1 /* $NetBSD: tmpfs.h,v 1.14 2006/02/10 16:00:02 christos 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);
64 #include <fs/tmpfs/tmpfs_uma.h>
66 /* --------------------------------------------------------------------- */
69 * Internal representation of a tmpfs directory entry.
72 TAILQ_ENTRY(tmpfs_dirent) td_entries;
74 /* Length of the name stored in this directory entry. This avoids
75 * the need to recalculate it every time the name is used. */
78 /* The name of the entry, allocated from a string pool. This
79 * string is not required to be zero-terminated; therefore, the
80 * td_namelen field must always be used when accessing its value. */
83 /* Pointer to the node this entry refers to. */
84 struct tmpfs_node * td_node;
87 /* A directory in tmpfs holds a sorted list of directory entries, which in
88 * turn point to other files (which can be directories themselves).
90 * In tmpfs, this list is managed by a tail queue, whose head is defined by
91 * the struct tmpfs_dir type.
93 * It is imporant to notice that directories do not have entries for . and
94 * .. as other file systems do. These can be generated when requested
95 * based on information available by other means, such as the pointer to
96 * the node itself in the former case or the pointer to the parent directory
97 * in the latter case. This is done to simplify tmpfs's code and, more
98 * importantly, to remove redundancy. */
99 TAILQ_HEAD(tmpfs_dir, tmpfs_dirent);
101 #define TMPFS_DIRCOOKIE(dirent) ((off_t)(uintptr_t)(dirent))
102 #define TMPFS_DIRCOOKIE_DOT 0
103 #define TMPFS_DIRCOOKIE_DOTDOT 1
104 #define TMPFS_DIRCOOKIE_EOF 2
106 /* --------------------------------------------------------------------- */
109 * Internal representation of a tmpfs file system node.
111 * This structure is splitted in two parts: one holds attributes common
112 * to all file types and the other holds data that is only applicable to
113 * a particular type. The code must be careful to only access those
114 * attributes that are actually allowed by the node's type.
117 * Below is the key of locks used to protected the fields in the following
122 /* Doubly-linked list entry which links all existing nodes for a
123 * single file system. This is provided to ease the removal of
124 * all nodes during the unmount operation. */
125 LIST_ENTRY(tmpfs_node) tn_entries;
127 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
128 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode
129 * types instead of a custom enumeration is to make things simpler
130 * and faster, as we do not need to convert between two types. */
133 /* Node identifier. */
136 /* Node's internal status. This is used by several file system
137 * operations to do modifications to the node in a delayed
140 #define TMPFS_NODE_ACCESSED (1 << 1)
141 #define TMPFS_NODE_MODIFIED (1 << 2)
142 #define TMPFS_NODE_CHANGED (1 << 3)
144 /* The node size. It does not necessarily match the real amount
145 * of memory consumed by it. */
148 /* Generic node attributes. */
154 struct timespec tn_atime;
155 struct timespec tn_mtime;
156 struct timespec tn_ctime;
157 struct timespec tn_birthtime;
158 unsigned long tn_gen;
160 /* Head of byte-level lock list (used by tmpfs_advlock). */
161 struct lockf * tn_lockf;
163 /* As there is a single vnode for each active file within the
164 * system, care has to be taken to avoid allocating more than one
165 * vnode per file. In order to do this, a bidirectional association
166 * is kept between vnodes and nodes.
168 * Whenever a vnode is allocated, its v_data field is updated to
169 * point to the node it references. At the same time, the node's
170 * tn_vnode field is modified to point to the new vnode representing
171 * it. Further attempts to allocate a vnode for this same node will
172 * result in returning a new reference to the value stored in
175 * May be NULL when the node is unused (that is, no vnode has been
176 * allocated for it or it has been reclaimed). */
177 struct vnode * tn_vnode;
179 /* Pointer to the node returned by tmpfs_lookup() after doing a
180 * delete or a rename lookup; its value is only valid in these two
181 * situations. In case we were looking up . or .., it holds a null
183 struct tmpfs_dirent * tn_lookup_dirent;
185 /* interlock to protect tn_vpstate */
186 struct mtx tn_interlock;
188 /* Identify if current node has vnode assiocate with
189 * or allocating vnode.
193 /* misc data field for different tn_type node */
195 /* Valid when tn_type == VBLK || tn_type == VCHR. */
198 /* Valid when tn_type == VDIR. */
200 /* Pointer to the parent directory. The root
201 * directory has a pointer to itself in this field;
202 * this property identifies the root node. */
203 struct tmpfs_node * tn_parent;
205 /* Head of a tail-queue that links the contents of
206 * the directory together. See above for a
207 * description of its contents. */
208 struct tmpfs_dir tn_dirhead;
210 /* Number and pointer of the first directory entry
211 * returned by the readdir operation if it were
212 * called again to continue reading data from the
213 * same directory as before. This is used to speed
214 * up reads of long directories, assuming that no
215 * more than one read is in progress at a given time.
