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 MALLOC_DECLARE(M_TMPFSNAME); 59 60 /* --------------------------------------------------------------------- */ 61 62 /* 63 * Internal representation of a tmpfs directory entry. 64 */ 65 struct tmpfs_dirent { 66 TAILQ_ENTRY(tmpfs_dirent) td_entries; 67 68 /* Length of the name stored in this directory entry. This avoids 69 * the need to recalculate it every time the name is used. */ 70 uint16_t td_namelen; 71 72 /* The name of the entry, allocated from a string pool. This 73 * string is not required to be zero-terminated; therefore, the 74 * td_namelen field must always be used when accessing its value. */ 75 char * td_name; 76 77 /* Pointer to the node this entry refers to. */ 78 struct tmpfs_node * td_node; 79 }; 80 81 /* A directory in tmpfs holds a sorted list of directory entries, which in 82 * turn point to other files (which can be directories themselves). 83 * 84 * In tmpfs, this list is managed by a tail queue, whose head is defined by 85 * the struct tmpfs_dir type. 86 * 87 * It is imporant to notice that directories do not have entries for . and 88 * .. as other file systems do. These can be generated when requested 89 * based on information available by other means, such as the pointer to 90 * the node itself in the former case or the pointer to the parent directory 91 * in the latter case. This is done to simplify tmpfs's code and, more 92 * importantly, to remove redundancy. */ 93 TAILQ_HEAD(tmpfs_dir, tmpfs_dirent); 94 95 /* Each entry in a directory has a cookie that identifies it. Cookies 96 * supersede offsets within directories because, given how tmpfs stores 97 * directories in memory, there is no such thing as an offset. (Emulating 98 * a real offset could be very difficult.) 99 * 100 * The '.', '..' and the end of directory markers have fixed cookies which 101 * cannot collide with the cookies generated by other entries. The cookies 102 * for the other entries are generated based on the memory address on which 103 * stores their information is stored. 104 * 105 * Ideally, using the entry's memory pointer as the cookie would be enough 106 * to represent it and it wouldn't cause collisions in any system. 107 * Unfortunately, this results in "offsets" with very large values which 108 * later raise problems in the Linux compatibility layer (and maybe in other 109 * places) as described in PR kern/32034. Hence we need to workaround this 110 * with a rather ugly hack. 111 * 112 * Linux 32-bit binaries, unless built with _FILE_OFFSET_BITS=64, have off_t 113 * set to 'long', which is a 32-bit *signed* long integer. Regardless of 114 * the macro value, GLIBC (2.3 at least) always uses the getdents64 115 * system call (when calling readdir) which internally returns off64_t 116 * offsets. In order to make 32-bit binaries work, *GLIBC* converts the 117 * 64-bit values returned by the kernel to 32-bit ones and aborts with 118 * EOVERFLOW if the conversion results in values that won't fit in 32-bit 119 * integers (which it assumes is because the directory is extremely large). 120 * This wouldn't cause problems if we were dealing with unsigned integers, 121 * but as we have signed integers, this check fails due to sign expansion. 122 * 123 * For example, consider that the kernel returns the 0xc1234567 cookie to 124 * userspace in a off64_t integer. Later on, GLIBC casts this value to 125 * off_t (remember, signed) with code similar to: 126 * system call returns the offset in kernel_value; 127 * off_t casted_value = kernel_value; 128 * if (sizeof(off_t) != sizeof(off64_t) && 129 * kernel_value != casted_value) 130 * error! 131 * In this case, casted_value still has 0xc1234567, but when it is compared 132 * for equality against kernel_value, it is promoted to a 64-bit integer and 133 * becomes 0xffffffffc1234567, which is different than 0x00000000c1234567. 134 * Then, GLIBC assumes this is because the directory is very large. 135 * 136 * Given that all the above happens in user-space, we have no control over 137 * it; therefore we must workaround the issue here. We do this by 138 * truncating the pointer value to a 32-bit integer and hope that there 139 * won't be collisions. In fact, this will not cause any problems in 140 * 32-bit platforms but some might arise in 64-bit machines (I'm not sure 141 * if they can happen at all in practice). 