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 * $FreeBSD$ 33 */ 34 35 #ifndef _FS_TMPFS_TMPFS_H_ 36 #define _FS_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/mutex.h> 48 49 /* --------------------------------------------------------------------- */ 50 #include <sys/malloc.h> 51 #include <sys/systm.h> 52 #include <sys/vmmeter.h> 53 #include <vm/swap_pager.h> 54 55 MALLOC_DECLARE(M_TMPFSMNT); 56 MALLOC_DECLARE(M_TMPFSNAME); 57 58 /* --------------------------------------------------------------------- */ 59 60 /* 61 * Internal representation of a tmpfs directory entry. 62 */ 63 struct tmpfs_dirent { 64 TAILQ_ENTRY(tmpfs_dirent) td_entries; 65 66 /* Length of the name stored in this directory entry. This avoids 67 * the need to recalculate it every time the name is used. */ 68 uint16_t td_namelen; 69 70 /* The name of the entry, allocated from a string pool. This 71 * string is not required to be zero-terminated; therefore, the 72 * td_namelen field must always be used when accessing its value. */ 73 char * td_name; 74 75 /* Pointer to the node this entry refers to. In case this field 76 * is NULL, the node is a whiteout. */ 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 * fot 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 MPASS(cookie != TMPFS_DIRCOOKIE_DOT); 155 MPASS(cookie != TMPFS_DIRCOOKIE_DOTDOT); 156 MPASS(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 struct timespec tn_atime; 211 struct timespec tn_mtime; 212 struct timespec tn_ctime; 213 struct timespec tn_birthtime; 214 unsigned long tn_gen; 215 216 /* As there is a single vnode for each active file within the 217 * system, care has to be taken to avoid allocating more than one 218 * vnode per file. In order to do this, a bidirectional association 219 * is kept between vnodes and nodes. 220 * 221 * Whenever a vnode is allocated, its v_data field is updated to 222 * point to the node it references. At the same time, the node's 223 * tn_vnode field is modified to point to the new vnode representing 224 * it. Further attempts to allocate a vnode for this same node will 225 * result in returning a new reference to the value stored in 226 * tn_vnode. 227 * 228 * May be NULL when the node is unused (that is, no vnode has been 229 * allocated for it or it has been reclaimed). */ 230 struct vnode * tn_vnode; 231 232 /* interlock to protect tn_vpstate */ 233 struct mtx tn_interlock; 234 235 /* Identify if current node has vnode assiocate with 236 * or allocating vnode. 237 */ 238 int tn_vpstate; 239 240 /* misc data field for different tn_type node */ 241 union { 242 /* Valid when tn_type == VBLK || tn_type == VCHR. */ 243 dev_t tn_rdev; 244 245 /* Valid when tn_type == VDIR. */ 246 struct tn_dir{ 247 /* Pointer to the parent directory. The root 248 * directory has a pointer to itself in this field; 249 * this property identifies the root node. */ 250 struct tmpfs_node * tn_parent; 251 252 /* Head of a tail-queue that links the contents of 253 * the directory together. See above for a 254 * description of its contents. */ 255 struct tmpfs_dir tn_dirhead; 256 257 /* Number and pointer of the first directory entry 258 * returned by the readdir operation if it were 259 * called again to continue reading data from the 260 * same directory as before. This is used to speed 261 * up reads of long directories, assuming that no 262 * more than one read is in progress at a given time. 263 * Otherwise, these values are discarded and a linear 264 * scan is performed from the beginning up to the 265 * point where readdir starts returning values. */ 266 off_t tn_readdir_lastn; 267 struct tmpfs_dirent * tn_readdir_lastp; 268 }tn_dir; 269 270 /* Valid when tn_type == VLNK. */ 271 /* The link's target, allocated from a string pool. */ 272 char * tn_link; 273 274 /* Valid when tn_type == VREG. */ 275 struct tn_reg { 276 /* The contents of regular files stored in a tmpfs 277 * file system are represented by a single anonymous 278 * memory object (aobj, for short). The aobj provides 279 * direct access to any position within the file, 280 * because its contents are always mapped in a 281 * contiguous region of virtual memory. It is a task 282 * of the memory management subsystem (see uvm(9)) to 283 * issue the required page ins or page outs whenever 284 * a position within the file is accessed. */ 285 vm_object_t tn_aobj; 286 287 }tn_reg; 