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
50 /* --------------------------------------------------------------------- */
51 #include <sys/malloc.h>
52 #ifdef _KERNEL
53 #include <sys/systm.h>
54 #endif
55 #include <sys/vmmeter.h>
56 #include <vm/swap_pager.h>
57
58 #ifdef MALLOC_DECLARE
59 MALLOC_DECLARE(M_TMPFSMNT);
60 #endif
61
62 /* --------------------------------------------------------------------- */
63
64 /*
65 * Internal representation of a tmpfs directory entry.
66 */
67 struct tmpfs_dirent {
68 RB_ENTRY(tmpfs_dirent) rb_node;
69 RB_ENTRY(tmpfs_dirent) rb_cookienode;
70
71 /* Length of the name stored in this directory entry. This avoids
72 * the need to recalculate it every time the name is used. */
73 uint16_t td_namelen;
74
75 /* The name of the entry, allocated from a string pool. This
76 * string is not required to be zero-terminated; therefore, the
77 * td_namelen field must always be used when accessing its value. */
78 char *td_name;
79
80 /* Pointer to the node this entry refers to. */
81 struct tmpfs_node *td_node;
82 };
83
84 struct tmpfs_dirtree;
85 RB_HEAD(tmpfs_dirtree, tmpfs_dirent);
86 RB_PROTOTYPE(tmpfs_dirtree, tmpfs_dirent, rb_node,
87 tmpfs_dirtree_compare);
88
89 RB_HEAD(tmpfs_dirtree_cookie, tmpfs_dirent);
90 RB_PROTOTYPE(tmpfs_dirtree_cookie, tmpfs_dirent, rb_cookienode,
91 tmpfs_dirtree_cookie_compare);
92
93
94 /*
95 * A directory in tmpfs holds a set of directory entries, which in
96 * turn point to other files (which can be directories themselves).
97 *
98 * In tmpfs, this set is managed by a red-black tree, whose root is defined
99 * by the struct tmpfs_dirtree type.
100 *
101 * It is important to notice that directories do not have entries for . and
102 * .. as other file systems do. These can be generated when requested
103 * based on information available by other means, such as the pointer to
104 * the node itself in the former case or the pointer to the parent directory
105 * in the latter case. This is done to simplify tmpfs's code and, more
106 * importantly, to remove redundancy.
107 *
108 * Each entry in a directory has a cookie that identifies it. Cookies
109 * supersede offsets within directories because, given how tmpfs stores
110 * directories in memory, there is no such thing as an offset. (Emulating
111 * a real offset could be very difficult.)
112 *
113 * The '.', '..' and the end of directory markers have fixed cookies which
114 * cannot collide with the cookies generated by other entries. The cookies
115 * for the other entries are generated based on the memory address on which
116 * stores their information is stored.
117 *
118 * DragonFly binaries use 64-bit cookies. We mask-off the signed bit to
119 * ensure that cookie 'offsets' are positive.
120 */
121 #ifdef _KERNEL
122
123 #define TMPFS_ROOTINO ((ino_t)2)
124
125 #define TMPFS_DIRCOOKIE_DOT 0
126 #define TMPFS_DIRCOOKIE_DOTDOT 1
127 #define TMPFS_DIRCOOKIE_EOF ((off_t)0x7FFFFFFFFFFFFFFFLLU)
128
129 static __inline
130 off_t
tmpfs_dircookie(struct tmpfs_dirent * de)131 tmpfs_dircookie(struct tmpfs_dirent *de)
132 {
133 return ((off_t)((uintptr_t)de >> 1) & 0x7FFFFFFFFFFFFFFFLLU);
134 }
135
136 /*
137 * WARNING! Caller should never try to actually access a tmpfs dirent
138 * structure from a cookiedir() conversion. It is used strictly
139 * for RBTREE operations.
140 */
141 static __inline
142 void *
tmpfs_cookiedir(off_t off)143 tmpfs_cookiedir(off_t off)
144 {
145 return ((void *)((uintptr_t)off << 1));
146 }
147
148 #endif /* _KERNEL */
149
150 /* --------------------------------------------------------------------- */
151
152 /*
153 * Internal representation of a tmpfs file system node.
154 *
155 * This structure is splitted in two parts: one holds attributes common
156 * to all file types and the other holds data that is only applicable to
157 * a particular type. The code must be careful to only access those
158 * attributes that are actually allowed by the node's type.
