xref: /freebsd/sys/fs/tmpfs/tmpfs.h (revision 1de7b4b805ddbf2429da511c053686ac4591ed89)
1 /*	$NetBSD: tmpfs.h,v 1.26 2007/02/22 06:37:00 thorpej Exp $	*/
2 
3 /*-
4  * SPDX-License-Identifier: BSD-2-Clause-NetBSD
5  *
6  * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to The NetBSD Foundation
10  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11  * 2005 program.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGE.
33  *
34  * $FreeBSD$
35  */
36 
37 #ifndef _FS_TMPFS_TMPFS_H_
38 #define _FS_TMPFS_TMPFS_H_
39 
40 #include <sys/dirent.h>
41 #include <sys/mount.h>
42 #include <sys/queue.h>
43 #include <sys/vnode.h>
44 #include <sys/file.h>
45 #include <sys/lock.h>
46 #include <sys/mutex.h>
47 
48 #include <sys/malloc.h>
49 #include <sys/systm.h>
50 #include <sys/tree.h>
51 #include <sys/vmmeter.h>
52 #include <vm/swap_pager.h>
53 
54 MALLOC_DECLARE(M_TMPFSMNT);
55 MALLOC_DECLARE(M_TMPFSNAME);
56 
57 /*
58  * Internal representation of a tmpfs directory entry.
59  */
60 
61 LIST_HEAD(tmpfs_dir_duphead, tmpfs_dirent);
62 
63 struct tmpfs_dirent {
64 	/*
65 	 * Depending on td_cookie flag entry can be of 3 types:
66 	 * - regular -- no hash collisions, stored in RB-Tree
67 	 * - duphead -- synthetic linked list head for dup entries
68 	 * - dup -- stored in linked list instead of RB-Tree
69 	 */
70 	union {
71 		/* regular and duphead entry types */
72 		RB_ENTRY(tmpfs_dirent)		td_entries;
73 
74 		/* dup entry type */
75 		struct {
76 			LIST_ENTRY(tmpfs_dirent) entries;
77 			LIST_ENTRY(tmpfs_dirent) index_entries;
78 		} td_dup;
79 	} uh;
80 
81 	uint32_t			td_cookie;
82 	uint32_t			td_hash;
83 	u_int				td_namelen;
84 
85 	/*
86 	 * Pointer to the node this entry refers to.  In case this field
87 	 * is NULL, the node is a whiteout.
88 	 */
89 	struct tmpfs_node *		td_node;
90 
91 	union {
92 		/*
93 		 * The name of the entry, allocated from a string pool.  This
94 		 * string is not required to be zero-terminated.
95 		 */
96 		char *			td_name;	/* regular, dup */
97 		struct tmpfs_dir_duphead td_duphead;	/* duphead */
98 	} ud;
99 };
100 
101 /*
102  * A directory in tmpfs holds a collection of directory entries, which
103  * in turn point to other files (which can be directories themselves).
104  *
105  * In tmpfs, this collection is managed by a RB-Tree, whose head is
106  * defined by the struct tmpfs_dir type.
107  *
108  * It is important to notice that directories do not have entries for . and
109  * .. as other file systems do.  These can be generated when requested
110  * based on information available by other means, such as the pointer to
111  * the node itself in the former case or the pointer to the parent directory
112  * in the latter case.  This is done to simplify tmpfs's code and, more
113  * importantly, to remove redundancy.
114  */
115 RB_HEAD(tmpfs_dir, tmpfs_dirent);
116 
117 /*
118  * Each entry in a directory has a cookie that identifies it.  Cookies
119  * supersede offsets within directories because, given how tmpfs stores
120  * directories in memory, there is no such thing as an offset.
121  *
122  * The '.', '..' and the end of directory markers have fixed cookies which
123  * cannot collide with the cookies generated by other entries.  The cookies
124  * for the other entries are generated based on the file name hash value or
125  * unique number in case of name hash collision.
126  *
127  * To preserve compatibility cookies are limited to 31 bits.
