sys/vfs/hammer/hammer_vfsops.c

   1 /*
   2  * Copyright (c) 2007-2008 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  * $DragonFly: src/sys/vfs/hammer/hammer_vfsops.c,v 1.62 2008/07/12 23:04:50 dillon Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/vnode.h>
  41 #include <sys/mount.h>
  42 #include <sys/malloc.h>
  43 #include <sys/nlookup.h>
  44 #include <sys/fcntl.h>
  45 #include <sys/sysctl.h>
  46 #include <sys/buf.h>
  47 #include <sys/buf2.h>
  48 #include "hammer.h"
  49
  50 int hammer_debug_io;
  51 int hammer_debug_general;
  52 int hammer_debug_debug;
  53 int hammer_debug_inode;
  54 int hammer_debug_locks;
  55 int hammer_debug_btree;
  56 int hammer_debug_tid;
  57 int hammer_debug_recover;               /* -1 will disable, +1 will force */
  58 int hammer_debug_recover_faults;
  59 int hammer_debug_cluster_enable = 1;    /* enable read clustering by default */
  60 int hammer_count_fsyncs;
  61 int hammer_count_inodes;
  62 int hammer_count_iqueued;
  63 int hammer_count_reclaiming;
  64 int hammer_count_records;
  65 int hammer_count_record_datas;
  66 int hammer_count_volumes;
  67 int hammer_count_buffers;
  68 int hammer_count_nodes;
  69 int64_t hammer_count_extra_space_used;
  70 int64_t hammer_stats_btree_lookups;
  71 int64_t hammer_stats_btree_searches;
  72 int64_t hammer_stats_btree_inserts;
  73 int64_t hammer_stats_btree_deletes;
  74 int64_t hammer_stats_btree_elements;
  75 int64_t hammer_stats_btree_splits;
  76 int64_t hammer_stats_btree_iterations;
  77 int64_t hammer_stats_record_iterations;
  78 int hammer_count_dirtybufspace;         /* global */
  79 int hammer_count_refedbufs;             /* global */
  80 int hammer_count_reservations;
  81 int hammer_count_io_running_read;
  82 int hammer_count_io_running_write;
  83 int hammer_count_io_locked;
  84 int hammer_limit_dirtybufspace;         /* per-mount */
  85 int hammer_limit_recs;                  /* as a whole XXX */
  86 int hammer_limit_iqueued;               /* per-mount */
  87 int hammer_bio_count;
  88 int hammer_verify_zone;
  89 int hammer_write_mode;
  90 int64_t hammer_contention_count;
  91 int64_t hammer_zone_limit;
  92
  93 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
  94 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
  95            &hammer_debug_general, 0, "");
  96 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
  97            &hammer_debug_io, 0, "");
  98 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
  99            &hammer_debug_debug, 0, "");
 100 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 101            &hammer_debug_inode, 0, "");
 102 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 103            &hammer_debug_locks, 0, "");
 104 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 105            &hammer_debug_btree, 0, "");
 106 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 107            &hammer_debug_tid, 0, "");
 108 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 109            &hammer_debug_recover, 0, "");
 110 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 111            &hammer_debug_recover_faults, 0, "");
 112 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_cluster_enable, CTLFLAG_RW,
 113            &hammer_debug_cluster_enable, 0, "");
 114
 115 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 116            &hammer_limit_dirtybufspace, 0, "");
 117 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 118            &hammer_limit_recs, 0, "");
 119 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_iqueued, CTLFLAG_RW,
 120            &hammer_limit_iqueued, 0, "");
 121
 122 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 123            &hammer_count_fsyncs, 0, "");
 124 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 125            &hammer_count_inodes, 0, "");
 126 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 127            &hammer_count_iqueued, 0, "");
 128 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 129            &hammer_count_reclaiming, 0, "");
 130 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 131            &hammer_count_records, 0, "");
 132 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 133            &hammer_count_record_datas, 0, "");
 134 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 135            &hammer_count_volumes, 0, "");
 136 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 137            &hammer_count_buffers, 0, "");
 138 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 139            &hammer_count_nodes, 0, "");
 140 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
 141            &hammer_count_extra_space_used, 0, "");
 142 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 143            &hammer_stats_btree_searches, 0, "");
 144 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 145            &hammer_stats_btree_lookups, 0, "");
 146 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 147            &hammer_stats_btree_inserts, 0, "");
 148 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 149            &hammer_stats_btree_deletes, 0, "");
 150 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 151            &hammer_stats_btree_elements, 0, "");
 152 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 153            &hammer_stats_btree_splits, 0, "");
 154 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 155            &hammer_stats_btree_iterations, 0, "");
