Home | History | Annotate | Line # | Download | only in tmpfs
tmpfs.h revision 1.8.2.1
      1 /*	$NetBSD: tmpfs.h,v 1.8.2.1 2005/10/20 07:13:14 yamt Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2005 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  * 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.
     26  *
     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.
     38  */
     39 
     40 #if !defined(_TMPFS_H_)
     41 #  define _TMPFS_H_
     42 #else
     43 #  error "tmpfs.h cannot be included multiple times."
     44 #endif
     45 
     46 /* ---------------------------------------------------------------------
     47  * KERNEL-SPECIFIC DEFINITIONS
     48  * --------------------------------------------------------------------- */
     49 
     50 #if defined(_KERNEL)
     51 
     52 #include <sys/dirent.h>
     53 #include <sys/mount.h>
     54 #include <sys/queue.h>
     55 #include <sys/vnode.h>
     56 
     57 #include <fs/tmpfs/tmpfs_pool.h>
     58 
     59 /* --------------------------------------------------------------------- */
     60 
     61 /*
     62  * Internal representation of a tmpfs directory entry.
     63  */
     64 struct tmpfs_dirent {
     65 	TAILQ_ENTRY(tmpfs_dirent)	td_entries;
     66 
     67 	/* Length of the name stored in this directory entry.  This avoids
     68 	 * the need to recalculate it every time the name is used. */
     69 	uint16_t			td_namelen;
     70 
     71 	/* The name of the entry, allocated from a string pool.  This
     72 	* string is not required to be zero-terminated; therefore, the
     73 	* td_namelen field must always be used when accessing its value. */
     74 	char *				td_name;
     75 
     76 	/* Pointer to the node this entry refers to. */
     77 	struct tmpfs_node *		td_node;
     78 };
     79 
     80 /* A directory in tmpfs holds a sorted list of directory entries, which in
     81  * turn point to other files (which can be directories themselves).
     82  *
     83  * In tmpfs, this list is managed by a tail queue, whose head is defined by
     84  * the struct tmpfs_dir type.
     85  *
     86  * It is imporant to notice that directories do not have entries for . and
     87  * .. as other file systems do.  These can be generated when requested
     88  * based on information available by other means, such as the pointer to
     89  * the node itself in the former case or the pointer to the parent directory
     90  * in the latter case.  This is done to simplify tmpfs's code and, more
     91  * importantly, to remove redundancy. */
     92 TAILQ_HEAD(tmpfs_dir, tmpfs_dirent);
     93 
     94 #define	TMPFS_DIRCOOKIE(dirent)	((off_t)(uintptr_t)(dirent))
     95 #define	TMPFS_DIRCOOKIE_DOT	0
     96 #define	TMPFS_DIRCOOKIE_DOTDOT	1
     97 #define	TMPFS_DIRCOOKIE_EOF	2
     98 
     99 /* --------------------------------------------------------------------- */
    100 
    101 /*
    102  * Internal representation of a tmpfs file system node.
    103  *
    104  * This structure is splitted in two parts: one holds attributes common
    105  * to all file types and the other holds data that is only applicable to
    106  * a particular type.  The code must be careful to only access those
    107  * attributes that are actually allowed by the node's type.
