tmpfs.h revision 1.22 1 /* $NetBSD: tmpfs.h,v 1.22 2006/11/05 16:59:18 jmmv 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 * 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 #define TMPFS_DIRCOOKIE_DOT 0
140 #define TMPFS_DIRCOOKIE_DOTDOT 1
141 #define TMPFS_DIRCOOKIE_EOF 2
142 static __inline
143 off_t
144 tmpfs_dircookie(struct tmpfs_dirent *de)
145 {
146 off_t cookie;
147
148 cookie = ((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFF;
149 KASSERT(cookie != TMPFS_DIRCOOKIE_DOT);
150 KASSERT(cookie != TMPFS_DIRCOOKIE_DOTDOT);
151 KASSERT(cookie != TMPFS_DIRCOOKIE_EOF);
152
153 return cookie;
154 }
155
156 /* --------------------------------------------------------------------- */
157
158 /*
159 * Internal representation of a tmpfs file system node.
160 *
161 * This structure is splitted in two parts: one holds attributes common
162 * to all file types and the other holds data that is only applicable to
163 * a particular type. The code must be careful to only access those
164 * attributes that are actually allowed by the node's type.
165 */
166 struct tmpfs_node {
167 /* Doubly-linked list entry which links all existing nodes for a
168 * single file system. This is provided to ease the removal of
169 * all nodes during the unmount operation. */
170 LIST_ENTRY(tmpfs_node) tn_entries;
171
172 /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
173 * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode
174 * types instead of a custom enumeration is to make things simpler
175 * and faster, as we do not need to convert between two types. */
176 enum vtype tn_type;
177
178 /* Node identifier. */
179 ino_t tn_id;
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 int tn_status;
185 #define TMPFS_NODE_ACCESSED (1 << 1)
186 #define TMPFS_NODE_MODIFIED (1 << 2)
187 #define TMPFS_NODE_CHANGED (1 << 3)
188
189 /* The node size. It does not necessarily match the real amount
190 * of memory consumed by it. */
191 off_t tn_size;
192
193 /* Generic node attributes. */
194 uid_t tn_uid;
195 gid_t tn_gid;
196 mode_t tn_mode;
197 int tn_flags;
198 nlink_t tn_links;
199 struct timespec tn_atime;
200 struct timespec tn_mtime;
201 struct timespec tn_ctime;
202 struct timespec tn_birthtime;
203 unsigned long tn_gen;
204
205 /* Head of byte-level lock list (used by tmpfs_advlock). */
206 struct lockf * tn_lockf;
207
208 /* As there is a single vnode for each active file within the
209 * system, care has to be taken to avoid allocating more than one
210 * vnode per file. In order to do this, a bidirectional association
211 * is kept between vnodes and nodes.
212 *
213 * Whenever a vnode is allocated, its v_data field is updated to
214 * point to the node it references. At the same time, the node's
215 * tn_vnode field is modified to point to the new vnode representing
216 * it. Further attempts to allocate a vnode for this same node will
217 * result in returning a new reference to the value stored in
218 * tn_vnode.
219 *
220 * May be NULL when the node is unused (that is, no vnode has been
221 * allocated for it or it has been reclaimed). */
222 struct vnode * tn_vnode;
223
224 /* Pointer to the node returned by tmpfs_lookup() after doing a
225 * delete or a rename lookup; its value is only valid in these two
226 * situations. In case we were looking up . or .., it holds a null
227 * pointer. */
228 struct tmpfs_dirent * tn_lookup_dirent;
229
230 union {
231 /* Valid when tn_type == VBLK || tn_type == VCHR. */
232 struct {
233 dev_t tn_rdev;
234 } tn_dev;
235
236 /* Valid when tn_type == VDIR. */
237 struct {
238 /* Pointer to the parent directory. The root
239 * directory has a pointer to itself in this field;
240 * this property identifies the root node. */
241 struct tmpfs_node * tn_parent;
242
243 /* Head of a tail-queue that links the contents of
244 * the directory together. See above for a
245 * description of its contents. */
246 struct tmpfs_dir tn_dir;
247
248 /* Number and pointer of the first directory entry
249 * returned by the readdir operation if it were
250 * called again to continue reading data from the
251 * same directory as before. This is used to speed
252 * up reads of long directories, assuming that no
253 * more than one read is in progress at a given time.
254 * Otherwise, these values are discarded and a linear
255 * scan is performed from the beginning up to the
256 * point where readdir starts returning values. */
257 off_t tn_readdir_lastn;
258 struct tmpfs_dirent * tn_readdir_lastp;
259 } tn_dir;
260
261 /* Valid when tn_type == VLNK. */
262 struct tn_lnk {
263 /* The link's target, allocated from a string pool. */
264 char * tn_link;
265 } tn_lnk;
266
267 /* Valid when tn_type == VREG. */
268 struct tn_reg {
269 /* The contents of regular files stored in a tmpfs
270 * file system are represented by a single anonymous
271 * memory object (aobj, for short). The aobj provides
272 * direct access to any position within the file,
273 * because its contents are always mapped in a
274 * contiguous region of virtual memory. It is a task
275 * of the memory management subsystem (see uvm(9)) to
276 * issue the required page ins or page outs whenever
277 * a position within the file is accessed. */
278 struct uvm_object * tn_aobj;
279 size_t tn_aobj_pages;
280 } tn_reg;
281 } tn_spec;
282 };
283
284 #if defined(_KERNEL)
285
286 LIST_HEAD(tmpfs_node_list, tmpfs_node);
287
288 /* --------------------------------------------------------------------- */
289
290 /*
291 * Internal representation of a tmpfs mount point.