159 */
160 struct tmpfs_node {
161 /* Doubly-linked list entry which links all existing nodes for a
162 * single file system. This is provided to ease the removal of
163 * all nodes during the unmount operation. */
164 LIST_ENTRY(tmpfs_node) tn_entries;
165
166 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
167 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode
168 * types instead of a custom enumeration is to make things simpler
169 * and faster, as we do not need to convert between two types. */
170 enum vtype tn_type;
171
172 /* Node identifier. */
173 ino_t tn_id;
174
175 /* Node's internal status. This is used by several file system
176 * operations to do modifications to the node in a delayed
177 * fashion. */
178 int tn_blksize; /* small file optimization */
179 int tn_status;
180 #define TMPFS_NODE_ACCESSED (1 << 1)
181 #define TMPFS_NODE_MODIFIED (1 << 2)
182 #define TMPFS_NODE_CHANGED (1 << 3)
183
184 /* The node size. It does not necessarily match the real amount
185 * of memory consumed by it. */
186 off_t tn_size;
187
188 /* Generic node attributes. */
189 uid_t tn_uid;
190 gid_t tn_gid;
191 mode_t tn_mode;
192 u_int tn_flags;
193 nlink_t tn_links; /* atomic ops req */
194 long tn_atime;
195 long tn_atimensec;
196 long tn_mtime;
197 long tn_mtimensec;
198 long tn_ctime;
199 long tn_ctimensec;
200 unsigned long tn_gen;
201 struct lockf tn_advlock;
202
203 /* As there is a single vnode for each active file within the
204 * system, care has to be taken to avoid allocating more than one
205 * vnode per file. In order to do this, a bidirectional association
206 * is kept between vnodes and nodes.
207 *
208 * Whenever a vnode is allocated, its v_data field is updated to
209 * point to the node it references. At the same time, the node's
210 * tn_vnode field is modified to point to the new vnode representing
211 * it. Further attempts to allocate a vnode for this same node will
212 * result in returning a new reference to the value stored in
213 * tn_vnode.
214 *
215 * May be NULL when the node is unused (that is, no vnode has been
216 * allocated for it or it has been reclaimed). */
217 struct vnode * tn_vnode;
218
219 /* interlock to protect structure */
220 struct lock tn_interlock;
221
222 /*
223 * tmpfs vnode state, may specify an allocation in-progress.
224 */
225 int tn_vpstate;
226
227 /* misc data field for different tn_type node */
228 union {
229 /* Valid when tn_type == VBLK || tn_type == VCHR. */
230 dev_t tn_rdev; /*int32_t ?*/
231
232 /* Valid when tn_type == VDIR. */
233 struct tn_dir {
234 /*
235 * Pointer to the parent directory. The root
236 * directory has a pointer to itself in this field;
237 * this property identifies the root node.
238 */
239 struct tmpfs_node * tn_parent;
240
241 /*
242 * Directory entries are indexed by name and also
243 * indexed by cookie.
244 */
245 struct tmpfs_dirtree tn_dirtree;
246 struct tmpfs_dirtree_cookie tn_cookietree;
247 } tn_dir;
248
249 /* Valid when tn_type == VLNK. */
250 /* The link's target, allocated from a string pool. */
251 char * tn_link;
252
253 /*
254 * Valid when tn_type == VREG.
255 *
256 * aobj is used as backing store for the vnode object. It
257 * typically only contains swap assignments, but we also use
258 * it to save the vnode object's vm_page's when the vnode
259 * becomes inactive.
