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. */ 76 struct tmpfs_node * td_node; 77 }; 78 79 /* A directory in tmpfs holds a sorted list of directory entries, which in 80 * turn point to other files (which can be directories themselves). 81 * 82 * In tmpfs, this list is managed by a tail queue, whose head is defined by 83 * the struct tmpfs_dir type. 84 * 85 * It is imporant to notice that directories do not have entries for . and 86 * .. as other file systems do. These can be generated when requested 87 * based on information available by other means, such as the pointer to 88 * the node itself in the former case or the pointer to the parent directory 89 * in the latter case. This is done to simplify tmpfs's code and, more 90 * importantly, to remove redundancy. */ 91 TAILQ_HEAD(tmpfs_dir, tmpfs_dirent); 92 93 /* Each entry in a directory has a cookie that identifies it. Cookies 94 * supersede offsets within directories because, given how tmpfs stores 95 * directories in memory, there is no such thing as an offset. (Emulating 96 * a real offset could be very difficult.) 97 * 98 * The '.', '..' and the end of directory markers have fixed cookies which 99 * cannot collide with the cookies generated by other entries. The cookies 100 * fot the other entries are generated based on the memory address on which 101 * stores their information is stored. 102 * 103 * Ideally, using the entry's memory pointer as the cookie would be enough 104 * to represent it and it wouldn't cause collisions in any system. 105 * Unfortunately, this results in "offsets" with very large values which 106 * later raise problems in the Linux compatibility layer (and maybe in other 107 * places) as described in PR kern/32034. Hence we need to workaround this 108 * with a rather ugly hack. 109 * 110 * Linux 32-bit binaries, unless built with _FILE_OFFSET_BITS=64, have off_t 111 * set to 'long', which is a 32-bit *signed* long integer. Regardless of 112 * the macro value, GLIBC (2.3 at least) always uses the getdents64 113 * system call (when calling readdir) which internally returns off64_t 114 * offsets. In order to make 32-bit binaries work, *GLIBC* converts the 115 * 64-bit values returned by the kernel to 32-bit ones and aborts with 116 * EOVERFLOW if the conversion results in values that won't fit in 32-bit 117 * integers (which it assumes is because the directory is extremely large). 118 * This wouldn't cause problems if we were dealing with unsigned integers, 119 * but as we have signed integers, this check fails due to sign expansion. 120 * 121 * For example, consider that the kernel returns the 0xc1234567 cookie to 122 * userspace in a off64_t integer. Later on, GLIBC casts this value to 123 * off_t (remember, signed) with code similar to: 124 * system call returns the offset in kernel_value; 125 * off_t casted_value = kernel_value; 126 * if (sizeof(off_t) != sizeof(off64_t) && 127 * kernel_value != casted_value) 128 * error! 129 * In this case, casted_value still has 0xc1234567, but when it is compared 130 * for equality against kernel_value, it is promoted to a 64-bit integer and 131 * becomes 0xffffffffc1234567, which is different than 0x00000000c1234567. 132 * Then, GLIBC assumes this is because the directory is very large. 133 * 134 * Given that all the above happens in user-space, we have no control over 135 * it; therefore we must workaround the issue here. We do this by 136 * truncating the pointer value to a 32-bit integer and hope that there 137 * won't be collisions. In fact, this will not cause any problems in 138 * 32-bit platforms but some might arise in 64-bit machines (I'm not sure 139 * if they can happen at all in practice). 140 * 141 * XXX A nicer solution shall be attempted. */ 142 #ifdef _KERNEL 143 #define TMPFS_DIRCOOKIE_DOT 0 144 #define TMPFS_DIRCOOKIE_DOTDOT 1 145 #define TMPFS_DIRCOOKIE_EOF 2 146 static __inline 147 off_t 148 tmpfs_dircookie(struct tmpfs_dirent *de) 149 { 150 off_t cookie; 151 152 cookie = ((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFF; 153 MPASS(cookie != TMPFS_DIRCOOKIE_DOT); 154 MPASS(cookie != TMPFS_DIRCOOKIE_DOTDOT); 155 MPASS(cookie != TMPFS_DIRCOOKIE_EOF); 156 157 return cookie; 158 } 159 #endif 160 161 /* --------------------------------------------------------------------- */ 162 163 /* 164 * Internal representation of a tmpfs file system node. 165 * 166 * This structure is splitted in two parts: one holds attributes common 167 * to all file types and the other holds data that is only applicable to 168 * a particular type. The code must be careful to only access those 169 * attributes that are actually allowed by the node's type. 170 * 171 * 172 * Below is the key of locks used to protected the fields in the following 173 * structures. 