1 /* $NetBSD: tmpfs.h,v 1.26 2007/02/22 06:37:00 thorpej Exp $ */
4 * Copyright (c) 2005, 2006 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);
63 MALLOC_DECLARE(M_TMPFSNAME);
65 /* --------------------------------------------------------------------- */
68 * Internal representation of a tmpfs directory entry.
71 TAILQ_ENTRY(tmpfs_dirent) td_entries;
73 /* Length of the name stored in this directory entry. This avoids
74 * the need to recalculate it every time the name is used. */
77 /* The name of the entry, allocated from a string pool. This
78 * string is not required to be zero-terminated; therefore, the
79 * td_namelen field must always be used when accessing its value. */
82 /* Pointer to the node this entry refers to. */
83 struct tmpfs_node * td_node;
86 /* A directory in tmpfs holds a sorted list of directory entries, which in
87 * turn point to other files (which can be directories themselves).
89 * In tmpfs, this list is managed by a tail queue, whose head is defined by
90 * the struct tmpfs_dir type.
92 * It is imporant to notice that directories do not have entries for . and
93 * .. as other file systems do. These can be generated when requested
94 * based on information available by other means, such as the pointer to
95 * the node itself in the former case or the pointer to the parent directory
96 * in the latter case. This is done to simplify tmpfs's code and, more
97 * importantly, to remove redundancy. */
98 TAILQ_HEAD(tmpfs_dir, tmpfs_dirent);
100 /* Each entry in a directory has a cookie that identifies it. Cookies
101 * supersede offsets within directories because, given how tmpfs stores
102 * directories in memory, there is no such thing as an offset. (Emulating
103 * a real offset could be very difficult.)
105 * The '.', '..' and the end of directory markers have fixed cookies which
106 * cannot collide with the cookies generated by other entries. The cookies
107 * fot the other entries are generated based on the memory address on which
108 * stores their information is stored.
110 * Ideally, using the entry's memory pointer as the cookie would be enough
111 * to represent it and it wouldn't cause collisions in any system.
112 * Unfortunately, this results in "offsets" with very large values which
113 * later raise problems in the Linux compatibility layer (and maybe in other
114 * places) as described in PR kern/32034. Hence we need to workaround this
115 * with a rather ugly hack.
117 * Linux 32-bit binaries, unless built with _FILE_OFFSET_BITS=64, have off_t
118 * set to 'long', which is a 32-bit *signed* long integer. Regardless of
119 * the macro value, GLIBC (2.3 at least) always uses the getdents64
120 * system call (when calling readdir) which internally returns off64_t
121 * offsets. In order to make 32-bit binaries work, *GLIBC* converts the
122 * 64-bit values returned by the kernel to 32-bit ones and aborts with
123 * EOVERFLOW if the conversion results in values that won't fit in 32-bit
124 * integers (which it assumes is because the directory is extremely large).
125 * This wouldn't cause problems if we were dealing with unsigned integers,
126 * but as we have signed integers, this check fails due to sign expansion.
128 * For example, consider that the kernel returns the 0xc1234567 cookie to
129 * userspace in a off64_t integer. Later on, GLIBC casts this value to
130 * off_t (remember, signed) with code similar to:
131 * system call returns the offset in kernel_value;
132 * off_t casted_value = kernel_value;
133 * if (sizeof(off_t) != sizeof(off64_t) &&
134 * kernel_value != casted_value)
136 * In this case, casted_value still has 0xc1234567, but when it is compared
137 * for equality against kernel_value, it is promoted to a 64-bit integer and
138 * becomes 0xffffffffc1234567, which is different than 0x00000000c1234567.
139 * Then, GLIBC assumes this is because the directory is very large.
141 * Given that all the above happens in user-space, we have no control over
142 * it; therefore we must workaround the issue here. We do this by
143 * truncating the pointer value to a 32-bit integer and hope that there
144 * won't be collisions. In fact, this will not cause any problems in
145 * 32-bit platforms but some might arise in 64-bit machines (I'm not sure
146 * if they can happen at all in practice).
148 * XXX A nicer solution shall be attempted. */
150 #define TMPFS_DIRCOOKIE_DOT 0
151 #define TMPFS_DIRCOOKIE_DOTDOT 1
152 #define TMPFS_DIRCOOKIE_EOF 2
155 tmpfs_dircookie(struct tmpfs_dirent *de)
159 cookie = ((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFF;
160 MPASS(cookie != TMPFS_DIRCOOKIE_DOT);
161 MPASS(cookie != TMPFS_DIRCOOKIE_DOTDOT);
162 MPASS(cookie != TMPFS_DIRCOOKIE_EOF);
168 /* --------------------------------------------------------------------- */
171 * Internal representation of a tmpfs file system node.
