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 long hammer2_iod_file_read;
  62 long hammer2_iod_meta_read;
  63 long hammer2_iod_indr_read;
  64 long hammer2_iod_file_write;
  65 long hammer2_iod_meta_write;
  66 long hammer2_iod_indr_write;
  67 long hammer2_iod_volu_write;
  68 long hammer2_ioa_file_read;
  69 long hammer2_ioa_meta_read;
  70 long hammer2_ioa_indr_read;
  71 long hammer2_ioa_file_write;
  72 long hammer2_ioa_meta_write;
  73 long hammer2_ioa_indr_write;
  74 long hammer2_ioa_volu_write;
  75
  76 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
  77
  78 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
  79            &hammer2_debug, 0, "");
  80 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
  81            &hammer2_cluster_enable, 0, "");
  82 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
  83            &hammer2_iod_file_read, 0, "");
  84 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
  85            &hammer2_iod_meta_read, 0, "");
  86 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
  87            &hammer2_iod_indr_read, 0, "");
  88 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
  89            &hammer2_iod_file_write, 0, "");
  90 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
  91            &hammer2_iod_meta_write, 0, "");
  92 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
  93            &hammer2_iod_indr_write, 0, "");
  94 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
  95            &hammer2_iod_volu_write, 0, "");
  96 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
  97            &hammer2_ioa_file_read, 0, "");
  98 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
  99            &hammer2_ioa_meta_read, 0, "");
 100 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
 101            &hammer2_ioa_indr_read, 0, "");
 102 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
 103            &hammer2_ioa_file_write, 0, "");
 104 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
 105            &hammer2_ioa_meta_write, 0, "");
 106 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
 107            &hammer2_ioa_indr_write, 0, "");
 108 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
 109            &hammer2_ioa_volu_write, 0, "");
 110
 111 static int hammer2_vfs_init(struct vfsconf *conf);
 112 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
 113                                 struct ucred *cred);
 114 static int hammer2_remount(struct mount *, char *, struct vnode *,
 115                                 struct ucred *);
 116 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
 117 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
 118 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
 119                                 struct ucred *cred);
 120 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 121                                 struct ucred *cred);
 122 static int hammer2_vfs_sync(struct mount *mp, int waitfor);
 123 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
 124                                 ino_t ino, struct vnode **vpp);
 125 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 126                                 struct fid *fhp, struct vnode **vpp);
 127 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 128 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 129                                 int *exflagsp, struct ucred **credanonp);
 130
 131 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
 132 static int hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
 133 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
 134
 135 /*
 136  * HAMMER2 vfs operations.
 137  */
 138 static struct vfsops hammer2_vfsops = {
 139         .vfs_init       = hammer2_vfs_init,
 140         .vfs_sync       = hammer2_vfs_sync,
 141         .vfs_mount      = hammer2_vfs_mount,
 142         .vfs_unmount    = hammer2_vfs_unmount,
 143         .vfs_root       = hammer2_vfs_root,
 144         .vfs_statfs     = hammer2_vfs_statfs,
 145         .vfs_statvfs    = hammer2_vfs_statvfs,
 146         .vfs_vget       = hammer2_vfs_vget,
 147         .vfs_vptofh     = hammer2_vfs_vptofh,
 148         .vfs_fhtovp     = hammer2_vfs_fhtovp,
 149         .vfs_checkexp   = hammer2_vfs_checkexp
 150 };
 151
 152 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
 153
 154 VFS_SET(hammer2_vfsops, hammer2, 0);
 155 MODULE_VERSION(hammer2, 1);
 156
 157 static
 158 int
 159 hammer2_vfs_init(struct vfsconf *conf)
 160 {
 161         int error;
 162
 163         error = 0;
 164
