xref: /freebsd/sys/fs/tmpfs/tmpfs.h (revision cc16dea626cf2fc80cde667ac4798065108e596c)
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/tree.h>
53 #include <sys/vmmeter.h>
54 #include <vm/swap_pager.h>
55 
56 MALLOC_DECLARE(M_TMPFSMNT);
57 MALLOC_DECLARE(M_TMPFSNAME);
58 
59 /* --------------------------------------------------------------------- */
60 
61 /*
62  * Internal representation of a tmpfs directory entry.
63  */
64 
65 LIST_HEAD(tmpfs_dir_duphead, tmpfs_dirent);
66 
67 struct tmpfs_dirent {
68 	/*
69 	 * Depending on td_cookie flag entry can be of 3 types:
70 	 * - regular -- no hash collisions, stored in RB-Tree
71 	 * - duphead -- synthetic linked list head for dup entries
72 	 * - dup -- stored in linked list instead of RB-Tree
73 	 */
74 	union {
75 		/* regular and duphead entry types */
76 		RB_ENTRY(tmpfs_dirent)		td_entries;
77 
78 		/* dup entry type */
79 		struct {
80 			LIST_ENTRY(tmpfs_dirent) entries;
81 			LIST_ENTRY(tmpfs_dirent) index_entries;
82 		} td_dup;
83 	} uh;
84 
85 	uint32_t			td_cookie;
86 	uint32_t			td_hash;
87 	u_int				td_namelen;
88 
89 	/* Pointer to the node this entry refers to.  In case this field
90 	 * is NULL, the node is a whiteout. */
91 	struct tmpfs_node *		td_node;
92 
93 	union {
94 		/*
95 		 * The name of the entry, allocated from a string pool.  This
96 		 * string is not required to be zero-terminated.
97 		 */
98 		char *			td_name;	/* regular, dup */
99 		struct tmpfs_dir_duphead td_duphead;	/* duphead */
100 	} ud;
101 };
102 
103 /* A directory in tmpfs holds a list of directory entries, which in
104  * turn point to other files (which can be directories themselves).
105  *
106  * In tmpfs, this list is managed by a RB-Tree, whose head is defined by
107  * the struct tmpfs_dir type.
108  *
109  * It is important to notice that directories do not have entries for . and
110  * .. as other file systems do.  These can be generated when requested
111  * based on information available by other means, such as the pointer to
112  * the node itself in the former case or the pointer to the parent directory
113  * in the latter case.  This is done to simplify tmpfs's code and, more
114  * importantly, to remove redundancy. */
115 RB_HEAD(tmpfs_dir, tmpfs_dirent);
116 
117 /* Each entry in a directory has a cookie that identifies it.  Cookies
118  * supersede offsets within directories because, given how tmpfs stores
119  * directories in memory, there is no such thing as an offset.
120  *
121  * The '.', '..' and the end of directory markers have fixed cookies which
122  * cannot collide with the cookies generated by other entries.  The cookies
123  * for the other entries are generated based on the file name hash value or
124  * unique number in case of name hash collision.
125  *
126  * To preserve compatibility cookies are limited to 31 bits.
127  */
128 
129 #define	TMPFS_DIRCOOKIE_DOT		0
130 #define	TMPFS_DIRCOOKIE_DOTDOT		1
131 #define	TMPFS_DIRCOOKIE_EOF		2
132 #define	TMPFS_DIRCOOKIE_MASK		((off_t)0x3fffffffU)
133 #define	TMPFS_DIRCOOKIE_MIN		((off_t)0x00000004U)
134 #define	TMPFS_DIRCOOKIE_DUP		((off_t)0x40000000U)
135 #define	TMPFS_DIRCOOKIE_DUPHEAD		((off_t)0x80000000U)
136 #define	TMPFS_DIRCOOKIE_DUP_MIN		TMPFS_DIRCOOKIE_DUP
137 #define	TMPFS_DIRCOOKIE_DUP_MAX		\
138 	(TMPFS_DIRCOOKIE_DUP | TMPFS_DIRCOOKIE_MASK)
139 
140 /* --------------------------------------------------------------------- */
141 
142 /*
143  * Internal representation of a tmpfs file system node.
144  *
145  * This structure is splitted in two parts: one holds attributes common
146  * to all file types and the other holds data that is only applicable to
147  * a particular type.  The code must be careful to only access those
148  * attributes that are actually allowed by the node's type.
