1 /* $NetBSD: tmpfs.h,v 1.18 2006/03/31 20:27:49 riz Exp $ */
4 * Copyright (c) 2005 The NetBSD Foundation, Inc.
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
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
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 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the NetBSD
22 * Foundation, Inc. and its contributors.
23 * 4. Neither the name of The NetBSD Foundation nor the names of its
24 * contributors may be used to endorse or promote products derived
25 * from this software without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 * POSSIBILITY OF SUCH DAMAGE.
42 #ifndef _FS_TMPFS_TMPFS_H_
43 #define _FS_TMPFS_TMPFS_H_
45 /* ---------------------------------------------------------------------
46 * KERNEL-SPECIFIC DEFINITIONS
47 * --------------------------------------------------------------------- */
48 #include <sys/dirent.h>
49 #include <sys/mount.h>
50 #include <sys/queue.h>
51 #include <sys/vnode.h>
54 #include <sys/mutex.h>
56 /* --------------------------------------------------------------------- */
57 #include <sys/malloc.h>
58 #include <sys/systm.h>
59 #include <sys/vmmeter.h>
60 #include <vm/swap_pager.h>
62 MALLOC_DECLARE(M_TMPFSMNT);
63 MALLOC_DECLARE(M_TMPFSNAME);
65 /* --------------------------------------------------------------------- */
68 * Internal representation of a tmpfs directory entry.
71 TAILQ_ENTRY(tmpfs_dirent) td_entries;
73 /* Length of the name stored in this directory entry. This avoids
74 * the need to recalculate it every time the name is used. */
77 /* The name of the entry, allocated from a string pool. This
78 * string is not required to be zero-terminated; therefore, the
79 * td_namelen field must always be used when accessing its value. */
82 /* Pointer to the node this entry refers to. */
83 struct tmpfs_node * td_node;
86 /* A directory in tmpfs holds a sorted list of directory entries, which in
87 * turn point to other files (which can be directories themselves).
89 * In tmpfs, this list is managed by a tail queue, whose head is defined by
90 * the struct tmpfs_dir type.
92 * It is imporant to notice that directories do not have entries for . and
93 * .. as other file systems do. These can be generated when requested
94 * based on information available by other means, such as the pointer to
95 * the node itself in the former case or the pointer to the parent directory
96 * in the latter case. This is done to simplify tmpfs's code and, more
97 * importantly, to remove redundancy. */
98 TAILQ_HEAD(tmpfs_dir, tmpfs_dirent);
100 #define TMPFS_DIRCOOKIE(dirent) ((off_t)(uintptr_t)(dirent))
101 #define TMPFS_DIRCOOKIE_DOT 0
102 #define TMPFS_DIRCOOKIE_DOTDOT 1
103 #define TMPFS_DIRCOOKIE_EOF 2
105 /* --------------------------------------------------------------------- */
108 * Internal representation of a tmpfs file system node.
110 * This structure is splitted in two parts: one holds attributes common
111 * to all file types and the other holds data that is only applicable to
112 * a particular type. The code must be careful to only access those
113 * attributes that are actually allowed by the node's type.
116 * Below is the key of locks used to protected the fields in the following
121 /* Doubly-linked list entry which links all existing nodes for a
122 * single file system. This is provided to ease the removal of
123 * all nodes during the unmount operation. */
124 LIST_ENTRY(tmpfs_node) tn_entries;
126 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
127 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode
128 * types instead of a custom enumeration is to make things simpler
129 * and faster, as we do not need to convert between two types. */
132 /* Node identifier. */
135 /* Node's internal status. This is used by several file system
136 * operations to do modifications to the node in a delayed
139 #define TMPFS_NODE_ACCESSED (1 << 1)
140 #define TMPFS_NODE_MODIFIED (1 << 2)
141 #define TMPFS_NODE_CHANGED (1 << 3)
143 /* The node size. It does not necessarily match the real amount
144 * of memory consumed by it. */
147 /* Generic node attributes. */
153 struct timespec tn_atime;
154 struct timespec tn_mtime;
155 struct timespec tn_ctime;
156 struct timespec tn_birthtime;
157 unsigned long tn_gen;
159 /* Head of byte-level lock list (used by tmpfs_advlock). */
160 struct lockf * tn_lockf;
162 /* As there is a single vnode for each active file within the
163 * system, care has to be taken to avoid allocating more than one
164 * vnode per file. In order to do this, a bidirectional association
165 * is kept between vnodes and nodes.
