xref: /dragonfly/sys/vfs/tmpfs/tmpfs.h (revision 78478697)
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/tree.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 #ifdef _KERNEL
54 #include <sys/systm.h>
55 #endif
56 #include <sys/vmmeter.h>
57 #include <vm/swap_pager.h>
58 
59 MALLOC_DECLARE(M_TMPFSMNT);
60 
61 /* --------------------------------------------------------------------- */
62 
63 /*
64  * Internal representation of a tmpfs directory entry.
65  */
66 struct tmpfs_dirent {
67 	RB_ENTRY(tmpfs_dirent)	rb_node;
68 	RB_ENTRY(tmpfs_dirent)	rb_cookienode;
69 
70 	/* Length of the name stored in this directory entry.  This avoids
71 	 * the need to recalculate it every time the name is used. */
72 	uint16_t		td_namelen;
73 
74 	/* The name of the entry, allocated from a string pool.  This
75 	* string is not required to be zero-terminated; therefore, the
76 	* td_namelen field must always be used when accessing its value. */
77 	char 			*td_name;
78 
79 	/* Pointer to the node this entry refers to. */
80 	struct tmpfs_node 	*td_node;
81 };
82 
83 struct tmpfs_dirtree;
84 RB_HEAD(tmpfs_dirtree, tmpfs_dirent);
85 RB_PROTOTYPE(tmpfs_dirtree, tmpfs_dirent, rb_node,
86 	tmpfs_dirtree_compare);
87 
88 RB_HEAD(tmpfs_dirtree_cookie, tmpfs_dirent);
89 RB_PROTOTYPE(tmpfs_dirtree_cookie, tmpfs_dirent, rb_cookienode,
90 	tmpfs_dirtree_cookie_compare);
91 
92 
93 /*
94  * A directory in tmpfs holds a set of directory entries, which in
95  * turn point to other files (which can be directories themselves).
96  *
97  * In tmpfs, this set is managed by a red-black tree, whose root is defined
98  * by the struct tmpfs_dirtree type.
99  *
100  * It is important to notice that directories do not have entries for . and
101  * .. as other file systems do.  These can be generated when requested
102  * based on information available by other means, such as the pointer to
103  * the node itself in the former case or the pointer to the parent directory
104  * in the latter case.  This is done to simplify tmpfs's code and, more
105  * importantly, to remove redundancy.
106  *
107  * Each entry in a directory has a cookie that identifies it.  Cookies
108  * supersede offsets within directories because, given how tmpfs stores
109  * directories in memory, there is no such thing as an offset.  (Emulating
110  * a real offset could be very difficult.)
111  *
112  * The '.', '..' and the end of directory markers have fixed cookies which
113  * cannot collide with the cookies generated by other entries.  The cookies
114  * for the other entries are generated based on the memory address on which
115  * stores their information is stored.
116  *
117  * DragonFly binaries use 64-bit cookies.  We mask-off the signed bit to
118  * ensure that cookie 'offsets' are positive.
119  */
120 #ifdef _KERNEL
121 
122 #define	TMPFS_ROOTINO	((ino_t)2)
123 
124 #define	TMPFS_DIRCOOKIE_DOT	0
125 #define	TMPFS_DIRCOOKIE_DOTDOT	1
126 #define	TMPFS_DIRCOOKIE_EOF	2
127 
128 static __inline
129 off_t
130 tmpfs_dircookie(struct tmpfs_dirent *de)
131 {
132 	return (((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFFFFFFFFFFLLU);
133 }
134 
135 #endif  /* _KERNEL */
136 
137 /* --------------------------------------------------------------------- */
138 
139 /*
140  * Internal representation of a tmpfs file system node.
141  *
142  * This structure is splitted in two parts: one holds attributes common
143  * to all file types and the other holds data that is only applicable to
144  * a particular type.  The code must be careful to only access those
145  * attributes that are actually allowed by the node's type.