 165         if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
 166                 error = EINVAL;
 167         if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
 168                 error = EINVAL;
 169         if (HAMMER2_ALLOCREF_BYTES != sizeof(struct hammer2_allocref))
 170                 error = EINVAL;
 171         if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
 172                 error = EINVAL;
 173
 174         if (error)
 175                 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
 176
 177         lockinit(&hammer2_mntlk, "mntlk", 0, 0);
 178         TAILQ_INIT(&hammer2_mntlist);
 179
 180         return (error);
 181 }
 182
 183 /*
 184  * Mount or remount HAMMER2 fileystem from physical media
 185  *
 186  *      mountroot
 187  *              mp              mount point structure
 188  *              path            NULL
 189  *              data            <unused>
 190  *              cred            <unused>
 191  *
 192  *      mount
 193  *              mp              mount point structure
 194  *              path            path to mount point
 195  *              data            pointer to argument structure in user space
 196  *                      volume  volume path (device@LABEL form)
 197  *                      hflags  user mount flags
 198  *              cred            user credentials
 199  *
 200  * RETURNS:     0       Success
 201  *              !0      error number
 202  */
 203 static
 204 int
 205 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
 206               struct ucred *cred)
 207 {
 208         struct hammer2_mount_info info;
 209         hammer2_pfsmount_t *pmp;
 210         hammer2_mount_t *hmp;
 211         hammer2_key_t lhc;
 212         struct vnode *devvp;
 213         struct nlookupdata nd;
 214         hammer2_chain_t *parent;
 215         hammer2_chain_t *schain;
 216         hammer2_chain_t *rchain;
 217         char devstr[MNAMELEN];
 218         size_t size;
 219         size_t done;
 220         char *dev;
 221         char *label;
 222         int ronly = 1;
 223         int create_hmp;
 224         int error;
 225
 226         hmp = NULL;
 227         pmp = NULL;
 228         dev = NULL;
 229         label = NULL;
 230         devvp = NULL;
 231
 232         kprintf("hammer2_mount\n");
 233
 234         if (path == NULL) {
 235                 /*
 236                  * Root mount
 237                  */
 238                 return (EOPNOTSUPP);
 239         } else {
 240                 /*
 241                  * Non-root mount or updating a mount
 242                  */
 243                 error = copyin(data, &info, sizeof(info));
 244                 if (error)
 245                         return (error);
 246
 247                 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
 248                 if (error)
 249                         return (error);
 250
 251                 /* Extract device and label */
 252                 dev = devstr;
 253                 label = strchr(devstr, '@');
 254                 if (label == NULL ||
 255                     ((label + 1) - dev) > done) {
 256                         return (EINVAL);
 257                 }
 258                 *label = '\0';
 259                 label++;
 260                 if (*label == '\0')
 261                         return (EINVAL);
 262
 263                 if (mp->mnt_flag & MNT_UPDATE) {
 264                         /* Update mount */
 265                         /* HAMMER2 implements NFS export via mountctl */
 266                         hmp = MPTOHMP(mp);
 267                         devvp = hmp->devvp;
 268                         error = hammer2_remount(mp, path, devvp, cred);
 269                         return error;
 270                 }
 271         }
 272
 273         /*
 274          * New non-root mount
 275          */
 276         /* Lookup name and verify it refers to a block device */
 277         error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
 278         if (error == 0)
 279                 error = nlookup(&nd);
 280         if (error == 0)
 281                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
 282         nlookup_done(&nd);
 283
 284         if (error == 0) {
 285                 if (vn_isdisk(devvp, &error))
 286                         error = vfs_mountedon(devvp);
 287         }
 288
 289         /*
 290          * Determine if the device has already been mounted.  After this
 291          * check hmp will be non-NULL if we are doing the second or more
 292          * hammer2 mounts from the same device.