167 * Whenever a vnode is allocated, its v_data field is updated to
168 * point to the node it references. At the same time, the node's
169 * tn_vnode field is modified to point to the new vnode representing
170 * it. Further attempts to allocate a vnode for this same node will
171 * result in returning a new reference to the value stored in
174 * May be NULL when the node is unused (that is, no vnode has been
175 * allocated for it or it has been reclaimed). */
176 struct vnode * tn_vnode;
178 /* Pointer to the node returned by tmpfs_lookup() after doing a
179 * delete or a rename lookup; its value is only valid in these two
180 * situations. In case we were looking up . or .., it holds a null
182 struct tmpfs_dirent * tn_lookup_dirent;
184 /* interlock to protect tn_vpstate */
185 struct mtx tn_interlock;
187 /* Identify if current node has vnode assiocate with
188 * or allocating vnode.
192 /* misc data field for different tn_type node */
194 /* Valid when tn_type == VBLK || tn_type == VCHR. */
197 /* Valid when tn_type == VDIR. */
199 /* Pointer to the parent directory. The root
200 * directory has a pointer to itself in this field;
201 * this property identifies the root node. */
202 struct tmpfs_node * tn_parent;
204 /* Head of a tail-queue that links the contents of
205 * the directory together. See above for a
206 * description of its contents. */
207 struct tmpfs_dir tn_dirhead;
209 /* Number and pointer of the first directory entry
210 * returned by the readdir operation if it were
211 * called again to continue reading data from the
212 * same directory as before. This is used to speed
213 * up reads of long directories, assuming that no
214 * more than one read is in progress at a given time.
215 * Otherwise, these values are discarded and a linear
216 * scan is performed from the beginning up to the
217 * point where readdir starts returning values. */
218 off_t tn_readdir_lastn;
219 struct tmpfs_dirent * tn_readdir_lastp;
222 /* Valid when tn_type == VLNK. */
223 /* The link's target, allocated from a string pool. */
226 /* Valid when tn_type == VREG. */
228 /* The contents of regular files stored in a tmpfs
229 * file system are represented by a single anonymous
230 * memory object (aobj, for short). The aobj provides
231 * direct access to any position within the file,
232 * because its contents are always mapped in a
233 * contiguous region of virtual memory. It is a task
234 * of the memory management subsystem (see uvm(9)) to
235 * issue the required page ins or page outs whenever
236 * a position within the file is accessed. */
238 size_t tn_aobj_pages;
242 /* Valid when tn_type = VFIFO */
244 fo_rdwr_t *tn_fo_read;
245 fo_rdwr_t *tn_fo_write;
249 LIST_HEAD(tmpfs_node_list, tmpfs_node);
251 #define tn_rdev tn_spec.tn_rdev
252 #define tn_dir tn_spec.tn_dir
253 #define tn_link tn_spec.tn_link
254 #define tn_reg tn_spec.tn_reg
255 #define tn_fifo tn_spec.tn_fifo
257 #define TMPFS_NODE_LOCK(node) mtx_lock(&(node)->tn_interlock)
258 #define TMPFS_NODE_UNLOCK(node) mtx_unlock(&(node)->tn_interlock)
260 #define TMPFS_VNODE_ALLOCATING 1
261 #define TMPFS_VNODE_WANT 2
262 /* --------------------------------------------------------------------- */
265 * Internal representation of a tmpfs mount point.