174 * 175 */ 176 struct tmpfs_node { 177 /* Doubly-linked list entry which links all existing nodes for a 178 * single file system. This is provided to ease the removal of 179 * all nodes during the unmount operation. */ 180 LIST_ENTRY(tmpfs_node) tn_entries; 181 182 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', 183 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode 184 * types instead of a custom enumeration is to make things simpler 185 * and faster, as we do not need to convert between two types. */ 186 enum vtype tn_type; 187 188 /* Node identifier. */ 189 ino_t tn_id; 190 191 /* Node's internal status. This is used by several file system 192 * operations to do modifications to the node in a delayed 193 * fashion. */ 194 int tn_status; 195 #define TMPFS_NODE_ACCESSED (1 << 1) 196 #define TMPFS_NODE_MODIFIED (1 << 2) 197 #define TMPFS_NODE_CHANGED (1 << 3) 198 199 /* The node size. It does not necessarily match the real amount 200 * of memory consumed by it. */ 201 off_t tn_size; 202 203 /* Generic node attributes. */ 204 uid_t tn_uid; 205 gid_t tn_gid; 206 mode_t tn_mode; 207 int tn_flags; 208 nlink_t tn_links; 209 struct timespec tn_atime; 210 struct timespec tn_mtime; 211 struct timespec tn_ctime; 212 struct timespec tn_birthtime; 213 unsigned long tn_gen; 214 215 /* As there is a single vnode for each active file within the 216 * system, care has to be taken to avoid allocating more than one 217 * vnode per file. In order to do this, a bidirectional association 218 * is kept between vnodes and nodes. 219 * 220 * Whenever a vnode is allocated, its v_data field is updated to 221 * point to the node it references. At the same time, the node's 222 * tn_vnode field is modified to point to the new vnode representing 223 * it. Further attempts to allocate a vnode for this same node will 224 * result in returning a new reference to the value stored in 225 * tn_vnode. 226 * 227 * May be NULL when the node is unused (that is, no vnode has been 228 * allocated for it or it has been reclaimed). */ 229 struct vnode * tn_vnode; 230 231 /* interlock to protect tn_vpstate */ 232 struct mtx tn_interlock; 233 234 /* Identify if current node has vnode assiocate with 235 * or allocating vnode. 236 */ 237 int tn_vpstate; 238 239 /* misc data field for different tn_type node */ 240 union { 241 /* Valid when tn_type == VBLK || tn_type == VCHR. */ 242 dev_t tn_rdev; 243 244 /* Valid when tn_type == VDIR. */ 245 struct tn_dir{ 246 /* Pointer to the parent directory. The root 247 * directory has a pointer to itself in this field; 248 * this property identifies the root node. */ 249 struct tmpfs_node * tn_parent; 250 251 /* Head of a tail-queue that links the contents of 252 * the directory together. See above for a 253 * description of its contents. */ 254 struct tmpfs_dir tn_dirhead; 255 256 /* Number and pointer of the first directory entry 257 * returned by the readdir operation if it were 258 * called again to continue reading data from the 259 * same directory as before. This is used to speed 260 * up reads of long directories, assuming that no 261 * more than one read is in progress at a given time. 262 * Otherwise, these values are discarded and a linear 263 * scan is performed from the beginning up to the 264 * point where readdir starts returning values. */ 265 off_t tn_readdir_lastn; 266 struct tmpfs_dirent * tn_readdir_lastp; 267 }tn_dir; 268 269 /* Valid when tn_type == VLNK. */ 270 /* The link's target, allocated from a string pool. */ 271 char * tn_link; 272 273 /* Valid when tn_type == VREG. */ 274 struct tn_reg { 275 /* The contents of regular files stored in a tmpfs 276 * file system are represented by a single anonymous 277 * memory object (aobj, for short). The aobj provides 278 * direct access to any position within the file, 279 * because its contents are always mapped in a 280 * contiguous region of virtual memory. It is a task 281 * of the memory management subsystem (see uvm(9)) to 282 * issue the required page ins or page outs whenever 283 * a position within the file is accessed. */ 284 vm_object_t tn_aobj; 285 size_t tn_aobj_pages; 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 #define TMPFS_VNODE_ALLOCATING 1 309 #define TMPFS_VNODE_WANT 2 310 /* --------------------------------------------------------------------- */ 311 312 /* 313 * Internal representation of a tmpfs mount point. 