sys/vfs/hammer2/hammer2_vfsops.c

   1 /*-
   2  * Copyright (c) 2011, 2012 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  */
  34 #include <sys/param.h>
  35 #include <sys/systm.h>
  36 #include <sys/kernel.h>
  37 #include <sys/nlookup.h>
  38 #include <sys/vnode.h>
  39 #include <sys/mount.h>
  40 #include <sys/fcntl.h>
  41 #include <sys/buf.h>
  42 #include <sys/uuid.h>
  43 #include <sys/vfsops.h>
  44 #include <sys/sysctl.h>
  45
  46 #include "hammer2.h"
  47 #include "hammer2_disk.h"
  48 #include "hammer2_mount.h"
  49
  50 struct hammer2_sync_info {
  51         int error;
  52         int waitfor;
  53 };
  54
  55 TAILQ_HEAD(hammer2_mntlist, hammer2_mount);
  56 static struct hammer2_mntlist hammer2_mntlist;
  57 static struct lock hammer2_mntlk;
  58
  59 int hammer2_debug;
  60 int hammer2_cluster_enable = 1;
  61 int hammer2_hardlink_enable = 1;
  62 long hammer2_iod_file_read;
  63 long hammer2_iod_meta_read;
  64 long hammer2_iod_indr_read;
  65 long hammer2_iod_file_write;
  66 long hammer2_iod_meta_write;
  67 long hammer2_iod_indr_write;
  68 long hammer2_iod_volu_write;
  69 long hammer2_ioa_file_read;
  70 long hammer2_ioa_meta_read;
  71 long hammer2_ioa_indr_read;
  72 long hammer2_ioa_file_write;
  73 long hammer2_ioa_meta_write;
  74 long hammer2_ioa_indr_write;
  75 long hammer2_ioa_volu_write;
  76
  77 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
  78
  79 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
  80            &hammer2_debug, 0, "");
  81 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
  82            &hammer2_cluster_enable, 0, "");
  83 SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
  84            &hammer2_hardlink_enable, 0, "");
  85 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
  86            &hammer2_iod_file_read, 0, "");
  87 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
  88            &hammer2_iod_meta_read, 0, "");
  89 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
  90            &hammer2_iod_indr_read, 0, "");
  91 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
  92            &hammer2_iod_file_write, 0, "");
  93 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
  94            &hammer2_iod_meta_write, 0, "");
  95 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
  96            &hammer2_iod_indr_write, 0, "");
  97 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
  98            &hammer2_iod_volu_write, 0, "");
  99 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
 100            &hammer2_ioa_file_read, 0, "");
 101 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
 102            &hammer2_ioa_meta_read, 0, "");
 103 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
 104            &hammer2_ioa_indr_read, 0, "");
 105 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
 106            &hammer2_ioa_file_write, 0, "");
 107 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
 108            &hammer2_ioa_meta_write, 0, "");
 109 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
 110            &hammer2_ioa_indr_write, 0, "");
 111 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
 112            &hammer2_ioa_volu_write, 0, "");
 113
 114 static int hammer2_vfs_init(struct vfsconf *conf);
 115 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
 116                                 struct ucred *cred);
 117 static int hammer2_remount(struct mount *, char *, struct vnode *,
 118                                 struct ucred *);
 119 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
 120 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
 121 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
 122                                 struct ucred *cred);
 123 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 124                                 struct ucred *cred);
 125 static int hammer2_vfs_sync(struct mount *mp, int waitfor);
 126 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
 127                                 ino_t ino, struct vnode **vpp);
 128 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 129                                 struct fid *fhp, struct vnode **vpp);
 130 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 131 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 132                                 int *exflagsp, struct ucred **credanonp);
 133
 134 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
 135 static int hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
 136 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
 137
 138 /*
 139  * HAMMER2 vfs operations.
