sys/vfs/tmpfs/tmpfs_vfsops.c

   1 /*      $NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $       */
   2
   3 /*-
   4  * Copyright (c) 2005 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
  33 /*
  34  * Efficient memory file system.
  35  *
  36  * tmpfs is a file system that uses NetBSD's virtual memory sub-system
  37  * (the well-known UVM) to store file data and metadata in an efficient
  38  * way.  This means that it does not follow the structure of an on-disk
  39  * file system because it simply does not need to.  Instead, it uses
  40  * memory-specific data structures and algorithms to automatically
  41  * allocate and release resources.
  42  */
  43 #include <sys/cdefs.h>
  44 #include <sys/conf.h>
  45 #include <sys/param.h>
  46 #include <sys/limits.h>
  47 #include <sys/lock.h>
  48 #include <sys/mutex.h>
  49 #include <sys/kernel.h>
  50 #include <sys/stat.h>
  51 #include <sys/systm.h>
  52 #include <sys/sysctl.h>
  53 #include <sys/objcache.h>
  54
  55 #include <vm/vm.h>
  56 #include <vm/vm_object.h>
  57 #include <vm/vm_param.h>
  58
  59 #include <vfs/tmpfs/tmpfs.h>
  60 #include <vfs/tmpfs/tmpfs_vnops.h>
  61
  62 /*
  63  * Default permission for root node
  64  */
  65 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
  66
  67 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
  68 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
  69 MALLOC_DEFINE(M_TMPFS_DIRENT, "tmpfs dirent", "tmpfs dirent structures");
  70 MALLOC_DEFINE(M_TMPFS_NODE, "tmpfs node", "tmpfs node structures");
  71
  72 /* --------------------------------------------------------------------- */
  73
  74 static int      tmpfs_mount(struct mount *, char *, caddr_t, struct ucred *);
  75 static int      tmpfs_unmount(struct mount *, int);
  76 static int      tmpfs_root(struct mount *, struct vnode **);
  77 static int      tmpfs_fhtovp(struct mount *, struct vnode *, struct fid *, struct vnode **);
  78 static int      tmpfs_statfs(struct mount *, struct statfs *, struct ucred *cred);
  79
  80 /* --------------------------------------------------------------------- */
  81
  82 #define SWI_MAXMIB      3
  83 static u_int
  84 get_swpgtotal(void)
  85 {
  86         struct swdevt swinfo;
  87         char *sname = "vm.swap_info";
  88         int soid[SWI_MAXMIB], oid[2];
  89         u_int unswdev, total, dmmax, nswapdev;
  90         size_t mibi, len;
  91
  92         total = 0;
  93
  94         len = sizeof(dmmax);
  95         if (kernel_sysctlbyname("vm.dmmax", &dmmax, &len,
  96                                 NULL, 0, NULL) != 0)
  97                 return total;
  98
  99         len = sizeof(nswapdev);
 100         if (kernel_sysctlbyname("vm.nswapdev", &nswapdev, &len,
 101                                 NULL, 0, NULL) != 0)
 102                 return total;
 103
 104         mibi = (SWI_MAXMIB - 1) * sizeof(int);
 105         oid[0] = 0;
 106         oid[1] = 3;
 107
 108         if (kernel_sysctl(oid, 2,
 109                         soid, &mibi, (void *)sname, strlen(sname),
 110                         NULL) != 0)
 111                 return total;
 112
 113         mibi = (SWI_MAXMIB - 1);
 114         for (unswdev = 0; unswdev < nswapdev; ++unswdev) {
 115                 soid[mibi] = unswdev;
 116                 len = sizeof(struct swdevt);
 117                 if (kernel_sysctl(soid, mibi + 1, &swinfo, &len, NULL, 0,
 118                                 NULL) != 0)
 119                         return total;
 120                 if (len == sizeof(struct swdevt))
 121                         total += (swinfo.sw_nblks - dmmax);
 122         }
 123
 124         return total;
 125 }
 126
 127 /* --------------------------------------------------------------------- */
 128 static int
 129 tmpfs_node_ctor(void *obj, void *privdata, int flags)
 130 {
 131         struct tmpfs_node *node = (struct tmpfs_node *)obj;
 132
 133         node->tn_gen++;
 134         node->tn_size = 0;
 135         node->tn_status = 0;