128  */
129 
130 #define	TMPFS_DIRCOOKIE_DOT		0
131 #define	TMPFS_DIRCOOKIE_DOTDOT		1
132 #define	TMPFS_DIRCOOKIE_EOF		2
133 #define	TMPFS_DIRCOOKIE_MASK		((off_t)0x3fffffffU)
134 #define	TMPFS_DIRCOOKIE_MIN		((off_t)0x00000004U)
135 #define	TMPFS_DIRCOOKIE_DUP		((off_t)0x40000000U)
136 #define	TMPFS_DIRCOOKIE_DUPHEAD		((off_t)0x80000000U)
137 #define	TMPFS_DIRCOOKIE_DUP_MIN		TMPFS_DIRCOOKIE_DUP
138 #define	TMPFS_DIRCOOKIE_DUP_MAX		\
139 	(TMPFS_DIRCOOKIE_DUP | TMPFS_DIRCOOKIE_MASK)
140 
141 /*
142  * Internal representation of a tmpfs file system node.
143  *
144  * This structure is splitted in two parts: one holds attributes common
145  * to all file types and the other holds data that is only applicable to
146  * a particular type.  The code must be careful to only access those
147  * attributes that are actually allowed by the node's type.
148  *
149  * Below is the key of locks used to protected the fields in the following
150  * structures.
151  * (v)  vnode lock in exclusive mode
152  * (vi) vnode lock in exclusive mode, or vnode lock in shared vnode and
153  *	tn_interlock
154  * (i)  tn_interlock
155  * (m)  tmpfs_mount tm_allnode_lock
156  * (c)  stable after creation
157  */
158 struct tmpfs_node {
159 	/*
160 	 * Doubly-linked list entry which links all existing nodes for
161 	 * a single file system.  This is provided to ease the removal
162 	 * of all nodes during the unmount operation, and to support
163 	 * the implementation of VOP_VNTOCNP().  tn_attached is false
164 	 * when the node is removed from list and unlocked.
165 	 */
166 	LIST_ENTRY(tmpfs_node)	tn_entries;	/* (m) */
167 	bool			tn_attached;	/* (m) */
168 
169 	/*
170 	 * The node's type.  Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
171 	 * 'VLNK', 'VREG' and 'VSOCK' is allowed.  The usage of vnode
172 	 * types instead of a custom enumeration is to make things simpler
173 	 * and faster, as we do not need to convert between two types.
174 	 */
175 	enum vtype		tn_type;	/* (c) */
176 
177 	/* Node identifier. */
178 	ino_t			tn_id;		/* (c) */
179 
180 	/*
181 	 * Node's internal status.  This is used by several file system
182 	 * operations to do modifications to the node in a delayed
183 	 * fashion.
184 	 */
185 	int			tn_status;	/* (vi) */
186 #define	TMPFS_NODE_ACCESSED	(1 << 1)
187 #define	TMPFS_NODE_MODIFIED	(1 << 2)
188 #define	TMPFS_NODE_CHANGED	(1 << 3)
189 
190 	/*
191 	 * The node size.  It does not necessarily match the real amount
192 	 * of memory consumed by it.
193 	 */
194 	off_t			tn_size;	/* (v) */
195 
196 	/* Generic node attributes. */
197 	uid_t			tn_uid;		/* (v) */
198 	gid_t			tn_gid;		/* (v) */
199 	mode_t			tn_mode;	/* (v) */
200 	u_long			tn_flags;	/* (v) */
201 	nlink_t			tn_links;	/* (v) */
202 	struct timespec		tn_atime;	/* (vi) */
203 	struct timespec		tn_mtime;	/* (vi) */
204 	struct timespec		tn_ctime;	/* (vi) */
205 	struct timespec		tn_birthtime;	/* (v) */
206 	unsigned long		tn_gen;		/* (c) */
207 
208 	/*
209 	 * As there is a single vnode for each active file within the
210 	 * system, care has to be taken to avoid allocating more than one
211 	 * vnode per file.  In order to do this, a bidirectional association
212 	 * is kept between vnodes and nodes.
213 	 *
214 	 * Whenever a vnode is allocated, its v_data field is updated to
215 	 * point to the node it references.  At the same time, the node's
216 	 * tn_vnode field is modified to point to the new vnode representing
217 	 * it.  Further attempts to allocate a vnode for this same node will
218 	 * result in returning a new reference to the value stored in
219 	 * tn_vnode.
220 	 *
221 	 * May be NULL when the node is unused (that is, no vnode has been
222 	 * allocated for it or it has been reclaimed).