 140  */
 141 static struct vfsops hammer2_vfsops = {
 142         .vfs_init       = hammer2_vfs_init,
 143         .vfs_sync       = hammer2_vfs_sync,
 144         .vfs_mount      = hammer2_vfs_mount,
 145         .vfs_unmount    = hammer2_vfs_unmount,
 146         .vfs_root       = hammer2_vfs_root,
 147         .vfs_statfs     = hammer2_vfs_statfs,
 148         .vfs_statvfs    = hammer2_vfs_statvfs,
 149         .vfs_vget       = hammer2_vfs_vget,
 150         .vfs_vptofh     = hammer2_vfs_vptofh,
 151         .vfs_fhtovp     = hammer2_vfs_fhtovp,
 152         .vfs_checkexp   = hammer2_vfs_checkexp
 153 };
 154
 155 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
 156
 157 VFS_SET(hammer2_vfsops, hammer2, 0);
 158 MODULE_VERSION(hammer2, 1);
 159
 160 static
 161 int
 162 hammer2_vfs_init(struct vfsconf *conf)
 163 {
 164         int error;
 165
 166         error = 0;
 167
 168         if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
 169                 error = EINVAL;
 170         if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
 171                 error = EINVAL;
 172         if (HAMMER2_ALLOCREF_BYTES != sizeof(struct hammer2_allocref))
 173                 error = EINVAL;
 174         if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
 175                 error = EINVAL;
 176
 177         if (error)
 178                 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
 179
 180         lockinit(&hammer2_mntlk, "mntlk", 0, 0);
 181         TAILQ_INIT(&hammer2_mntlist);
 182
 183         return (error);
 184 }
 185
 186 /*
 187  * Mount or remount HAMMER2 fileystem from physical media
 188  *
 189  *      mountroot
 190  *              mp              mount point structure
 191  *              path            NULL
 192  *              data            <unused>
 193  *              cred            <unused>
 194  *
 195  *      mount
 196  *              mp              mount point structure
 197  *              path            path to mount point
 198  *              data            pointer to argument structure in user space
 199  *                      volume  volume path (device@LABEL form)
 200  *                      hflags  user mount flags
 201  *              cred            user credentials
 202  *
 203  * RETURNS:     0       Success
 204  *              !0      error number
 205  */
 206 static
 207 int
 208 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
 209               struct ucred *cred)
 210 {
 211         struct hammer2_mount_info info;
 212         hammer2_pfsmount_t *pmp;
 213         hammer2_mount_t *hmp;
 214         hammer2_key_t lhc;
 215         struct vnode *devvp;
 216         struct nlookupdata nd;
 217         hammer2_chain_t *parent;
 218         hammer2_chain_t *schain;
 219         hammer2_chain_t *rchain;
 220         char devstr[MNAMELEN];
 221         size_t size;
 222         size_t done;
 223         char *dev;
 224         char *label;
 225         int ronly = 1;
 226         int create_hmp;
 227         int error;
 228
 229         hmp = NULL;
 230         pmp = NULL;
 231         dev = NULL;
 232         label = NULL;
 233         devvp = NULL;
 234
 235         kprintf("hammer2_mount\n");
 236
 237         if (path == NULL) {
 238                 /*
 239                  * Root mount
 240                  */
 241                 return (EOPNOTSUPP);
 242         } else {
 243                 /*
 244                  * Non-root mount or updating a mount
 245                  */
 246                 error = copyin(data, &info, sizeof(info));
 247                 if (error)
 248                         return (error);
 249
 250                 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
 251                 if (error)
 252                         return (error);
 253
 254                 /* Extract device and label */
 255                 dev = devstr;
 256                 label = strchr(devstr, '@');
 257                 if (label == NULL ||
 258                     ((label + 1) - dev) > done) {
 259                         return (EINVAL);
 260                 }
 261                 *label = '\0';
 262                 label++;
 263                 if (*label == '\0')
 264                         return (EINVAL);
 265
 266                 if (mp->mnt_flag & MNT_UPDATE) {
 267                         /* Update mount */
 268                         /* HAMMER2 implements NFS export via mountctl */
 269                         hmp = MPTOHMP(mp);
 270                         devvp = hmp->devvp;
 271                         error = hammer2_remount(mp, path, devvp, cred);
 272                         return error;
 273                 }
 274         }
 275
 276         /*
 277          * New non-root mount
 278          */
 279         /* Lookup name and verify it refers to a block device */
 280         error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
 281         if (error == 0)
 282                 error = nlookup(&nd);
 283         if (error == 0)
 284                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
 285         nlookup_done(&nd);
 286
 287         if (error == 0) {
 288                 if (vn_isdisk(devvp, &error))
 289                         error = vfs_mountedon(devvp);
 290         }
 291
 292         /*
 293          * Determine if the device has already been mounted.  After this
 294          * check hmp will be non-NULL if we are doing the second or more
 295          * hammer2 mounts from the same device.