 156 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 157            &hammer_stats_record_iterations, 0, "");
 158 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 159            &hammer_count_dirtybufspace, 0, "");
 160 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 161            &hammer_count_refedbufs, 0, "");
 162 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 163            &hammer_count_reservations, 0, "");
 164 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 165            &hammer_count_io_running_read, 0, "");
 166 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 167            &hammer_count_io_locked, 0, "");
 168 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 169            &hammer_count_io_running_write, 0, "");
 170 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 171            &hammer_zone_limit, 0, "");
 172 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 173            &hammer_contention_count, 0, "");
 174 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 175            &hammer_verify_zone, 0, "");
 176 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 177            &hammer_write_mode, 0, "");
 178
 179 KTR_INFO_MASTER(hammer);
 180
 181 /*
 182  * VFS ABI
 183  */
 184 static void     hammer_free_hmp(struct mount *mp);
 185
 186 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 187                                 struct ucred *cred);
 188 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 189 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 190 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 191                                 struct ucred *cred);
 192 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 193                                 struct ucred *cred);
 194 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 195 static int      hammer_vfs_vget(struct mount *mp, ino_t ino,
 196                                 struct vnode **vpp);
 197 static int      hammer_vfs_init(struct vfsconf *conf);
 198 static int      hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp,
 199                                 struct vnode **vpp);
 200 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 201 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 202                                 int *exflagsp, struct ucred **credanonp);
 203
 204
 205 static struct vfsops hammer_vfsops = {
 206         .vfs_mount      = hammer_vfs_mount,
 207         .vfs_unmount    = hammer_vfs_unmount,
 208         .vfs_root       = hammer_vfs_root,
 209         .vfs_statfs     = hammer_vfs_statfs,
 210         .vfs_statvfs    = hammer_vfs_statvfs,
 211         .vfs_sync       = hammer_vfs_sync,
 212         .vfs_vget       = hammer_vfs_vget,
 213         .vfs_init       = hammer_vfs_init,
 214         .vfs_vptofh     = hammer_vfs_vptofh,
 215         .vfs_fhtovp     = hammer_vfs_fhtovp,
 216         .vfs_checkexp   = hammer_vfs_checkexp
 217 };
 218
 219 MALLOC_DEFINE(M_HAMMER, "hammer-mount", "hammer mount");
 220
 221 VFS_SET(hammer_vfsops, hammer, 0);
 222 MODULE_VERSION(hammer, 1);
 223
 224 static int
 225 hammer_vfs_init(struct vfsconf *conf)
 226 {
 227         int n;
 228
 229         if (hammer_limit_recs == 0) {
 230                 hammer_limit_recs = nbuf * 25;
 231                 n = kmalloc_limit(M_HAMMER) / 512;
 232                 if (hammer_limit_recs > n)
 233                         hammer_limit_recs = n;
 234         }
 235         if (hammer_limit_dirtybufspace == 0) {
 236                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 237                 if (hammer_limit_dirtybufspace < 100)
 238                         hammer_limit_dirtybufspace = 100;
 239         }
 240         if (hammer_limit_iqueued == 0)
 241                 hammer_limit_iqueued = desiredvnodes / 5;
 242         return(0);
 243 }
 244
 245 static int
 246 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 247                  struct ucred *cred)
 248 {
 249         struct hammer_mount_info info;
 250         hammer_mount_t hmp;
 251         hammer_volume_t rootvol;
 252         struct vnode *rootvp;
 253         const char *upath;      /* volume name in userspace */
 254         char *path;             /* volume name in system space */
 255         int error;
 256         int i;
 257
 258         if ((error = copyin(data, &info, sizeof(info))) != 0)
 259                 return (error);
 260         if ((mp->mnt_flag & MNT_UPDATE) == 0) {
 261                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 262                         return (EINVAL);
 263         }
 264         if ((info.hflags & HMNT_MASTERID) &&
 265             (info.masterid < -1 || info.masterid >= HAMMER_MAX_MASTERS)) {
 266                         return (EINVAL);
 267         }
 268
 269         /*
 270          * Interal mount data structure
 271          */
 272         if (mp->mnt_flag & MNT_UPDATE) {
 273                 hmp = (void *)mp->mnt_data;
 274                 KKASSERT(hmp != NULL);
 275         } else {
 276                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 277                 mp->mnt_data = (qaddr_t)hmp;
 278                 hmp->mp = mp;
 279                 hmp->namekey_iterator = mycpu->gd_time_seconds;
 280                 /*TAILQ_INIT(&hmp->recycle_list);*/
 281
 282                 hmp->root_btree_beg.localization = 0x00000000U;
 283                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 284                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 285                 hmp->root_btree_beg.create_tid = 1;