268 /* Maximum number of memory pages available for use by the file
269 * system, set during mount time. This variable must never be
270 * used directly as it may be bigger than the current amount of
271 * free memory; in the extreme case, it will hold the SIZE_MAX
272 * value. Instead, use the TMPFS_PAGES_MAX macro. */
275 /* Number of pages in use by the file system. Cannot be bigger
276 * than the value returned by TMPFS_PAGES_MAX in any case. */
277 size_t tm_pages_used;
279 /* Pointer to the node representing the root directory of this
281 struct tmpfs_node * tm_root;
283 /* Maximum number of possible nodes for this file system; set
284 * during mount time. We need a hard limit on the maximum number
285 * of nodes to avoid allocating too much of them; their objects
286 * cannot be released until the file system is unmounted.
287 * Otherwise, we could easily run out of memory by creating lots
288 * of empty files and then simply removing them. */
291 /* unrhdr used to allocate inode numbers */
292 struct unrhdr * tm_ino_unr;
294 /* Number of nodes currently that are in use. */
295 ino_t tm_nodes_inuse;
297 /* maximum representable file size */
298 u_int64_t tm_maxfilesize;
300 /* Nodes are organized in two different lists. The used list
301 * contains all nodes that are currently used by the file system;
302 * i.e., they refer to existing files. The available list contains
303 * all nodes that are currently available for use by new files.
304 * Nodes must be kept in this list (instead of deleting them)
305 * because we need to keep track of their generation number (tn_gen
308 * Note that nodes are lazily allocated: if the available list is
309 * empty and we have enough space to create more nodes, they will be
310 * created and inserted in the used list. Once these are released,
311 * they will go into the available list, remaining alive until the
312 * file system is unmounted. */
313 struct tmpfs_node_list tm_nodes_used;
315 /* All node lock to protect the node list and tmp_pages_used */
316 struct mtx allnode_lock;
318 /* Pools used to store file system meta data. These are not shared
319 * across several instances of tmpfs for the reasons described in
321 uma_zone_t tm_dirent_pool;
322 uma_zone_t tm_node_pool;
324 #define TMPFS_LOCK(tm) mtx_lock(&(tm)->allnode_lock)
325 #define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->allnode_lock)
327 /* --------------------------------------------------------------------- */
330 * This structure maps a file identifier to a tmpfs node. Used by the
337 unsigned long tf_gen;
340 /* --------------------------------------------------------------------- */
344 * Prototypes for tmpfs_subr.c.
347 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype,
348 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *,
349 char *, dev_t, struct thread *, struct tmpfs_node **);
350 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
351 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
352 const char *, uint16_t, struct tmpfs_dirent **);
353 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *,
355 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, struct vnode **,
357 void tmpfs_free_vp(struct vnode *);
358 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
359 struct componentname *, char *);
360 void tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *);
361 void tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *);
362 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node,
363 struct componentname *cnp);
364 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *);
365 int tmpfs_dir_getdotdotdent(struct tmpfs_node *, struct uio *);
366 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t);
367 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *);
368 int tmpfs_reg_resize(struct vnode *, off_t);
369 int tmpfs_chflags(struct vnode *, int, struct ucred *, struct thread *);
370 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct thread *);
371 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *,
373 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct thread *);
374 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *,
375 struct timespec *, int, struct ucred *, struct thread *);
376 void tmpfs_itimes(struct vnode *, const struct timespec *,
377 const struct timespec *);
379 void tmpfs_update(struct vnode *);
380 int tmpfs_truncate(struct vnode *, off_t);
382 /* --------------------------------------------------------------------- */
385 * Convenience macros to simplify some logical expressions.
387 #define IMPLIES(a, b) (!(a) || (b))
388 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a))
390 /* --------------------------------------------------------------------- */
393 * Checks that the directory entry pointed by 'de' matches the name 'name'
394 * with a length of 'len'.
396 #define TMPFS_DIRENT_MATCHES(de, name, len) \
397 (de->td_namelen == (uint16_t)len && \
398 memcmp((de)->td_name, (name), (de)->td_namelen) == 0)
400 /* --------------------------------------------------------------------- */
403 * Ensures that the node pointed by 'node' is a directory and that its
404 * contents are consistent with respect to directories.