149  *
150  *
151  * Below is the key of locks used to protected the fields in the following
152  * structures.
153  *
154  */
155 struct tmpfs_node {
156 	/* Doubly-linked list entry which links all existing nodes for a
157 	 * single file system.  This is provided to ease the removal of
158 	 * all nodes during the unmount operation. */
159 	LIST_ENTRY(tmpfs_node)	tn_entries;
160 
161 	/* The node's type.  Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
162 	 * 'VLNK', 'VREG' and 'VSOCK' is allowed.  The usage of vnode
163 	 * types instead of a custom enumeration is to make things simpler
164 	 * and faster, as we do not need to convert between two types. */
165 	enum vtype		tn_type;
166 
167 	/* Node identifier. */
168 	ino_t			tn_id;
169 
170 	/* Node's internal status.  This is used by several file system
171 	 * operations to do modifications to the node in a delayed
172 	 * fashion. */
173 	int			tn_status;
174 #define	TMPFS_NODE_ACCESSED	(1 << 1)
175 #define	TMPFS_NODE_MODIFIED	(1 << 2)
176 #define	TMPFS_NODE_CHANGED	(1 << 3)
177 
178 	/* The node size.  It does not necessarily match the real amount
179 	 * of memory consumed by it. */
180 	off_t			tn_size;
181 
182 	/* Generic node attributes. */
183 	uid_t			tn_uid;
184 	gid_t			tn_gid;
185 	mode_t			tn_mode;
186 	u_long			tn_flags;
187 	nlink_t			tn_links;
188 	struct timespec		tn_atime;
189 	struct timespec		tn_mtime;
190 	struct timespec		tn_ctime;
191 	struct timespec		tn_birthtime;
192 	unsigned long		tn_gen;
193 
194 	/* As there is a single vnode for each active file within the
195 	 * system, care has to be taken to avoid allocating more than one
196 	 * vnode per file.  In order to do this, a bidirectional association
197 	 * is kept between vnodes and nodes.
198 	 *
199 	 * Whenever a vnode is allocated, its v_data field is updated to
200 	 * point to the node it references.  At the same time, the node's
201 	 * tn_vnode field is modified to point to the new vnode representing
202 	 * it.  Further attempts to allocate a vnode for this same node will
203 	 * result in returning a new reference to the value stored in
204 	 * tn_vnode.
205 	 *
206 	 * May be NULL when the node is unused (that is, no vnode has been
207 	 * allocated for it or it has been reclaimed). */
208 	struct vnode *		tn_vnode;
209 
210 	/* interlock to protect tn_vpstate */
211 	struct mtx	tn_interlock;
212 
213 	/* Identify if current node has vnode assiocate with
214 	 * or allocating vnode.
215 	 */
216 	int		tn_vpstate;
217 
218 	/* misc data field for different tn_type node */
219 	union {
220 		/* Valid when tn_type == VBLK || tn_type == VCHR. */
221 		dev_t			tn_rdev;
222 
223 		/* Valid when tn_type == VDIR. */
224 		struct tn_dir {
225 			/* Pointer to the parent directory.  The root
226 			 * directory has a pointer to itself in this field;
227 			 * this property identifies the root node. */
228 			struct tmpfs_node *	tn_parent;
229 
230 			/* Head of a tree that links the contents of
231 			 * the directory together. */
232 			struct tmpfs_dir	tn_dirhead;
233 
234 			/* Head of a list the contains fake directory entries
235 			 * heads, i.e. entries with TMPFS_DIRCOOKIE_DUPHEAD
236 			 * flag. */
237 			struct tmpfs_dir_duphead tn_dupindex;
238 
239 			/* Number and pointer of the first directory entry
240 			 * returned by the readdir operation if it were
241 			 * called again to continue reading data from the
242 			 * same directory as before.  This is used to speed
243 			 * up reads of long directories, assuming that no
244 			 * more than one read is in progress at a given time.
