sys/vfs/tmpfs/tmpfs.h

   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 #include <sys/objcache.h>
  50
  51 /* --------------------------------------------------------------------- */
  52 #include <sys/malloc.h>
  53 #ifdef _KERNEL
  54 #include <sys/systm.h>
  55 #endif
  56 #include <sys/vmmeter.h>
  57 #include <vm/swap_pager.h>
  58
  59 MALLOC_DECLARE(M_TMPFSMNT);
  60
  61 /* --------------------------------------------------------------------- */
  62
  63 /*
  64  * Internal representation of a tmpfs directory entry.
  65  */
  66 struct tmpfs_dirent {
  67         RB_ENTRY(tmpfs_dirent)  rb_node;
  68         RB_ENTRY(tmpfs_dirent)  rb_cookienode;
  69
  70         /* Length of the name stored in this directory entry.  This avoids
  71          * the need to recalculate it every time the name is used. */
  72         uint16_t                td_namelen;
  73
  74         /* The name of the entry, allocated from a string pool.  This
  75         * string is not required to be zero-terminated; therefore, the
  76         * td_namelen field must always be used when accessing its value. */
  77         char                    *td_name;
  78
  79         /* Pointer to the node this entry refers to. */
  80         struct tmpfs_node       *td_node;
  81 };
  82
  83 struct tmpfs_dirtree;
  84 RB_HEAD(tmpfs_dirtree, tmpfs_dirent);
  85 RB_PROTOTYPE(tmpfs_dirtree, tmpfs_dirent, rb_node,
  86         tmpfs_dirtree_compare);
  87
  88 RB_HEAD(tmpfs_dirtree_cookie, tmpfs_dirent);
  89 RB_PROTOTYPE(tmpfs_dirtree_cookie, tmpfs_dirent, rb_cookienode,
  90         tmpfs_dirtree_cookie_compare);
  91
  92
  93 /*
  94  * A directory in tmpfs holds a set of directory entries, which in
  95  * turn point to other files (which can be directories themselves).
  96  *
  97  * In tmpfs, this set is managed by a red-black tree, whose root is defined
  98  * by the struct tmpfs_dirtree type.
  99  *
 100  * It is important to notice that directories do not have entries for . and
 101  * .. as other file systems do.  These can be generated when requested
 102  * based on information available by other means, such as the pointer to
 103  * the node itself in the former case or the pointer to the parent directory
 104  * in the latter case.  This is done to simplify tmpfs's code and, more
 105  * importantly, to remove redundancy.
 106  *
 107  * Each entry in a directory has a cookie that identifies it.  Cookies
 108  * supersede offsets within directories because, given how tmpfs stores
 109  * directories in memory, there is no such thing as an offset.  (Emulating
 110  * a real offset could be very difficult.)
 111  *
 112  * The '.', '..' and the end of directory markers have fixed cookies which
 113  * cannot collide with the cookies generated by other entries.  The cookies
 114  * for the other entries are generated based on the memory address on which
 115  * stores their information is stored.
 116  *
 117  * DragonFly binaries use 64-bit cookies.  We mask-off the signed bit to
 118  * ensure that cookie 'offsets' are positive.
 119  */
 120 #ifdef _KERNEL
 121
 122 #define TMPFS_ROOTINO   ((ino_t)2)
 123
 124 #define TMPFS_DIRCOOKIE_DOT     0
 125 #define TMPFS_DIRCOOKIE_DOTDOT  1
 126 #define TMPFS_DIRCOOKIE_EOF     2
 127
 128 static __inline
 129 off_t
 130 tmpfs_dircookie(struct tmpfs_dirent *de)
 131 {
 132         return (((off_t)(uintptr_t)de >> 1) & 0x7FFFFFFFFFFFFFFFLLU);
 133 }
 134
 135 #endif  /* _KERNEL */
 136
 137 /* --------------------------------------------------------------------- */
 138
 139 /*
 140  * Internal representation of a tmpfs file system node.
 141  *
 142  * This structure is splitted in two parts: one holds attributes common
 143  * to all file types and the other holds data that is only applicable to
 144  * a particular type.  The code must be careful to only access those
 145  * attributes that are actually allowed by the node's type.