 296          */
 297         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
 298         TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
 299                 if (hmp->devvp == devvp)
 300                         break;
 301         }
 302
 303         /*
 304          * Open the device if this isn't a secondary mount
 305          */
 306         if (hmp) {
 307                 create_hmp = 0;
 308         } else {
 309                 create_hmp = 1;
 310                 if (error == 0 && vcount(devvp) > 0)
 311                         error = EBUSY;
 312
 313                 /*
 314                  * Now open the device
 315                  */
 316                 if (error == 0) {
 317                         ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 318                         vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
 319                         error = vinvalbuf(devvp, V_SAVE, 0, 0);
 320                         if (error == 0) {
 321                                 error = VOP_OPEN(devvp,
 322                                                  ronly ? FREAD : FREAD | FWRITE,
 323                                                  FSCRED, NULL);
 324                         }
 325                         vn_unlock(devvp);
 326                 }
 327                 if (error && devvp) {
 328                         vrele(devvp);
 329                         devvp = NULL;
 330                 }
 331                 if (error) {
 332                         lockmgr(&hammer2_mntlk, LK_RELEASE);
 333                         return error;
 334                 }
 335         }
 336
 337         /*
 338          * Block device opened successfully, finish initializing the
 339          * mount structure.
 340          *
 341          * From this point on we have to call hammer2_unmount() on failure.
 342          */
 343         pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
 344         mp->mnt_data = (qaddr_t)pmp;
 345         pmp->mp = mp;
 346
 347         if (create_hmp) {
 348                 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
 349                 hmp->ronly = ronly;
 350                 hmp->devvp = devvp;
 351                 kmalloc_create(&hmp->minode, "HAMMER2-inodes");
 352                 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
 353                 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
 354         }
 355         pmp->hmp = hmp;
 356         ++hmp->pmp_count;
 357         lockmgr(&hammer2_mntlk, LK_RELEASE);
 358         kprintf("hammer2_mount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
 359
 360         mp->mnt_flag = MNT_LOCAL;
 361         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;   /* all entry pts are SMP */
 362
 363         if (create_hmp) {
 364                 /*
 365                  * vchain setup. vchain.data is special cased to NULL.
 366                  * vchain.refs is initialized and will never drop to 0.
 367                  */
 368                 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
 369                 hmp->vchain.refs = 1;
 370                 hmp->vchain.data = (void *)&hmp->voldata;
 371                 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
 372                 /* hmp->vchain.u.xxx is left NULL */
 373                 lockinit(&hmp->vchain.lk, "volume", 0, LK_CANRECURSE);
 374                 lockinit(&hmp->alloclk, "h2alloc", 0, 0);
 375                 lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
 376
 377                 /*
 378                  * Install the volume header
 379                  */
 380                 error = hammer2_install_volume_header(hmp);
 381                 if (error) {
 382                         hammer2_vfs_unmount(mp, MNT_FORCE);
 383                         return error;
 384                 }
 385         }
 386
 387         /*
 388          * required mount structure initializations
 389          */
 390         mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
 391         mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
 392
 393         mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
 394         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
 395
 396         /*
 397          * Optional fields
 398          */
 399         mp->mnt_iosize_max = MAXPHYS;
 400
 401         /*
 402          * First locate the super-root inode, which is key 0 relative to the
 403          * volume header's blockset.
 404          *
 405          * Then locate the root inode by scanning the directory keyspace
 406          * represented by the label.
