tmpfs.h revision 1.28.2.1 1 /* $NetBSD: tmpfs.h,v 1.28.2.1 2007/12/04 13:03:09 ad Exp $ */
2
3 /*
4 * Copyright (c) 2005, 2006, 2007 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 #ifndef _FS_TMPFS_TMPFS_H_
41 #define _FS_TMPFS_TMPFS_H_
42
43 /* ---------------------------------------------------------------------
44 * KERNEL-SPECIFIC DEFINITIONS
45 * --------------------------------------------------------------------- */
46 #include <sys/dirent.h>
47 #include <sys/mount.h>
48 #include <sys/queue.h>
49 #include <sys/vnode.h>
50
51 #if defined(_KERNEL)
52 #include <fs/tmpfs/tmpfs_pool.h>
53 #endif /* defined(_KERNEL) */
54
55 /* --------------------------------------------------------------------- */
56
57 /*
58 * Internal representation of a tmpfs directory entry.
59 */
60 struct tmpfs_dirent {
61 TAILQ_ENTRY(tmpfs_dirent) td_entries;
62
63 /* Length of the name stored in this directory entry. This avoids
64 * the need to recalculate it every time the name is used. */
65 uint16_t td_namelen;
66
67 /* The name of the entry, allocated from a string pool. This
68 * string is not required to be zero-terminated; therefore, the
69 * td_namelen field must always be used when accessing its value. */
70 char * td_name;
71
72 /* Pointer to the node this entry refers to. */
73 struct tmpfs_node * td_node;
74 };
75
76 /* A directory in tmpfs holds a sorted list of directory entries, which in
77 * turn point to other files (which can be directories themselves).
78 *
79 * In tmpfs, this list is managed by a tail queue, whose head is defined by
80 * the struct tmpfs_dir type.
81 *
82 * It is imporant to notice that directories do not have entries for . and
83 * .. as other file systems do. These can be generated when requested
84 * based on information available by other means, such as the pointer to
85 * the node itself in the former case or the pointer to the parent directory
86 * in the latter case. This is done to simplify tmpfs's code and, more
87 * importantly, to remove redundancy. */
88 TAILQ_HEAD(tmpfs_dir, tmpfs_dirent);
89
90 /* Each entry in a directory has a cookie that identifies it. Cookies
91 * supersede offsets within directories because, given how tmpfs stores
92 * directories in memory, there is no such thing as an offset. (Emulating
93 * a real offset could be very difficult.)
94 *
95 * The '.', '..' and the end of directory markers have fixed cookies which
96 * cannot collide with the cookies generated by other entries. The cookies
97 * fot the other entries are generated based on the memory address on which
98 * stores their information is stored.
99 *
100 * Ideally, using the entry's memory pointer as the cookie would be enough
101 * to represent it and it wouldn't cause collisions in any system.
102 * Unfortunately, this results in "offsets" with very large values which
103 * later raise problems in the Linux compatibility layer (and maybe in other
104 * places) as described in PR kern/32034. Hence we need to workaround this
105 * with a rather ugly hack.
106 *
107 * Linux 32-bit binaries, unless built with _FILE_OFFSET_BITS=64, have off_t
108 * set to 'long', which is a 32-bit *signed* long integer. Regardless of
109 * the macro value, GLIBC (2.3 at least) always uses the getdents64
110 * system call (when calling readdir) which internally returns off64_t
111 * offsets. In order to make 32-bit binaries work, *GLIBC* converts the
112 * 64-bit values returned by the kernel to 32-bit ones and aborts with
113 * EOVERFLOW if the conversion results in values that won't fit in 32-bit
114 * integers (which it assumes is because the directory is extremely large).
115 * This wouldn't cause problems if we were dealing with unsigned integers,
116 * but as we have signed integers, this check fails due to sign expansion.