 286                 hmp->root_btree_beg.delete_tid = 1;
 287                 hmp->root_btree_beg.rec_type = 0;
 288                 hmp->root_btree_beg.obj_type = 0;
 289
 290                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 291                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 292                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 293                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 294                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 295                 hmp->root_btree_end.rec_type = 0xFFFFU;
 296                 hmp->root_btree_end.obj_type = 0;
 297
 298                 hmp->sync_lock.refs = 1;
 299                 hmp->free_lock.refs = 1;
 300                 hmp->undo_lock.refs = 1;
 301                 hmp->blkmap_lock.refs = 1;
 302
 303                 TAILQ_INIT(&hmp->delay_list);
 304                 TAILQ_INIT(&hmp->flush_group_list);
 305                 TAILQ_INIT(&hmp->objid_cache_list);
 306                 TAILQ_INIT(&hmp->undo_lru_list);
 307                 TAILQ_INIT(&hmp->reclaim_list);
 308         }
 309         hmp->hflags &= ~HMNT_USERFLAGS;
 310         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 311         if (info.hflags & HMNT_MASTERID)
 312                 hmp->masterid = info.masterid;
 313         else
 314                 hmp->masterid = -1;
 315         if (info.asof) {
 316                 kprintf("ASOF\n");
 317                 mp->mnt_flag |= MNT_RDONLY;
 318                 hmp->asof = info.asof;
 319         } else {
 320                 hmp->asof = HAMMER_MAX_TID;
 321         }
 322
 323         /*
 324          * Re-open read-write if originally read-only, or vise-versa.
 325          */
 326         if (mp->mnt_flag & MNT_UPDATE) {
 327                 error = 0;
 328                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 329                         kprintf("HAMMER read-only -> read-write\n");
 330                         hmp->ronly = 0;
 331                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 332                                 hammer_adjust_volume_mode, NULL);
 333                         rootvol = hammer_get_root_volume(hmp, &error);
 334                         if (rootvol) {
 335                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 336                                 bcopy(rootvol->ondisk->vol0_blockmap,
 337                                       hmp->blockmap,
 338                                       sizeof(hmp->blockmap));
 339                                 hammer_rel_volume(rootvol, 0);
 340                         }
 341                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 342                                 hammer_reload_inode, NULL);
 343                         /* kernel clears MNT_RDONLY */
 344                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 345                         kprintf("HAMMER read-write -> read-only\n");
 346                         hmp->ronly = 1; /* messy */
 347                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 348                                 hammer_reload_inode, NULL);
 349                         hmp->ronly = 0;
 350                         hammer_flusher_sync(hmp);
 351                         hammer_flusher_sync(hmp);
 352                         hammer_flusher_sync(hmp);
 353                         hmp->ronly = 1;
 354                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 355                                 hammer_adjust_volume_mode, NULL);
 356                 }
 357                 return(error);
 358         }
 359
 360         RB_INIT(&hmp->rb_vols_root);
 361         RB_INIT(&hmp->rb_inos_root);
 362         RB_INIT(&hmp->rb_nods_root);
 363         RB_INIT(&hmp->rb_undo_root);
 364         RB_INIT(&hmp->rb_resv_root);
 365         RB_INIT(&hmp->rb_bufs_root);
 366         RB_INIT(&hmp->rb_pfsm_root);
 367
 368         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 369
 370         TAILQ_INIT(&hmp->volu_list);
 371         TAILQ_INIT(&hmp->undo_list);
 372         TAILQ_INIT(&hmp->data_list);
 373         TAILQ_INIT(&hmp->meta_list);
 374         TAILQ_INIT(&hmp->lose_list);
 375
 376         /*
 377          * Load volumes
 378          */
 379         path = objcache_get(namei_oc, M_WAITOK);
 380         hmp->nvolumes = info.nvolumes;
 381         for (i = 0; i < info.nvolumes; ++i) {
 382                 error = copyin(&info.volumes[i], &upath, sizeof(char *));
 383                 if (error == 0)
 384                         error = copyinstr(upath, path, MAXPATHLEN, NULL);
 385                 if (error == 0)
 386                         error = hammer_install_volume(hmp, path);
 387                 if (error)
 388                         break;
 389         }
 390         objcache_put(namei_oc, path);
 391
 392         /*
 393          * Make sure we found a root volume
 394          */
 395         if (error == 0 && hmp->rootvol == NULL) {
 396                 kprintf("hammer_mount: No root volume found!\n");
 397                 error = EINVAL;
 398         }
 399         if (error) {
 400                 hammer_free_hmp(mp);
 401                 return (error);
 402         }
 403
 404         /*
 405          * No errors, setup enough of the mount point so we can lookup the
 406          * root vnode.