245 			 * Otherwise, these values are discarded. */
246 			off_t			tn_readdir_lastn;
247 			struct tmpfs_dirent *	tn_readdir_lastp;
248 		} tn_dir;
249 
250 		/* Valid when tn_type == VLNK. */
251 		/* The link's target, allocated from a string pool. */
252 		char *			tn_link;
253 
254 		/* Valid when tn_type == VREG. */
255 		struct tn_reg {
256 			/* The contents of regular files stored in a tmpfs
257 			 * file system are represented by a single anonymous
258 			 * memory object (aobj, for short).  The aobj provides
259 			 * direct access to any position within the file,
260 			 * because its contents are always mapped in a
261 			 * contiguous region of virtual memory.  It is a task
262 			 * of the memory management subsystem (see uvm(9)) to
263 			 * issue the required page ins or page outs whenever
264 			 * a position within the file is accessed. */
265 			vm_object_t		tn_aobj;
266 
267 		}tn_reg;
268 
269 		/* Valid when tn_type = VFIFO */
270 		struct tn_fifo {
271 			fo_rdwr_t		*tn_fo_read;
272 			fo_rdwr_t		*tn_fo_write;
273 		}tn_fifo;
274 	}tn_spec;
275 };
276 LIST_HEAD(tmpfs_node_list, tmpfs_node);
277 
278 #define tn_rdev tn_spec.tn_rdev
279 #define tn_dir tn_spec.tn_dir
280 #define tn_link tn_spec.tn_link
281 #define tn_reg tn_spec.tn_reg
282 #define tn_fifo tn_spec.tn_fifo
283 
284 #define TMPFS_NODE_LOCK(node) mtx_lock(&(node)->tn_interlock)
285 #define TMPFS_NODE_UNLOCK(node) mtx_unlock(&(node)->tn_interlock)
286 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock)
287 
288 #ifdef INVARIANTS
289 #define TMPFS_ASSERT_LOCKED(node) do {					\
290 		MPASS(node != NULL);					\
291 		MPASS(node->tn_vnode != NULL);				\
292 		if (!VOP_ISLOCKED(node->tn_vnode) &&			\
293 		    !mtx_owned(TMPFS_NODE_MTX(node)))			\
294 			panic("tmpfs: node is not locked: %p", node);	\
295 	} while (0)
296 #define TMPFS_ASSERT_ELOCKED(node) do {					\
297 		MPASS((node) != NULL);					\
298 		MPASS((node)->tn_vnode != NULL);			\
299 		mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED);		\
300 		ASSERT_VOP_LOCKED((node)->tn_vnode, "tmpfs");		\
301 	} while (0)
302 #else
303 #define TMPFS_ASSERT_LOCKED(node) (void)0
304 #define TMPFS_ASSERT_ELOCKED(node) (void)0
305 #endif
306 
307 #define TMPFS_VNODE_ALLOCATING	1
308 #define TMPFS_VNODE_WANT	2
309 #define TMPFS_VNODE_DOOMED	4
310 #define	TMPFS_VNODE_WRECLAIM	8
311 /* --------------------------------------------------------------------- */
312 
313 /*
314  * Internal representation of a tmpfs mount point.
315  */
316 struct tmpfs_mount {
317 	/* Maximum number of memory pages available for use by the file
318 	 * system, set during mount time.  This variable must never be
319 	 * used directly as it may be bigger than the current amount of
320 	 * free memory; in the extreme case, it will hold the SIZE_MAX
321 	 * value. */
322 	size_t			tm_pages_max;
323 
324 	/* Number of pages in use by the file system. */
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 	/* Read-only status. */
373 	int			tm_ronly;
374 };
375 #define TMPFS_LOCK(tm) mtx_lock(&(tm)->allnode_lock)
376 #define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->allnode_lock)
377 
378 /* --------------------------------------------------------------------- */
379 
380 /*
381  * This structure maps a file identifier to a tmpfs node.  Used by the
382  * NFS code.
383  */
384 struct tmpfs_fid {
385 	uint16_t		tf_len;
386 	uint16_t		tf_pad;
387 	ino_t			tf_id;
388 	unsigned long		tf_gen;
389 };
390 
391 /* --------------------------------------------------------------------- */
392 
393 #ifdef _KERNEL
394 /*
395  * Prototypes for tmpfs_subr.c.