 407          */
 408         if (create_hmp) {
 409                 parent = &hmp->vchain;
 410                 hammer2_chain_lock(hmp, parent, HAMMER2_RESOLVE_ALWAYS);
 411                 schain = hammer2_chain_lookup(hmp, &parent,
 412                                       HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY, 0);
 413                 hammer2_chain_unlock(hmp, parent);
 414                 if (schain == NULL) {
 415                         kprintf("hammer2_mount: invalid super-root\n");
 416                         hammer2_vfs_unmount(mp, MNT_FORCE);
 417                         return EINVAL;
 418                 }
 419                 hammer2_chain_ref(hmp, schain); /* for hmp->schain */
 420                 hmp->schain = schain;           /* left locked */
 421         } else {
 422                 schain = hmp->schain;
 423                 hammer2_chain_lock(hmp, schain, HAMMER2_RESOLVE_ALWAYS);
 424         }
 425
 426         parent = schain;
 427         lhc = hammer2_dirhash(label, strlen(label));
 428         rchain = hammer2_chain_lookup(hmp, &parent,
 429                                       lhc, lhc + HAMMER2_DIRHASH_LOMASK,
 430                                       0);
 431         while (rchain) {
 432                 if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
 433                     rchain->u.ip &&
 434                     strcmp(label, rchain->data->ipdata.filename) == 0) {
 435                         break;
 436                 }
 437                 rchain = hammer2_chain_next(hmp, &parent, rchain,
 438                                             lhc, lhc + HAMMER2_DIRHASH_LOMASK,
 439                                             0);
 440         }
 441         hammer2_chain_unlock(hmp, parent);
 442         if (rchain == NULL) {
 443                 kprintf("hammer2_mount: PFS label not found\n");
 444                 hammer2_vfs_unmount(mp, MNT_FORCE);
 445                 return EINVAL;
 446         }
 447         if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
 448                 hammer2_chain_unlock(hmp, rchain);
 449                 kprintf("hammer2_mount: PFS label already mounted!\n");
 450                 hammer2_vfs_unmount(mp, MNT_FORCE);
 451                 return EBUSY;
 452         }
 453         atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
 454
 455         hammer2_chain_ref(hmp, rchain); /* for pmp->rchain */
 456         hammer2_chain_unlock(hmp, rchain);
 457         pmp->rchain = rchain;           /* left held & unlocked */
 458         pmp->iroot = rchain->u.ip;      /* implied hold from rchain */
 459         pmp->iroot->pmp = pmp;
 460         kprintf("iroot %p\n", pmp->iroot);
 461
 462         vfs_getnewfsid(mp);
 463         vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
 464         vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
 465         vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
 466
 467         copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
 468         bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
 469         bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
 470         copyinstr(path, mp->mnt_stat.f_mntonname,
 471                   sizeof(mp->mnt_stat.f_mntonname) - 1,
 472                   &size);
 473
 474         hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
 475
 476         return 0;
 477 }
 478
 479 static
 480 int
 481 hammer2_remount(struct mount *mp, char *path, struct vnode *devvp,
 482                 struct ucred *cred)
 483 {
 484         return (0);
 485 }
 486
 487 static
 488 int
 489 hammer2_vfs_unmount(struct mount *mp, int mntflags)
 490 {
 491         hammer2_pfsmount_t *pmp;
 492         hammer2_mount_t *hmp;
 493         int flags;
 494         int error = 0;
 495         int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 496         struct vnode *devvp;
 497
 498         pmp = MPTOPMP(mp);
 499         hmp = pmp->hmp;
 500         flags = 0;
 501
 502         if (mntflags & MNT_FORCE)
 503                 flags |= FORCECLOSE;
 504
 505         hammer2_mount_exlock(hmp);
 506
 507         /*
 508          * If mount initialization proceeded far enough we must flush
 509          * its vnodes.
 510          */
 511         if (pmp->iroot)
 512                 error = vflush(mp, 0, flags);
 513
 514         if (error)
 515                 return error;
 516
 517         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
 518         --hmp->pmp_count;
 519         kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
 520
 521         /*
 522          * Flush any left over chains.  The voldata lock is only used
 523          * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
 524          */
 525         hammer2_voldata_lock(hmp);
 526         if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
 527                                  HAMMER2_CHAIN_MODIFIED_AUX |
 528                                  HAMMER2_CHAIN_SUBMODIFIED)) {
 529                 hammer2_voldata_unlock(hmp);
 530                 hammer2_vfs_sync(mp, MNT_WAIT);
 531         } else {
 532                 hammer2_voldata_unlock(hmp);
 533         }
 534         if (hmp->pmp_count == 0) {
 535                 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
 536                                          HAMMER2_CHAIN_MODIFIED_AUX |
 537                                          HAMMER2_CHAIN_SUBMODIFIED)) {
 538                         kprintf("hammer2_unmount: chains left over after "
 539                                 "final sync\n");
 540                         if (hammer2_debug & 0x0010)
 541                                 Debugger("entered debugger");
 542                 }
 543         }
 544
 545         /*
 546          * Cleanup the root and super-root chain elements (which should be
 547          * clean).