    108  */
    109 struct tmpfs_node {
    110 	/* Doubly-linked list entry which links all existing nodes for a
    111 	 * single file system.  This is provided to ease the removal of
    112 	 * all nodes during the unmount operation. */
    113 	LIST_ENTRY(tmpfs_node)	tn_entries;
    114 
    115 	/* The node's type.  Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
    116 	 * 'VLNK', 'VREG' and 'VSOCK' is allowed.  The usage of vnode
    117 	 * types instead of a custom enumeration is to make things simpler
    118 	 * and faster, as we do not need to convert between two types. */
    119 	enum vtype		tn_type;
    120 
    121 	/* Node identifier. */
    122 	ino_t			tn_id;
    123 
    124 	/* Node's internal status.  This is used by several file system
    125 	 * operations to do modifications to the node in a delayed
    126 	 * fashion. */
    127 	int			tn_status;
    128 #define	TMPFS_NODE_ACCESSED	(1 << 1)
    129 #define	TMPFS_NODE_MODIFIED	(1 << 2)
    130 #define	TMPFS_NODE_CHANGED	(1 << 3)
    131 
    132 	/* The node size.  It does not necessarily match the real amount
    133 	 * of memory consumed by it. */
    134 	off_t			tn_size;
    135 
    136 	/* Generic node attributes. */
    137 	uid_t			tn_uid;
    138 	gid_t			tn_gid;
    139 	mode_t			tn_mode;
    140 	int			tn_flags;
    141 	nlink_t			tn_links;
    142 	struct timespec		tn_atime;
    143 	struct timespec		tn_mtime;
    144 	struct timespec		tn_ctime;
    145 	struct timespec		tn_birthtime;
    146 	unsigned long		tn_gen;
    147 
    148 	/* Head of byte-level lock list (used by tmpfs_advlock). */
    149 	struct lockf *		tn_lockf;
    150 
    151 	/* As there is a single vnode for each active file within the
    152 	 * system, care has to be taken to avoid allocating more than one
    153 	 * vnode per file.  In order to do this, a bidirectional association
    154 	 * is kept between vnodes and nodes.
    155 	 *
    156 	 * Whenever a vnode is allocated, its v_data field is updated to
    157 	 * point to the node it references.  At the same time, the node's
    158 	 * tn_vnode field is modified to point to the new vnode representing
    159 	 * it.  Further attempts to allocate a vnode for this same node will
    160 	 * result in returning a new reference to the value stored in
    161 	 * tn_vnode.
    162 	 *
    163 	 * May be NULL when the node is unused (that is, no vnode has been
    164 	 * allocated for it or it has been reclaimed). */
    165 	struct vnode *		tn_vnode;
    166 
    167 	/* Pointer to the node returned by tmpfs_lookup() after doing a
    168 	 * delete or a rename lookup; its value is only valid in these two
    169 	 * situations.  In case we were looking up . or .., it holds a null
    170 	 * pointer. */
    171 	struct tmpfs_dirent *	tn_lookup_dirent;
    172 
    173 	union {
    174 		/* Valid when tn_type == VBLK || tn_type == VCHR. */
    175 		struct {
    176 			dev_t			tn_rdev;
    177 		};
    178 
    179 		/* Valid when tn_type == VDIR. */
    180 		struct {
    181 			/* Pointer to the parent directory.  The root
    182 			 * directory has a pointer to itself in this field;
    183 			 * this property identifies the root node. */
    184 			struct tmpfs_node *	tn_parent;
    185 
    186 			/* Head of a tail-queue that links the contents of
    187 			 * the directory together.  See above for a
    188 			 * description of its contents. */
    189 			struct tmpfs_dir	tn_dir;
    190 
    191 			/* Number and pointer of the first directory entry
    192 			 * returned by the readdir operation if it were
    193 			 * called again to continue reading data from the
    194 			 * same directory as before.  This is used to speed
    195 			 * up reads of long directories, assuming that no
    196 			 * more than one read is in progress at a given time.