146  */
147 struct tmpfs_node {
148 	/* Doubly-linked list entry which links all existing nodes for a
149 	 * single file system.  This is provided to ease the removal of
150 	 * all nodes during the unmount operation. */
151 	LIST_ENTRY(tmpfs_node)	tn_entries;
152 
153 	/* The node's type.  Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
154 	 * 'VLNK', 'VREG' and 'VSOCK' is allowed.  The usage of vnode
155 	 * types instead of a custom enumeration is to make things simpler
156 	 * and faster, as we do not need to convert between two types. */
157 	enum vtype		tn_type;
158 
159 	/* Node identifier. */
160 	ino_t			tn_id;
161 
162 	/* Node's internal status.  This is used by several file system
163 	 * operations to do modifications to the node in a delayed
164 	 * fashion. */
165 	int			tn_status;
166 #define	TMPFS_NODE_ACCESSED	(1 << 1)
167 #define	TMPFS_NODE_MODIFIED	(1 << 2)
168 #define	TMPFS_NODE_CHANGED	(1 << 3)
169 
170 	/* The node size.  It does not necessarily match the real amount
171 	 * of memory consumed by it. */
172 	off_t			tn_size;
173 
174 	/* Generic node attributes. */
175 	uid_t			tn_uid;
176 	gid_t			tn_gid;
177 	mode_t			tn_mode;
178 	int			tn_flags;
179 	nlink_t			tn_links;
180 	int32_t			tn_atime;
181 	int32_t			tn_atimensec;
182 	int32_t			tn_mtime;
183 	int32_t			tn_mtimensec;
184 	int32_t			tn_ctime;
185 	int32_t			tn_ctimensec;
186 	unsigned long		tn_gen;
187 	struct lockf		tn_advlock;
188 
189 	/* As there is a single vnode for each active file within the
190 	 * system, care has to be taken to avoid allocating more than one
191 	 * vnode per file.  In order to do this, a bidirectional association
192 	 * is kept between vnodes and nodes.
193 	 *
194 	 * Whenever a vnode is allocated, its v_data field is updated to
195 	 * point to the node it references.  At the same time, the node's
196 	 * tn_vnode field is modified to point to the new vnode representing
197 	 * it.  Further attempts to allocate a vnode for this same node will
198 	 * result in returning a new reference to the value stored in
199 	 * tn_vnode.
200 	 *
201 	 * May be NULL when the node is unused (that is, no vnode has been
202 	 * allocated for it or it has been reclaimed). */
203 	struct vnode *		tn_vnode;
204 
205 	/* interlock to protect tn_vpstate */
206 	struct lock		tn_interlock;
207 
208 	/* Identify if current node has vnode assiocate with
209 	 * or allocating vnode.
210 	 */
211 	int		tn_vpstate;
212 
213 	/* misc data field for different tn_type node */
214 	union {
215 		/* Valid when tn_type == VBLK || tn_type == VCHR. */
216 		dev_t			tn_rdev; /*int32_t ?*/
217 
218 		/* Valid when tn_type == VDIR. */
219 		struct tn_dir {
220 			/*
221 			 * Pointer to the parent directory.  The root
222 			 * directory has a pointer to itself in this field;
223 			 * this property identifies the root node.
224 			 */
225 			struct tmpfs_node *	tn_parent;
226 
227 			/*
228 			 * Directory entries are indexed by name and also
229 			 * indexed by cookie.
230 			 */
231 			struct tmpfs_dirtree		tn_dirtree;
232 			struct tmpfs_dirtree_cookie	tn_cookietree;
233 		} tn_dir;
234 
235 		/* Valid when tn_type == VLNK. */
236 		/* The link's target, allocated from a string pool. */
237 		char *			tn_link;
238 
239 		/* Valid when tn_type == VREG. */
240 		struct tn_reg {
241 			/* The contents of regular files stored in a tmpfs
242 			 * file system are represented by a single anonymous
243 			 * memory object (aobj, for short).  The aobj provides
244 			 * direct access to any position within the file,
245 			 * because its contents are always mapped in a
246 			 * contiguous region of virtual memory.  It is a task
247 			 * of the memory management subsystem (see uvm(9)) to
248 			 * issue the required page ins or page outs whenever
249 			 * a position within the file is accessed. */
250 			vm_object_t		tn_aobj;
251 			size_t			tn_aobj_pages;
252 
253 		} tn_reg;
254 
255 		/* Valid when tn_type = VFIFO */
256 		struct tn_fifo {
257 			int (*tn_fo_read)  (struct file *fp, struct uio *uio,
258 			        struct ucred *cred, int flags);
259 			int (*tn_fo_write) (struct file *fp, struct uio *uio,
260 			        struct ucred *cred, int flags);
261 		} tn_fifo;
262 	} tn_spec;
263 };
264 
265 /* Only userspace needs this */
266 #define VTOI(vp)	((struct tmpfs_node *)(vp)->v_data)
267 
268 #ifdef _KERNEL
269 LIST_HEAD(tmpfs_node_list, tmpfs_node);
270 
271 #define tn_rdev tn_spec.tn_rdev