 136         node->tn_flags = 0;
 137         node->tn_links = 0;
 138         node->tn_vnode = NULL;
 139         node->tn_vpstate = TMPFS_VNODE_WANT;
 140
 141         return (1);
 142 }
 143
 144 static void
 145 tmpfs_node_dtor(void *obj, void *privdata)
 146 {
 147         struct tmpfs_node *node = (struct tmpfs_node *)obj;
 148         node->tn_type = VNON;
 149         node->tn_vpstate = TMPFS_VNODE_DOOMED;
 150 }
 151
 152 static void*
 153 tmpfs_node_init(void *args, int flags)
 154 {
 155         struct tmpfs_node *node = (struct tmpfs_node *)objcache_malloc_alloc(args, flags);
 156         node->tn_id = 0;
 157
 158         lockinit(&node->tn_interlock, "tmpfs node interlock", 0, LK_CANRECURSE);
 159         node->tn_gen = karc4random();
 160
 161         return node;
 162 }
 163
 164 static void
 165 tmpfs_node_fini(void *obj, void *args)
 166 {
 167         struct tmpfs_node *node = (struct tmpfs_node *)obj;
 168         lockuninit(&node->tn_interlock);
 169         objcache_malloc_free(obj, args);
 170 }
 171
 172 struct objcache_malloc_args tmpfs_dirent_pool_malloc_args =
 173         { sizeof(struct tmpfs_dirent), M_TMPFS_DIRENT };
 174 struct objcache_malloc_args tmpfs_node_pool_malloc_args =
 175         { sizeof(struct tmpfs_node), M_TMPFS_NODE };
 176
 177 static int
 178 tmpfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
 179 {
 180         struct tmpfs_mount *tmp;
 181         struct tmpfs_node *root;
 182         size_t pages, mem_size;
 183         ino_t nodes;
 184         int error;
 185         /* Size counters. */
 186         ino_t   nodes_max = 0;
 187         size_t  size_max = 0;
 188         size_t size;
 189
 190         /* Root node attributes. */
 191         uid_t   root_uid = cred->cr_uid;
 192         gid_t   root_gid = cred->cr_gid;
 193         mode_t  root_mode = (VREAD | VWRITE);
 194
 195         if (mp->mnt_flag & MNT_UPDATE) {
 196                 /* XXX: There is no support yet to update file system
 197                  * settings.  Should be added. */
 198
 199                 return EOPNOTSUPP;
 200         }
 201
 202         kprintf("WARNING: TMPFS is considered to be a highly experimental "
 203             "feature in DragonFly.\n");
 204
 205         /* Do not allow mounts if we do not have enough memory to preserve
 206          * the minimum reserved pages. */
 207         mem_size = vmstats.v_free_count + vmstats.v_inactive_count + get_swpgtotal();
 208         mem_size -= mem_size > vmstats.v_wire_count ? vmstats.v_wire_count : mem_size;
 209         if (mem_size < TMPFS_PAGES_RESERVED)
 210                 return ENOSPC;
 211
 212         /*
 213          * If mount by non-root, then verify that user has necessary
 214          * permissions on the device.
 215          */
 216         if (cred->cr_uid != 0) {
 217                 root_mode = VREAD;
 218                 if ((mp->mnt_flag & MNT_RDONLY) == 0)
 219                         root_mode |= VWRITE;
 220         }
 221
 222         /* Get the maximum number of memory pages this file system is
 223          * allowed to use, based on the maximum size the user passed in
 224          * the mount structure.  A value of zero is treated as if the
 225          * maximum available space was requested. */
 226         if (size_max < PAGE_SIZE || size_max >= SIZE_MAX)
 227                 pages = SIZE_MAX;
 228         else
 229                 pages = howmany(size_max, PAGE_SIZE);
 230         KKASSERT(pages > 0);
 231
 232         if (nodes_max <= 3)
 233                 nodes = 3 + pages * PAGE_SIZE / 1024;
 234         else
 235                 nodes = nodes_max;
 236         KKASSERT(nodes >= 3);
 237
 238         /* Allocate the tmpfs mount structure and fill it. */
 239         tmp = (struct tmpfs_mount *)kmalloc(sizeof(struct tmpfs_mount),
 240             M_TMPFSMNT, M_WAITOK | M_ZERO);
 241
 242         lockinit(&(tmp->allnode_lock), "tmpfs allnode lock", 0, LK_CANRECURSE);
 243         tmp->tm_nodes_max = nodes;