 548          */
 549         pmp->iroot = NULL;
 550         if (pmp->rchain) {
 551                 atomic_clear_int(&pmp->rchain->flags, HAMMER2_CHAIN_MOUNTED);
 552                 KKASSERT(pmp->rchain->refs == 1);
 553                 hammer2_chain_drop(hmp, pmp->rchain);
 554                 pmp->rchain = NULL;
 555         }
 556         if (hmp->pmp_count == 0) {
 557                 if (hmp->schain) {
 558                         KKASSERT(hmp->schain->refs == 1);
 559                         hammer2_chain_drop(hmp, hmp->schain);
 560                         hmp->schain = NULL;
 561                 }
 562
 563                 /*
 564                  * Finish up with the device vnode
 565                  */
 566                 if ((devvp = hmp->devvp) != NULL) {
 567                         vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
 568                         hmp->devvp = NULL;
 569                         VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE));
 570                         vrele(devvp);
 571                         devvp = NULL;
 572                 }
 573                 kmalloc_destroy(&hmp->minode);
 574                 kmalloc_destroy(&hmp->mchain);
 575         }
 576         hammer2_mount_unlock(hmp);
 577
 578         pmp->mp = NULL;
 579         pmp->hmp = NULL;
 580         mp->mnt_data = NULL;
 581
 582         kfree(pmp, M_HAMMER2);
 583         if (hmp->pmp_count == 0) {
 584                 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
 585                 kfree(hmp, M_HAMMER2);
 586         }
 587         lockmgr(&hammer2_mntlk, LK_RELEASE);
 588         return (error);
 589 }
 590
 591 static
 592 int
 593 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
 594              ino_t ino, struct vnode **vpp)
 595 {
 596         kprintf("hammer2_vget\n");
 597         return (EOPNOTSUPP);
 598 }
 599
 600 static
 601 int
 602 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
 603 {
 604         hammer2_pfsmount_t *pmp;
 605         int error;
 606         struct vnode *vp;
 607
 608         pmp = MPTOPMP(mp);
 609         hammer2_mount_exlock(pmp->hmp);
 610         if (pmp->iroot == NULL) {
 611                 *vpp = NULL;
 612                 error = EINVAL;
 613         } else {
 614                 vp = hammer2_igetv(pmp->iroot, &error);
 615                 *vpp = vp;
 616                 if (vp == NULL)
 617                         kprintf("vnodefail\n");
 618         }
 619         hammer2_mount_unlock(pmp->hmp);
 620
 621         return (error);
 622 }
 623
 624 /*
 625  * Filesystem status
 626  *
 627  * XXX incorporate pmp->iroot->ip_data.inode_quota and data_quota
 628  */
 629 static
 630 int
 631 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 632 {
 633         hammer2_pfsmount_t *pmp;
 634         hammer2_mount_t *hmp;
 635
 636         pmp = MPTOPMP(mp);
 637         hmp = MPTOHMP(mp);
 638
 639         mp->mnt_stat.f_files = pmp->iroot->ip_data.inode_count +
 640                                pmp->iroot->delta_icount;
 641         mp->mnt_stat.f_ffree = 0;
 642         mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
 643         mp->mnt_stat.f_bfree = (hmp->voldata.allocator_size -
 644                                 hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
 645         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 646
 647         *sbp = mp->mnt_stat;
 648         return (0);
 649 }
 650
 651 static
 652 int
 653 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 654 {
 655         hammer2_pfsmount_t *pmp;
 656         hammer2_mount_t *hmp;
 657
 658         pmp = MPTOPMP(mp);
 659         hmp = MPTOHMP(mp);
 660
 661         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
 662         mp->mnt_vstat.f_files = pmp->iroot->ip_data.inode_count +
 663                                 pmp->iroot->delta_icount;
 664         mp->mnt_vstat.f_ffree = 0;
 665         mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
 666         mp->mnt_vstat.f_bfree = (hmp->voldata.allocator_size -
 667                                  hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
 668         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
 669
 670         *sbp = mp->mnt_vstat;
 671         return (0);
 672 }
 673
 674 /*
 675  * Sync the entire filesystem; this is called from the filesystem syncer
 676  * process periodically and whenever a user calls sync(1) on the hammer
 677  * mountpoint.