173 * This structure is splitted in two parts: one holds attributes common
174 * to all file types and the other holds data that is only applicable to
175 * a particular type. The code must be careful to only access those
176 * attributes that are actually allowed by the node's type.
179 * Below is the key of locks used to protected the fields in the following
184 /* Doubly-linked list entry which links all existing nodes for a
185 * single file system. This is provided to ease the removal of
186 * all nodes during the unmount operation. */
187 LIST_ENTRY(tmpfs_node) tn_entries;
189 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
190 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode
191 * types instead of a custom enumeration is to make things simpler
192 * and faster, as we do not need to convert between two types. */
195 /* Node identifier. */
198 /* Node's internal status. This is used by several file system
199 * operations to do modifications to the node in a delayed
202 #define TMPFS_NODE_ACCESSED (1 << 1)
203 #define TMPFS_NODE_MODIFIED (1 << 2)
204 #define TMPFS_NODE_CHANGED (1 << 3)
206 /* The node size. It does not necessarily match the real amount
207 * of memory consumed by it. */
210 /* Generic node attributes. */
216 struct timespec tn_atime;
217 struct timespec tn_mtime;
218 struct timespec tn_ctime;
219 struct timespec tn_birthtime;
220 unsigned long tn_gen;
222 /* As there is a single vnode for each active file within the
223 * system, care has to be taken to avoid allocating more than one
224 * vnode per file. In order to do this, a bidirectional association
225 * is kept between vnodes and nodes.
227 * Whenever a vnode is allocated, its v_data field is updated to
228 * point to the node it references. At the same time, the node's
229 * tn_vnode field is modified to point to the new vnode representing
230 * it. Further attempts to allocate a vnode for this same node will
231 * result in returning a new reference to the value stored in
234 * May be NULL when the node is unused (that is, no vnode has been
235 * allocated for it or it has been reclaimed). */
236 struct vnode * tn_vnode;
238 /* interlock to protect tn_vpstate */
239 struct mtx tn_interlock;
241 /* Identify if current node has vnode assiocate with
242 * or allocating vnode.
246 /* misc data field for different tn_type node */
248 /* Valid when tn_type == VBLK || tn_type == VCHR. */
251 /* Valid when tn_type == VDIR. */
253 /* Pointer to the parent directory. The root
254 * directory has a pointer to itself in this field;
255 * this property identifies the root node. */
256 struct tmpfs_node * tn_parent;
258 /* Head of a tail-queue that links the contents of
259 * the directory together. See above for a
260 * description of its contents. */
261 struct tmpfs_dir tn_dirhead;
263 /* Number and pointer of the first directory entry
264 * returned by the readdir operation if it were
265 * called again to continue reading data from the
266 * same directory as before. This is used to speed
267 * up reads of long directories, assuming that no
268 * more than one read is in progress at a given time.
269 * Otherwise, these values are discarded and a linear
270 * scan is performed from the beginning up to the
271 * point where readdir starts returning values. */
272 off_t tn_readdir_lastn;
273 struct tmpfs_dirent * tn_readdir_lastp;
276 /* Valid when tn_type == VLNK. */
277 /* The link's target, allocated from a string pool. */
280 /* Valid when tn_type == VREG. */
282 /* The contents of regular files stored in a tmpfs
283 * file system are represented by a single anonymous
284 * memory object (aobj, for short). The aobj provides
285 * direct access to any position within the file,
286 * because its contents are always mapped in a
287 * contiguous region of virtual memory. It is a task
288 * of the memory management subsystem (see uvm(9)) to
289 * issue the required page ins or page outs whenever
290 * a position within the file is accessed. */
292 size_t tn_aobj_pages;
296 /* Valid when tn_type = VFIFO */
298 fo_rdwr_t *tn_fo_read;
299 fo_rdwr_t *tn_fo_write;
303 LIST_HEAD(tmpfs_node_list, tmpfs_node);
305 #define tn_rdev tn_spec.tn_rdev
306 #define tn_dir tn_spec.tn_dir
307 #define tn_link tn_spec.tn_link
308 #define tn_reg tn_spec.tn_reg
309 #define tn_fifo tn_spec.tn_fifo
311 #define TMPFS_NODE_LOCK(node) mtx_lock(&(node)->tn_interlock)
312 #define TMPFS_NODE_UNLOCK(node) mtx_unlock(&(node)->tn_interlock)
313 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock)
315 #define TMPFS_VNODE_ALLOCATING 1
316 #define TMPFS_VNODE_WANT 2
317 /* --------------------------------------------------------------------- */
320 * Internal representation of a tmpfs mount point.