 293          */
 294         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
 295         TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
 296                 if (hmp->devvp == devvp)
 297                         break;
 298         }
 299
 300         /*
 301          * Open the device if this isn't a secondary mount
 302          */
 303         if (hmp) {
 304                 create_hmp = 0;
 305         } else {
 306                 create_hmp = 1;
 307                 if (error == 0 && vcount(devvp) > 0)
 308                         error = EBUSY;
 309
 310                 /*
 311                  * Now open the device
 312                  */
 313                 if (error == 0) {
 314                         ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 315                         vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
 316                         error = vinvalbuf(devvp, V_SAVE, 0, 0);
 317                         if (error == 0) {
 318                                 error = VOP_OPEN(devvp,
 319                                                  ronly ? FREAD : FREAD | FWRITE,
 320                                                  FSCRED, NULL);
 321                         }
 322                         vn_unlock(devvp);
 323                 }
 324                 if (error && devvp) {
 325                         vrele(devvp);
 326                         devvp = NULL;
 327                 }
 328                 if (error) {
 329                         lockmgr(&hammer2_mntlk, LK_RELEASE);
 330                         return error;
 331                 }
 332         }
 333
 334         /*
 335          * Block device opened successfully, finish initializing the
 336          * mount structure.
 337          *
 338          * From this point on we have to call hammer2_unmount() on failure.
 339          */
 340         pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
 341         mp->mnt_data = (qaddr_t)pmp;
 342         pmp->mp = mp;
 343
 344         if (create_hmp) {
 345                 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
 346                 hmp->ronly = ronly;
 347                 hmp->devvp = devvp;
 348                 kmalloc_create(&hmp->minode, "HAMMER2-inodes");
 349                 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
 350                 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
 351         }
 352         pmp->hmp = hmp;
 353         ++hmp->pmp_count;
 354         lockmgr(&hammer2_mntlk, LK_RELEASE);
 355         kprintf("hammer2_mount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
 356
 357         mp->mnt_flag = MNT_LOCAL;
 358         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;   /* all entry pts are SMP */
 359
 360         if (create_hmp) {
 361                 /*
 362                  * vchain setup. vchain.data is special cased to NULL.
 363                  * vchain.refs is initialized and will never drop to 0.
 364                  */
 365                 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
 366                 hmp->vchain.refs = 1;
 367                 hmp->vchain.data = (void *)&hmp->voldata;
 368                 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
 369                 /* hmp->vchain.u.xxx is left NULL */
 370                 lockinit(&hmp->vchain.lk, "volume", 0, LK_CANRECURSE);
 371                 lockinit(&hmp->alloclk, "h2alloc", 0, 0);
 372                 lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
 373
 374                 /*
 375                  * Install the volume header
 376                  */
 377                 error = hammer2_install_volume_header(hmp);
 378                 if (error) {
 379                         hammer2_vfs_unmount(mp, MNT_FORCE);
 380                         return error;
 381                 }
 382         }
 383
 384         /*
 385          * required mount structure initializations
 386          */
 387         mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
 388         mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
 389
 390         mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
 391         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
 392
 393         /*
 394          * Optional fields
 395          */
 396         mp->mnt_iosize_max = MAXPHYS;
 397
 398         /*
 399          * First locate the super-root inode, which is key 0 relative to the
 400          * volume header's blockset.
 401          *
 402          * Then locate the root inode by scanning the directory keyspace
 403          * represented by the label.