 146  */
 147 struct tmpfs_node {
 148         /* Doubly-linked list entry which links all existing nodes for a
 149          * single file system.  This is provided to ease the removal of
 150          * all nodes during the unmount operation. */
 151         LIST_ENTRY(tmpfs_node)  tn_entries;
 152
 153         /* The node's type.  Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO',
 154          * 'VLNK', 'VREG' and 'VSOCK' is allowed.  The usage of vnode
 155          * types instead of a custom enumeration is to make things simpler
 156          * and faster, as we do not need to convert between two types. */
 157         enum vtype              tn_type;
 158
 159         /* Node identifier. */
 160         ino_t                   tn_id;
 161
 162         /* Node's internal status.  This is used by several file system
 163          * operations to do modifications to the node in a delayed
 164          * fashion. */
 165         int                     tn_status;
 166 #define TMPFS_NODE_ACCESSED     (1 << 1)
 167 #define TMPFS_NODE_MODIFIED     (1 << 2)
 168 #define TMPFS_NODE_CHANGED      (1 << 3)
 169
 170         /* The node size.  It does not necessarily match the real amount
 171          * of memory consumed by it. */
 172         off_t                   tn_size;
 173
 174         /* Generic node attributes. */
 175         uid_t                   tn_uid;
 176         gid_t                   tn_gid;
 177         mode_t                  tn_mode;
 178         int                     tn_flags;
 179         nlink_t                 tn_links;
 180         long                    tn_atime;
 181         long                    tn_atimensec;
 182         long                    tn_mtime;
 183         long                    tn_mtimensec;
 184         long                    tn_ctime;
 185         long                    tn_ctimensec;
 186         unsigned long           tn_gen;
 187         struct lockf            tn_advlock;
 188
 189         /* As there is a single vnode for each active file within the
 190          * system, care has to be taken to avoid allocating more than one
 191          * vnode per file.  In order to do this, a bidirectional association
 192          * is kept between vnodes and nodes.
 193          *
 194          * Whenever a vnode is allocated, its v_data field is updated to
 195          * point to the node it references.  At the same time, the node's
 196          * tn_vnode field is modified to point to the new vnode representing
 197          * it.  Further attempts to allocate a vnode for this same node will
 198          * result in returning a new reference to the value stored in
 199          * tn_vnode.
 200          *
 201          * May be NULL when the node is unused (that is, no vnode has been
 202          * allocated for it or it has been reclaimed). */
 203         struct vnode *          tn_vnode;
 204
 205         /* interlock to protect tn_vpstate */
 206         struct lock             tn_interlock;
 207
 208         /* Identify if current node has vnode assiocate with
 209          * or allocating vnode.
 210          */
 211         int             tn_vpstate;
 212
 213         /* misc data field for different tn_type node */
 214         union {
 215                 /* Valid when tn_type == VBLK || tn_type == VCHR. */
 216                 dev_t                   tn_rdev; /*int32_t ?*/
 217
 218                 /* Valid when tn_type == VDIR. */
 219                 struct tn_dir {
 220                         /*
 221                          * Pointer to the parent directory.  The root
 222                          * directory has a pointer to itself in this field;
 223                          * this property identifies the root node.
 224                          */
 225                         struct tmpfs_node *     tn_parent;
 226
 227                         /*
 228                          * Directory entries are indexed by name and also
 229                          * indexed by cookie.
 230                          */
 231                         struct tmpfs_dirtree            tn_dirtree;
 232                         struct tmpfs_dirtree_cookie     tn_cookietree;
 233                 } tn_dir;
 234
 235                 /* Valid when tn_type == VLNK. */
 236                 /* The link's target, allocated from a string pool. */
 237                 char *                  tn_link;
 238
 239                 /*
 240                  * Valid when tn_type == VREG.
 241                  *
 242                  * aobj is used as backing store for the vnode object.  It
 243                  * typically only contains swap assignments, but we also use
 244                  * it to save the vnode object's vm_page's when the vnode
 245                  * becomes inactive.