 678  *
 679  * Currently is actually called from the syncer! \o/
 680  *
 681  * This task will have to snapshot the state of the dirty inode chain.
 682  * From that, it will have to make sure all of the inodes on the dirty
 683  * chain have IO initiated. We make sure that io is initiated for the root
 684  * block.
 685  *
 686  * If waitfor is set, we wait for media to acknowledge the new rootblock.
 687  *
 688  * THINKS: side A vs side B, to have sync not stall all I/O?
 689  */
 690 static
 691 int
 692 hammer2_vfs_sync(struct mount *mp, int waitfor)
 693 {
 694         struct hammer2_sync_info info;
 695         hammer2_mount_t *hmp;
 696         int flags;
 697         int error;
 698         int haswork;
 699
 700         hmp = MPTOHMP(mp);
 701
 702         flags = VMSC_GETVP;
 703         if (waitfor & MNT_LAZY)
 704                 flags |= VMSC_ONEPASS;
 705
 706         info.error = 0;
 707         info.waitfor = MNT_NOWAIT;
 708         vmntvnodescan(mp, flags | VMSC_NOWAIT,
 709                       hammer2_sync_scan1,
 710                       hammer2_sync_scan2, &info);
 711         if (info.error == 0 && (waitfor & MNT_WAIT)) {
 712                 info.waitfor = waitfor;
 713                     vmntvnodescan(mp, flags,
 714                                   hammer2_sync_scan1,
 715                                   hammer2_sync_scan2, &info);
 716
 717         }
 718 #if 0
 719         if (waitfor == MNT_WAIT) {
 720                 /* XXX */
 721         } else {
 722                 /* XXX */
 723         }
 724 #endif
 725         hammer2_chain_lock(hmp, &hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
 726         if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
 727                                  HAMMER2_CHAIN_MODIFIED_AUX |
 728                                  HAMMER2_CHAIN_SUBMODIFIED)) {
 729                 hammer2_chain_flush(hmp, &hmp->vchain, 0);
 730                 haswork = 1;
 731         } else {
 732                 haswork = 0;
 733         }
 734         hammer2_chain_unlock(hmp, &hmp->vchain);
 735
 736         error = 0;
 737
 738         if ((waitfor & MNT_LAZY) == 0) {
 739                 waitfor = MNT_NOWAIT;
 740                 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
 741                 error = VOP_FSYNC(hmp->devvp, waitfor, 0);
 742                 vn_unlock(hmp->devvp);
 743         }
 744
 745         if (error == 0 && haswork) {
 746                 struct buf *bp;
 747
 748                 /*
 749                  * Synchronize the disk before flushing the volume
 750                  * header.
 751                  */
 752                 bp = getpbuf(NULL);
 753                 bp->b_bio1.bio_offset = 0;
 754                 bp->b_bufsize = 0;
 755                 bp->b_bcount = 0;
 756                 bp->b_cmd = BUF_CMD_FLUSH;
 757                 bp->b_bio1.bio_done = biodone_sync;
 758                 bp->b_bio1.bio_flags |= BIO_SYNC;
 759                 vn_strategy(hmp->devvp, &bp->b_bio1);
 760                 biowait(&bp->b_bio1, "h2vol");
 761                 relpbuf(bp, NULL);
 762
 763                 /*
 764                  * Then we can safely flush the volume header.  Volume
 765                  * data is locked separately to prevent ioctl functions
 766                  * from deadlocking due to a configuration issue.
 767                  */
 768                 bp = getblk(hmp->devvp, 0, HAMMER2_PBUFSIZE, 0, 0);
 769                 hammer2_voldata_lock(hmp);
 770                 bcopy(&hmp->voldata, bp->b_data, HAMMER2_PBUFSIZE);
 771                 hammer2_voldata_unlock(hmp);
 772                 bawrite(bp);
 773         }
 774         return (error);
 775 }
 776
 777 /*
 778  * Sync passes.
 779  *
 780  * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
 781  *       won't flush on those flags.  The syncer code above will do a
 782  *       general meta-data flush globally that will catch these flags.
