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 virtual memory to store file data and
  37  * metadata efficiently. It does not follow the structure of an on-disk
  38  * file system because it simply does not need to. Instead, it uses
  39  * memory-specific data structures and algorithms to automatically
  40  * allocate and release resources.
  41  */
  42
  43 #include <sys/conf.h>
  44 #include <sys/param.h>
  45 #include <sys/limits.h>
  46 #include <sys/lock.h>
  47 #include <sys/mutex.h>
  48 #include <sys/kernel.h>
  49 #include <sys/stat.h>
  50 #include <sys/systm.h>
  51 #include <sys/sysctl.h>
  52 #include <sys/objcache.h>
  53
  54 #include <vm/vm.h>
  55 #include <vm/vm_object.h>
  56 #include <vm/vm_param.h>
  57
  58 #if 0
  59 #include <vfs/tmpfs/tmpfs.h>
  60 #endif
  61 #include "tmpfs.h"
  62 #include <vfs/tmpfs/tmpfs_vnops.h>
  63 #include <vfs/tmpfs/tmpfs_mount.h>
  64
  65 /*
  66  * Default permission for root node
  67  */
  68 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
  69
  70 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount 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 boolean_t
  82 tmpfs_node_ctor(void *obj, void *privdata, int flags)
  83 {
  84         struct tmpfs_node *node = obj;
  85
  86         node->tn_gen++;
  87         node->tn_size = 0;
  88         node->tn_status = 0;
  89         node->tn_flags = 0;
  90         node->tn_links = 0;
  91         node->tn_vnode = NULL;
  92         node->tn_vpstate = TMPFS_VNODE_WANT;
  93         bzero(&node->tn_spec, sizeof(node->tn_spec));
  94
  95         return (TRUE);
  96 }
  97
  98 static void
  99 tmpfs_node_dtor(void *obj, void *privdata)
 100 {
 101         struct tmpfs_node *node = (struct tmpfs_node *)obj;
 102         node->tn_type = VNON;
 103         node->tn_vpstate = TMPFS_VNODE_DOOMED;
 104 }
 105
 106 static void *
 107 tmpfs_node_init(void *args, int flags)
 108 {
 109         struct tmpfs_node *node = objcache_malloc_alloc(args, flags);
 110         if (node == NULL)
 111                 return (NULL);
 112         node->tn_id = 0;
 113
 114         lockinit(&node->tn_interlock, "tmpfs node interlock", 0, LK_CANRECURSE);
 115         node->tn_gen = karc4random();
 116
 117         return node;
 118 }
 119
 120 static void
 121 tmpfs_node_fini(void *obj, void *args)
 122 {
 123         struct tmpfs_node *node = (struct tmpfs_node *)obj;
 124         lockuninit(&node->tn_interlock);
 125         objcache_malloc_free(obj, args);
 126 }
 127
 128 static int
 129 tmpfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
 130 {
 131         struct tmpfs_mount *tmp;
 132         struct tmpfs_node *root;
 133         struct tmpfs_mount_info args;
 134         vm_pindex_t pages;
 135         vm_pindex_t pages_limit;
 136         ino_t nodes;
 137         u_int64_t       maxfsize;
 138         int error;
 139         /* Size counters. */
 140         ino_t   nodes_max;
 141         off_t   size_max;
 142         size_t  maxfsize_max;
 143         size_t  size;
 144
 145         /* Root node attributes. */
 146         uid_t   root_uid = cred->cr_uid;
 147         gid_t   root_gid = cred->cr_gid;
 148         mode_t  root_mode = (VREAD | VWRITE);
 149
 150         if (mp->mnt_flag & MNT_UPDATE) {
 151                 /* XXX: There is no support yet to update file system
 152                  * settings.  Should be added. */
 153
 154                 return EOPNOTSUPP;
 155         }
 156
 157         /*
 158          * mount info
 159          */
 160         bzero(&args, sizeof(args));
 161         size_max  = 0;
 162         nodes_max = 0;
 163         maxfsize_max = 0;
 164
 165         if (path) {
 166                 if (data) {
 167                         error = copyin(data, &args, sizeof(args));
 168                         if (error)
 169                                 return (error);
 170                 }
 171                 size_max = args.ta_size_max;
 172                 nodes_max = args.ta_nodes_max;
 173                 maxfsize_max = args.ta_maxfsize_max;
 174                 root_uid = args.ta_root_uid;
 175                 root_gid = args.ta_root_gid;
 176                 root_mode = args.ta_root_mode;
 177         }
 178
 179         /*
 180          * If mount by non-root, then verify that user has necessary
 181          * permissions on the device.