142 * 143 * XXX A nicer solution shall be attempted. */ 144 #ifdef _KERNEL 145 #define TMPFS_DIRCOOKIE_DOT 0 146 #define TMPFS_DIRCOOKIE_DOTDOT 1 147 #define TMPFS_DIRCOOKIE_EOF 2 148 static __inline 149 off_t 150 tmpfs_dircookie(struct tmpfs_dirent *de) 151 { 152 off_t cookie; 153 154 cookie = ((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFF; 155 KKASSERT(cookie != TMPFS_DIRCOOKIE_DOT); 156 KKASSERT(cookie != TMPFS_DIRCOOKIE_DOTDOT); 157 KKASSERT(cookie != TMPFS_DIRCOOKIE_EOF); 158 159 return cookie; 160 } 161 #endif 162 163 /* --------------------------------------------------------------------- */ 164 165 /* 166 * Internal representation of a tmpfs file system node. 167 * 168 * This structure is splitted in two parts: one holds attributes common 169 * to all file types and the other holds data that is only applicable to 170 * a particular type. The code must be careful to only access those 171 * attributes that are actually allowed by the node's type. 172 * 173 * 174 * Below is the key of locks used to protected the fields in the following 175 * structures. 176 * 177 */ 178 struct tmpfs_node { 179 /* Doubly-linked list entry which links all existing nodes for a 180 * single file system. This is provided to ease the removal of 181 * all nodes during the unmount operation. */ 182 LIST_ENTRY(tmpfs_node) tn_entries; 183 184 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', 185 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode 186 * types instead of a custom enumeration is to make things simpler 187 * and faster, as we do not need to convert between two types. */ 188 enum vtype tn_type; 189 190 /* Node identifier. */ 191 ino_t tn_id; 192 193 /* Node's internal status. This is used by several file system 194 * operations to do modifications to the node in a delayed 195 * fashion. */ 196 int tn_status; 197 #define TMPFS_NODE_ACCESSED (1 << 1) 198 #define TMPFS_NODE_MODIFIED (1 << 2) 199 #define TMPFS_NODE_CHANGED (1 << 3) 200 201 /* The node size. It does not necessarily match the real amount 202 * of memory consumed by it. */ 203 off_t tn_size; 204 205 /* Generic node attributes. */ 206 uid_t tn_uid; 207 gid_t tn_gid; 208 mode_t tn_mode; 209 int tn_flags; 210 nlink_t tn_links; 211 int32_t tn_atime; 212 int32_t tn_atimensec; 213 int32_t tn_mtime; 214 int32_t tn_mtimensec; 215 int32_t tn_ctime; 216 int32_t tn_ctimensec; 217 unsigned long tn_gen; 218 struct lockf tn_advlock; 219 220 /* As there is a single vnode for each active file within the 221 * system, care has to be taken to avoid allocating more than one 222 * vnode per file. In order to do this, a bidirectional association 223 * is kept between vnodes and nodes. 224 * 225 * Whenever a vnode is allocated, its v_data field is updated to 226 * point to the node it references. At the same time, the node's 227 * tn_vnode field is modified to point to the new vnode representing 228 * it. Further attempts to allocate a vnode for this same node will 229 * result in returning a new reference to the value stored in 230 * tn_vnode. 231 * 232 * May be NULL when the node is unused (that is, no vnode has been 233 * allocated for it or it has been reclaimed). */ 234 struct vnode * tn_vnode; 235 236 /* interlock to protect tn_vpstate */ 237 struct lock tn_interlock; 238 239 /* Identify if current node has vnode assiocate with 240 * or allocating vnode. 241 */ 242 int tn_vpstate; 243 244 /* misc data field for different tn_type node */ 245 union { 246 /* Valid when tn_type == VBLK || tn_type == VCHR. */ 247 dev_t tn_rdev; /*int32_t ?*/ 248 249 /* Valid when tn_type == VDIR. */ 250 struct tn_dir{ 251 /* Pointer to the parent directory. The root 252 * directory has a pointer to itself in this field; 253 * this property identifies the root node. */ 254 struct tmpfs_node * tn_parent; 255 256 /* Head of a tail-queue that links the contents of 257 * the directory together. See above for a 258 * description of its contents. */ 259 struct tmpfs_dir tn_dirhead; 260 261 /* Number and pointer of the first directory entry 262 * returned by the readdir operation if it were 263 * called again to continue reading data from the 264 * same directory as before. This is used to speed 265 * up reads of long directories, assuming that no 266 * more than one read is in progress at a given time. 267 * Otherwise, these values are discarded and a linear 268 * scan is performed from the beginning up to the 269 * point where readdir starts returning values. */ 270 off_t tn_readdir_lastn; 271 struct tmpfs_dirent * tn_readdir_lastp; 272 }tn_dir; 273 274 /* Valid when tn_type == VLNK. */ 275 /* The link's target, allocated from a string pool. */ 276 char * tn_link; 277 278 /* Valid when tn_type == VREG. */ 279 struct tn_reg { 280 /* The contents of regular files stored in a tmpfs 281 * file system