406 #define TMPFS_VALIDATE_DIR(node) \
407 MPASS((node)->tn_type == VDIR); \
408 MPASS((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
409 MPASS((node)->tn_dir.tn_readdir_lastp == NULL || \
410 TMPFS_DIRCOOKIE((node)->tn_dir.tn_readdir_lastp) == (node)->tn_dir.tn_readdir_lastn);
412 /* --------------------------------------------------------------------- */
415 * Memory management stuff.
418 /* Amount of memory pages to reserve for the system (e.g., to not use by
420 * XXX: Should this be tunable through sysctl, for instance? */
421 #define TMPFS_PAGES_RESERVED (4 * 1024 * 1024 / PAGE_SIZE)
424 * Returns information about the number of available memory pages,
425 * including physical and virtual ones.
427 * If 'total' is TRUE, the value returned is the total amount of memory
428 * pages configured for the system (either in use or free).
429 * If it is FALSE, the value returned is the amount of free memory pages.
431 * Remember to remove TMPFS_PAGES_RESERVED from the returned value to avoid
432 * excessive memory usage.
435 static __inline size_t
440 size = swap_pager_avail + cnt.v_free_count + cnt.v_inactive_count;
441 size -= size > cnt.v_wire_count ? cnt.v_wire_count : size;
445 /* Returns the maximum size allowed for a tmpfs file system. This macro
446 * must be used instead of directly retrieving the value from tm_pages_max.
447 * The reason is that the size of a tmpfs file system is dynamic: it lets
448 * the user store files as long as there is enough free memory (including
449 * physical memory and swap space). Therefore, the amount of memory to be
450 * used is either the limit imposed by the user during mount time or the
451 * amount of available memory, whichever is lower. To avoid consuming all
452 * the memory for a given mount point, the system will always reserve a
453 * minimum of TMPFS_PAGES_RESERVED pages, which is also taken into account
454 * by this macro (see above). */
455 static __inline size_t
456 TMPFS_PAGES_MAX(struct tmpfs_mount *tmp)
460 freepages = tmpfs_mem_info();
461 freepages -= freepages < TMPFS_PAGES_RESERVED ?
462 freepages : TMPFS_PAGES_RESERVED;
464 return MIN(tmp->tm_pages_max, freepages + tmp->tm_pages_used);
467 /* Returns the available space for the given file system. */
468 #define TMPFS_META_PAGES(tmp) (howmany((tmp)->tm_nodes_inuse * (sizeof(struct tmpfs_node) \
469 + sizeof(struct tmpfs_dirent)), PAGE_SIZE))
470 #define TMPFS_FILE_PAGES(tmp) ((tmp)->tm_pages_used)
472 #define TMPFS_PAGES_AVAIL(tmp) (TMPFS_PAGES_MAX(tmp) > \
473 TMPFS_META_PAGES(tmp)+TMPFS_FILE_PAGES(tmp)? \
474 TMPFS_PAGES_MAX(tmp) - TMPFS_META_PAGES(tmp) \
475 - TMPFS_FILE_PAGES(tmp):0)
479 /* --------------------------------------------------------------------- */
482 * Macros/functions to convert from generic data structures to tmpfs
488 VFS_TO_TMPFS(struct mount *mp)
490 struct tmpfs_mount *tmp;
492 MPASS((mp) != NULL && (mp)->mnt_data != NULL);
493 tmp = (struct tmpfs_mount *)(mp)->mnt_data;
499 VP_TO_TMPFS_NODE(struct vnode *vp)
501 struct tmpfs_node *node;
503 MPASS((vp) != NULL && (vp)->v_data != NULL);
504 node = (struct tmpfs_node *)vp->v_data;
510 VP_TO_TMPFS_DIR(struct vnode *vp)
512 struct tmpfs_node *node;
514 node = VP_TO_TMPFS_NODE(vp);
515 TMPFS_VALIDATE_DIR(node);
519 #endif /* _FS_TMPFS_TMPFS_H_ */