117 *
118 * For example, consider that the kernel returns the 0xc1234567 cookie to
119 * userspace in a off64_t integer. Later on, GLIBC casts this value to
120 * off_t (remember, signed) with code similar to:
121 * system call returns the offset in kernel_value;
122 * off_t casted_value = kernel_value;
123 * if (sizeof(off_t) != sizeof(off64_t) &&
124 * kernel_value != casted_value)
125 * error!
126 * In this case, casted_value still has 0xc1234567, but when it is compared
127 * for equality against kernel_value, it is promoted to a 64-bit integer and
128 * becomes 0xffffffffc1234567, which is different than 0x00000000c1234567.
129 * Then, GLIBC assumes this is because the directory is very large.
130 *
131 * Given that all the above happens in user-space, we have no control over
132 * it; therefore we must workaround the issue here. We do this by
133 * truncating the pointer value to a 32-bit integer and hope that there
134 * won't be collisions. In fact, this will not cause any problems in
135 * 32-bit platforms but some might arise in 64-bit machines (I'm not sure
136 * if they can happen at all in practice).
137 *
138 * XXX A nicer solution shall be attempted. */
139 #if defined(_KERNEL)
140 #define TMPFS_DIRCOOKIE_DOT 0
141 #define TMPFS_DIRCOOKIE_DOTDOT 1
142 #define TMPFS_DIRCOOKIE_EOF 2
143 static __inline
144 off_t
145 tmpfs_dircookie(struct tmpfs_dirent *de)
146 {
147 off_t cookie;
148
149 cookie = ((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFF;
150 KASSERT(cookie != TMPFS_DIRCOOKIE_DOT);
151 KASSERT(cookie != TMPFS_DIRCOOKIE_DOTDOT);
152 KASSERT(cookie != TMPFS_DIRCOOKIE_EOF);
153
154 return cookie;
155 }
156 #endif /* defined(_KERNEL) */
157
158 /* --------------------------------------------------------------------- */
159
160 /*
161 * Internal representation of a tmpfs file system node.
162 *
163 * This structure is splitted in two parts: one holds attributes common
164 * to all file types and the other holds data that is only applicable to
165 * a particular type. The code must be careful to only access those
166 * attributes that are actually allowed by the node's type.
167 */
168 struct tmpfs_node {
169 /* Doubly-linked list entry which links all existing nodes for a
170 * single file system. This is provided to ease the removal of
171 * all nodes during the unmount operation. */
172 LIST_ENTRY(tmpfs_node) tn_entries;
173
174 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
175 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode
176 * types instead of a custom enumeration is to make things simpler
177 * and faster, as we do not need to convert between two types. */
178 enum vtype tn_type;
179
180 /* Node identifier. */
181 ino_t tn_id;
182
183 /* Node's internal status. This is used by several file system
184 * operations to do modifications to the node in a delayed
185 * fashion. */
186 int tn_status;
187 #define TMPFS_NODE_ACCESSED (1 << 1)
188 #define TMPFS_NODE_MODIFIED (1 << 2)
189 #define TMPFS_NODE_CHANGED (1 << 3)
190
191 /* The node size. It does not necessarily match the real amount
192 * of memory consumed by it. */
193 off_t tn_size;
194
195 /* Generic node attributes. */
196 uid_t tn_uid;
197 gid_t tn_gid;
198 mode_t tn_mode;
199 int tn_flags;
200 nlink_t tn_links;
201 struct timespec tn_atime;
202 struct timespec tn_mtime;
203 struct timespec tn_ctime;
204 struct timespec tn_birthtime;
205 unsigned long tn_gen;
206
207 /* Head of byte-level lock list (used by tmpfs_advlock). */
208 struct lockf * tn_lockf;
209
210 /* As there is a single vnode for each active file within the
211 * system, care has to be taken to avoid allocating more than one
212 * vnode per file. In order to do this, a bidirectional association
213 * is kept between vnodes and nodes.