 182          */
 183         if (cred->cr_uid != 0) {
 184                 root_mode = VREAD;
 185                 if ((mp->mnt_flag & MNT_RDONLY) == 0)
 186                         root_mode |= VWRITE;
 187         }
 188
 189         pages_limit = vm_swap_max + vmstats.v_page_count / 2;
 190
 191         if (size_max == 0) {
 192                 pages = pages_limit / 2;
 193         } else if (size_max < PAGE_SIZE) {
 194                 pages = 1;
 195         } else if (OFF_TO_IDX(size_max) > pages_limit) {
 196                 /*
 197                  * do not force pages = pages_limit for this case, otherwise
 198                  * we might not honor tmpfs size requests from /etc/fstab
 199                  * during boot because they are mounted prior to swap being
 200                  * turned on.
 201                  */
 202                 pages = OFF_TO_IDX(size_max);
 203         } else {
 204                 pages = OFF_TO_IDX(size_max);
 205         }
 206
 207         if (nodes_max == 0)
 208                 nodes = 3 + pages * PAGE_SIZE / 1024;
 209         else if (nodes_max < 3)
 210                 nodes = 3;
 211         else if (nodes_max > pages)
 212                 nodes = pages;
 213         else
 214                 nodes = nodes_max;
 215
 216         maxfsize = 0x7FFFFFFFFFFFFFFFLLU - TMPFS_BLKSIZE;
 217         if (maxfsize_max != 0 && maxfsize > maxfsize_max)
 218                 maxfsize = maxfsize_max;
 219
 220         /* Allocate the tmpfs mount structure and fill it. */
 221         tmp = kmalloc(sizeof(*tmp), M_TMPFSMNT, M_WAITOK | M_ZERO);
 222
 223         tmp->tm_mount = mp;
 224         tmp->tm_nodes_max = nodes;
 225         tmp->tm_nodes_inuse = 0;
 226         tmp->tm_maxfilesize = maxfsize;
 227         LIST_INIT(&tmp->tm_nodes_used);
 228
 229         tmp->tm_pages_max = pages;
 230         tmp->tm_pages_used = 0;
 231
 232         kmalloc_create(&tmp->tm_node_zone, "tmpfs node");
 233         kmalloc_create(&tmp->tm_dirent_zone, "tmpfs dirent");
 234         kmalloc_create(&tmp->tm_name_zone, "tmpfs name zone");
 235
 236         kmalloc_raise_limit(tmp->tm_node_zone, sizeof(struct tmpfs_node) *
 237                             tmp->tm_nodes_max);
 238
 239         tmp->tm_node_zone_malloc_args.objsize = sizeof(struct tmpfs_node);
 240         tmp->tm_node_zone_malloc_args.mtype = tmp->tm_node_zone;
 241
 242         tmp->tm_dirent_zone_malloc_args.objsize = sizeof(struct tmpfs_dirent);
 243         tmp->tm_dirent_zone_malloc_args.mtype = tmp->tm_dirent_zone;
 244
 245         tmp->tm_dirent_pool =  objcache_create( "tmpfs dirent cache",
 246             0, 0,
 247             NULL, NULL, NULL,
 248             objcache_malloc_alloc, objcache_malloc_free,
 249             &tmp->tm_dirent_zone_malloc_args);
 250         tmp->tm_node_pool = objcache_create( "tmpfs node cache",
 251             0, 0,
 252             tmpfs_node_ctor, tmpfs_node_dtor, NULL,
 253             tmpfs_node_init, tmpfs_node_fini,
 254             &tmp->tm_node_zone_malloc_args);
 255
 256         tmp->tm_ino = TMPFS_ROOTINO;
 257
 258         /* Allocate the root node. */
 259         error = tmpfs_alloc_node(tmp, VDIR, root_uid, root_gid,
 260                                  root_mode & ALLPERMS, NULL,
 261                                  VNOVAL, VNOVAL, &root);
 262
 263         /*
 264          * We are backed by swap, set snocache chflags flag so we
 265          * don't trip over swapcache.