323 /* Maximum number of memory pages available for use by the file
324 * system, set during mount time. This variable must never be
325 * used directly as it may be bigger than the current amount of
326 * free memory; in the extreme case, it will hold the SIZE_MAX
327 * value. Instead, use the TMPFS_PAGES_MAX macro. */
330 /* Number of pages in use by the file system. Cannot be bigger
331 * than the value returned by TMPFS_PAGES_MAX in any case. */
332 size_t tm_pages_used;
334 /* Pointer to the node representing the root directory of this
336 struct tmpfs_node * tm_root;
338 /* Maximum number of possible nodes for this file system; set
339 * during mount time. We need a hard limit on the maximum number
340 * of nodes to avoid allocating too much of them; their objects
341 * cannot be released until the file system is unmounted.
342 * Otherwise, we could easily run out of memory by creating lots
343 * of empty files and then simply removing them. */
346 /* unrhdr used to allocate inode numbers */
347 struct unrhdr * tm_ino_unr;
349 /* Number of nodes currently that are in use. */
350 ino_t tm_nodes_inuse;
352 /* maximum representable file size */
353 u_int64_t tm_maxfilesize;
355 /* Nodes are organized in two different lists. The used list
356 * contains all nodes that are currently used by the file system;
357 * i.e., they refer to existing files. The available list contains
358 * all nodes that are currently available for use by new files.
359 * Nodes must be kept in this list (instead of deleting them)
360 * because we need to keep track of their generation number (tn_gen
363 * Note that nodes are lazily allocated: if the available list is
364 * empty and we have enough space to create more nodes, they will be
365 * created and inserted in the used list. Once these are released,
366 * they will go into the available list, remaining alive until the
367 * file system is unmounted. */
368 struct tmpfs_node_list tm_nodes_used;
370 /* All node lock to protect the node list and tmp_pages_used */
371 struct mtx allnode_lock;
373 /* Pools used to store file system meta data. These are not shared
374 * across several instances of tmpfs for the reasons described in
376 uma_zone_t tm_dirent_pool;
377 uma_zone_t tm_node_pool;
379 #define TMPFS_LOCK(tm) mtx_lock(&(tm)->allnode_lock)
380 #define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->allnode_lock)
382 /* --------------------------------------------------------------------- */
385 * This structure maps a file identifier to a tmpfs node. Used by the
392 unsigned long tf_gen;
395 /* --------------------------------------------------------------------- */
399 * Prototypes for tmpfs_subr.c.
402 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype,
403 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *,
404 char *, dev_t, struct thread *, struct tmpfs_node **);
405 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
406 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
407 const char *, uint16_t, struct tmpfs_dirent **);
408 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *,
410 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int,
411 struct vnode **, struct thread *);
412 void tmpfs_free_vp(struct vnode *);
413 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
414 struct componentname *, char *);
415 void tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *);
416 void tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *);
417 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node,
418 struct componentname *cnp);
419 struct tmpfs_dirent *tmpfs_dir_search(struct tmpfs_node *node,
420 struct tmpfs_node *f);
421 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *);
422 int tmpfs_dir_getdotdotdent(struct tmpfs_node *, struct uio *);
423 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t);
424 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *);
425 int tmpfs_reg_resize(struct vnode *, off_t);
426 int tmpfs_chflags(struct vnode *, int, struct ucred *, struct thread *);
427 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *, struct thread *);
428 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *,
430 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *, struct thread *);
431 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *,
432 struct timespec *, int, struct ucred *, struct thread *);
433 void tmpfs_itimes(struct vnode *, const struct timespec *,
434 const struct timespec *);
436 void tmpfs_update(struct vnode *);
437 int tmpfs_truncate(struct vnode *, off_t);
439 /* --------------------------------------------------------------------- */
442 * Convenience macros to simplify some logical expressions.
444 #define IMPLIES(a, b) (!(a) || (b))
445 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a))
447 /* --------------------------------------------------------------------- */
450 * Checks that the directory entry pointed by 'de' matches the name 'name'
451 * with a length of 'len'.