214 *
215 * Whenever a vnode is allocated, its v_data field is updated to
216 * point to the node it references. At the same time, the node's
217 * tn_vnode field is modified to point to the new vnode representing
218 * it. Further attempts to allocate a vnode for this same node will
219 * result in returning a new reference to the value stored in
220 * tn_vnode.
221 *
222 * May be NULL when the node is unused (that is, no vnode has been
223 * allocated for it or it has been reclaimed). */
224 struct vnode * tn_vnode;
225
226 /* Lock on tn_vnode. */
227 kmutex_t tn_vlock;
228
229 /* Pointer to the node returned by tmpfs_lookup() after doing a
230 * delete or a rename lookup; its value is only valid in these two
231 * situations. In case we were looking up . or .., it holds a null
232 * pointer. */
233 struct tmpfs_dirent * tn_lookup_dirent;
234
235 union {
236 /* Valid when tn_type == VBLK || tn_type == VCHR. */
237 struct {
238 dev_t tn_rdev;
239 } tn_dev;
240
241 /* Valid when tn_type == VDIR. */
242 struct {
243 /* Pointer to the parent directory. The root
244 * directory has a pointer to itself in this field;
245 * this property identifies the root node. */
246 struct tmpfs_node * tn_parent;
247
248 /* Head of a tail-queue that links the contents of
249 * the directory together. See above for a
250 * description of its contents. */
251 struct tmpfs_dir tn_dir;
252
253 /* Number and pointer of the first directory entry
254 * returned by the readdir operation if it were
255 * called again to continue reading data from the
256 * same directory as before. This is used to speed
257 * up reads of long directories, assuming that no
258 * more than one read is in progress at a given time.
259 * Otherwise, these values are discarded and a linear
260 * scan is performed from the beginning up to the
261 * point where readdir starts returning values. */
262 off_t tn_readdir_lastn;
263 struct tmpfs_dirent * tn_readdir_lastp;
264 } tn_dir;
265
266 /* Valid when tn_type == VLNK. */
267 struct tn_lnk {
268 /* The link's target, allocated from a string pool. */
269 char * tn_link;
270 } tn_lnk;
271
272 /* Valid when tn_type == VREG. */
273 struct tn_reg {
274 /* The contents of regular files stored in a tmpfs
275 * file system are represented by a single anonymous
276 * memory object (aobj, for short). The aobj provides
277 * direct access to any position within the file,
278 * because its contents are always mapped in a
279 * contiguous region of virtual memory. It is a task
280 * of the memory management subsystem (see uvm(9)) to
281 * issue the required page ins or page outs whenever
282 * a position within the file is accessed. */
283 struct uvm_object * tn_aobj;
284 size_t tn_aobj_pages;
285 } tn_reg;
286 } tn_spec;
287 };
288
289 #if defined(_KERNEL)
290
291 LIST_HEAD(tmpfs_node_list, tmpfs_node);
292
293 /* --------------------------------------------------------------------- */
294
295 /*
296 * Internal representation of a tmpfs mount point.
297 */
298 struct tmpfs_mount {
299 /* Lock on global data */
300 kmutex_t tm_lock;
301
302 /* Maximum number of memory pages available for use by the file
303 * system, set during mount time. This variable must never be
304 * used directly as it may be bigger than the current amount of
305 * free memory; in the extreme case, it will hold the SIZE_MAX
306 * value. Instead, use the TMPFS_PAGES_MAX macro. */
307 size_t tm_pages_max;
308
309 /* Number of pages in use by the file system. Cannot be bigger
310 * than the value returned by TMPFS_PAGES_MAX in any case. */
311 size_t tm_pages_used;
312
313 /* Pointer to the node representing the root directory of this
314 * file system. */
315 struct tmpfs_node * tm_root;
316
317 /* Maximum number of possible nodes for this file system; set
318 * during mount time. We need a hard limit on the maximum number
319 * of nodes to avoid allocating too much of them; their objects
320 * cannot be released until the file system is unmounted.