 244         tmp->tm_nodes_inuse = 0;
 245         tmp->tm_maxfilesize = (u_int64_t)(vmstats.v_page_count + get_swpgtotal()) * PAGE_SIZE;
 246         LIST_INIT(&tmp->tm_nodes_used);
 247
 248         tmp->tm_pages_max = pages;
 249         tmp->tm_pages_used = 0;
 250         tmp->tm_dirent_pool =  objcache_create( "tmpfs dirent cache",
 251             0, 0,
 252             NULL, NULL, NULL,
 253             objcache_malloc_alloc, objcache_malloc_free,
 254             &tmpfs_dirent_pool_malloc_args);
 255         tmp->tm_node_pool = objcache_create( "tmpfs node cache",
 256             0, 0,
 257             tmpfs_node_ctor, tmpfs_node_dtor, NULL,
 258             tmpfs_node_init, tmpfs_node_fini,
 259             &tmpfs_node_pool_malloc_args);
 260
 261         /* Allocate the root node. */
 262         error = tmpfs_alloc_node(tmp, VDIR, root_uid,
 263             root_gid, root_mode & ALLPERMS, NULL, NULL,
 264             VNOVAL, VNOVAL, &root);
 265
 266         if (error != 0 || root == NULL) {
 267             objcache_destroy(tmp->tm_node_pool);
 268             objcache_destroy(tmp->tm_dirent_pool);
 269             kfree(tmp, M_TMPFSMNT);
 270             return error;
 271         }
 272         KASSERT(root->tn_id >= 0, ("tmpfs root with invalid ino: %d", (int)root->tn_id));
 273         tmp->tm_root = root;
 274
 275         mp->mnt_flag |= MNT_LOCAL;
 276         mp->mnt_kern_flag |= MNTK_RD_MPSAFE | MNTK_WR_MPSAFE | MNTK_GA_MPSAFE  |
 277                              MNTK_IN_MPSAFE | MNTK_SG_MPSAFE;
 278         mp->mnt_data = (qaddr_t)tmp;
 279         vfs_getnewfsid(mp);
 280
 281
 282         vfs_add_vnodeops(mp, &tmpfs_vnode_vops, &mp->mnt_vn_norm_ops);
 283         vfs_add_vnodeops(mp, &tmpfs_fifo_vops, &mp->mnt_vn_fifo_ops);
 284
 285         copystr("tmpfs", mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
 286         bzero(mp->mnt_stat.f_mntfromname +size, MNAMELEN - size);
 287         bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
 288         copyinstr(path, mp->mnt_stat.f_mntonname,
 289                   sizeof(mp->mnt_stat.f_mntonname) -1,
 290                   &size);
 291
 292         tmpfs_statfs(mp, &mp->mnt_stat, cred);
 293
 294         return 0;
 295 }
 296
 297 /* --------------------------------------------------------------------- */
 298
 299 /* ARGSUSED2 */
 300 static int
 301 tmpfs_unmount(struct mount *mp, int mntflags)
 302 {
 303         int error;
 304         int flags = 0;
 305         struct tmpfs_mount *tmp;
 306         struct tmpfs_node *node;
 307         struct vnode *vp;
 308
 309         /* Handle forced unmounts. */
 310         if (mntflags & MNT_FORCE)
 311                 flags |= FORCECLOSE;
 312
 313         /* Finalize all pending I/O. */
 314         error = vflush(mp, 0, flags);
 315         if (error != 0)
 316                 return error;
 317
 318         tmp = VFS_TO_TMPFS(mp);
 319
 320         /* Free all associated data.  The loop iterates over the linked list
 321          * we have containing all used nodes.  For each of them that is
 322          * a directory, we free all its directory entries.  Note that after
 323          * freeing a node, it will automatically go to the available list,
 324          * so we will later have to iterate over it to release its items. */
 325         node = LIST_FIRST(&tmp->tm_nodes_used);
 326         while (node != NULL) {
 327                 struct tmpfs_node *next;
 328
 329                 if (node->tn_type == VDIR) {
 330                         struct tmpfs_dirent *de;
 331
 332                         de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
 333                         while (de != NULL) {
 334                                 struct tmpfs_dirent *nde;
 335
 336                                 nde = TAILQ_NEXT(de, td_entries);
 337                                 tmpfs_free_dirent(tmp, de, FALSE);
 338                                 de = nde;