272 #define tn_dir tn_spec.tn_dir
273 #define tn_link tn_spec.tn_link
274 #define tn_reg tn_spec.tn_reg
275 #define tn_fifo tn_spec.tn_fifo
276 
277 #define TMPFS_NODE_LOCK(node) lockmgr(&(node)->tn_interlock, LK_EXCLUSIVE|LK_RETRY)
278 #define TMPFS_NODE_LOCK_SH(node) lockmgr(&(node)->tn_interlock, LK_SHARED|LK_RETRY)
279 #define TMPFS_NODE_UNLOCK(node) lockmgr(&(node)->tn_interlock, LK_RELEASE)
280 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock)
281 
282 #ifdef INVARIANTS
283 #define TMPFS_ASSERT_LOCKED(node) do {					\
284 		KKASSERT(node != NULL);					\
285 		KKASSERT(node->tn_vnode != NULL);			\
286 		if (!vn_islocked(node->tn_vnode) &&			\
287 		    (lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE ))		\
288 			panic("tmpfs: node is not locked: %p", node);	\
289 	} while (0)
290 #define TMPFS_ASSERT_ELOCKED(node) do {					\
291 		KKASSERT((node) != NULL);				\
292 		KKASSERT(lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE);		\
293 	} while (0)
294 #else
295 #define TMPFS_ASSERT_LOCKED(node) (void)0
296 #define TMPFS_ASSERT_ELOCKED(node) (void)0
297 #endif  /* INVARIANTS */
298 
299 #define TMPFS_VNODE_ALLOCATING	1
300 #define TMPFS_VNODE_WANT	2
301 #define TMPFS_VNODE_DOOMED	4
302 /* --------------------------------------------------------------------- */
303 
304 /*
305  * Internal representation of a tmpfs mount point.
306  */
307 struct tmpfs_mount {
308 	struct mount		*tm_mount;
309 
310 	/* Maximum number of memory pages available for use by the file
311 	 * system, set during mount time.  This variable must never be
312 	 * used directly as it may be bigger than the current amount of
313 	 * free memory; in the extreme case, it will hold the SIZE_MAX
314 	 * value.  Instead, use the TMPFS_PAGES_MAX macro. */
315 	long			tm_pages_max;
316 
317 	/* Number of pages in use by the file system.  Cannot be bigger
318 	 * than the value returned by TMPFS_PAGES_MAX in any case. */
319 	long			tm_pages_used;
320 
321 	/* Pointer to the node representing the root directory of this
322 	 * file system. */
323 	struct tmpfs_node *	tm_root;
324 
325 	/* Maximum number of possible nodes for this file system; set
326 	 * during mount time.  We need a hard limit on the maximum number
327 	 * of nodes to avoid allocating too much of them; their objects
328 	 * cannot be released until the file system is unmounted.
329 	 * Otherwise, we could easily run out of memory by creating lots
330 	 * of empty files and then simply removing them. */
331 	ino_t			tm_nodes_max;
332 
333 	/* Number of nodes currently that are in use. */
334 	ino_t			tm_nodes_inuse;
335 
336 	/* maximum representable file size */
337 	u_int64_t		tm_maxfilesize;
338 
339 	/* Nodes are organized in two different lists.  The used list
340 	 * contains all nodes that are currently used by the file system;
341 	 * i.e., they refer to existing files.  The available list contains
342 	 * all nodes that are currently available for use by new files.
343 	 * Nodes must be kept in this list (instead of deleting them)
344 	 * because we need to keep track of their generation number (tn_gen
345 	 * field).
346 	 *
347 	 * Note that nodes are lazily allocated: if the available list is
348 	 * empty and we have enough space to create more nodes, they will be
349 	 * created and inserted in the used list.  Once these are released,
350 	 * they will go into the available list, remaining alive until the
351 	 * file system is unmounted. */
352 	struct tmpfs_node_list	tm_nodes_used;
353 
354 	/* Per-mount malloc zones for tmpfs nodes, names, and dirents */
355 	struct malloc_type	*tm_node_zone;
356 	struct malloc_type	*tm_dirent_zone;
357 	struct malloc_type	*tm_name_zone;
358 
359 	struct objcache_malloc_args tm_node_zone_malloc_args;
360 	struct objcache_malloc_args tm_dirent_zone_malloc_args;
361 
362 	/* Pools used to store file system meta data. */
363 	struct objcache		*tm_dirent_pool;
364 	struct objcache		*tm_node_pool;
365 
366 	ino_t			tm_ino;
367 	int			tm_flags;
368 
369 	struct netexport	tm_export;
370 
371 	struct mount		*tm_mnt;
372 };
373 
374 #define TMPFS_LOCK(tm) lwkt_gettoken(&(tm)->tm_mount->mnt_token)
375 #define TMPFS_UNLOCK(tm) lwkt_reltoken(&(tm)->tm_mount->mnt_token)
376 
377 /* --------------------------------------------------------------------- */
378 
379 /*
380  * This structure maps a file identifier to a tmpfs node.  Used by the
381  * NFS code.