321 * Otherwise, we could easily run out of memory by creating lots
322 * of empty files and then simply removing them. */
323 ino_t tm_nodes_max;
324
325 /* Number of nodes currently allocated. This number only grows.
326 * When it reaches tm_nodes_max, no more new nodes can be allocated.
327 * Of course, the old, unused ones can be reused. */
328 ino_t tm_nodes_cnt;
329
330 /* Node list. */
331 struct tmpfs_node_list tm_nodes;
332
333 /* Pools used to store file system meta data. These are not shared
334 * across several instances of tmpfs for the reasons described in
335 * tmpfs_pool.c. */
336 struct tmpfs_pool tm_dirent_pool;
337 struct tmpfs_pool tm_node_pool;
338 struct tmpfs_str_pool tm_str_pool;
339 };
340
341 /* --------------------------------------------------------------------- */
342
343 /*
344 * This structure maps a file identifier to a tmpfs node. Used by the
345 * NFS code.
346 */
347 struct tmpfs_fid {
348 uint16_t tf_len;
349 uint16_t tf_pad;
350 uint32_t tf_gen;
351 ino_t tf_id;
352 };
353
354 /* --------------------------------------------------------------------- */
355
356 /*
357 * Prototypes for tmpfs_subr.c.
358 */
359
360 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype,
361 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *,
362 char *, dev_t, struct proc *, struct tmpfs_node **);
363 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
364 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
365 const char *, uint16_t, struct tmpfs_dirent **);
366 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *,
367 bool);
368 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, struct vnode **);
369 void tmpfs_free_vp(struct vnode *);
370 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
371 struct componentname *, char *);
372 void tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *);
373 void tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *);
374 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node,
375 struct componentname *cnp);
376 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *);
377 int tmpfs_dir_getdotdotdent(struct tmpfs_node *, struct uio *);
378 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t);
379 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *);
380 int tmpfs_reg_resize(struct vnode *, off_t);
381 size_t tmpfs_mem_info(bool);
382 int tmpfs_chflags(struct vnode *, int, kauth_cred_t, struct lwp *);
383 int tmpfs_chmod(struct vnode *, mode_t, kauth_cred_t, struct lwp *);
384 int tmpfs_chown(struct vnode *, uid_t, gid_t, kauth_cred_t, struct lwp *);
385 int tmpfs_chsize(struct vnode *, u_quad_t, kauth_cred_t, struct lwp *);
386 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *,
387 int, kauth_cred_t, struct lwp *);
388 void tmpfs_itimes(struct vnode *, const struct timespec *,
389 const struct timespec *);
390
391 void tmpfs_update(struct vnode *, const struct timespec *,
392 const struct timespec *, int);
393 int tmpfs_truncate(struct vnode *, off_t);
394
395 /* --------------------------------------------------------------------- */
396
397 /*
398 * Convenience macros to simplify some logical expressions.
399 */
400 #define IMPLIES(a, b) (!(a) || (b))
401 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a))
402
403 /* --------------------------------------------------------------------- */
404
405 /*
406 * Checks that the directory entry pointed by 'de' matches the name 'name'
407 * with a length of 'len'.
408 */
409 #define TMPFS_DIRENT_MATCHES(de, name, len) \
410 (de->td_namelen == (uint16_t)len && \
411 memcmp((de)->td_name, (name), (de)->td_namelen) == 0)
412
413 /* --------------------------------------------------------------------- */
414
415 /*
416 * Ensures that the node pointed by 'node' is a directory and that its
417 * contents are consistent with respect to directories.
418 */
419 #define TMPFS_VALIDATE_DIR(node) \
420 KASSERT((node)->tn_type == VDIR); \
421 KASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
422 KASSERT((node)->tn_spec.tn_dir.tn_readdir_lastp == NULL || \
423 tmpfs_dircookie((node)->tn_spec.tn_dir.tn_readdir_lastp) == \
424 (node)->tn_spec.tn_dir.tn_readdir_lastn);
425
426 /* --------------------------------------------------------------------- */
427
428 /*
429 * Memory management stuff.