223 	 */
224 	struct vnode *		tn_vnode;	/* (i) */
225 
226 	/*
227 	 * Interlock to protect tn_vpstate, and tn_status under shared
228 	 * vnode lock.
229 	 */
230 	struct mtx	tn_interlock;
231 
232 	/*
233 	 * Identify if current node has vnode assiocate with
234 	 * or allocating vnode.
235 	 */
236 	int		tn_vpstate;		/* (i) */
237 
238 	/* Transient refcounter on this node. */
239 	u_int		tn_refcount;		/* (m) + (i) */
240 
241 	/* misc data field for different tn_type node */
242 	union {
243 		/* Valid when tn_type == VBLK || tn_type == VCHR. */
244 		dev_t			tn_rdev;	/* (c) */
245 
246 		/* Valid when tn_type == VDIR. */
247 		struct tn_dir {
248 			/*
249 			 * Pointer to the parent directory.  The root
250 			 * directory has a pointer to itself in this field;
251 			 * this property identifies the root node.
252 			 */
253 			struct tmpfs_node *	tn_parent;
254 
255 			/*
256 			 * Head of a tree that links the contents of
257 			 * the directory together.
258 			 */
259 			struct tmpfs_dir	tn_dirhead;
260 
261 			/*
262 			 * Head of a list the contains fake directory entries
263 			 * heads, i.e. entries with TMPFS_DIRCOOKIE_DUPHEAD
264 			 * flag.
265 			 */
266 			struct tmpfs_dir_duphead tn_dupindex;
267 
268 			/*
269 			 * Number and pointer of the first directory entry
270 			 * returned by the readdir operation if it were
271 			 * called again to continue reading data from the
272 			 * same directory as before.  This is used to speed
273 			 * up reads of long directories, assuming that no
274 			 * more than one read is in progress at a given time.
275 			 * Otherwise, these values are discarded.
276 			 */
277 			off_t			tn_readdir_lastn;
278 			struct tmpfs_dirent *	tn_readdir_lastp;
279 		} tn_dir;
280 
281 		/* Valid when tn_type == VLNK. */
282 		/* The link's target, allocated from a string pool. */
283 		char *			tn_link;	/* (c) */
284 
285 		/* Valid when tn_type == VREG. */
286 		struct tn_reg {
287 			/*
288 			 * The contents of regular files stored in a
289 			 * tmpfs file system are represented by a
290 			 * single anonymous memory object (aobj, for
291 			 * short).  The aobj provides direct access to
292 			 * any position within the file.  It is a task
293 			 * of the memory management subsystem to issue
294 			 * the required page ins or page outs whenever
295 			 * a position within the file is accessed.
296 			 */
297 			vm_object_t		tn_aobj;	/* (c) */
298 		} tn_reg;
299 	} tn_spec;	/* (v) */
300 };
301 LIST_HEAD(tmpfs_node_list, tmpfs_node);
302 
303 #define tn_rdev tn_spec.tn_rdev
304 #define tn_dir tn_spec.tn_dir
305 #define tn_link tn_spec.tn_link
306 #define tn_reg tn_spec.tn_reg
307 #define tn_fifo tn_spec.tn_fifo
308 
309 #define TMPFS_NODE_LOCK(node) mtx_lock(&(node)->tn_interlock)
310 #define TMPFS_NODE_UNLOCK(node) mtx_unlock(&(node)->tn_interlock)
311 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock)
312 #define	TMPFS_NODE_ASSERT_LOCKED(node) mtx_assert(TMPFS_NODE_MTX(node), \
313     MA_OWNED)
314 
315 #ifdef INVARIANTS
316 #define TMPFS_ASSERT_LOCKED(node) do {					\
317 		MPASS((node) != NULL);					\
318 		MPASS((node)->tn_vnode != NULL);			\
319 		ASSERT_VOP_LOCKED((node)->tn_vnode, "tmpfs assert");	\
320 	} while (0)
321 #else
322 #define TMPFS_ASSERT_LOCKED(node) (void)0
323 #endif
324 
325 #define TMPFS_VNODE_ALLOCATING	1
326 #define TMPFS_VNODE_WANT	2
327 #define TMPFS_VNODE_DOOMED	4
328 #define	TMPFS_VNODE_WRECLAIM	8
329 
330 /*
331  * Internal representation of a tmpfs mount point.