 404          */
 405         if (create_hmp) {
 406                 parent = &hmp->vchain;
 407                 hammer2_chain_lock(hmp, parent, HAMMER2_RESOLVE_ALWAYS);
 408                 schain = hammer2_chain_lookup(hmp, &parent,
 409                                       HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY, 0);
 410                 hammer2_chain_unlock(hmp, parent);
 411                 if (schain == NULL) {
 412                         kprintf("hammer2_mount: invalid super-root\n");
 413                         hammer2_vfs_unmount(mp, MNT_FORCE);
 414                         return EINVAL;
 415                 }
 416                 hammer2_chain_ref(hmp, schain); /* for hmp->schain */
 417                 hmp->schain = schain;           /* left locked */
 418         } else {
 419                 schain = hmp->schain;
 420                 hammer2_chain_lock(hmp, schain, HAMMER2_RESOLVE_ALWAYS);
 421         }
 422
 423         parent = schain;
 424         lhc = hammer2_dirhash(label, strlen(label));
 425         rchain = hammer2_chain_lookup(hmp, &parent,
 426                                       lhc, lhc + HAMMER2_DIRHASH_LOMASK,
 427                                       0);
 428         while (rchain) {
 429                 if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
 430                     rchain->u.ip &&
 431                     strcmp(label, rchain->data->ipdata.filename) == 0) {
 432                         break;
 433                 }
 434                 rchain = hammer2_chain_next(hmp, &parent, rchain,
 435                                             lhc, lhc + HAMMER2_DIRHASH_LOMASK,
 436                                             0);
 437         }
 438         hammer2_chain_unlock(hmp, parent);
 439         if (rchain == NULL) {
 440                 kprintf("hammer2_mount: PFS label not found\n");
 441                 hammer2_vfs_unmount(mp, MNT_FORCE);
 442                 return EINVAL;
 443         }
 444         if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
 445                 hammer2_chain_unlock(hmp, rchain);
 446                 kprintf("hammer2_mount: PFS label already mounted!\n");
 447                 hammer2_vfs_unmount(mp, MNT_FORCE);
 448                 return EINVAL;
 449         }
 450         atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
 451
 452         hammer2_chain_ref(hmp, rchain); /* for pmp->rchain */
 453         hammer2_chain_unlock(hmp, rchain);
 454         pmp->rchain = rchain;           /* left held & unlocked */
 455         pmp->iroot = rchain->u.ip;      /* implied hold from rchain */
 456         pmp->iroot->pmp = pmp;
 457         kprintf("iroot %p\n", pmp->iroot);
 458
 459         vfs_getnewfsid(mp);
 460         vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
 461         vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
 462         vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
 463
 464         copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
 465         bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
 466         bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
 467         copyinstr(path, mp->mnt_stat.f_mntonname,
 468                   sizeof(mp->mnt_stat.f_mntonname) - 1,
 469                   &size);
 470
 471         hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
 472
 473         return 0;
 474 }
 475
 476 static
 477 int
 478 hammer2_remount(struct mount *mp, char *path, struct vnode *devvp,
 479                 struct ucred *cred)
 480 {
 481         return (0);
 482 }
 483
 484 static
 485 int
 486 hammer2_vfs_unmount(struct mount *mp, int mntflags)
 487 {
 488         hammer2_pfsmount_t *pmp;
 489         hammer2_mount_t *hmp;
 490         int flags;
 491         int error = 0;
 492         int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 493         struct vnode *devvp;
 494
 495         pmp = MPTOPMP(mp);
 496         hmp = pmp->hmp;
 497         flags = 0;
 498
 499         if (mntflags & MNT_FORCE)
 500                 flags |= FORCECLOSE;
 501
 502         hammer2_mount_exlock(hmp);
 503
 504         /*
 505          * If mount initialization proceeded far enough we must flush
 506          * its vnodes.
 507          */
 508         if (pmp->iroot)
 509                 error = vflush(mp, 0, flags);
 510
 511         if (error)
 512                 return error;
 513
 514         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
 515         --hmp->pmp_count;
 516         kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
 517
 518         /*
 519          * Flush any left over chains.  The voldata lock is only used
 520          * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
 521          */
 522         hammer2_voldata_lock(hmp);
 523         if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
 524                                  HAMMER2_CHAIN_MODIFIED_AUX |
 525                                  HAMMER2_CHAIN_SUBMODIFIED)) {
 526                 hammer2_voldata_unlock(hmp);
 527                 hammer2_vfs_sync(mp, MNT_WAIT);
 528         } else {
 529                 hammer2_voldata_unlock(hmp);
 530         }
 531         if (hmp->pmp_count == 0) {
 532                 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
 533                                          HAMMER2_CHAIN_MODIFIED_AUX |
 534                                          HAMMER2_CHAIN_SUBMODIFIED)) {
 535                         kprintf("hammer2_unmount: chains left over after "
 536                                 "final sync\n");
 537                         if (hammer2_debug & 0x0010)
 538                                 Debugger("entered debugger");
 539                 }
 540         }
 541
 542         /*
 543          * Cleanup the root and super-root chain elements (which should be
 544          * clean).