216 * Otherwise, these values are discarded and a linear
217 * scan is performed from the beginning up to the
218 * point where readdir starts returning values. */
219 off_t tn_readdir_lastn;
220 struct tmpfs_dirent * tn_readdir_lastp;
223 /* Valid when tn_type == VLNK. */
224 /* The link's target, allocated from a string pool. */
227 /* Valid when tn_type == VREG. */
229 /* The contents of regular files stored in a tmpfs
230 * file system are represented by a single anonymous
231 * memory object (aobj, for short). The aobj provides
232 * direct access to any position within the file,
233 * because its contents are always mapped in a
234 * contiguous region of virtual memory. It is a task
235 * of the memory management subsystem (see uvm(9)) to
236 * issue the required page ins or page outs whenever
237 * a position within the file is accessed. */
239 size_t tn_aobj_pages;
243 /* Valid when tn_type = VFIFO */
245 fo_rdwr_t *tn_fo_read;
246 fo_rdwr_t *tn_fo_write;
250 LIST_HEAD(tmpfs_node_list, tmpfs_node);
252 #define tn_rdev tn_spec.tn_rdev
253 #define tn_dir tn_spec.tn_dir
254 #define tn_link tn_spec.tn_link
255 #define tn_reg tn_spec.tn_reg
256 #define tn_fifo tn_spec.tn_fifo
258 #define TMPFS_NODE_LOCK(node) mtx_lock(&(node)->tn_interlock)
259 #define TMPFS_NODE_UNLOCK(node) mtx_unlock(&(node)->tn_interlock)
261 #define TMPFS_VNODE_ALLOCATING 1
262 #define TMPFS_VNODE_WANT 2
263 /* --------------------------------------------------------------------- */
266 * Internal representation of a tmpfs mount point.
269 /* Maximum number of memory pages available for use by the file
270 * system, set during mount time. This variable must never be
271 * used directly as it may be bigger that the current amount of
272 * free memory; in the extreme case, it will hold the SIZE_MAX
273 * value. Instead, use the TMPFS_PAGES_MAX macro. */
276 /* Number of pages in use by the file system. Cannot be bigger
277 * than the value returned by TMPFS_PAGES_MAX in any case. */
278 size_t tm_pages_used;
280 /* Pointer to the node representing the root directory of this
282 struct tmpfs_node * tm_root;
284 /* Maximum number of possible nodes for this file system; set
285 * during mount time. We need a hard limit on the maximum number
286 * of nodes to avoid allocating too much of them; their objects
287 * cannot be released until the file system is unmounted.
288 * Otherwise, we could easily run out of memory by creating lots
289 * of empty files and then simply removing them. */
292 /* Number of nodes currently allocated. This number only grows.
293 * When it reaches tm_nodes_max, no more new nodes can be allocated.
294 * Of course, the old, unused ones can be reused. */
297 /* Number of nodes currently that are in use. */
298 ino_t tm_nodes_inuse;
300 /* maximum representable file size */
301 u_int64_t tm_maxfilesize;
303 /* Nodes are organized in two different lists. The used list
304 * contains all nodes that are currently used by the file system;
305 * i.e., they refer to existing files. The available list contains
306 * all nodes that are currently available for use by new files.
307 * Nodes must be kept in this list (instead of deleting them)
308 * because we need to keep track of their generation number (tn_gen
311 * Note that nodes are lazily allocated: if the available list is
312 * empty and we have enough space to create more nodes, they will be
313 * created and inserted in the used list. Once these are released,
314 * they will go into the available list, remaining alive until the
315 * file system is unmounted. */
316 struct tmpfs_node_list tm_nodes_used;
317 struct tmpfs_node_list tm_nodes_avail;
319 /* All node lock to protect the node list and tmp_pages_used */
320 struct mtx allnode_lock;
322 /* Pools used to store file system meta data. These are not shared
323 * across several instances of tmpfs for the reasons described in
325 uma_zone_t tm_dirent_pool;
326 uma_zone_t tm_node_pool;
327 struct tmpfs_str_zone tm_str_pool;
329 #define TMPFS_LOCK(tm) mtx_lock(&(tm)->allnode_lock)
330 #define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->allnode_lock)
332 /* --------------------------------------------------------------------- */
335 * This structure maps a file identifier to a tmpfs node. Used by the
341 unsigned long tf_gen;
345 /* --------------------------------------------------------------------- */
349 * Prototypes for tmpfs_subr.c.
352 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype,
353 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *,
354 char *, dev_t, struct thread *, struct tmpfs_node **);
355 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
356 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
357 const char *, uint16_t, struct tmpfs_dirent **);
358 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *,
360 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, struct vnode **,
362 void tmpfs_free_vp(struct vnode *);
363 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
364 struct componentname *, char *);
365 void tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *);
366 void tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *);
367 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node,
368 struct componentname *cnp);
369 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *);
370 int tmpfs_dir_getdotdotdent(struct tmpfs_node *, struct uio *);
371 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t);
372 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *);
373 int tmpfs_reg_resize(struct vnode *, off_t);
374 int tmpfs_chflags(struct vnode *, int, struct ucred *, struct thread *);
375 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct thread *);
376 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *,
378 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct thread *);
379 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *,
380 struct timespec *, int, struct ucred *, struct thread *);
381 void tmpfs_itimes(struct vnode *, const struct timespec *,
382 const struct timespec *);
384 void tmpfs_update(struct vnode *);
385 int tmpfs_truncate(struct vnode *, off_t);
387 /* --------------------------------------------------------------------- */
390 * Convenience macros to simplify some logical expressions.