382  */
383 struct tmpfs_fid {
384 	uint16_t		tf_len;
385 	uint16_t		tf_pad;
386 	ino_t			tf_id;
387 	unsigned long		tf_gen;
388 };
389 
390 /* --------------------------------------------------------------------- */
391 
392 /*
393  * Prototypes for tmpfs_subr.c.
394  */
395 
396 int	tmpfs_alloc_node(struct tmpfs_mount *, enum vtype,
397 	    uid_t uid, gid_t gid, mode_t mode, char *, int, int,
398 	    struct tmpfs_node **);
399 void	tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
400 int	tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
401 	    const char *, uint16_t, struct tmpfs_dirent **);
402 void	tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *);
403 int	tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int,
404 	    struct vnode **);
405 void	tmpfs_free_vp(struct vnode *);
406 int	tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
407 	    struct namecache *, struct ucred *, char *);
408 void	tmpfs_dir_attach(struct tmpfs_node *, struct tmpfs_dirent *);
409 void	tmpfs_dir_detach(struct tmpfs_node *, struct tmpfs_dirent *);
410 struct tmpfs_dirent *	tmpfs_dir_lookup(struct tmpfs_node *node,
411 			    struct tmpfs_node *f,
412 			    struct namecache *ncp);
413 int	tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *);
414 int	tmpfs_dir_getdotdotdent(struct tmpfs_mount *,
415 			    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, int);
419 int	tmpfs_chflags(struct vnode *, int, struct ucred *);
420 int	tmpfs_chmod(struct vnode *, mode_t, struct ucred *);
421 int	tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *);
422 int	tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *);
423 int	tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *,
424 	    int, struct ucred *);
425 void	tmpfs_itimes(struct vnode *, const struct timespec *,
426 	    const struct timespec *);
427 
428 void	tmpfs_update(struct vnode *);
429 int	tmpfs_truncate(struct vnode *, off_t);
430 boolean_t tmpfs_node_ctor(void *obj, void *privdata, int flags);
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) do {	\
457     KKASSERT((node)->tn_type == VDIR);	\
458     KKASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
459 } while(0)
460 
461 /* --------------------------------------------------------------------- */
462 
463 /*
464  * Macros/functions to convert from generic data structures to tmpfs
465  * specific ones.  Kernel code use VP_TO_TMPFS_NODE() instead of VTOI().
466  */
467 
468 static inline
469 struct tmpfs_mount *
470 VFS_TO_TMPFS(struct mount *mp)
471 {
472 	struct tmpfs_mount *tmp;
473 
474 	KKASSERT((mp) != NULL && (mp)->mnt_data != NULL);
475 	tmp = (struct tmpfs_mount *)(mp)->mnt_data;
476 	return tmp;
477 }
478 
479 static inline
480 struct tmpfs_node *
481 VP_TO_TMPFS_NODE(struct vnode *vp)
482 {
483 	struct tmpfs_node *node;
484 
485 	KKASSERT((vp) != NULL && (vp)->v_data != NULL);
486 	node = (struct tmpfs_node *)vp->v_data;
487 	return node;
488 }
489 
490 static inline
491 struct tmpfs_node *
492 VP_TO_TMPFS_DIR(struct vnode *vp)
493 {
494 	struct tmpfs_node *node;
495 
496 	node = VP_TO_TMPFS_NODE(vp);
497 	TMPFS_VALIDATE_DIR(node);
498 	return node;
499 }
500 
501 /* --------------------------------------------------------------------- */
502 /*
503  * buffer cache size
504  */
505 #define TMPFS_BLKSIZE	16384			/* buffer cache size*/
506 #define TMPFS_BLKMASK	(TMPFS_BLKSIZE - 1)
507 #define TMPFS_BLKMASK64	((off_t)(TMPFS_BLKSIZE - 1))
508 #endif /* _KERNEL */
509 
510 #endif /* _VFS_TMPFS_TMPFS_H_ */
511