 266          */
 267         root->tn_flags = SF_NOCACHE;
 268
 269         if (error != 0 || root == NULL) {
 270             objcache_destroy(tmp->tm_node_pool);
 271             objcache_destroy(tmp->tm_dirent_pool);
 272             kfree(tmp, M_TMPFSMNT);
 273             return error;
 274         }
 275         KASSERT(root->tn_id == TMPFS_ROOTINO,
 276                 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
 277
 278         ++root->tn_links;       /* prevent destruction */
 279         tmp->tm_root = root;
 280
 281         mp->mnt_flag |= MNT_LOCAL;
 282         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
 283         mp->mnt_kern_flag |= MNTK_NOMSYNC;
 284         mp->mnt_kern_flag |= MNTK_THR_SYNC;     /* new vsyncscan semantics */
 285         mp->mnt_kern_flag |= MNTK_QUICKHALT;    /* no teardown needed on halt */
 286         mp->mnt_data = (qaddr_t)tmp;
 287         vfs_getnewfsid(mp);
 288
 289         vfs_add_vnodeops(mp, &tmpfs_vnode_vops, &mp->mnt_vn_norm_ops);
 290         vfs_add_vnodeops(mp, &tmpfs_fifo_vops, &mp->mnt_vn_fifo_ops);
 291
 292         copystr("tmpfs", mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
 293         bzero(mp->mnt_stat.f_mntfromname +size, MNAMELEN - size);
 294         bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
 295         copyinstr(path, mp->mnt_stat.f_mntonname,
 296                   sizeof(mp->mnt_stat.f_mntonname) -1,
 297                   &size);
 298
 299         tmpfs_statfs(mp, &mp->mnt_stat, cred);
 300
 301         return 0;
 302 }
 303
 304 /* --------------------------------------------------------------------- */
 305
 306 /* ARGSUSED2 */
 307 static int
 308 tmpfs_unmount(struct mount *mp, int mntflags)
 309 {
 310         int error;
 311         int flags = 0;
 312         struct tmpfs_mount *tmp;
 313         struct tmpfs_node *node;
 314         struct vnode *vp;
 315         int isok;
 316
 317         tmp = VFS_TO_TMPFS(mp);
 318         TMPFS_LOCK(tmp);
 319
 320         /* Handle forced unmounts. */
 321         if (mntflags & MNT_FORCE)
 322                 flags |= FORCECLOSE;
 323
 324         /*
 325          * Finalize all pending I/O.  In the case of tmpfs we want
 326          * to throw all the data away so clean out the buffer cache
 327          * and vm objects before calling vflush().
 328          */
 329         LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
 330                 /*
 331                  * tn_links is mnt_token protected
 332                  */
 333                 TMPFS_NODE_LOCK(node);
 334                 ++node->tn_links;
 335                 TMPFS_NODE_UNLOCK(node);
 336
 337                 while (node->tn_type == VREG && node->tn_vnode) {
 338                         vp = node->tn_vnode;
 339                         vhold(vp);
 340                         lwkt_yield();
 341
 342                         /*
 343                          * vx_get/vx_put and tmpfs_truncate may block,
 344                          * releasing the tmpfs mountpoint token.
 345                          *
 346                          * Make sure the lock order is correct.
 347                          */
 348                         vx_get(vp);             /* held vnode */
 349                         TMPFS_NODE_LOCK(node);
 350                         if (node->tn_vnode == vp) {
 351                                 tmpfs_truncate(vp, 0);
 352                                 isok = 1;
 353                         } else {
 354                                 isok = 0;
 355                         }
 356                         TMPFS_NODE_UNLOCK(node);
 357                         vx_put(vp);
 358                         vdrop(vp);
 359                         if (isok)
 360                                 break;
 361                         /* retry */
 362                 }
 363
 364                 TMPFS_NODE_LOCK(node);
 365                 --node->tn_links;
 366                 TMPFS_NODE_UNLOCK(node);
 367         }
 368
 369         /*
 370          * Flush all vnodes on the unmount.