260 */
261 struct tn_reg {
262 vm_object_t tn_aobj;
263 size_t tn_aobj_pages;
264 int tn_pages_in_aobj;
265 } tn_reg;
266
267 /* Valid when tn_type = VFIFO */
268 struct tn_fifo {
269 int (*tn_fo_read) (struct file *fp, struct uio *uio,
270 struct ucred *cred, int flags);
271 int (*tn_fo_write) (struct file *fp, struct uio *uio,
272 struct ucred *cred, int flags);
273 } tn_fifo;
274 } tn_spec;
275 };
276
277 /* Only userspace needs this */
278 #define VTOI(vp) ((struct tmpfs_node *)(vp)->v_data)
279
280 #ifdef _KERNEL
281 LIST_HEAD(tmpfs_node_list, tmpfs_node);
282
283 #define tn_rdev tn_spec.tn_rdev
284 #define tn_dir tn_spec.tn_dir
285 #define tn_link tn_spec.tn_link
286 #define tn_reg tn_spec.tn_reg
287 #define tn_fifo tn_spec.tn_fifo
288
289 #define TMPFS_NODE_LOCK(node) lockmgr(&(node)->tn_interlock, LK_EXCLUSIVE|LK_RETRY)
290 #define TMPFS_NODE_LOCK_SH(node) lockmgr(&(node)->tn_interlock, LK_SHARED|LK_RETRY)
291 #define TMPFS_NODE_UNLOCK(node) lockmgr(&(node)->tn_interlock, LK_RELEASE)
292 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock)
293
294 #ifdef INVARIANTS
295 #define TMPFS_ASSERT_LOCKED(node) do { \
296 KKASSERT(node != NULL); \
297 KKASSERT(node->tn_vnode != NULL); \
298 if (!vn_islocked(node->tn_vnode) && \
299 (lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE )) \
300 panic("tmpfs: node is not locked: %p", node); \
301 } while (0)
302 #define TMPFS_ASSERT_ELOCKED(node) do { \
303 KKASSERT((node) != NULL); \
304 KKASSERT(lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE); \
305 } while (0)
306 #else
307 #define TMPFS_ASSERT_LOCKED(node) (void)0
308 #define TMPFS_ASSERT_ELOCKED(node) (void)0
309 #endif /* INVARIANTS */
310
311 #define TMPFS_VNODE_DOOMED 0x0001
312 /* --------------------------------------------------------------------- */
313
314 /*
315 * Internal representation of a tmpfs mount point.
316 */
317 struct tmpfs_mount {
318 struct mount *tm_mount;
319
320 /* Maximum number of memory pages available for use by the file
321 * system, set during mount time. This variable must never be
322 * used directly as it may be bigger than the current amount of
323 * free memory; in the extreme case, it will hold the SIZE_MAX
324 * value. Instead, use the TMPFS_PAGES_MAX macro. */
325 long tm_pages_max;
326
327 /* Number of pages in use by the file system. Cannot be bigger
328 * than the value returned by TMPFS_PAGES_MAX in any case. */
329 long tm_pages_used;
330
331 /* Pointer to the node representing the root directory of this
332 * file system. */
333 struct tmpfs_node * tm_root;
334
335 /* Maximum number of possible nodes for this file system; set
336 * during mount time. We need a hard limit on the maximum number
337 * of nodes to avoid allocating too much of them; their objects
338 * cannot be released until the file system is unmounted.
339 * Otherwise, we could easily run out of memory by creating lots
340 * of empty files and then simply removing them. */
341 ino_t tm_nodes_max;
342
343 /* Number of nodes currently that are in use. */
344 ino_t tm_nodes_inuse;
345
346 /* maximum representable file size */
347 u_int64_t tm_maxfilesize;
348
349 /* Nodes are organized in two different lists. The used list
350 * contains all nodes that are currently used by the file system;
351 * i.e., they refer to existing files. The available list contains
352 * all nodes that are currently available for use by new files.
353 * Nodes must be kept in this list (instead of deleting them)
354 * because we need to keep track of their generation number (tn_gen
355 * field).
356 *
357 * Note that nodes are lazily allocated: if the available list is
358 * empty and we have enough space to create more nodes, they will be
359 * created and inserted in the used list. Once these are released,
360 * they will go into the available list, remaining alive until the
361 * file system is unmounted. */
362 struct tmpfs_node_list tm_nodes_used;
363
364 /* Per-mount malloc zones for tmpfs nodes, names, and dirents */
365 struct malloc_type *tm_node_zone_obj;
366 struct malloc_type *tm_dirent_zone_obj;
367 struct malloc_type *tm_name_zone;
368
369 ino_t tm_ino;
370 int tm_flags;
371
372 struct netexport tm_export;
373
374 struct mount *tm_mnt;
375 };
376
377 #define TMPFS_LOCK(tm) lwkt_gettoken(&(tm)->tm_mount->mnt_token)
378 #define TMPFS_UNLOCK(tm) lwkt_reltoken(&(tm)->tm_mount->mnt_token)
379
380 /* --------------------------------------------------------------------- */
381
382 /*
383 * This structure maps a file identifier to a tmpfs node. Used by the
384 * NFS code.
385 */
386 struct tmpfs_fid {
387 uint16_t tf_len;
388 uint16_t tf_pad;
389 ino_t tf_id;
390 unsigned long tf_gen;
391 } __packed;
392
393 /* --------------------------------------------------------------------- */
394
395 /*
396 * Prototypes for tmpfs_subr.c.