314 */ 315 struct tmpfs_mount { 316 /* Maximum number of memory pages available for use by the file 317 * system, set during mount time. This variable must never be 318 * used directly as it may be bigger than the current amount of 319 * free memory; in the extreme case, it will hold the SIZE_MAX 320 * value. Instead, use the TMPFS_PAGES_MAX macro. */ 321 size_t tm_pages_max; 322 323 /* Number of pages in use by the file system. Cannot be bigger 324 * than the value returned by TMPFS_PAGES_MAX in any case. */ 325 size_t tm_pages_used; 326 327 /* Pointer to the node representing the root directory of this 328 * file system. */ 329 struct tmpfs_node * tm_root; 330 331 /* Maximum number of possible nodes for this file system; set 332 * during mount time. We need a hard limit on the maximum number 333 * of nodes to avoid allocating too much of them; their objects 334 * cannot be released until the file system is unmounted. 335 * Otherwise, we could easily run out of memory by creating lots 336 * of empty files and then simply removing them. */ 337 ino_t tm_nodes_max; 338 339 /* unrhdr used to allocate inode numbers */ 340 struct unrhdr * tm_ino_unr; 341 342 /* Number of nodes currently that are in use. */ 343 ino_t tm_nodes_inuse; 344 345 /* maximum representable file size */ 346 u_int64_t tm_maxfilesize; 347 348 /* Nodes are organized in two different lists. The used list 349 * contains all nodes that are currently used by the file system; 350 * i.e., they refer to existing files. The available list contains 351 * all nodes that are currently available for use by new files. 352 * Nodes must be kept in this list (instead of deleting them) 353 * because we need to keep track of their generation number (tn_gen 354 * field). 355 * 356 * Note that nodes are lazily allocated: if the available list is 357 * empty and we have enough space to create more nodes, they will be 358 * created and inserted in the used list. Once these are released, 359 * they will go into the available list, remaining alive until the 360 * file system is unmounted. */ 361 struct tmpfs_node_list tm_nodes_used; 362 363 /* All node lock to protect the node list and tmp_pages_used */ 364 struct mtx allnode_lock; 365 366 /* Pools used to store file system meta data. These are not shared 367 * across several instances of tmpfs for the reasons described in 368 * tmpfs_pool.c. */ 369 uma_zone_t tm_dirent_pool; 370 uma_zone_t tm_node_pool; 371 }; 372 #define TMPFS_LOCK(tm) mtx_lock(&(tm)->allnode_lock) 373 #define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->allnode_lock) 374 375 /* --------------------------------------------------------------------- */ 376 377 /* 378 * This structure maps a file identifier to a tmpfs node. Used by the 379 * NFS code. 380 */ 381 struct tmpfs_fid { 382 uint16_t tf_len; 383 uint16_t tf_pad; 384 ino_t tf_id; 385 unsigned long tf_gen; 386 }; 387 388 /* --------------------------------------------------------------------- */ 389 390 #ifdef _KERNEL 391 /* 392 * Prototypes for tmpfs_subr.c. 393 */ 394 395 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype, 396 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *, 397 char *, dev_t, struct thread *, struct tmpfs_node **); 398 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); 399 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, 400 const char *, uint16_t, struct tmpfs_dirent **); 401 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *, 402 boolean_t); 403 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int, 404 struct vnode **, struct thread *); 405 void tmpfs_free_vp(struct vnode *); 406 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *, 407 struct componentname *, char *); 408 void tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *); 409 void tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *); 410 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node, 411 struct componentname *cnp); 412 struct tmpfs_dirent *tmpfs_dir_search(struct tmpfs_node *node, 413 struct tmpfs_node *f); 414 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *); 415 int tmpfs_dir_getdotdotdent(struct tmpfs_node *, struct uio *); 416 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t); 417 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *); 418 int tmpfs_reg_resize(struct vnode *, off_t); 419 int tmpfs_chflags(struct vnode *, int, struct ucred *, struct thread *); 420 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct thread *); 421 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *, 422 struct thread *); 423 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct thread *); 424 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *, 425 struct timespec *, int, struct ucred *, struct thread *); 426 void tmpfs_itimes(struct vnode *, const struct timespec *, 427 const struct timespec *); 428 429 void tmpfs_update(struct vnode *); 430 int tmpfs_truncate(struct vnode *, off_t); 431 432 /* --------------------------------------------------------------------- */ 433 434 /* 435 * Convenience macros to simplify some logical expressions. 