288 289 /* Valid when tn_type = VFIFO */ 290 struct tn_fifo { 291 fo_rdwr_t *tn_fo_read; 292 fo_rdwr_t *tn_fo_write; 293 }tn_fifo; 294 }tn_spec; 295 }; 296 LIST_HEAD(tmpfs_node_list, tmpfs_node); 297 298 #define tn_rdev tn_spec.tn_rdev 299 #define tn_dir tn_spec.tn_dir 300 #define tn_link tn_spec.tn_link 301 #define tn_reg tn_spec.tn_reg 302 #define tn_fifo tn_spec.tn_fifo 303 304 #define TMPFS_NODE_LOCK(node) mtx_lock(&(node)->tn_interlock) 305 #define TMPFS_NODE_UNLOCK(node) mtx_unlock(&(node)->tn_interlock) 306 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock) 307 308 #ifdef INVARIANTS 309 #define TMPFS_ASSERT_LOCKED(node) do { \ 310 MPASS(node != NULL); \ 311 MPASS(node->tn_vnode != NULL); \ 312 if (!VOP_ISLOCKED(node->tn_vnode) && \ 313 !mtx_owned(TMPFS_NODE_MTX(node))) \ 314 panic("tmpfs: node is not locked: %p", node); \ 315 } while (0) 316 #define TMPFS_ASSERT_ELOCKED(node) do { \ 317 MPASS((node) != NULL); \ 318 MPASS((node)->tn_vnode != NULL); \ 319 mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED); \ 320 ASSERT_VOP_LOCKED((node)->tn_vnode, "tmpfs"); \ 321 } while (0) 322 #else 323 #define TMPFS_ASSERT_LOCKED(node) (void)0 324 #define TMPFS_ASSERT_ELOCKED(node) (void)0 325 #endif 326 327 #define TMPFS_VNODE_ALLOCATING 1 328 #define TMPFS_VNODE_WANT 2 329 #define TMPFS_VNODE_DOOMED 4 330 /* --------------------------------------------------------------------- */ 331 332 /* 333 * Internal representation of a tmpfs mount point. 334 */ 335 struct tmpfs_mount { 336 /* Maximum number of memory pages available for use by the file 337 * system, set during mount time. This variable must never be 338 * used directly as it may be bigger than the current amount of 339 * free memory; in the extreme case, it will hold the SIZE_MAX 340 * value. */ 341 size_t tm_pages_max; 342 343 /* Number of pages in use by the file system. */ 344 size_t tm_pages_used; 345 346 /* Pointer to the node representing the root directory of this 347 * file system. */ 348 struct tmpfs_node * tm_root; 349 350 /* Maximum number of possible nodes for this file system; set 351 * during mount time. We need a hard limit on the maximum number 352 * of nodes to avoid allocating too much of them; their objects 353 * cannot be released until the file system is unmounted. 354 * Otherwise, we could easily run out of memory by creating lots 355 * of empty files and then simply removing them. */ 356 ino_t tm_nodes_max; 357 358 /* unrhdr used to allocate inode numbers */ 359 struct unrhdr * tm_ino_unr; 360 361 /* Number of nodes currently that are in use. */ 362 ino_t tm_nodes_inuse; 363 364 /* maximum representable file size */ 365 u_int64_t tm_maxfilesize; 366 367 /* Nodes are organized in two different lists. The used list 368 * contains all nodes that are currently used by the file system; 369 * i.e., they refer to existing files. The available list contains 370 * all nodes that are currently available for use by new files. 371 * Nodes must be kept in this list (instead of deleting them) 372 * because we need to keep track of their generation number (tn_gen 373 * field). 374 * 375 * Note that nodes are lazily allocated: if the available list is 376 * empty and we have enough space to create more nodes, they will be 377 * created and inserted in the used list. Once these are released, 378 * they will go into the available list, remaining alive until the 379 * file system is unmounted. */ 380 struct tmpfs_node_list tm_nodes_used; 381 382 /* All node lock to protect the node list and tmp_pages_used */ 383 struct mtx allnode_lock; 384 385 /* Pools used to store file system meta data. These are not shared 386 * across several instances of tmpfs for the reasons described in 387 * tmpfs_pool.c. */ 388 uma_zone_t tm_dirent_pool; 389 uma_zone_t tm_node_pool; 390 391 /* Read-only status. */ 392 int tm_ronly; 393 }; 394 #define TMPFS_LOCK(tm) mtx_lock(&(tm)->allnode_lock) 395 #define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->allnode_lock) 396 397 /* --------------------------------------------------------------------- */ 398 399 /* 400 * This structure maps a file identifier to a tmpfs node. Used by the 401 * NFS code. 402 */ 403 struct tmpfs_fid { 404 uint16_t tf_len; 405 uint16_t tf_pad; 406 ino_t tf_id; 407 unsigned long tf_gen; 408 }; 409 410 /* --------------------------------------------------------------------- */ 