332  */
333 struct tmpfs_mount {
334 	/*
335 	 * Maximum number of memory pages available for use by the file
336 	 * system, set during mount time.  This variable must never be
337 	 * used directly as it may be bigger than the current amount of
338 	 * free memory; in the extreme case, it will hold the ULONG_MAX
339 	 * value.
340 	 */
341 	u_long			tm_pages_max;
342 
343 	/* Number of pages in use by the file system. */
344 	u_long			tm_pages_used;
345 
346 	/*
347 	 * Pointer to the node representing the root directory of this
348 	 * file system.
349 	 */
350 	struct tmpfs_node *	tm_root;
351 
352 	/*
353 	 * Maximum number of possible nodes for this file system; set
354 	 * during mount time.  We need a hard limit on the maximum number
355 	 * of nodes to avoid allocating too much of them; their objects
356 	 * cannot be released until the file system is unmounted.
357 	 * Otherwise, we could easily run out of memory by creating lots
358 	 * of empty files and then simply removing them.
359 	 */
360 	ino_t			tm_nodes_max;
361 
362 	/* unrhdr used to allocate inode numbers */
363 	struct unrhdr *		tm_ino_unr;
364 
365 	/* Number of nodes currently that are in use. */
366 	ino_t			tm_nodes_inuse;
367 
368 	/* Refcounter on this struct tmpfs_mount. */
369 	uint64_t		tm_refcount;
370 
371 	/* maximum representable file size */
372 	u_int64_t		tm_maxfilesize;
373 
374 	/*
375 	 * The used list contains all nodes that are currently used by
376 	 * the file system; i.e., they refer to existing files.
377 	 */
378 	struct tmpfs_node_list	tm_nodes_used;
379 
380 	/* All node lock to protect the node list and tmp_pages_used. */
381 	struct mtx		tm_allnode_lock;
382 
383 	/* Zones used to store file system meta data, per tmpfs mount. */
384 	uma_zone_t		tm_dirent_pool;
385 	uma_zone_t		tm_node_pool;
386 
387 	/* Read-only status. */
388 	bool			tm_ronly;
389 	/* Do not use namecache. */
390 	bool			tm_nonc;
391 };
392 #define	TMPFS_LOCK(tm) mtx_lock(&(tm)->tm_allnode_lock)
393 #define	TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->tm_allnode_lock)
394 #define	TMPFS_MP_ASSERT_LOCKED(tm) mtx_assert(&(tm)->tm_allnode_lock, MA_OWNED)
395 
396 /*
397  * This structure maps a file identifier to a tmpfs node.  Used by the
398  * NFS code.
399  */
400 struct tmpfs_fid {
401 	uint16_t		tf_len;
402 	uint16_t		tf_pad;
403 	ino_t			tf_id;
404 	unsigned long		tf_gen;
405 };
406 
407 struct tmpfs_dir_cursor {
408 	struct tmpfs_dirent	*tdc_current;
409 	struct tmpfs_dirent	*tdc_tree;
410 };
411 
412 #ifdef _KERNEL
413 /*
414  * Prototypes for tmpfs_subr.c.