 371          *
 372          * If we fail to flush, we cannot unmount, but all the nodes have
 373          * already been truncated. Erroring out is the best we can do.
 374          */
 375         error = vflush(mp, 0, flags);
 376         if (error != 0) {
 377                 TMPFS_UNLOCK(tmp);
 378                 return (error);
 379         }
 380
 381         /*
 382          * First pass get rid of all the directory entries and
 383          * vnode associations.  This will also destroy the
 384          * directory topology and should drop all link counts
 385          * to 0 except for the root.
 386          *
 387          * No vnodes should remain after the vflush above.
 388          */
 389         LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
 390                 lwkt_yield();
 391
 392                 TMPFS_NODE_LOCK(node);
 393                 ++node->tn_links;
 394                 if (node->tn_type == VDIR) {
 395                         struct tmpfs_dirent *de;
 396
 397                         while ((de = RB_ROOT(&node->tn_dir.tn_dirtree)) != NULL)                        {
 398                                 tmpfs_dir_detach(node, de);
 399                                 tmpfs_free_dirent(tmp, de);
 400                         }
 401                 }
 402                 KKASSERT(node->tn_vnode == NULL);
 403
 404                 --node->tn_links;
 405                 TMPFS_NODE_UNLOCK(node);
 406         }
 407
 408         /*
 409          * Allow the root node to be destroyed by dropping the link count
 410          * we bumped in the mount code.
 411          */
 412         KKASSERT(tmp->tm_root);
 413         TMPFS_NODE_LOCK(tmp->tm_root);
 414         --tmp->tm_root->tn_links;
 415         TMPFS_NODE_UNLOCK(tmp->tm_root);
 416
 417         /*
 418          * At this point all nodes, including the root node, should have a
 419          * link count of 0.  The root is not necessarily going to be last.
 420          */
 421         while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
 422                 if (node->tn_links)
 423                         panic("tmpfs: Dangling nodes during umount (%p)!\n",
 424                               node);
 425
 426                 TMPFS_NODE_LOCK(node);
 427                 tmpfs_free_node(tmp, node);
 428                 /* eats lock */
 429                 lwkt_yield();
 430         }
 431         KKASSERT(tmp->tm_root == NULL);
 432
 433         objcache_destroy(tmp->tm_dirent_pool);
 434         objcache_destroy(tmp->tm_node_pool);
 435
 436         kmalloc_destroy(&tmp->tm_name_zone);
 437         kmalloc_destroy(&tmp->tm_dirent_zone);
 438         kmalloc_destroy(&tmp->tm_node_zone);
 439
 440         tmp->tm_node_zone = tmp->tm_dirent_zone = NULL;
 441
 442         KKASSERT(tmp->tm_pages_used == 0);
 443         KKASSERT(tmp->tm_nodes_inuse == 0);
 444
 445         TMPFS_UNLOCK(tmp);
 446
 447         /* Throw away the tmpfs_mount structure. */
 448         kfree(tmp, M_TMPFSMNT);
 449         mp->mnt_data = NULL;
 450
 451         mp->mnt_flag &= ~MNT_LOCAL;
 452         return 0;
 453 }
 454
 455 /* --------------------------------------------------------------------- */
 456
 457 static int
 458 tmpfs_root(struct mount *mp, struct vnode **vpp)
 459 {
 460         struct tmpfs_mount *tmp;
 461         int error;
 462
 463         tmp = VFS_TO_TMPFS(mp);
 464         if (tmp->tm_root == NULL) {
 465                 kprintf("tmpfs_root: called without root node %p\n", mp);
 466                 print_backtrace(-1);
 467                 *vpp = NULL;
 468                 error = EINVAL;
 469         } else {
 470                 error = tmpfs_alloc_vp(mp, tmp->tm_root, LK_EXCLUSIVE, vpp);
 471                 (*vpp)->v_flag |= VROOT;
 472                 (*vpp)->v_type = VDIR;
 473         }
 474         return error;
 475 }
 476