396  */
397 
398 int	tmpfs_alloc_node(struct tmpfs_mount *, enum vtype,
399 	    uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *,
400 	    char *, dev_t, struct tmpfs_node **);
401 void	tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
402 int	tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
403 	    const char *, u_int, struct tmpfs_dirent **);
404 void	tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *);
405 void	tmpfs_dirent_init(struct tmpfs_dirent *, const char *, u_int);
406 void	tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj);
407 int	tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int,
408 	    struct vnode **);
409 void	tmpfs_free_vp(struct vnode *);
410 int	tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
411 	    struct componentname *, char *);
412 void	tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *);
413 void	tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *);
414 void	tmpfs_dir_destroy(struct tmpfs_mount *, struct tmpfs_node *);
415 struct tmpfs_dirent *	tmpfs_dir_lookup(struct tmpfs_node *node,
416 			    struct tmpfs_node *f,
417 			    struct componentname *cnp);
418 int	tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, int,
419 	    u_long *, int *);
420 int	tmpfs_dir_whiteout_add(struct vnode *, struct componentname *);
421 void	tmpfs_dir_whiteout_remove(struct vnode *, struct componentname *);
422 int	tmpfs_reg_resize(struct vnode *, off_t, boolean_t);
423 int	tmpfs_chflags(struct vnode *, u_long, struct ucred *, struct thread *);
424 int	tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct thread *);
425 int	tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *,
426 	    struct thread *);
427 int	tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct thread *);
428 int	tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *,
429 	    struct timespec *, int, struct ucred *, struct thread *);
430 void	tmpfs_itimes(struct vnode *, const struct timespec *,
431 	    const struct timespec *);
432 
433 void	tmpfs_update(struct vnode *);
434 int	tmpfs_truncate(struct vnode *, off_t);
435 
436 /* --------------------------------------------------------------------- */
437 
438 /*
439  * Convenience macros to simplify some logical expressions.
440  */
441 #define IMPLIES(a, b) (!(a) || (b))
442 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a))
443 
444 /* --------------------------------------------------------------------- */
445 
446 /*
447  * Checks that the directory entry pointed by 'de' matches the name 'name'
448  * with a length of 'len'.
449  */
450 #define TMPFS_DIRENT_MATCHES(de, name, len) \
451     (de->td_namelen == len && \
452     bcmp((de)->ud.td_name, (name), (de)->td_namelen) == 0)
453 
454 /* --------------------------------------------------------------------- */
455 
456 /*
457  * Ensures that the node pointed by 'node' is a directory and that its
458  * contents are consistent with respect to directories.
459  */
460 #define TMPFS_VALIDATE_DIR(node) do { \
461 	MPASS((node)->tn_type == VDIR); \
462 	MPASS((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
463 } while (0)
464 
465 /* --------------------------------------------------------------------- */
466 
467 /*
468  * Memory management stuff.
469  */
470 
471 /*
472  * Amount of memory pages to reserve for the system (e.g., to not use by
473  * tmpfs).
474  */
475 #define TMPFS_PAGES_MINRESERVED		(4 * 1024 * 1024 / PAGE_SIZE)
476 
477 size_t tmpfs_mem_avail(void);
478 
479 size_t tmpfs_pages_used(struct tmpfs_mount *tmp);
480 
481 #endif
482 
483 /* --------------------------------------------------------------------- */
484 
485 /*
486  * Macros/functions to convert from generic data structures to tmpfs
487  * specific ones.
488  */
489 
490 static inline
491 struct tmpfs_mount *
492 VFS_TO_TMPFS(struct mount *mp)
493 {
494 	struct tmpfs_mount *tmp;
495 
496 	MPASS((mp) != NULL && (mp)->mnt_data != NULL);
497 	tmp = (struct tmpfs_mount *)(mp)->mnt_data;
498 	return tmp;
499 }
500 
501 static inline
502 struct tmpfs_node *
503 VP_TO_TMPFS_NODE(struct vnode *vp)
504 {
505 	struct tmpfs_node *node;
506 
507 	MPASS((vp) != NULL && (vp)->v_data != NULL);
508 	node = (struct tmpfs_node *)vp->v_data;
509 	return node;
510 }
511 
512 static inline
513 struct tmpfs_node *
514 VP_TO_TMPFS_DIR(struct vnode *vp)
515 {
516 	struct tmpfs_node *node;
517 
518 	node = VP_TO_TMPFS_NODE(vp);
519 	TMPFS_VALIDATE_DIR(node);
520 	return node;
521 }
522 
523 #endif /* _FS_TMPFS_TMPFS_H_ */
524