 407          */
 408         mp->mnt_iosize_max = MAXPHYS;
 409         mp->mnt_kern_flag |= MNTK_FSMID;
 410
 411         /*
 412          * note: f_iosize is used by vnode_pager_haspage() when constructing
 413          * its VOP_BMAP call.
 414          */
 415         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 416         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 417
 418         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 419         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 420
 421         mp->mnt_maxsymlinklen = 255;
 422         mp->mnt_flag |= MNT_LOCAL;
 423
 424         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 425         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 426         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 427
 428         /*
 429          * The root volume's ondisk pointer is only valid if we hold a
 430          * reference to it.
 431          */
 432         rootvol = hammer_get_root_volume(hmp, &error);
 433         if (error)
 434                 goto failed;
 435
 436         /*
 437          * Perform any necessary UNDO operations.  The recovery code does
 438          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 439          * and then re-copy it again after recovery is complete.
 440          *
 441          * If this is a read-only mount the UNDO information is retained
 442          * in memory in the form of dirty buffer cache buffers, and not
 443          * written back to the media.
 444          */
 445         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 446               sizeof(hmp->blockmap));
 447
 448         /*
 449          * The undo_rec_limit limits the size of flush groups to avoid
 450          * blowing out the UNDO FIFO.  This calculation is typically in
 451          * the tens of thousands and is designed primarily when small
 452          * HAMMER filesystems are created.
 453          */
 454         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 455         if (hammer_debug_general & 0x0001)
 456                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 457
 458         error = hammer_recover(hmp, rootvol);
 459         if (error) {
 460                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 461                 goto done;
 462         }
 463
 464         /*
 465          * Finish setup now that we have a good root volume.
 466          *
 467          * The top 16 bits of fsid.val[1] is a pfs id.
 468          */
 469         ksnprintf(mp->mnt_stat.f_mntfromname,
 470                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 471                   rootvol->ondisk->vol_name);
 472         mp->mnt_stat.f_fsid.val[0] =
 473                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 474         mp->mnt_stat.f_fsid.val[1] =
 475                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 476         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 477
 478         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 479         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 480                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 481
 482         /*
 483          * Certain often-modified fields in the root volume are cached in
 484          * the hammer_mount structure so we do not have to generate lots
 485          * of little UNDO structures for them.
 486          *
 487          * Recopy after recovery.  This also has the side effect of
 488          * setting our cached undo FIFO's first_offset, which serves to
 489          * placemark the FIFO start for the NEXT flush cycle while the
 490          * on-disk first_offset represents the LAST flush cycle.
 491          */
 492         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 493         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 494               sizeof(hmp->blockmap));
 495         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 496
 497         hammer_flusher_create(hmp);
 498
 499         /*
 500          * Locate the root directory using the root cluster's B-Tree as a
 501          * starting point.  The root directory uses an obj_id of 1.
 502          *
 503          * FUTURE: Leave the root directory cached referenced but unlocked
 504          * in hmp->rootvp (need to flush it on unmount).
 505          */
 506         error = hammer_vfs_vget(mp, 1, &rootvp);
 507         if (error)
 508                 goto done;
 509         vput(rootvp);
 510         /*vn_unlock(hmp->rootvp);*/
 511
 512 done:
 513         hammer_rel_volume(rootvol, 0);
 514 failed:
 515         /*
 516          * Cleanup and return.
 517          */
 518         if (error)
 519                 hammer_free_hmp(mp);
 520         return (error);
 521 }
 522
 523 static int
 524 hammer_vfs_unmount(struct mount *mp, int mntflags)
 525 {
 526 #if 0
 527         struct hammer_mount *hmp = (void *)mp->mnt_data;
 528 #endif
 529         int flags;
 530         int error;
 531
 532         /*
 533          * Clean out the vnodes
 534          */
 535         flags = 0;
 536         if (mntflags & MNT_FORCE)
 537                 flags |= FORCECLOSE;
 538         if ((error = vflush(mp, 0, flags)) != 0)
 539                 return (error);
 540
 541         /*
 542          * Clean up the internal mount structure and related entities.  This
 543          * may issue I/O.