415  */
416 
417 void	tmpfs_ref_node(struct tmpfs_node *node);
418 void	tmpfs_ref_node_locked(struct tmpfs_node *node);
419 int	tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *, enum vtype,
420 	    uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *,
421 	    char *, dev_t, struct tmpfs_node **);
422 void	tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
423 bool	tmpfs_free_node_locked(struct tmpfs_mount *, struct tmpfs_node *, bool);
424 void	tmpfs_free_tmp(struct tmpfs_mount *);
425 int	tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
426 	    const char *, u_int, struct tmpfs_dirent **);
427 void	tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *);
428 void	tmpfs_dirent_init(struct tmpfs_dirent *, const char *, u_int);
429 void	tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj);
430 int	tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int,
431 	    struct vnode **);
432 void	tmpfs_free_vp(struct vnode *);
433 int	tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
434 	    struct componentname *, char *);
435 void	tmpfs_check_mtime(struct vnode *);
436 void	tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *);
437 void	tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *);
438 void	tmpfs_dir_destroy(struct tmpfs_mount *, struct tmpfs_node *);
439 struct tmpfs_dirent *	tmpfs_dir_lookup(struct tmpfs_node *node,
440 			    struct tmpfs_node *f,
441 			    struct componentname *cnp);
442 int	tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, int,
443 	    u_long *, int *);
444 int	tmpfs_dir_whiteout_add(struct vnode *, struct componentname *);
445 void	tmpfs_dir_whiteout_remove(struct vnode *, struct componentname *);
446 int	tmpfs_reg_resize(struct vnode *, off_t, boolean_t);
447 int	tmpfs_chflags(struct vnode *, u_long, struct ucred *, struct thread *);
448 int	tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct thread *);
449 int	tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *,
450 	    struct thread *);
451 int	tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct thread *);
452 int	tmpfs_chtimes(struct vnode *, struct vattr *, struct ucred *cred,
453 	    struct thread *);
454 void	tmpfs_itimes(struct vnode *, const struct timespec *,
455 	    const struct timespec *);
456 
457 void	tmpfs_set_status(struct tmpfs_node *node, int status);
458 void	tmpfs_update(struct vnode *);
459 int	tmpfs_truncate(struct vnode *, off_t);
460 struct tmpfs_dirent *tmpfs_dir_first(struct tmpfs_node *dnode,
461 	    struct tmpfs_dir_cursor *dc);
462 struct tmpfs_dirent *tmpfs_dir_next(struct tmpfs_node *dnode,
463 	    struct tmpfs_dir_cursor *dc);
464 
465 /*
466  * Convenience macros to simplify some logical expressions.
467  */
468 #define IMPLIES(a, b) (!(a) || (b))
469 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a))
470 
471 /*
472  * Checks that the directory entry pointed by 'de' matches the name 'name'
473  * with a length of 'len'.
474  */
475 #define TMPFS_DIRENT_MATCHES(de, name, len) \
476     (de->td_namelen == len && \
477     bcmp((de)->ud.td_name, (name), (de)->td_namelen) == 0)
478 
479 /*
480  * Ensures that the node pointed by 'node' is a directory and that its
481  * contents are consistent with respect to directories.
482  */
483 #define TMPFS_VALIDATE_DIR(node) do { \
484 	MPASS((node)->tn_type == VDIR); \
485 	MPASS((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
486 } while (0)
487 
488 /*
489  * Amount of memory pages to reserve for the system (e.g., to not use by
490  * tmpfs).
491  */
492 #define TMPFS_PAGES_MINRESERVED		(4 * 1024 * 1024 / PAGE_SIZE)
493 
494 size_t tmpfs_mem_avail(void);
495 
496 size_t tmpfs_pages_used(struct tmpfs_mount *tmp);
497 
498 #endif
499 
500 /*
501  * Macros/functions to convert from generic data structures to tmpfs
502  * specific ones.
503  */
504 
505 static inline struct tmpfs_mount *
506 VFS_TO_TMPFS(struct mount *mp)
507 {
508 	struct tmpfs_mount *tmp;
509 
510 	MPASS(mp != NULL && mp->mnt_data != NULL);
511 	tmp = (struct tmpfs_mount *)mp->mnt_data;
512 	return (tmp);
513 }
514 
515 static inline struct tmpfs_node *
516 VP_TO_TMPFS_NODE(struct vnode *vp)
517 {
518 	struct tmpfs_node *node;
519 
520 	MPASS(vp != NULL && vp->v_data != NULL);
521 	node = (struct tmpfs_node *)vp->v_data;
522 	return (node);
523 }
524 
525 static inline struct tmpfs_node *
526 VP_TO_TMPFS_DIR(struct vnode *vp)
527 {
528 	struct tmpfs_node *node;
529 
530 	node = VP_TO_TMPFS_NODE(vp);
531 	TMPFS_VALIDATE_DIR(node);
532 	return (node);
533 }
534 
535 static inline bool
536 tmpfs_use_nc(struct vnode *vp)
537 {
538 
539 	return (!(VFS_TO_TMPFS(vp->v_mount)->tm_nonc));
540 }
541 
542 #endif /* _FS_TMPFS_TMPFS_H_ */
543