 545          */
 546         pmp->iroot = NULL;
 547         if (pmp->rchain) {
 548                 atomic_clear_int(&pmp->rchain->flags, HAMMER2_CHAIN_MOUNTED);
 549                 KKASSERT(pmp->rchain->refs == 1);
 550                 hammer2_chain_drop(hmp, pmp->rchain);
 551                 pmp->rchain = NULL;
 552         }
 553         if (hmp->pmp_count == 0) {
 554                 if (hmp->schain) {
 555                         KKASSERT(hmp->schain->refs == 1);
 556                         hammer2_chain_drop(hmp, hmp->schain);
 557                         hmp->schain = NULL;
 558                 }
 559
 560                 /*
 561                  * Finish up with the device vnode
 562                  */
 563                 if ((devvp = hmp->devvp) != NULL) {
 564                         vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
 565                         hmp->devvp = NULL;
 566                         VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE));
 567                         vrele(devvp);
 568                         devvp = NULL;
 569                 }
 570                 kmalloc_destroy(&hmp->minode);
 571                 kmalloc_destroy(&hmp->mchain);
 572         }
 573         hammer2_mount_unlock(hmp);
 574
 575         pmp->mp = NULL;
 576         pmp->hmp = NULL;
 577         mp->mnt_data = NULL;
 578
 579         kfree(pmp, M_HAMMER2);
 580         if (hmp->pmp_count == 0) {
 581                 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
 582                 kfree(hmp, M_HAMMER2);
 583         }
 584         lockmgr(&hammer2_mntlk, LK_RELEASE);
 585         return (error);
 586 }
 587
 588 static
 589 int
 590 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
 591              ino_t ino, struct vnode **vpp)
 592 {
 593         kprintf("hammer2_vget\n");
 594         return (EOPNOTSUPP);
 595 }
 596
 597 static
 598 int
 599 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
 600 {
 601         hammer2_pfsmount_t *pmp;
 602         int error;
 603         struct vnode *vp;
 604
 605         pmp = MPTOPMP(mp);
 606         hammer2_mount_exlock(pmp->hmp);
 607         if (pmp->iroot == NULL) {
 608                 *vpp = NULL;
 609                 error = EINVAL;
 610         } else {
 611                 vp = hammer2_igetv(pmp->iroot, &error);
 612                 *vpp = vp;
 613                 if (vp == NULL)
 614                         kprintf("vnodefail\n");
 615         }
 616         hammer2_mount_unlock(pmp->hmp);
 617
 618         return (error);
 619 }
 620
 621 static
 622 int
 623 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 624 {
 625         hammer2_mount_t *hmp;
 626
 627         hmp = MPTOHMP(mp);
 628
 629         mp->mnt_stat.f_files = 10;
 630         mp->mnt_stat.f_bfree = 10;
 631         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 632
 633         *sbp = mp->mnt_stat;
 634         return (0);
 635 }
 636
 637 static
 638 int
 639 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 640 {
 641         hammer2_mount_t *hmp;
 642
 643         hmp = MPTOHMP(mp);
 644
 645         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
 646         mp->mnt_vstat.f_files = 0;
 647         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 648
 649         *sbp = mp->mnt_vstat;
 650         return (0);
 651 }
 652
 653 /*
 654  * Sync the entire filesystem; this is called from the filesystem syncer
 655  * process periodically and whenever a user calls sync(1) on the hammer
 656  * mountpoint.
 657  *
 658  * Currently is actually called from the syncer! \o/
 659  *
 660  * This task will have to snapshot the state of the dirty inode chain.
 661  * From that, it will have to make sure all of the inodes on the dirty
 662  * chain have IO initiated. We make sure that io is initiated for the root
 663  * block.
 664  *
 665  * If waitfor is set, we wait for media to acknowledge the new rootblock.
 666  *
 667  * THINKS: side A vs side B, to have sync not stall all I/O?