430 */
431
432 /* Amount of memory pages to reserve for the system (e.g., to not use by
433 * tmpfs).
434 * XXX: Should this be tunable through sysctl, for instance? */
435 #define TMPFS_PAGES_RESERVED (4 * 1024 * 1024 / PAGE_SIZE)
436
437 /* Returns the maximum size allowed for a tmpfs file system. This macro
438 * must be used instead of directly retrieving the value from tm_pages_max.
439 * The reason is that the size of a tmpfs file system is dynamic: it lets
440 * the user store files as long as there is enough free memory (including
441 * physical memory and swap space). Therefore, the amount of memory to be
442 * used is either the limit imposed by the user during mount time or the
443 * amount of available memory, whichever is lower. To avoid consuming all
444 * the memory for a given mount point, the system will always reserve a
445 * minimum of TMPFS_PAGES_RESERVED pages, which is also taken into account
446 * by this macro (see above). */
447 static __inline size_t
448 TMPFS_PAGES_MAX(struct tmpfs_mount *tmp)
449 {
450 size_t freepages;
451
452 freepages = tmpfs_mem_info(false);
453 if (freepages < TMPFS_PAGES_RESERVED)
454 freepages = 0;
455 else
456 freepages -= TMPFS_PAGES_RESERVED;
457
458 return MIN(tmp->tm_pages_max, freepages + tmp->tm_pages_used);
459 }
460
461 /* Returns the available space for the given file system. */
462 #define TMPFS_PAGES_AVAIL(tmp) (TMPFS_PAGES_MAX(tmp) - (tmp)->tm_pages_used)
463
464 /* --------------------------------------------------------------------- */
465
466 /*
467 * Macros/functions to convert from generic data structures to tmpfs
468 * specific ones.
469 */
470
471 static __inline
472 struct tmpfs_mount *
473 VFS_TO_TMPFS(struct mount *mp)
474 {
475 struct tmpfs_mount *tmp;
476
477 #ifdef KASSERT
478 KASSERT((mp) != NULL && (mp)->mnt_data != NULL);
479 #endif
480 tmp = (struct tmpfs_mount *)(mp)->mnt_data;
481 return tmp;
482 }
483
484 #endif /* defined(_KERNEL) */
485
486 static __inline
487 struct tmpfs_node *
488 VP_TO_TMPFS_NODE(struct vnode *vp)
489 {
490 struct tmpfs_node *node;
491
492 #ifdef KASSERT
493 KASSERT((vp) != NULL && (vp)->v_data != NULL);
494 #endif
495 node = (struct tmpfs_node *)vp->v_data;
496 return node;
497 }
498
499 #if defined(_KERNEL)
500
501 static __inline
502 struct tmpfs_node *
503 VP_TO_TMPFS_DIR(struct vnode *vp)
504 {
505 struct tmpfs_node *node;
506
507 node = VP_TO_TMPFS_NODE(vp);
508 #ifdef KASSERT
509 TMPFS_VALIDATE_DIR(node);
510 #endif
511 return node;
512 }
513
514 #endif /* defined(_KERNEL) */
515
516 /* ---------------------------------------------------------------------
517 * USER AND KERNEL DEFINITIONS
518 * --------------------------------------------------------------------- */
519
520 /*
521 * This structure is used to communicate mount parameters between userland
522 * and kernel space.
523 */
524 #define TMPFS_ARGS_VERSION 1
525 struct tmpfs_args {
526 int ta_version;
527
528 /* Size counters. */
529 ino_t ta_nodes_max;
530 off_t ta_size_max;
531
532 /* Root node attributes. */
533 uid_t ta_root_uid;
534 gid_t ta_root_gid;
535 mode_t ta_root_mode;
536 };
537 #endif /* _FS_TMPFS_TMPFS_H_ */
538