    197 			 * Otherwise, these values are discarded and a linear
    198 			 * scan is performed from the beginning up to the
    199 			 * point where readdir starts returning values. */
    200 			off_t			tn_readdir_lastn;
    201 			struct tmpfs_dirent *	tn_readdir_lastp;
    202 		};
    203 
    204 		/* Valid when tn_type == VLNK. */
    205 		struct {
    206 			/* The link's target, allocated from a string pool. */
    207 			char *			tn_link;
    208 		};
    209 
    210 		/* Valid when tn_type == VREG. */
    211 		struct {
    212 			/* The contents of regular files stored in a tmpfs
    213 			 * file system are represented by a single anonymous
    214 			 * memory object (aobj, for short).  The aobj provides
    215 			 * direct access to any position within the file,
    216 			 * because its contents are always mapped in a
    217 			 * contiguous region of virtual memory.  It is a task
    218 			 * of the memory management subsystem (see uvm(9)) to
    219 			 * issue the required page ins or page outs whenever
    220 			 * a position within the file is accessed. */
    221 			struct uvm_object *	tn_aobj;
    222 			size_t			tn_aobj_pages;
    223 		};
    224 	};
    225 };
    226 LIST_HEAD(tmpfs_node_list, tmpfs_node);
    227 
    228 /* --------------------------------------------------------------------- */
    229 
    230 /*
    231  * Internal representation of a tmpfs mount point.
    232  */
    233 struct tmpfs_mount {
    234 	/* Maximum number of memory pages available for use by the file
    235 	 * system, set during mount time.  This variable must never be
    236 	 * used directly as it may be bigger that the current amount of
    237 	 * free memory; in the extreme case, it will hold the SIZE_MAX
    238 	 * value.  Instead, use the TMPFS_PAGES_MAX macro. */
    239 	size_t			tm_pages_max;
    240 
    241 	/* Number of pages in use by the file system.  Cannot be bigger
    242 	 * than the value returned by TMPFS_PAGES_MAX in any case. */
    243 	size_t			tm_pages_used;
    244 
    245 	/* Pointer to the node representing the root directory of this
    246 	 * file system. */
    247 	struct tmpfs_node *	tm_root;
    248 
    249 	/* Maximum number of possible nodes for this file system; set
    250 	 * during mount time.  We need a hard limit on the maximum number
    251 	 * of nodes to avoid allocating too much of them; their objects
    252 	 * cannot be released until the file system is unmounted.
    253 	 * Otherwise, we could easily run out of memory by creating lots
    254 	 * of empty files and then simply removing them. */
    255 	ino_t			tm_nodes_max;
    256 
    257 	/* Number of nodes currently allocated.  This number only grows.
    258 	 * When it reaches tm_nodes_max, no more new nodes can be allocated.
    259 	 * Of course, the old, unused ones can be reused. */
    260 	ino_t			tm_nodes_last;
    261 
    262 	/* Nodes are organized in two different lists.  The used list
    263 	 * contains all nodes that are currently used by the file system;
    264 	 * i.e., they refer to existing files.  The available list contains
    265 	 * all nodes that are currently available for use by new files.
    266 	 * Nodes must be kept in this list (instead of deleting them)
    267 	 * because we need to keep track of their generation number (tn_gen
    268 	 * field).
    269 	 *
    270 	 * Note that nodes are lazily allocated: if the available list is
    271 	 * empty and we have enough space to create more nodes, they will be
    272 	 * created and inserted in the used list.  Once these are released,
    273 	 * they will go into the available list, remaining alive until the
    274 	 * file system is unmounted. */
    275 	struct tmpfs_node_list	tm_nodes_used;
    276 	struct tmpfs_node_list	tm_nodes_avail;
    277 
    278 	/* Pools used to store file system meta data.  These are not shared
    279 	 * across several instances of tmpfs for the reasons described in
    280 	 * tmpfs_pool.c. */
    281 	struct tmpfs_pool	tm_dirent_pool;
    282 	struct tmpfs_pool	tm_node_pool;
    283 	struct tmpfs_str_pool	tm_str_pool;
    284 };
    285 
    286 /* --------------------------------------------------------------------- */
    287 
    288 /*
    289  * This structure maps a file identifier to a tmpfs node.  Used by the
    290  * NFS code.
    291  */
    292 struct tmpfs_fid {
    293 	uint16_t		tf_len;
    294 	uint16_t		tf_pad;
    295 	ino_t			tf_id;
    296 	unsigned long		tf_gen;
    297 };
    298 
    299 /* --------------------------------------------------------------------- */
    300 
    301 /*
    302  * Prototypes for tmpfs_subr.c.