are represented by a single anonymous 282 * memory object (aobj, for short). The aobj provides 283 * direct access to any position within the file, 284 * because its contents are always mapped in a 285 * contiguous region of virtual memory. It is a task 286 * of the memory management subsystem (see uvm(9)) to 287 * issue the required page ins or page outs whenever 288 * a position within the file is accessed. */ 289 vm_object_t tn_aobj; 290 size_t tn_aobj_pages; 291 292 }tn_reg; 293 294 /* Valid when tn_type = VFIFO */ 295 struct tn_fifo { 296 int (*tn_fo_read) (struct file *fp, struct uio *uio, 297 struct ucred *cred, int flags); 298 int (*tn_fo_write) (struct file *fp, struct uio *uio, 299 struct ucred *cred, int flags); 300 }tn_fifo; 301 }tn_spec; 302 }; 303 LIST_HEAD(tmpfs_node_list, tmpfs_node); 304 305 #define tn_rdev tn_spec.tn_rdev 306 #define tn_dir tn_spec.tn_dir 307 #define tn_link tn_spec.tn_link 308 #define tn_reg tn_spec.tn_reg 309 #define tn_fifo tn_spec.tn_fifo 310 311 #define TMPFS_NODE_LOCK(node) lockmgr(&(node)->tn_interlock, LK_EXCLUSIVE|LK_RETRY) 312 #define TMPFS_NODE_UNLOCK(node) lockmgr(&(node)->tn_interlock, LK_RELEASE) 313 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock) 314 315 #ifdef INVARIANTS 316 #define TMPFS_ASSERT_LOCKED(node) do { \ 317 KKASSERT(node != NULL); \ 318 KKASSERT(node->tn_vnode != NULL); \ 319 if (!vn_islocked(node->tn_vnode) && \ 320 (lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE )) \ 321 panic("tmpfs: node is not locked: %p", node); \ 322 } while (0) 323 #define TMPFS_ASSERT_ELOCKED(node) do { \ 324 KKASSERT((node) != NULL); \ 325 KKASSERT(lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE); \ 326 } while (0) 327 #else 328 #define TMPFS_ASSERT_LOCKED(node) (void)0 329 #define TMPFS_ASSERT_ELOCKED(node) (void)0 330 #endif 331 332 #define TMPFS_VNODE_ALLOCATING 1 333 #define TMPFS_VNODE_WANT 2 334 #define TMPFS_VNODE_DOOMED 4 335 /* --------------------------------------------------------------------- */ 336 337 /* 338 * Internal representation of a tmpfs mount point. 339 */ 340 struct tmpfs_mount { 341 /* Maximum number of memory pages available for use by the file 342 * system, set during mount time. This variable must never be 343 * used directly as it may be bigger than the current amount of 344 * free memory; in the extreme case, it will hold the SIZE_MAX 345 * value. Instead, use the TMPFS_PAGES_MAX macro. */ 346 vm_pindex_t tm_pages_max; 347 348 /* Number of pages in use by the file system. Cannot be bigger 349 * than the value returned by TMPFS_PAGES_MAX in any case. */ 350 vm_pindex_t tm_pages_used; 351 352 /* Pointer to the node representing the root directory of this 353 * file system. */ 354 struct tmpfs_node * tm_root; 355 356 /* Maximum number of possible nodes for this file system; set 357 * during mount time. We need a hard limit on the maximum number 358 * of nodes to avoid allocating too much of them; their objects 359 * cannot be released until the file system is unmounted. 360 * Otherwise, we could easily run out of memory by creating lots 361 * of empty files and then simply removing them. */ 362 ino_t tm_nodes_max; 363 364 /* Number of nodes currently that are in use. */ 365 ino_t tm_nodes_inuse; 366 367 /* maximum representable file size */ 368 u_int64_t tm_maxfilesize; 369 370 /* Nodes are organized in two different lists. The used list 371 * contains all nodes that are currently used by the file system; 372 * i.e., they refer to existing files. The available list contains 373 * all nodes that are currently available for use by new files. 374 * Nodes must be kept in this list (instead of deleting them) 375 * because we need to keep track of their generation number (tn_gen 376 * field). 377 * 378 * Note that nodes are lazily allocated: if the available list is 379 * empty and we have enough space to create more nodes, they will be 380 * created and inserted in the used list. Once these are released, 381 * they will go into the available list, remaining alive until the 382 * file system is unmounted. */ 383 struct tmpfs_node_list tm_nodes_used; 384 385 /* All node lock to protect the node list and tmp_pages_used */ 386 struct lock allnode_lock; 387 388 /* Pools used to store file system meta data. These are not shared 389 * across several instances of tmpfs for the reasons described in 390 * tmpfs_pool.c. */ 391 struct objcache *tm_dirent_pool; 392 struct objcache *tm_node_pool; 393 394 int tm_flags; 395 }; 396 397 #define TMPFS_LOCK(tm) lockmgr(&(tm)->allnode_lock, LK_EXCLUSIVE|LK_RETRY) 398 #define