397 */
398
399 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype,
400 uid_t uid, gid_t gid, mode_t mode, char *, int, int,
401 struct tmpfs_node **);
402 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
403 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
404 const char *, uint16_t, struct tmpfs_dirent **);
405 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *);
406 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *,
407 struct tmpfs_node *, int, struct vnode **);
408 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
409 struct namecache *, struct ucred *, char *);
410 void tmpfs_dir_attach_locked(struct tmpfs_node *, struct tmpfs_dirent *);
411 void tmpfs_dir_detach_locked(struct tmpfs_node *, struct tmpfs_dirent *);
412 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node,
413 struct tmpfs_node *f,
414 struct namecache *ncp);
415 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *);
416 int tmpfs_dir_getdotdotdent(struct tmpfs_mount *,
417 struct tmpfs_node *, struct uio *);
418 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t,
419 int);
420 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *);
421 int tmpfs_reg_resize(struct vnode *, off_t, int);
422 int tmpfs_chflags(struct vnode *, u_long, struct ucred *);
423 int tmpfs_chmod(struct vnode *, mode_t, struct ucred *);
424 int tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *);
425 int tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *);
426 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *,
427 int, struct ucred *);
428 void tmpfs_itimes(struct vnode *, const struct timespec *,
429 const struct timespec *);
430
431 void tmpfs_node_init(struct tmpfs_node *node);
432 void tmpfs_node_uninit(struct tmpfs_node *node);
433 void tmpfs_update(struct vnode *);
434 int tmpfs_truncate(struct vnode *, off_t);
435 void tmpfs_lock4(struct tmpfs_node *node1, struct tmpfs_node *node2,
436 struct tmpfs_node *node3, struct tmpfs_node *node4);
437 void tmpfs_unlock4(struct tmpfs_node *node1, struct tmpfs_node *node2,
438 struct tmpfs_node *node3, struct tmpfs_node *node4);
439
440 /* --------------------------------------------------------------------- */
441
442 /*
443 * Convenience macros to simplify some logical expressions.
444 */
445 #define IMPLIES(a, b) (!(a) || (b))
446 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a))
447
448 /* --------------------------------------------------------------------- */
449
450 /*
451 * Checks that the directory entry pointed by 'de' matches the name 'name'
452 * with a length of 'len'.
453 */
454 #define TMPFS_DIRENT_MATCHES(de, name, len) \
455 (de->td_namelen == (uint16_t)len && \
456 bcmp((de)->td_name, (name), (de)->td_namelen) == 0)
457
458 /* --------------------------------------------------------------------- */
459
460 /*
461 * Ensures that the node pointed by 'node' is a directory and that its
462 * contents are consistent with respect to directories.
463 */
464 #define TMPFS_VALIDATE_DIR(node) do { \
465 KKASSERT((node)->tn_type == VDIR); \
466 KKASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
467 } while(0)
468
469 /* --------------------------------------------------------------------- */
470
471 /*
472 * Macros/functions to convert from generic data structures to tmpfs
473 * specific ones. Kernel code use VP_TO_TMPFS_NODE() instead of VTOI().
474 */
475
476 static inline
477 struct tmpfs_mount *
VFS_TO_TMPFS(struct mount * mp)478 VFS_TO_TMPFS(struct mount *mp)
479 {
480 struct tmpfs_mount *tmp;
481
482 KKASSERT((mp) != NULL && (mp)->mnt_data != NULL);
483 tmp = (struct tmpfs_mount *)(mp)->mnt_data;
484 return tmp;
485 }
486
487 static inline
488 struct tmpfs_node *
VP_TO_TMPFS_NODE(struct vnode * vp)489 VP_TO_TMPFS_NODE(struct vnode *vp)
490 {
491 struct tmpfs_node *node;
492
493 KKASSERT((vp) != NULL && (vp)->v_data != NULL);
494 node = (struct tmpfs_node *)vp->v_data;
495 return node;
496 }
497
498 static inline
499 struct tmpfs_node *
VP_TO_TMPFS_DIR(struct vnode * vp)500 VP_TO_TMPFS_DIR(struct vnode *vp)
501 {
502 struct tmpfs_node *node;
503
504 node = VP_TO_TMPFS_NODE(vp);
505 TMPFS_VALIDATE_DIR(node);
506 return node;
507 }
508
509 /* --------------------------------------------------------------------- */
510 /*
511 * buffer cache size
512 */
513 #define TMPFS_BLKSIZE 16384 /* buffer cache size*/
514 #define TMPFS_BLKMASK (TMPFS_BLKSIZE - 1)
515 #define TMPFS_BLKMASK64 ((off_t)(TMPFS_BLKSIZE - 1))
516 #endif /* _KERNEL */
517
518 #endif /* _VFS_TMPFS_TMPFS_H_ */
519