 339                                 node->tn_size -= sizeof(struct tmpfs_dirent);
 340                         }
 341                 }
 342
 343                 next = LIST_NEXT(node, tn_entries);
 344                 vp = node->tn_vnode;
 345                 if (vp != NULL) {
 346                         tmpfs_free_vp(vp);
 347                         vrecycle(vp);
 348                          node->tn_vnode = NULL;
 349                 }
 350                 tmpfs_free_node(tmp, node);
 351                 node = next;
 352         }
 353
 354         objcache_destroy(tmp->tm_dirent_pool);
 355         objcache_destroy(tmp->tm_node_pool);
 356
 357         lockuninit(&tmp->allnode_lock);
 358         KKASSERT(tmp->tm_pages_used == 0);
 359         KKASSERT(tmp->tm_nodes_inuse == 0);
 360
 361         /* Throw away the tmpfs_mount structure. */
 362         kfree(mp->mnt_data, M_TMPFSMNT);
 363         mp->mnt_data = NULL;
 364
 365         mp->mnt_flag &= ~MNT_LOCAL;
 366         return 0;
 367 }
 368
 369 /* --------------------------------------------------------------------- */
 370
 371 static int
 372 tmpfs_root(struct mount *mp, struct vnode **vpp)
 373 {
 374         int error;
 375         error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, LK_EXCLUSIVE, vpp);
 376         (*vpp)->v_flag |= VROOT;
 377         (*vpp)->v_type = VDIR;
 378
 379         return error;
 380 }
 381
 382 /* --------------------------------------------------------------------- */
 383
 384 static int
 385 tmpfs_fhtovp(struct mount *mp, struct vnode *rootvp, struct fid *fhp, struct vnode **vpp)
 386 {
 387         boolean_t found;
 388         struct tmpfs_fid *tfhp;
 389         struct tmpfs_mount *tmp;
 390         struct tmpfs_node *node;
 391
 392         tmp = VFS_TO_TMPFS(mp);
 393
 394         tfhp = (struct tmpfs_fid *)fhp;
 395         if (tfhp->tf_len != sizeof(struct tmpfs_fid))
 396                 return EINVAL;
 397
 398         if (tfhp->tf_id >= tmp->tm_nodes_max)
 399                 return EINVAL;
 400
 401         found = FALSE;
 402
 403         TMPFS_LOCK(tmp);
 404         LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
 405                 if (node->tn_id == tfhp->tf_id &&
 406                     node->tn_gen == tfhp->tf_gen) {
 407                         found = TRUE;
 408                         break;
 409                 }
 410         }
 411         TMPFS_UNLOCK(tmp);
 412
 413         if (found)
 414                 return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp));
 415
 416         return (EINVAL);
 417 }
 418
 419 /* --------------------------------------------------------------------- */
 420
 421 /* ARGSUSED2 */
 422 static int
 423 tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 424 {
 425         fsfilcnt_t freenodes;
 426         struct tmpfs_mount *tmp;
 427
 428         tmp = VFS_TO_TMPFS(mp);
 429
 430         sbp->f_iosize = PAGE_SIZE;
 431         sbp->f_bsize = PAGE_SIZE;
 432
 433         sbp->f_blocks = TMPFS_PAGES_MAX(tmp);
 434         sbp->f_bavail = sbp->f_bfree = TMPFS_PAGES_AVAIL(tmp);
 435
 436         freenodes = MIN(tmp->tm_nodes_max - tmp->tm_nodes_inuse,
 437             TMPFS_PAGES_AVAIL(tmp) * PAGE_SIZE / sizeof(struct tmpfs_node));
 438
 439         sbp->f_files = freenodes + tmp->tm_nodes_inuse;
 440         sbp->f_ffree = freenodes;
 441         /* sbp->f_owner = tmp->tn_uid; */
 442
 443         return 0;
 444 }
 445
 446 /* --------------------------------------------------------------------- */
 447
 448 /*
 449  * tmpfs vfs operations.
 450  */
 451
 452 static struct vfsops tmpfs_vfsops = {
 453         .vfs_mount =                    tmpfs_mount,
 454         .vfs_unmount =                  tmpfs_unmount,
 455         .vfs_root =                     tmpfs_root,
 456         .vfs_statfs =                   tmpfs_statfs,
 457         .vfs_fhtovp =                   tmpfs_fhtovp,
 458         .vfs_sync =                     vfs_stdsync
 459 };
 460
 461 VFS_SET(tmpfs_vfsops, tmpfs, 0);