436 */ 437 #define IMPLIES(a, b) (!(a) || (b)) 438 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a)) 439 440 /* --------------------------------------------------------------------- */ 441 442 /* 443 * Checks that the directory entry pointed by 'de' matches the name 'name' 444 * with a length of 'len'. 445 */ 446 #define TMPFS_DIRENT_MATCHES(de, name, len) \ 447 (de->td_namelen == (uint16_t)len && \ 448 bcmp((de)->td_name, (name), (de)->td_namelen) == 0) 449 450 /* --------------------------------------------------------------------- */ 451 452 /* 453 * Ensures that the node pointed by 'node' is a directory and that its 454 * contents are consistent with respect to directories. 455 */ 456 #define TMPFS_VALIDATE_DIR(node) \ 457 MPASS((node)->tn_type == VDIR); \ 458 MPASS((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \ 459 MPASS((node)->tn_dir.tn_readdir_lastp == NULL || \ 460 tmpfs_dircookie((node)->tn_dir.tn_readdir_lastp) == (node)->tn_dir.tn_readdir_lastn); 461 462 /* --------------------------------------------------------------------- */ 463 464 /* 465 * Memory management stuff. 466 */ 467 468 /* Amount of memory pages to reserve for the system (e.g., to not use by 469 * tmpfs). 470 * XXX: Should this be tunable through sysctl, for instance? */ 471 #define TMPFS_PAGES_RESERVED (4 * 1024 * 1024 / PAGE_SIZE) 472 473 /* 474 * Returns information about the number of available memory pages, 475 * including physical and virtual ones. 476 * 477 * If 'total' is TRUE, the value returned is the total amount of memory 478 * pages configured for the system (either in use or free). 479 * If it is FALSE, the value returned is the amount of free memory pages. 480 * 481 * Remember to remove TMPFS_PAGES_RESERVED from the returned value to avoid 482 * excessive memory usage. 483 * 484 */ 485 static __inline size_t 486 tmpfs_mem_info(void) 487 { 488 size_t size; 489 490 size = swap_pager_avail + cnt.v_free_count + cnt.v_inactive_count; 491 size -= size > cnt.v_wire_count ? cnt.v_wire_count : size; 492 return size; 493 } 494 495 /* Returns the maximum size allowed for a tmpfs file system. This macro 496 * must be used instead of directly retrieving the value from tm_pages_max. 497 * The reason is that the size of a tmpfs file system is dynamic: it lets 498 * the user store files as long as there is enough free memory (including 499 * physical memory and swap space). Therefore, the amount of memory to be 500 * used is either the limit imposed by the user during mount time or the 501 * amount of available memory, whichever is lower. To avoid consuming all 502 * the memory for a given mount point, the system will always reserve a 503 * minimum of TMPFS_PAGES_RESERVED pages, which is also taken into account 504 * by this macro (see above). */ 505 static __inline size_t 506 TMPFS_PAGES_MAX(struct tmpfs_mount *tmp) 507 { 508 size_t freepages; 509 510 freepages = tmpfs_mem_info(); 511 freepages -= freepages < TMPFS_PAGES_RESERVED ? 512 freepages : TMPFS_PAGES_RESERVED; 513 514 return MIN(tmp->tm_pages_max, freepages + tmp->tm_pages_used); 515 } 516 517 /* Returns the available space for the given file system. */ 518 #define TMPFS_META_PAGES(tmp) (howmany((tmp)->tm_nodes_inuse * (sizeof(struct tmpfs_node) \ 519 + sizeof(struct tmpfs_dirent)), PAGE_SIZE)) 520 #define TMPFS_FILE_PAGES(tmp) ((tmp)->tm_pages_used) 521 522 #define TMPFS_PAGES_AVAIL(tmp) (TMPFS_PAGES_MAX(tmp) > \ 523 TMPFS_META_PAGES(tmp)+TMPFS_FILE_PAGES(tmp)? \ 524 TMPFS_PAGES_MAX(tmp) - TMPFS_META_PAGES(tmp) \ 525 - TMPFS_FILE_PAGES(tmp):0) 526 527 #endif 528 529 /* --------------------------------------------------------------------- */ 530 531 /* 532 * Macros/functions to convert from generic data structures to tmpfs 533 * specific ones. 534 */ 535 536 static inline 537 struct tmpfs_mount * 538 VFS_TO_TMPFS(struct mount *mp) 539 { 540 struct tmpfs_mount *tmp; 541 542 MPASS((mp) != NULL && (mp)->mnt_data != NULL); 543 tmp = (struct tmpfs_mount *)(mp)->mnt_data; 544 return tmp; 545 } 546 547 static inline 548 struct tmpfs_node * 549 VP_TO_TMPFS_NODE(struct vnode *vp) 550 { 551 struct tmpfs_node *node; 552 553 MPASS((vp) != NULL && (vp)->v_data != NULL); 554 node = (struct tmpfs_node *)vp->v_data; 555 return node; 556 } 557 558 static inline 559 struct tmpfs_node * 560 VP_TO_TMPFS_DIR(struct vnode *vp) 561 { 562 struct tmpfs_node *node; 563 564 node = VP_TO_TMPFS_NODE(vp); 565 TMPFS_VALIDATE_DIR(node); 566 return node; 567 } 568 569 #endif /* _FS_TMPFS_TMPFS_H_ */ 570