    303  */
    304 
    305 int	tmpfs_alloc_node(struct tmpfs_mount *, enum vtype,
    306 	    uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *,
    307 	    char *, dev_t, struct proc *, struct tmpfs_node **);
    308 void	tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
    309 int	tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
    310 	    const char *, uint16_t, struct tmpfs_dirent **);
    311 void	tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *,
    312 	    boolean_t);
    313 int	tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, struct vnode **);
    314 void	tmpfs_free_vp(struct vnode *);
    315 int	tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
    316 	    struct componentname *, char *);
    317 void	tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *);
    318 void	tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *);
    319 struct tmpfs_dirent *	tmpfs_dir_lookup(struct tmpfs_node *node,
    320 			    struct componentname *cnp);
    321 int	tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *);
    322 int	tmpfs_dir_getdotdotdent(struct tmpfs_node *, struct uio *);
    323 struct tmpfs_dirent *	tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t);
    324 int	tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *);
    325 int	tmpfs_reg_resize(struct vnode *, off_t);
    326 size_t	tmpfs_mem_info(boolean_t);
    327 int	tmpfs_chflags(struct vnode *, int, struct ucred *, struct proc *);
    328 int	tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct proc *);
    329 int	tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *,
    330 	    struct proc *);
    331 int	tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct proc *);
    332 int	tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *,
    333 	    int, struct ucred *, struct proc *);
    334 void	tmpfs_itimes(struct vnode *, const struct timespec *,
    335 	    const struct timespec *);
    336 
    337 int	tmpfs_update(struct vnode *, const struct timespec *,
    338 	    const struct timespec *, int);
    339 int	tmpfs_truncate(struct vnode *, off_t);
    340 
    341 /* --------------------------------------------------------------------- */
    342 
    343 /*
    344  * Convenience macros to simplify some logical expressions.
    345  */
    346 #define IMPLIES(a, b) (!(a) || (b))
    347 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a))
    348 
    349 /* --------------------------------------------------------------------- */
    350 
    351 /*
    352  * Checks that the directory entry pointed by 'de' matches the name 'name'
    353  * with a length of 'len'.
    354  */
    355 #define TMPFS_DIRENT_MATCHES(de, name, len) \
    356     (de->td_namelen == (uint16_t)len && \
    357     memcmp((de)->td_name, (name), (de)->td_namelen) == 0)
    358 
    359 /* --------------------------------------------------------------------- */
    360 
    361 /*
    362  * Ensures that the node pointed by 'node' is a directory and that its
    363  * contents are consistent with respect to directories.
    364  */
    365 #define TMPFS_VALIDATE_DIR(node) \
    366     KASSERT((node)->tn_type == VDIR); \
    367     KASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
    368     KASSERT((node)->tn_readdir_lastp == NULL || \
    369 	TMPFS_DIRCOOKIE((node)->tn_readdir_lastp) == (node)->tn_readdir_lastn);
    370 
    371 /* --------------------------------------------------------------------- */
    372 
    373 /*
    374  * Memory management stuff.
    375  */
    376 
    377 /* Amount of memory pages to reserve for the system (e.g., to not use by
    378  * tmpfs).
    379  * XXX: Should this be tunable through sysctl, for instance? */
    380 #define TMPFS_PAGES_RESERVED (4 * 1024 * 1024 / PAGE_SIZE)
    381 
    382 /* Returns the maximum size allowed for a tmpfs file system.  This macro
    383  * must be used instead of directly retrieving the value from tm_pages_max.