 246                  */
 247                 struct tn_reg {
 248                         vm_object_t             tn_aobj;
 249                         size_t                  tn_aobj_pages;
 250                         int                     tn_pages_in_aobj;
 251                 } tn_reg;
 252
 253                 /* Valid when tn_type = VFIFO */
 254                 struct tn_fifo {
 255                         int (*tn_fo_read)  (struct file *fp, struct uio *uio,
 256                                 struct ucred *cred, int flags);
 257                         int (*tn_fo_write) (struct file *fp, struct uio *uio,
 258                                 struct ucred *cred, int flags);
 259                 } tn_fifo;
 260         } tn_spec;
 261 };
 262
 263 /* Only userspace needs this */
 264 #define VTOI(vp)        ((struct tmpfs_node *)(vp)->v_data)
 265
 266 #ifdef _KERNEL
 267 LIST_HEAD(tmpfs_node_list, tmpfs_node);
 268
 269 #define tn_rdev tn_spec.tn_rdev
 270 #define tn_dir tn_spec.tn_dir
 271 #define tn_link tn_spec.tn_link
 272 #define tn_reg tn_spec.tn_reg
 273 #define tn_fifo tn_spec.tn_fifo
 274
 275 #define TMPFS_NODE_LOCK(node) lockmgr(&(node)->tn_interlock, LK_EXCLUSIVE|LK_RETRY)
 276 #define TMPFS_NODE_LOCK_SH(node) lockmgr(&(node)->tn_interlock, LK_SHARED|LK_RETRY)
 277 #define TMPFS_NODE_UNLOCK(node) lockmgr(&(node)->tn_interlock, LK_RELEASE)
 278 #define TMPFS_NODE_MTX(node) (&(node)->tn_interlock)
 279
 280 #ifdef INVARIANTS
 281 #define TMPFS_ASSERT_LOCKED(node) do {                                  \
 282                 KKASSERT(node != NULL);                                 \
 283                 KKASSERT(node->tn_vnode != NULL);                       \
 284                 if (!vn_islocked(node->tn_vnode) &&                     \
 285                     (lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE ))             \
 286                         panic("tmpfs: node is not locked: %p", node);   \
 287         } while (0)
 288 #define TMPFS_ASSERT_ELOCKED(node) do {                                 \
 289                 KKASSERT((node) != NULL);                               \
 290                 KKASSERT(lockstatus(TMPFS_NODE_MTX(node), curthread) == LK_EXCLUSIVE);          \
 291         } while (0)
 292 #else
 293 #define TMPFS_ASSERT_LOCKED(node) (void)0
 294 #define TMPFS_ASSERT_ELOCKED(node) (void)0
 295 #endif  /* INVARIANTS */
 296
 297 #define TMPFS_VNODE_ALLOCATING  1
 298 #define TMPFS_VNODE_WANT        2
 299 #define TMPFS_VNODE_DOOMED      4
 300 /* --------------------------------------------------------------------- */
 301
 302 /*
 303  * Internal representation of a tmpfs mount point.
 304  */
 305 struct tmpfs_mount {
 306         struct mount            *tm_mount;
 307
 308         /* Maximum number of memory pages available for use by the file
 309          * system, set during mount time.  This variable must never be
 310          * used directly as it may be bigger than the current amount of
 311          * free memory; in the extreme case, it will hold the SIZE_MAX
 312          * value.  Instead, use the TMPFS_PAGES_MAX macro. */
 313         long                    tm_pages_max;
 314
 315         /* Number of pages in use by the file system.  Cannot be bigger
 316          * than the value returned by TMPFS_PAGES_MAX in any case. */
 317         long                    tm_pages_used;
 318
 319         /* Pointer to the node representing the root directory of this
 320          * file system. */
 321         struct tmpfs_node *     tm_root;
 322
 323         /* Maximum number of possible nodes for this file system; set
 324          * during mount time.  We need a hard limit on the maximum number
 325          * of nodes to avoid allocating too much of them; their objects
 326          * cannot be released until the file system is unmounted.