411 412 #ifdef _KERNEL 413 /* 414 * Prototypes for tmpfs_subr.c. 415 */ 416 417 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype, 418 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *, 419 char *, dev_t, struct tmpfs_node **); 420 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 421 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 422 const char *, uint16_t, struct tmpfs_dirent **); 423 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *, 424 boolean_t); 425 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int, 426 struct vnode **); 427 void tmpfs_free_vp(struct vnode *); 428 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 429 struct componentname *, char *); 430 void tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *); 431 void tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *); 432 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 433 struct tmpfs_node *f, 434 struct componentname *cnp); 435 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *); 436 int tmpfs_dir_getdotdotdent(struct tmpfs_node *, struct uio *); 437 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t); 438 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *); 439 int tmpfs_dir_whiteout_add(struct vnode *, struct componentname *); 440 void tmpfs_dir_whiteout_remove(struct vnode *, struct componentname *); 441 int tmpfs_reg_resize(struct vnode *, off_t, boolean_t); 442 int tmpfs_chflags(struct vnode *, int, struct ucred *, struct thread *); 443 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct thread *); 444 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *, 445 struct thread *); 446 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct thread *); 447 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *, 448 struct timespec *, int, struct ucred *, struct thread *); 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); 454 455 /* --------------------------------------------------------------------- */ 456 457 /* 458 * Convenience macros to simplify some logical expressions. 459 */ 460 #define IMPLIES(a, b) (!(a) || (b)) 461 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 462 463 /* --------------------------------------------------------------------- */ 464 465 /* 466 * Checks that the directory entry pointed by 'de' matches the name 'name' 467 * with a length of 'len'. 468 */ 469 #define TMPFS_DIRENT_MATCHES(de, name, len) \ 470 (de->td_namelen == (uint16_t)len && \ 471 bcmp((de)->td_name, (name), (de)->td_namelen) == 0) 472 473 /* --------------------------------------------------------------------- */ 474 475 /* 476 * Ensures that the node pointed by 'node' is a directory and that its 477 * contents are consistent with respect to directories. 478 */ 479 #define TMPFS_VALIDATE_DIR(node) \ 480 MPASS((node)->tn_type == VDIR); \ 481 MPASS((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 482 MPASS((node)->tn_dir.tn_readdir_lastp == NULL || \ 483 tmpfs_dircookie((node)->tn_dir.tn_readdir_lastp) == (node)->tn_dir.tn_readdir_lastn); 484 485 /* --------------------------------------------------------------------- */ 486 487 /* 488 * Memory management stuff. 489 */ 490 491 /* 492 * Amount of memory pages to reserve for the system (e.g., to not use by 493 * tmpfs). 494 */ 495 #define TMPFS_PAGES_MINRESERVED (4 * 1024 * 1024 / PAGE_SIZE) 496 497 size_t tmpfs_mem_avail(void); 498 499 size_t tmpfs_pages_used(struct tmpfs_mount *tmp); 500 501 #endif 502 503 /* --------------------------------------------------------------------- */ 504 505 /* 506 * Macros/functions to convert from generic data structures to tmpfs 507 * specific ones. 508 */ 509 510 static inline 511 struct tmpfs_mount * 512 VFS_TO_TMPFS(struct mount *mp) 513 { 514 struct tmpfs_mount *tmp; 515 516 MPASS((mp) != NULL && (mp)->mnt_data != NULL); 517 tmp = (struct tmpfs_mount *)(mp)->mnt_data; 518 return tmp; 519 } 520 521 static inline 522 struct tmpfs_node * 523 VP_TO_TMPFS_NODE(struct vnode *vp) 524 { 525 struct tmpfs_node *node; 526 527 MPASS((vp) != NULL && (vp)->v_data != NULL); 528 node = (struct tmpfs_node *)vp->v_data; 529 return node; 530 } 531 532 static inline 533 struct tmpfs_node * 534 VP_TO_TMPFS_DIR(struct vnode *vp) 535 { 536 struct tmpfs_node *node; 537 538 node = VP_TO_TMPFS_NODE(vp); 539 TMPFS_VALIDATE_DIR(node); 540 return node; 541 } 542 543 #endif /* _FS_TMPFS_TMPFS_H_ */ 544