 477 /* --------------------------------------------------------------------- */
 478
 479 static int
 480 tmpfs_fhtovp(struct mount *mp, struct vnode *rootvp, struct fid *fhp,
 481              struct vnode **vpp)
 482 {
 483         boolean_t found;
 484         struct tmpfs_fid *tfhp;
 485         struct tmpfs_mount *tmp;
 486         struct tmpfs_node *node;
 487         int rc;
 488
 489         tmp = VFS_TO_TMPFS(mp);
 490
 491         tfhp = (struct tmpfs_fid *) fhp;
 492         if (tfhp->tf_len != sizeof(struct tmpfs_fid))
 493                 return EINVAL;
 494
 495         rc = EINVAL;
 496         found = FALSE;
 497
 498         TMPFS_LOCK(tmp);
 499         LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
 500                 if (node->tn_id == tfhp->tf_id &&
 501                     node->tn_gen == tfhp->tf_gen) {
 502                         found = TRUE;
 503                         break;
 504                 }
 505         }
 506
 507         if (found)
 508                 rc = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
 509
 510         TMPFS_UNLOCK(tmp);
 511
 512         return (rc);
 513 }
 514
 515 /* --------------------------------------------------------------------- */
 516
 517 /* ARGSUSED2 */
 518 static int
 519 tmpfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 520 {
 521         fsfilcnt_t freenodes;
 522         struct tmpfs_mount *tmp;
 523
 524         tmp = VFS_TO_TMPFS(mp);
 525
 526         TMPFS_LOCK(tmp);
 527         sbp->f_iosize = PAGE_SIZE;
 528         sbp->f_bsize = PAGE_SIZE;
 529
 530         sbp->f_blocks = tmp->tm_pages_max;
 531         sbp->f_bavail = tmp->tm_pages_max - tmp->tm_pages_used;
 532         sbp->f_bfree = sbp->f_bavail;
 533
 534         freenodes = tmp->tm_nodes_max - tmp->tm_nodes_inuse;
 535
 536         sbp->f_files = freenodes + tmp->tm_nodes_inuse;
 537         sbp->f_ffree = freenodes;
 538         sbp->f_owner = tmp->tm_root->tn_uid;
 539
 540         TMPFS_UNLOCK(tmp);
 541
 542         return 0;
 543 }
 544
 545 /* --------------------------------------------------------------------- */
 546
 547 static int
 548 tmpfs_vptofh(struct vnode *vp, struct fid *fhp)
 549 {
 550         struct tmpfs_node *node;
 551         struct tmpfs_fid tfh;
 552         node = VP_TO_TMPFS_NODE(vp);
 553         memset(&tfh, 0, sizeof(tfh));
 554         tfh.tf_len = sizeof(struct tmpfs_fid);
 555         tfh.tf_gen = node->tn_gen;
 556         tfh.tf_id = node->tn_id;
 557         memcpy(fhp, &tfh, sizeof(tfh));
 558         return (0);
 559 }
 560
 561 /* --------------------------------------------------------------------- */
 562
 563 static int
 564 tmpfs_checkexp(struct mount *mp, struct sockaddr *nam, int *exflagsp,
 565                struct ucred **credanonp)
 566 {
 567         struct tmpfs_mount *tmp;
 568         struct netcred *nc;
 569
 570         tmp = (struct tmpfs_mount *) mp->mnt_data;
 571         nc = vfs_export_lookup(mp, &tmp->tm_export, nam);
 572         if (nc == NULL)
 573                 return (EACCES);
 574
 575         *exflagsp = nc->netc_exflags;
 576         *credanonp = &nc->netc_anon;
 577
 578         return (0);
 579 }
 580
 581 /* --------------------------------------------------------------------- */
 582
 583 /*
 584  * tmpfs vfs operations.
 585  */
 586
 587 static struct vfsops tmpfs_vfsops = {
 588         .vfs_mount =                    tmpfs_mount,
 589         .vfs_unmount =                  tmpfs_unmount,
 590         .vfs_root =                     tmpfs_root,
 591         .vfs_statfs =                   tmpfs_statfs,
 592         .vfs_fhtovp =                   tmpfs_fhtovp,
 593         .vfs_vptofh =                   tmpfs_vptofh,
 594         .vfs_checkexp =                 tmpfs_checkexp,
 595 };
 596
 597 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_MPSAFE);
 598 MODULE_VERSION(tmpfs, 1);