 327          * Otherwise, we could easily run out of memory by creating lots
 328          * of empty files and then simply removing them. */
 329         ino_t                   tm_nodes_max;
 330
 331         /* Number of nodes currently that are in use. */
 332         ino_t                   tm_nodes_inuse;
 333
 334         /* maximum representable file size */
 335         u_int64_t               tm_maxfilesize;
 336
 337         /* Nodes are organized in two different lists.  The used list
 338          * contains all nodes that are currently used by the file system;
 339          * i.e., they refer to existing files.  The available list contains
 340          * all nodes that are currently available for use by new files.
 341          * Nodes must be kept in this list (instead of deleting them)
 342          * because we need to keep track of their generation number (tn_gen
 343          * field).
 344          *
 345          * Note that nodes are lazily allocated: if the available list is
 346          * empty and we have enough space to create more nodes, they will be
 347          * created and inserted in the used list.  Once these are released,
 348          * they will go into the available list, remaining alive until the
 349          * file system is unmounted. */
 350         struct tmpfs_node_list  tm_nodes_used;
 351
 352         /* Per-mount malloc zones for tmpfs nodes, names, and dirents */
 353         struct malloc_type      *tm_node_zone;
 354         struct malloc_type      *tm_dirent_zone;
 355         struct malloc_type      *tm_name_zone;
 356
 357         struct objcache_malloc_args tm_node_zone_malloc_args;
 358         struct objcache_malloc_args tm_dirent_zone_malloc_args;
 359
 360         /* Pools used to store file system meta data. */
 361         struct objcache         *tm_dirent_pool;
 362         struct objcache         *tm_node_pool;
 363
 364         ino_t                   tm_ino;
 365         int                     tm_flags;
 366
 367         struct netexport        tm_export;
 368
 369         struct mount            *tm_mnt;
 370 };
 371
 372 #define TMPFS_LOCK(tm) lwkt_gettoken(&(tm)->tm_mount->mnt_token)
 373 #define TMPFS_UNLOCK(tm) lwkt_reltoken(&(tm)->tm_mount->mnt_token)
 374
 375 /* --------------------------------------------------------------------- */
 376
 377 /*
 378  * This structure maps a file identifier to a tmpfs node.  Used by the
 379  * NFS code.
 380  */
 381 struct tmpfs_fid {
 382         uint16_t                tf_len;
 383         uint16_t                tf_pad;
 384         ino_t                   tf_id;
 385         unsigned long           tf_gen;
 386 } __packed;
 387
 388 /* --------------------------------------------------------------------- */
 389
 390 /*
 391  * Prototypes for tmpfs_subr.c.
 392  */
 393
 394 int     tmpfs_alloc_node(struct tmpfs_mount *, enum vtype,
 395             uid_t uid, gid_t gid, mode_t mode, char *, int, int,
 396             struct tmpfs_node **);
 397 void    tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *);
 398 int     tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *,
 399             const char *, uint16_t, struct tmpfs_dirent **);
 400 void    tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *);
 401 int     tmpfs_alloc_vp(struct mount *, struct tmpfs_node *, int,
 402             struct vnode **);
 403 void    tmpfs_free_vp(struct vnode *);
 404 int     tmpfs_alloc_file(struct vnode *, struct vnode **, struct vattr *,
 405             struct namecache *, struct ucred *, char *);
 406 void    tmpfs_dir_attach(struct tmpfs_node *, struct tmpfs_dirent *);
 407 void    tmpfs_dir_detach(struct tmpfs_node *, struct tmpfs_dirent *);
 408 struct tmpfs_dirent *   tmpfs_dir_lookup(struct tmpfs_node *node,
 409                             struct tmpfs_node *f,
 410                             struct namecache *ncp);
 411 int     tmpfs_dir_getdotdent(struct tmpfs_node *, struct uio *);
 412 int     tmpfs_dir_getdotdotdent(struct tmpfs_mount *,
 413                             struct tmpfs_node *, struct uio *);
 414 struct tmpfs_dirent *   tmpfs_dir_lookupbycookie(struct tmpfs_node *, off_t);
 415 int     tmpfs_dir_getdents(struct tmpfs_node *, struct uio *, off_t *);
 416 int     tmpfs_reg_resize(struct vnode *, off_t, int);
 417 int     tmpfs_chflags(struct vnode *, int, struct ucred *);
 418 int     tmpfs_chmod(struct vnode *, mode_t, struct ucred *);
 419 int     tmpfs_chown(struct vnode *, uid_t, gid_t, struct ucred *);
 420 int     tmpfs_chsize(struct vnode *, u_quad_t, struct ucred *);
 421 int     tmpfs_chtimes(struct vnode *, struct timespec *, struct timespec *,
 422             int, struct ucred *);
 423 void    tmpfs_itimes(struct vnode *, const struct timespec *,
 424             const struct timespec *);
 425
 426 void    tmpfs_update(struct vnode *);
 427 int     tmpfs_truncate(struct vnode *, off_t);
 428 boolean_t tmpfs_node_ctor(void *obj, void *privdata, int flags);
 429
 430 /* --------------------------------------------------------------------- */
 431
 432 /*
 433  * Convenience macros to simplify some logical expressions.