TMPFS_UNLOCK(tm) lockmgr(&(tm)->allnode_lock, LK_RELEASE) 399 400 /* --------------------------------------------------------------------- */ 401 402 /* 403 * This structure maps a file identifier to a tmpfs node. Used by the 404 * NFS code. 405 */ 406 struct tmpfs_fid { 407 uint16_t tf_len; 408 uint16_t tf_pad; 409 ino_t tf_id; 410 unsigned long tf_gen; 411 }; 412 413 /* --------------------------------------------------------------------- */ 414 415 #ifdef _KERNEL 416 /* 417 * Prototypes for tmpfs_subr.c. 418 */ 419 420 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype, 421 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *, 422 char *, int, int, struct tmpfs_node **); 423 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 424 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 425 const char *, uint16_t, struct tmpfs_dirent **); 426 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *); 427 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int, 428 struct vnode **); 429 void tmpfs_free_vp(struct vnode *); 430 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 431 struct namecache *, struct ucred *, char *); 432 void tmpfs_dir_attach(struct tmpfs_node *, struct tmpfs_dirent *); 433 void tmpfs_dir_detach(struct tmpfs_node *, struct tmpfs_dirent *); 434 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 435 struct tmpfs_node *f, 436 struct namecache *ncp); 437 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *); 438 int tmpfs_dir_getdotdotdent(struct tmpfs_mount *, 439 struct tmpfs_node *, struct uio *); 440 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t); 441 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *); 442 int tmpfs_reg_resize(struct vnode *, off_t, int); 443 int tmpfs_chflags(struct vnode *, int, struct ucred *); 444 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *); 445 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *); 446 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *); 447 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *, 448 int, struct ucred *); 449 void tmpfs_itimes(struct vnode *, const struct timespec *, 450 const struct timespec *); 451 452 void tmpfs_update(struct vnode *); 453 int tmpfs_truncate(struct vnode *, off_t); 454 int tmpfs_node_ctor(void *obj, void *privdata, int flags); 455 456 /* --------------------------------------------------------------------- */ 457 458 /* 459 * Convenience macros to simplify some logical expressions. 460 */ 461 #define IMPLIES(a, b) (!(a) || (b)) 462 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 463 464 /* --------------------------------------------------------------------- */ 465 466 /* 467 * Checks that the directory entry pointed by 'de' matches the name 'name' 468 * with a length of 'len'. 469 */ 470 #define TMPFS_DIRENT_MATCHES(de, name, len) \ 471 (de->td_namelen == (uint16_t)len && \ 472 bcmp((de)->td_name, (name), (de)->td_namelen) == 0) 473 474 /* --------------------------------------------------------------------- */ 475 476 /* 477 * Ensures that the node pointed by 'node' is a directory and that its 478 * contents are consistent with respect to directories. 479 */ 480 #define TMPFS_VALIDATE_DIR(node) \ 481 KKASSERT((node)->tn_type == VDIR); \ 482 KKASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 483 KKASSERT((node)->tn_dir.tn_readdir_lastp == NULL || \ 484 tmpfs_dircookie((node)->tn_dir.tn_readdir_lastp) == (node)->tn_dir.tn_readdir_lastn); 485 486 #endif 487 488 /* --------------------------------------------------------------------- */ 489 490 /* 491 * Macros/functions to convert from generic data structures to tmpfs 492 * specific ones. 493 */ 494 495 static inline 496 struct tmpfs_mount * 497 VFS_TO_TMPFS(struct mount *mp) 498 { 499 struct tmpfs_mount *tmp; 500 501 KKASSERT((mp) != NULL && (mp)->mnt_data != NULL); 502 tmp = (struct tmpfs_mount *)(mp)->mnt_data; 503 return tmp; 504 } 505 506 static inline 507 struct tmpfs_node * 508 VP_TO_TMPFS_NODE(struct vnode *vp) 509 { 510 struct tmpfs_node *node; 511 512 KKASSERT((vp) != NULL && (vp)->v_data != NULL); 513 node = (struct tmpfs_node *)vp->v_data; 514 return node; 515 } 516 517 static inline 518 struct tmpfs_node * 519 VP_TO_TMPFS_DIR(struct vnode *vp) 520 { 521 struct tmpfs_node *node; 522 523 node = VP_TO_TMPFS_NODE(vp); 524 TMPFS_VALIDATE_DIR(node); 525 return node; 526 } 527 528 /* --------------------------------------------------------------------- */ 529 /* 530 * buffer cache size 531 */ 532 #define BSIZE (off_t)16384 /* buffer cache size*/ 533 #define BMASK (off_t)(BSIZE - 1) 534 535 #endif /* _VFS_TMPFS_TMPFS_H_ */ 536