number (tn_gen 375 * field). 376 * 377 * Note that nodes are lazily allocated: if the available list is 378 * empty and we have enough space to create more nodes, they will be 379 * created and inserted in the used list. Once these are released, 380 * they will go into the available list, remaining alive until the 381 * file system is unmounted. */ 382 struct tmpfs_node_list tm_nodes_used; 383 384 /* All node lock to protect the node list and tmp_pages_used */ 385 struct lock allnode_lock; 386 387 /* Per-mount malloc zones for tmpfs nodes, names, and dirents */ 388 struct malloc_type *tm_node_zone; 389 struct malloc_type *tm_dirent_zone; 390 struct malloc_type *tm_name_zone; 391 392 struct objcache_malloc_args tm_node_zone_malloc_args; 393 struct objcache_malloc_args tm_dirent_zone_malloc_args; 394 395 /* Pools used to store file system meta data. These are not shared 396 * across several instances of tmpfs for the reasons described in 397 * tmpfs_pool.c. */ 398 struct objcache *tm_dirent_pool; 399 struct objcache *tm_node_pool; 400 401 int tm_flags; 402 }; 403 404 #define TMPFS_LOCK(tm) lockmgr(&(tm)->allnode_lock, LK_EXCLUSIVE|LK_RETRY) 405 #define TMPFS_UNLOCK(tm) lockmgr(&(tm)->allnode_lock, LK_RELEASE) 406 407 /* --------------------------------------------------------------------- */ 408 409 /* 410 * This structure maps a file identifier to a tmpfs node. Used by the 411 * NFS code. 412 */ 413 struct tmpfs_fid { 414 uint16_t tf_len; 415 uint16_t tf_pad; 416 ino_t tf_id; 417 unsigned long tf_gen; 418 }; 419 420 /* --------------------------------------------------------------------- */ 421 422 #ifdef _KERNEL 423 /* 424 * Prototypes for tmpfs_subr.c. 425 */ 426 427 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype, 428 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *, 429 char *, int, int, struct tmpfs_node **); 430 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 431 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 432 const char *, uint16_t, struct tmpfs_dirent **); 433 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *); 434 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int, 435 struct vnode **); 436 void tmpfs_free_vp(struct vnode *); 437 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 438 struct namecache *, struct ucred *, char *); 439 void tmpfs_dir_attach(struct tmpfs_node *, struct tmpfs_dirent *); 440 void tmpfs_dir_detach(struct tmpfs_node *, struct tmpfs_dirent *); 441 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 442 struct tmpfs_node *f, 443 struct namecache *ncp); 444 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *); 445 int tmpfs_dir_getdotdotdent(struct tmpfs_mount *, 446 struct tmpfs_node *, struct uio *); 447 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t); 448 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *); 449 int tmpfs_reg_resize(struct vnode *, off_t, int); 450 int tmpfs_chflags(struct vnode *, int, struct ucred *); 451 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *); 452 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *); 453 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *); 454 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *, 455 int, struct ucred *); 456 void tmpfs_itimes(struct vnode *, const struct timespec *, 457 const struct timespec *); 458 459 void tmpfs_update(struct vnode *); 460 int tmpfs_truncate(struct vnode *, off_t); 461 int tmpfs_node_ctor(void *obj, void *privdata, int flags); 462 463 /* --------------------------------------------------------------------- */ 464 465 /* 466 * Convenience macros to simplify some logical expressions. 467 */ 468 #define IMPLIES(a, b) (!(a) || (b)) 469 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 470 471 /* --------------------------------------------------------------------- */ 472 473 /* 474 * Checks that the directory entry pointed by 'de' matches the name 'name' 475 * with a length of 'len'. 476 */ 477 #define TMPFS_DIRENT_MATCHES(de, name, len) \ 478 (de->td_namelen == (uint16_t)len && \ 479 bcmp((de)->td_name, (name), (de)->td_namelen) == 0) 480 481 /* --------------------------------------------------------------------- */ 482 483 /* 484 * Ensures that the node pointed by 'node' is a directory and that its 485 * contents are consistent with respect to directories. 486 */ 487 #define TMPFS_VALIDATE_DIR(node) \ 488 KKASSERT((node)->tn_type == VDIR); \ 489 KKASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 490 KKASSERT((node)->tn_dir.tn_readdir_lastp == NULL || \ 491 tmpfs_dircookie((node)->tn_dir.tn_readdir_lastp) == (node)->tn_dir.tn_readdir_lastn); 492 493 #endif 494 495 /* --------------------------------------------------------------------- */ 496 497 /* 498 * Macros/functions to convert from generic data structures to tmpfs 499 * specific ones. 500 */ 501 502 static inline 503 struct tmpfs_mount * 504 VFS_TO_TMPFS(struct mount *mp) 505 { 506 struct tmpfs_mount *tmp; 507 508 KKASSERT((mp) != NULL && (mp)->mnt_data != NULL); 509 tmp = (struct tmpfs_mount *)(mp)->mnt_data; 510 return tmp; 511 } 512 513 static inline 514 struct tmpfs_node * 515 VP_TO_TMPFS_NODE(struct vnode *vp) 516 { 517 struct tmpfs_node *node; 518 519 KKASSERT((vp) != NULL && (vp)->v_data != NULL); 520 node = (struct tmpfs_node *)vp->v_data; 521 return node; 522 } 523 524 static inline 525 struct tmpfs_node * 526 VP_TO_TMPFS_DIR(struct vnode *vp) 527 { 528 struct tmpfs_node *node; 529 530 node = VP_TO_TMPFS_NODE(vp); 531 TMPFS_VALIDATE_DIR(node); 532 return node; 533 } 534 535 /* --------------------------------------------------------------------- */ 536 /* 537 * buffer cache size 538 */ 539 #define BSIZE (off_t)16384 /* buffer cache size*/ 540 #define BMASK (off_t)(BSIZE - 1) 541 542 #endif /* _VFS_TMPFS_TMPFS_H_ */ 543