292 */
293 struct tmpfs_mount {
294 /* Maximum number of memory pages available for use by the file
295 * system, set during mount time. This variable must never be
296 * used directly as it may be bigger that the current amount of
297 * free memory; in the extreme case, it will hold the SIZE_MAX
298 * value. Instead, use the TMPFS_PAGES_MAX macro. */
299 size_t tm_pages_max;
300
301 /* Number of pages in use by the file system. Cannot be bigger
302 * than the value returned by TMPFS_PAGES_MAX in any case. */
303 size_t tm_pages_used;
304
305 /* Pointer to the node representing the root directory of this
306 * file system. */
307 struct tmpfs_node * tm_root;
308
309 /* Maximum number of possible nodes for this file system; set
310 * during mount time. We need a hard limit on the maximum number
311 * of nodes to avoid allocating too much of them; their objects
312 * cannot be released until the file system is unmounted.
313 * Otherwise, we could easily run out of memory by creating lots
314 * of empty files and then simply removing them. */
315 ino_t tm_nodes_max;
316
317 /* Number of nodes currently allocated. This number only grows.
318 * When it reaches tm_nodes_max, no more new nodes can be allocated.
319 * Of course, the old, unused ones can be reused. */
320 ino_t tm_nodes_last;
321
322 /* Nodes are organized in two different lists. The used list
323 * contains all nodes that are currently used by the file system;
324 * i.e., they refer to existing files. The available list contains
325 * all nodes that are currently available for use by new files.
326 * Nodes must be kept in this list (instead of deleting them)
327 * because we need to keep track of their generation number (tn_gen
328 * field).
329 *
330 * Note that nodes are lazily allocated: if the available list is
331 * empty and we have enough space to create more nodes, they will be
332 * created and inserted in the used list. Once these are released,
333 * they will go into the available list, remaining alive until the
334 * file system is unmounted. */
335 struct tmpfs_node_list tm_nodes_used;
336 struct tmpfs_node_list tm_nodes_avail;
337
338 /* Pools used to store file system meta data. These are not shared
339 * across several instances of tmpfs for the reasons described in
340 * tmpfs_pool.c. */
341 struct tmpfs_pool tm_dirent_pool;
342 struct tmpfs_pool tm_node_pool;
343 struct tmpfs_str_pool tm_str_pool;
344 };
345
346 /* --------------------------------------------------------------------- */
347
348 /*
349 * This structure maps a file identifier to a tmpfs node. Used by the
350 * NFS code.
351 */
352 struct tmpfs_fid {
353 uint16_t tf_len;
354 uint16_t tf_pad;
355 uint32_t tf_gen;
356 ino_t tf_id;
357 };
358
359 /* --------------------------------------------------------------------- */
360
361 /*
362 * Prototypes for tmpfs_subr.c.
363 */
364
365 int tmpfs_alloc_node(struct tmpfs_mount *, enum vtype,
366 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *,
367 char *, dev_t, struct proc *, struct tmpfs_node **);
368 void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
369 int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
370 const char *, uint16_t, struct tmpfs_dirent **);
371 void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *,
372 boolean_t);
373 int tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, struct vnode **);
374 void tmpfs_free_vp(struct vnode *);
375 int tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
376 struct componentname *, char *);
377 void tmpfs_dir_attach(struct vnode *, struct tmpfs_dirent *);
378 void tmpfs_dir_detach(struct vnode *, struct tmpfs_dirent *);
379 struct tmpfs_dirent * tmpfs_dir_lookup(struct tmpfs_node *node,
380 struct componentname *cnp);
381 int tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *);
382 int tmpfs_dir_getdotdotdent(struct tmpfs_node *, struct uio *);
383 struct tmpfs_dirent * tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t);
384 int tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *);
385 int tmpfs_reg_resize(struct vnode *, off_t);
386 size_t tmpfs_mem_info(boolean_t);
387 int tmpfs_chflags(struct vnode *, int, kauth_cred_t, struct lwp *);
388 int tmpfs_chmod(struct vnode *, mode_t, kauth_cred_t, struct lwp *);
389 int tmpfs_chown(struct vnode *, uid_t, gid_t, kauth_cred_t, struct lwp *);
390 int tmpfs_chsize(struct vnode *, u_quad_t, kauth_cred_t, struct lwp *);
391 int tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *,
392 int, kauth_cred_t, struct lwp *);
393 void tmpfs_itimes(struct vnode *, const struct timespec *,
394 const struct timespec *);
395
396 void tmpfs_update(struct vnode *, const struct timespec *,
397 const struct timespec *, int);
398 int tmpfs_truncate(struct vnode *, off_t);
399
400 /* --------------------------------------------------------------------- */
401
402 /*
403 * Convenience macros to simplify some logical expressions.