 668  */
 669 static
 670 int
 671 hammer2_vfs_sync(struct mount *mp, int waitfor)
 672 {
 673         struct hammer2_sync_info info;
 674         hammer2_mount_t *hmp;
 675         int flags;
 676         int error;
 677         int haswork;
 678
 679         hmp = MPTOHMP(mp);
 680
 681         flags = VMSC_GETVP;
 682         if (waitfor & MNT_LAZY)
 683                 flags |= VMSC_ONEPASS;
 684
 685         info.error = 0;
 686         info.waitfor = MNT_NOWAIT;
 687         vmntvnodescan(mp, flags | VMSC_NOWAIT,
 688                       hammer2_sync_scan1,
 689                       hammer2_sync_scan2, &info);
 690         if (info.error == 0 && (waitfor & MNT_WAIT)) {
 691                 info.waitfor = waitfor;
 692                     vmntvnodescan(mp, flags,
 693                                   hammer2_sync_scan1,
 694                                   hammer2_sync_scan2, &info);
 695
 696         }
 697 #if 0
 698         if (waitfor == MNT_WAIT) {
 699                 /* XXX */
 700         } else {
 701                 /* XXX */
 702         }
 703 #endif
 704         hammer2_chain_lock(hmp, &hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
 705         if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
 706                                  HAMMER2_CHAIN_MODIFIED_AUX |
 707                                  HAMMER2_CHAIN_SUBMODIFIED)) {
 708                 hammer2_chain_flush(hmp, &hmp->vchain);
 709                 haswork = 1;
 710         } else {
 711                 haswork = 0;
 712         }
 713         hammer2_chain_unlock(hmp, &hmp->vchain);
 714
 715         error = 0;
 716
 717         if ((waitfor & MNT_LAZY) == 0) {
 718                 waitfor = MNT_NOWAIT;
 719                 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
 720                 error = VOP_FSYNC(hmp->devvp, waitfor, 0);
 721                 vn_unlock(hmp->devvp);
 722         }
 723
 724         if (error == 0 && haswork) {
 725                 struct buf *bp;
 726
 727                 /*
 728                  * Synchronize the disk before flushing the volume
 729                  * header.
 730                  */
 731                 bp = getpbuf(NULL);
 732                 bp->b_bio1.bio_offset = 0;
 733                 bp->b_bufsize = 0;
 734                 bp->b_bcount = 0;
 735                 bp->b_cmd = BUF_CMD_FLUSH;
 736                 bp->b_bio1.bio_done = biodone_sync;
 737                 bp->b_bio1.bio_flags |= BIO_SYNC;
 738                 vn_strategy(hmp->devvp, &bp->b_bio1);
 739                 biowait(&bp->b_bio1, "h2vol");
 740                 relpbuf(bp, NULL);
 741
 742                 /*
 743                  * Then we can safely flush the volume header.  Volume
 744                  * data is locked separately to prevent ioctl functions
 745                  * from deadlocking due to a configuration issue.
 746                  */
 747                 bp = getblk(hmp->devvp, 0, HAMMER2_PBUFSIZE, 0, 0);
 748                 hammer2_voldata_lock(hmp);
 749                 bcopy(&hmp->voldata, bp->b_data, HAMMER2_PBUFSIZE);
 750                 hammer2_voldata_unlock(hmp);
 751                 bawrite(bp);
 752         }
 753         return (error);
 754 }
 755
 756 /*
 757  * Sync passes.
 758  *
 759  * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
 760  *       won't flush on those flags.  The syncer code above will do a
 761  *       general meta-data flush globally that will catch these flags.