    384  * The reason is that the size of a tmpfs file system is dynamic: it lets
    385  * the user store files as long as there is enough free memory (including
    386  * physical memory and swap space).  Therefore, the amount of memory to be
    387  * used is either the limit imposed by the user during mount time or the
    388  * amount of available memory, whichever is lower.  To avoid consuming all
    389  * the memory for a given mount point, the system will always reserve a
    390  * minimum of TMPFS_PAGES_RESERVED pages, which is also taken into account
    391  * by this macro (see above). */
    392 static inline size_t
    393 TMPFS_PAGES_MAX(struct tmpfs_mount *tmp)
    394 {
    395 	size_t freepages;
    396 
    397 	freepages = tmpfs_mem_info(FALSE);
    398 	if (freepages < TMPFS_PAGES_RESERVED)
    399 		freepages = 0;
    400 	else
    401 		freepages -= TMPFS_PAGES_RESERVED;
    402 
    403 	return MIN(tmp->tm_pages_max, freepages + tmp->tm_pages_used);
    404 }
    405 
    406 /* Returns the available space for the given file system. */
    407 #define TMPFS_PAGES_AVAIL(tmp) (TMPFS_PAGES_MAX(tmp) - (tmp)->tm_pages_used)
    408 
    409 /* --------------------------------------------------------------------- */
    410 
    411 /*
    412  * Macros/functions to convert from generic data structures to tmpfs
    413  * specific ones.
    414  *
    415  * Macros are used when no sanity checks have to be done, as they provide
    416  * the fastest conversion.  On the other hand, inlined functions are used
    417  * when expensive sanity checks are enabled, mostly because the checks
    418  * have to be done separately from the return value.
    419  */
    420 
    421 #if defined(DIAGNOSTIC)
    422 static inline
    423 struct tmpfs_mount *
    424 VFS_TO_TMPFS(struct mount *mp)
    425 {
    426 	struct tmpfs_mount *tmp;
    427 
    428 	KASSERT((mp) != NULL && (mp)->mnt_data != NULL);
    429 	tmp = (struct tmpfs_mount *)(mp)->mnt_data;
    430 	KASSERT(TMPFS_PAGES_MAX(tmp) >= tmp->tm_pages_used);
    431 	return tmp;
    432 }
    433 
    434 static inline
    435 struct tmpfs_node *
    436 VP_TO_TMPFS_NODE(struct vnode *vp)
    437 {
    438 	struct tmpfs_node *node;
    439 
    440 	KASSERT((vp) != NULL && (vp)->v_data != NULL);
    441 	node = (struct tmpfs_node *)vp->v_data;
    442 	return node;
    443 }
    444 
    445 static inline
    446 struct tmpfs_node *
    447 VP_TO_TMPFS_DIR(struct vnode *vp)
    448 {
    449 	struct tmpfs_node *node;
    450 
    451 	node = VP_TO_TMPFS_NODE(vp);
    452 	TMPFS_VALIDATE_DIR(node);
    453 	return node;
    454 }
    455 #else
    456 #	define VFS_TO_TMPFS(mp) ((struct tmpfs_mount *)mp->mnt_data)
    457 #	define VP_TO_TMPFS_NODE(vp) ((struct tmpfs_node *)vp->v_data)
    458 #	define VP_TO_TMPFS_DIR(vp) VP_TO_TMPFS_NODE(vp)
    459 #endif
    460 
    461 #endif /* _KERNEL */
    462 
    463 /* ---------------------------------------------------------------------
    464  * USER AND KERNEL DEFINITIONS
    465  * --------------------------------------------------------------------- */
    466 
    467 /*
    468  * This structure is used to communicate mount parameters between userland
    469  * and kernel space.
    470  */
    471 #define TMPFS_ARGS_VERSION	1
    472 struct tmpfs_args {
    473 	int			ta_version;
    474 
    475 	/* Size counters. */
    476 	ino_t			ta_nodes_max;
    477 	off_t			ta_size_max;
    478 
    479 	/* Root node attributes. */
    480 	uid_t			ta_root_uid;
    481 	gid_t			ta_root_gid;
    482 	mode_t			ta_root_mode;
    483 };
    484