392 #define IMPLIES(a, b) (!(a) || (b))
393 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a))
395 /* --------------------------------------------------------------------- */
398 * Checks that the directory entry pointed by 'de' matches the name 'name'
399 * with a length of 'len'.
401 #define TMPFS_DIRENT_MATCHES(de, name, len) \
402 (de->td_namelen == (uint16_t)len && \
403 memcmp((de)->td_name, (name), (de)->td_namelen) == 0)
405 /* --------------------------------------------------------------------- */
408 * Ensures that the node pointed by 'node' is a directory and that its
409 * contents are consistent with respect to directories.
411 #define TMPFS_VALIDATE_DIR(node) \
412 MPASS((node)->tn_type == VDIR); \
413 MPASS((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
414 MPASS((node)->tn_dir.tn_readdir_lastp == NULL || \
415 TMPFS_DIRCOOKIE((node)->tn_dir.tn_readdir_lastp) == (node)->tn_dir.tn_readdir_lastn);
417 /* --------------------------------------------------------------------- */
420 * Memory management stuff.
423 /* Amount of memory pages to reserve for the system (e.g., to not use by
425 * XXX: Should this be tunable through sysctl, for instance? */
426 #define TMPFS_PAGES_RESERVED (4 * 1024 * 1024 / PAGE_SIZE)
429 * Returns information about the number of available memory pages,
430 * including physical and virtual ones.
432 * If 'total' is TRUE, the value returned is the total amount of memory
433 * pages configured for the system (either in use or free).
434 * If it is FALSE, the value returned is the amount of free memory pages.
436 * Remember to remove TMPFS_PAGES_RESERVED from the returned value to avoid
437 * excessive memory usage.
440 static __inline size_t
445 size = swap_pager_avail + cnt.v_free_count + cnt.v_inactive_count;
446 size -= size > cnt.v_wire_count ? cnt.v_wire_count : size;
450 /* Returns the maximum size allowed for a tmpfs file system. This macro
451 * must be used instead of directly retrieving the value from tm_pages_max.
452 * The reason is that the size of a tmpfs file system is dynamic: it lets
453 * the user store files as long as there is enough free memory (including
454 * physical memory and swap space). Therefore, the amount of memory to be
455 * used is either the limit imposed by the user during mount time or the
456 * amount of available memory, whichever is lower. To avoid consuming all
457 * the memory for a given mount point, the system will always reserve a
458 * minimum of TMPFS_PAGES_RESERVED pages, which is also taken into account
459 * by this macro (see above). */
460 static __inline size_t
461 TMPFS_PAGES_MAX(struct tmpfs_mount *tmp)
465 freepages = tmpfs_mem_info();
466 freepages -= freepages < TMPFS_PAGES_RESERVED ?
467 freepages : TMPFS_PAGES_RESERVED;
469 return MIN(tmp->tm_pages_max, freepages + tmp->tm_pages_used);
472 /* Returns the available space for the given file system. */
473 #define TMPFS_META_SIZE(tmp) ((tmp)->tm_nodes_inuse * (sizeof(struct tmpfs_node) \
474 + sizeof(struct dirent)))
475 #define TMPFS_PAGES_AVAIL(tmp) (TMPFS_PAGES_MAX(tmp) - (tmp)->tm_pages_used - \
476 TMPFS_META_SIZE(tmp) / PAGE_SIZE - 1)
480 /* --------------------------------------------------------------------- */
483 * Macros/functions to convert from generic data structures to tmpfs
489 VFS_TO_TMPFS(struct mount *mp)
491 struct tmpfs_mount *tmp;
493 MPASS((mp) != NULL && (mp)->mnt_data != NULL);
494 tmp = (struct tmpfs_mount *)(mp)->mnt_data;
500 VP_TO_TMPFS_NODE(struct vnode *vp)
502 struct tmpfs_node *node;
504 MPASS((vp) != NULL && (vp)->v_data != NULL);
505 node = (struct tmpfs_node *)vp->v_data;
511 VP_TO_TMPFS_DIR(struct vnode *vp)
513 struct tmpfs_node *node;
515 node = VP_TO_TMPFS_NODE(vp);
516 TMPFS_VALIDATE_DIR(node);
520 /* ---------------------------------------------------------------------
521 * USER AND KERNEL DEFINITIONS
522 * --------------------------------------------------------------------- */
525 * This structure is used to communicate mount parameters between userland
528 #define TMPFS_ARGS_VERSION 1
536 /* Root node attributes. */
542 #endif /* _FS_TMPFS_TMPFS_H_ */