 783  */
 784 static int
 785 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
 786 {
 787         hammer2_inode_t *ip;
 788
 789         ip = VTOI(vp);
 790         if (vp->v_type == VNON || ip == NULL ||
 791             ((ip->chain.flags & (HAMMER2_CHAIN_MODIFIED |
 792                                  HAMMER2_CHAIN_DIRTYEMBED)) == 0 &&
 793              RB_EMPTY(&vp->v_rbdirty_tree))) {
 794                 return(-1);
 795         }
 796         return(0);
 797 }
 798
 799 static int
 800 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
 801 {
 802         struct hammer2_sync_info *info = data;
 803         hammer2_inode_t *ip;
 804         int error;
 805
 806         ip = VTOI(vp);
 807         if (vp->v_type == VNON || vp->v_type == VBAD ||
 808             ((ip->chain.flags & (HAMMER2_CHAIN_MODIFIED |
 809                                  HAMMER2_CHAIN_DIRTYEMBED)) == 0 &&
 810             RB_EMPTY(&vp->v_rbdirty_tree))) {
 811                 return(0);
 812         }
 813         error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
 814         if (error)
 815                 info->error = error;
 816         return(0);
 817 }
 818
 819 static
 820 int
 821 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 822 {
 823         return (0);
 824 }
 825
 826 static
 827 int
 828 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 829                struct fid *fhp, struct vnode **vpp)
 830 {
 831         return (0);
 832 }
 833
 834 static
 835 int
 836 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 837                  int *exflagsp, struct ucred **credanonp)
 838 {
 839         return (0);
 840 }
 841
 842 /*
 843  * Support code for hammer2_mount().  Read, verify, and install the volume
 844  * header into the HMP
 845  *
 846  * XXX read four volhdrs and use the one with the highest TID whos CRC
 847  *     matches.
 848  *
 849  * XXX check iCRCs.
 850  *
 851  * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
 852  *     nonexistant locations.
 853  *
 854  * XXX Record selected volhdr and ring updates to each of 4 volhdrs
 855  */
 856 static
 857 int
 858 hammer2_install_volume_header(hammer2_mount_t *hmp)
 859 {
 860         hammer2_volume_data_t *vd;
 861         struct buf *bp;
 862         hammer2_crc32_t crc0, crc, bcrc0, bcrc;
 863         int error_reported;
 864         int error;
 865         int valid;
 866         int i;
 867
 868         error_reported = 0;
 869         error = 0;
 870         valid = 0;
 871         bp = NULL;
 872
 873         /*
 874          * There are up to 4 copies of the volume header (syncs iterate
 875          * between them so there is no single master).  We don't trust the
 876          * volu_size field so we don't know precisely how large the filesystem
 877          * is, so depend on the OS to return an error if we go beyond the
 878          * block device's EOF.
 879          */
 880         for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
 881                 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
 882                               HAMMER2_VOLUME_BYTES, &bp);
 883                 if (error) {
 884                         brelse(bp);
 885                         bp = NULL;
 886                         continue;
 887                 }
 888
 889                 vd = (struct hammer2_volume_data *) bp->b_data;
 890                 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
 891                     (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
 892                         brelse(bp);
 893                         bp = NULL;
 894                         continue;
 895                 }
 896
 897                 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
 898                         /* XXX: Reversed-endianness filesystem */
 899                         kprintf("hammer2: reverse-endian filesystem detected");
 900                         brelse(bp);
 901                         bp = NULL;
 902                         continue;
 903                 }
 904
 905                 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
 906                 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
 907                                       HAMMER2_VOLUME_ICRC0_SIZE);
 908                 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
 909                 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
 910                                        HAMMER2_VOLUME_ICRC1_SIZE);
 911                 if ((crc0 != crc) || (bcrc0 != bcrc)) {
 912                         kprintf("hammer2 volume header crc "
 913                                 "mismatch copy #%d\t%08x %08x",
 914                                 i, crc0, crc);
 915                         error_reported = 1;
 916                         brelse(bp);
 917                         bp = NULL;
 918                         continue;
 919                 }
 920                 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
 921                         valid = 1;
 922                         hmp->voldata = *vd;
 923                 }
 924                 brelse(bp);
 925                 bp = NULL;
 926         }
 927         if (valid) {
 928                 error = 0;
 929                 if (error_reported)
 930                         kprintf("hammer2: a valid volume header was found\n");
 931         } else {
 932                 error = EINVAL;
 933                 kprintf("hammer2: no valid volume headers found!\n");
 934         }
 935         return (error);
 936 }
 937