 434  */
 435 #define IMPLIES(a, b) (!(a) || (b))
 436 #define IFF(a, b) (IMPLIES(a, b) && IMPLIES(b, a))
 437
 438 /* --------------------------------------------------------------------- */
 439
 440 /*
 441  * Checks that the directory entry pointed by 'de' matches the name 'name'
 442  * with a length of 'len'.
 443  */
 444 #define TMPFS_DIRENT_MATCHES(de, name, len) \
 445     (de->td_namelen == (uint16_t)len && \
 446     bcmp((de)->td_name, (name), (de)->td_namelen) == 0)
 447
 448 /* --------------------------------------------------------------------- */
 449
 450 /*
 451  * Ensures that the node pointed by 'node' is a directory and that its
 452  * contents are consistent with respect to directories.
 453  */
 454 #define TMPFS_VALIDATE_DIR(node) do {   \
 455     KKASSERT((node)->tn_type == VDIR);  \
 456     KKASSERT((node)->tn_size % sizeof(struct tmpfs_dirent) == 0); \
 457 } while(0)
 458
 459 /* --------------------------------------------------------------------- */
 460
 461 /*
 462  * Macros/functions to convert from generic data structures to tmpfs
 463  * specific ones.  Kernel code use VP_TO_TMPFS_NODE() instead of VTOI().
 464  */
 465
 466 static inline
 467 struct tmpfs_mount *
 468 VFS_TO_TMPFS(struct mount *mp)
 469 {
 470         struct tmpfs_mount *tmp;
 471
 472         KKASSERT((mp) != NULL && (mp)->mnt_data != NULL);
 473         tmp = (struct tmpfs_mount *)(mp)->mnt_data;
 474         return tmp;
 475 }
 476
 477 static inline
 478 struct tmpfs_node *
 479 VP_TO_TMPFS_NODE(struct vnode *vp)
 480 {
 481         struct tmpfs_node *node;
 482
 483         KKASSERT((vp) != NULL && (vp)->v_data != NULL);
 484         node = (struct tmpfs_node *)vp->v_data;
 485         return node;
 486 }
 487
 488 static inline
 489 struct tmpfs_node *
 490 VP_TO_TMPFS_DIR(struct vnode *vp)
 491 {
 492         struct tmpfs_node *node;
 493
 494         node = VP_TO_TMPFS_NODE(vp);
 495         TMPFS_VALIDATE_DIR(node);
 496         return node;
 497 }
 498
 499 /* --------------------------------------------------------------------- */
 500 /*
 501  * buffer cache size
 502  */
 503 #define TMPFS_BLKSIZE   16384                   /* buffer cache size*/
 504 #define TMPFS_BLKMASK   (TMPFS_BLKSIZE - 1)
 505 #define TMPFS_BLKMASK64 ((off_t)(TMPFS_BLKSIZE - 1))
 506 #endif /* _KERNEL */
 507
 508 #endif /* _VFS_TMPFS_TMPFS_H_ */