404 */
405 #define IMPLIES(a, b) (!(a) || (b))
406 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a))
407
408 /* --------------------------------------------------------------------- */
409
410 /*
411 * Checks that the directory entry pointed by 'de' matches the name 'name'
412 * with a length of 'len'.
413 */
414 #define TMPFS_DIRENT_MATCHES(de, name, len) \
415 (de->td_namelen == (uint16_t)len && \
416 memcmp((de)->td_name, (name), (de)->td_namelen) == 0)
417
418 /* --------------------------------------------------------------------- */
419
420 /*
421 * Ensures that the node pointed by 'node' is a directory and that its
422 * contents are consistent with respect to directories.
423 */
424 #define TMPFS_VALIDATE_DIR(node) \
425 KASSERT((node)->tn_type == VDIR); \
426 KASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
427 KASSERT((node)->tn_spec.tn_dir.tn_readdir_lastp == NULL || \
428 tmpfs_dircookie((node)->tn_spec.tn_dir.tn_readdir_lastp) == \
429 (node)->tn_spec.tn_dir.tn_readdir_lastn);
430
431 /* --------------------------------------------------------------------- */
432
433 /*
434 * Memory management stuff.
435 */
436
437 /* Amount of memory pages to reserve for the system (e.g., to not use by
438 * tmpfs).
439 * XXX: Should this be tunable through sysctl, for instance? */
440 #define TMPFS_PAGES_RESERVED (4 * 1024 * 1024 / PAGE_SIZE)
441
442 /* Returns the maximum size allowed for a tmpfs file system. This macro
443 * must be used instead of directly retrieving the value from tm_pages_max.
444 * The reason is that the size of a tmpfs file system is dynamic: it lets
445 * the user store files as long as there is enough free memory (including
446 * physical memory and swap space). Therefore, the amount of memory to be
447 * used is either the limit imposed by the user during mount time or the
448 * amount of available memory, whichever is lower. To avoid consuming all
449 * the memory for a given mount point, the system will always reserve a
450 * minimum of TMPFS_PAGES_RESERVED pages, which is also taken into account
451 * by this macro (see above). */
452 static __inline size_t
453 TMPFS_PAGES_MAX(struct tmpfs_mount *tmp)
454 {
455 size_t freepages;
456
457 freepages = tmpfs_mem_info(FALSE);
458 if (freepages < TMPFS_PAGES_RESERVED)
459 freepages = 0;
460 else
461 freepages -= TMPFS_PAGES_RESERVED;
462
463 return MIN(tmp->tm_pages_max, freepages + tmp->tm_pages_used);
464 }
465
466 /* Returns the available space for the given file system. */
467 #define TMPFS_PAGES_AVAIL(tmp) (TMPFS_PAGES_MAX(tmp) - (tmp)->tm_pages_used)
468
469 /* --------------------------------------------------------------------- */
470
471 /*
472 * Macros/functions to convert from generic data structures to tmpfs
473 * specific ones.
474 */
475
476 static __inline
477 struct tmpfs_mount *
478 VFS_TO_TMPFS(struct mount *mp)
479 {
480 struct tmpfs_mount *tmp;
481
482 #ifdef KASSERT
483 KASSERT((mp) != NULL && (mp)->mnt_data != NULL);
484 #endif
485 tmp = (struct tmpfs_mount *)(mp)->mnt_data;
486 return tmp;
487 }
488
489 #endif /* defined(_KERNEL) */
490
491 static __inline
492 struct tmpfs_node *
493 VP_TO_TMPFS_NODE(struct vnode *vp)
494 {
495 struct tmpfs_node *node;
496
497 #ifdef KASSERT
498 KASSERT((vp) != NULL && (vp)->v_data != NULL);
499 #endif
500 node = (struct tmpfs_node *)vp->v_data;
501 return node;
502 }
503
504 #if defined(_KERNEL)
505
506 static __inline
507 struct tmpfs_node *
508 VP_TO_TMPFS_DIR(struct vnode *vp)
509 {
510 struct tmpfs_node *node;
511
512 node = VP_TO_TMPFS_NODE(vp);
513 #ifdef KASSERT
514 TMPFS_VALIDATE_DIR(node);
515 #endif
516 return node;
517 }
518
519 #endif /* defined(_KERNEL) */
520
521 /* ---------------------------------------------------------------------
522 * USER AND KERNEL DEFINITIONS
523 * --------------------------------------------------------------------- */
524
525 /*
526 * This structure is used to communicate mount parameters between userland
527 * and kernel space.
528 */
529 #define TMPFS_ARGS_VERSION 1
530 struct tmpfs_args {
531 int ta_version;
532
533 /* Size counters. */
534 ino_t ta_nodes_max;
535 off_t ta_size_max;
536
537 /* Root node attributes. */
538 uid_t ta_root_uid;
539 gid_t ta_root_gid;
540 mode_t ta_root_mode;
541 };
542 #endif /* _FS_TMPFS_TMPFS_H_ */
543