 762  */
 763 static int
 764 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
 765 {
 766         hammer2_inode_t *ip;
 767
 768         ip = VTOI(vp);
 769         if (vp->v_type == VNON || ip == NULL ||
 770             ((ip->chain.flags & (HAMMER2_CHAIN_MODIFIED |
 771                                  HAMMER2_CHAIN_DIRTYEMBED)) == 0 &&
 772              RB_EMPTY(&vp->v_rbdirty_tree))) {
 773                 return(-1);
 774         }
 775         return(0);
 776 }
 777
 778 static int
 779 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
 780 {
 781         struct hammer2_sync_info *info = data;
 782         hammer2_inode_t *ip;
 783         int error;
 784
 785         ip = VTOI(vp);
 786         if (vp->v_type == VNON || vp->v_type == VBAD ||
 787             ((ip->chain.flags & (HAMMER2_CHAIN_MODIFIED |
 788                                  HAMMER2_CHAIN_DIRTYEMBED)) == 0 &&
 789             RB_EMPTY(&vp->v_rbdirty_tree))) {
 790                 return(0);
 791         }
 792         error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
 793         if (error)
 794                 info->error = error;
 795         return(0);
 796 }
 797
 798 static
 799 int
 800 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 801 {
 802         return (0);
 803 }
 804
 805 static
 806 int
 807 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 808                struct fid *fhp, struct vnode **vpp)
 809 {
 810         return (0);
 811 }
 812
 813 static
 814 int
 815 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 816                  int *exflagsp, struct ucred **credanonp)
 817 {
 818         return (0);
 819 }
 820
 821 /*
 822  * Support code for hammer2_mount().  Read, verify, and install the volume
 823  * header into the HMP
 824  *
 825  * XXX read four volhdrs and use the one with the highest TID whos CRC
 826  *     matches.
 827  *
 828  * XXX check iCRCs.
 829  *
 830  * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
 831  *     nonexistant locations.
 832  *
 833  * XXX Record selected volhdr and ring updates to each of 4 volhdrs
 834  */
 835 static
 836 int
 837 hammer2_install_volume_header(hammer2_mount_t *hmp)
 838 {
 839         hammer2_volume_data_t *vd;
 840         struct buf *bp;
 841         hammer2_crc32_t crc0, crc, bcrc0, bcrc;
 842         int error_reported;
 843         int error;
 844         int valid;
 845         int i;
 846
 847         error_reported = 0;
 848         error = 0;
 849         valid = 0;
 850         bp = NULL;
 851
 852         /*
 853          * There are up to 4 copies of the volume header (syncs iterate
 854          * between them so there is no single master).  We don't trust the
 855          * volu_size field so we don't know precisely how large the filesystem
 856          * is, so depend on the OS to return an error if we go beyond the
 857          * block device's EOF.
 858          */
 859         for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
 860                 error = bread(hmp->devvp, i * HAMMER2_RESERVE_BYTES64,
 861                               HAMMER2_VOLUME_BYTES, &bp);
 862                 if (error) {
 863                         brelse(bp);
 864                         bp = NULL;
 865                         continue;
 866                 }
 867
 868                 vd = (struct hammer2_volume_data *) bp->b_data;
 869                 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
 870                     (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
 871                         brelse(bp);
 872                         bp = NULL;
 873                         continue;
 874                 }
 875
 876                 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
 877                         /* XXX: Reversed-endianness filesystem */
 878                         kprintf("hammer2: reverse-endian filesystem detected");
 879                         brelse(bp);
 880                         bp = NULL;
 881                         continue;
 882                 }
 883
 884                 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
 885                 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
 886                                       HAMMER2_VOLUME_ICRC0_SIZE);
 887                 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
 888                 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
 889                                        HAMMER2_VOLUME_ICRC1_SIZE);
 890                 if ((crc0 != crc) || (bcrc0 != bcrc)) {
 891                         kprintf("hammer2 volume header crc "
 892                                 "mismatch copy #%d\t%08x %08x",
 893                                 i, crc0, crc);
 894                         error_reported = 1;
 895                         brelse(bp);
 896                         bp = NULL;
 897                         continue;
 898                 }
 899                 if (valid == 0 || hmp->voldata.last_tid < vd->last_tid) {
 900                         valid = 1;
 901                         hmp->voldata = *vd;
 902                 }
 903                 brelse(bp);
 904                 bp = NULL;
 905         }
 906         if (valid) {
 907                 error = 0;
 908                 if (error_reported)
 909                         kprintf("hammer2: a valid volume header was found\n");
 910         } else {
 911                 error = EINVAL;
 912                 kprintf("hammer2: no valid volume headers found!\n");
 913         }
 914         return (error);
 915 }
 916