453 #define TMPFS_DIRENT_MATCHES(de, name, len) \
454 (de->td_namelen == (uint16_t)len && \
455 memcmp((de)->td_name, (name), (de)->td_namelen) == 0)
457 /* --------------------------------------------------------------------- */
460 * Ensures that the node pointed by 'node' is a directory and that its
461 * contents are consistent with respect to directories.
463 #define TMPFS_VALIDATE_DIR(node) \
464 MPASS((node)->tn_type == VDIR); \
465 MPASS((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
466 MPASS((node)->tn_dir.tn_readdir_lastp == NULL || \
467 tmpfs_dircookie((node)->tn_dir.tn_readdir_lastp) == (node)->tn_dir.tn_readdir_lastn);
469 /* --------------------------------------------------------------------- */
472 * Memory management stuff.
475 /* Amount of memory pages to reserve for the system (e.g., to not use by
477 * XXX: Should this be tunable through sysctl, for instance? */
478 #define TMPFS_PAGES_RESERVED (4 * 1024 * 1024 / PAGE_SIZE)
481 * Returns information about the number of available memory pages,
482 * including physical and virtual ones.
484 * If 'total' is TRUE, the value returned is the total amount of memory
485 * pages configured for the system (either in use or free).
486 * If it is FALSE, the value returned is the amount of free memory pages.
488 * Remember to remove TMPFS_PAGES_RESERVED from the returned value to avoid
489 * excessive memory usage.
492 static __inline size_t
497 size = swap_pager_avail + cnt.v_free_count + cnt.v_inactive_count;
498 size -= size > cnt.v_wire_count ? cnt.v_wire_count : size;
502 /* Returns the maximum size allowed for a tmpfs file system. This macro
503 * must be used instead of directly retrieving the value from tm_pages_max.
504 * The reason is that the size of a tmpfs file system is dynamic: it lets
505 * the user store files as long as there is enough free memory (including
506 * physical memory and swap space). Therefore, the amount of memory to be
507 * used is either the limit imposed by the user during mount time or the
508 * amount of available memory, whichever is lower. To avoid consuming all
509 * the memory for a given mount point, the system will always reserve a
510 * minimum of TMPFS_PAGES_RESERVED pages, which is also taken into account
511 * by this macro (see above). */
512 static __inline size_t
513 TMPFS_PAGES_MAX(struct tmpfs_mount *tmp)
517 freepages = tmpfs_mem_info();
518 freepages -= freepages < TMPFS_PAGES_RESERVED ?
519 freepages : TMPFS_PAGES_RESERVED;
521 return MIN(tmp->tm_pages_max, freepages + tmp->tm_pages_used);
524 /* Returns the available space for the given file system. */
525 #define TMPFS_META_PAGES(tmp) (howmany((tmp)->tm_nodes_inuse * (sizeof(struct tmpfs_node) \
526 + sizeof(struct tmpfs_dirent)), PAGE_SIZE))
527 #define TMPFS_FILE_PAGES(tmp) ((tmp)->tm_pages_used)
529 #define TMPFS_PAGES_AVAIL(tmp) (TMPFS_PAGES_MAX(tmp) > \
530 TMPFS_META_PAGES(tmp)+TMPFS_FILE_PAGES(tmp)? \
531 TMPFS_PAGES_MAX(tmp) - TMPFS_META_PAGES(tmp) \
532 - TMPFS_FILE_PAGES(tmp):0)
536 /* --------------------------------------------------------------------- */
539 * Macros/functions to convert from generic data structures to tmpfs
545 VFS_TO_TMPFS(struct mount *mp)
547 struct tmpfs_mount *tmp;
549 MPASS((mp) != NULL && (mp)->mnt_data != NULL);
550 tmp = (struct tmpfs_mount *)(mp)->mnt_data;
556 VP_TO_TMPFS_NODE(struct vnode *vp)
558 struct tmpfs_node *node;
560 MPASS((vp) != NULL && (vp)->v_data != NULL);
561 node = (struct tmpfs_node *)vp->v_data;
567 VP_TO_TMPFS_DIR(struct vnode *vp)
569 struct tmpfs_node *node;
571 node = VP_TO_TMPFS_NODE(vp);
572 TMPFS_VALIDATE_DIR(node);
576 #endif /* _FS_TMPFS_TMPFS_H_ */