 544          */
 545         hammer_free_hmp(mp);
 546         return(0);
 547 }
 548
 549 /*
 550  * Clean up the internal mount structure and disassociate it from the mount.
 551  * This may issue I/O.
 552  */
 553 static void
 554 hammer_free_hmp(struct mount *mp)
 555 {
 556         struct hammer_mount *hmp = (void *)mp->mnt_data;
 557
 558 #if 0
 559         /*
 560          * Clean up the root vnode
 561          */
 562         if (hmp->rootvp) {
 563                 vrele(hmp->rootvp);
 564                 hmp->rootvp = NULL;
 565         }
 566 #endif
 567         hammer_flusher_sync(hmp);
 568         hammer_flusher_sync(hmp);
 569         hammer_flusher_destroy(hmp);
 570
 571         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 572
 573 #if 0
 574         /*
 575          * Unload & flush inodes
 576          *
 577          * XXX illegal to call this from here, it can only be done from
 578          * the flusher.
 579          */
 580         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 581                 hammer_unload_inode, (void *)MNT_WAIT);
 582
 583         /*
 584          * Unload & flush volumes
 585          */
 586 #endif
 587         /*
 588          * Unload buffers and then volumes
 589          */
 590         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 591                 hammer_unload_buffer, NULL);
 592         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 593                 hammer_unload_volume, NULL);
 594
 595         mp->mnt_data = NULL;
 596         mp->mnt_flag &= ~MNT_LOCAL;
 597         hmp->mp = NULL;
 598         hammer_destroy_objid_cache(hmp);
 599         kfree(hmp, M_HAMMER);
 600 }
 601
 602 /*
 603  * Obtain a vnode for the specified inode number.  An exclusively locked
 604  * vnode is returned.
 605  */
 606 int
 607 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 608 {
 609         struct hammer_transaction trans;
 610         struct hammer_mount *hmp = (void *)mp->mnt_data;
 611         struct hammer_inode *ip;
 612         int error;
 613
 614         hammer_simple_transaction(&trans, hmp);
 615
 616         /*
 617          * Lookup the requested HAMMER inode.  The structure must be
 618          * left unlocked while we manipulate the related vnode to avoid
 619          * a deadlock.
 620          */
 621         ip = hammer_get_inode(&trans, NULL, ino,
 622                               hmp->asof, HAMMER_DEF_LOCALIZATION,
 623                               0, &error);
 624         if (ip == NULL) {
 625                 *vpp = NULL;
 626                 return(error);
 627         }
 628         error = hammer_get_vnode(ip, vpp);
 629         hammer_rel_inode(ip, 0);
 630         hammer_done_transaction(&trans);
 631         return (error);
 632 }
 633
 634 /*
 635  * Return the root vnode for the filesystem.
 636  *
 637  * HAMMER stores the root vnode in the hammer_mount structure so
 638  * getting it is easy.
 639  */
 640 static int
 641 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 642 {
 643 #if 0
 644         struct hammer_mount *hmp = (void *)mp->mnt_data;
 645 #endif
 646         int error;
 647
 648         error = hammer_vfs_vget(mp, 1, vpp);
 649         return (error);
 650 }
 651
 652 static int
 653 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 654 {
 655         struct hammer_mount *hmp = (void *)mp->mnt_data;
 656         hammer_volume_t volume;
 657         hammer_volume_ondisk_t ondisk;
 658         int error;
 659         int64_t bfree;
 660
 661         volume = hammer_get_root_volume(hmp, &error);
 662         if (error)
 663                 return(error);
 664         ondisk = volume->ondisk;
 665
 666         /*
 667          * Basic stats
 668          */
 669         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 670         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 671         hammer_rel_volume(volume, 0);
 672
 673         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
 674         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 675         if (mp->mnt_stat.f_files < 0)
 676                 mp->mnt_stat.f_files = 0;
 677
 678         *sbp = mp->mnt_stat;
 679         return(0);
 680 }
 681
 682 static int
 683 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 684 {
 685         struct hammer_mount *hmp = (void *)mp->mnt_data;
 686         hammer_volume_t volume;
 687         hammer_volume_ondisk_t ondisk;
 688         int error;
 689         int64_t bfree;
 690
 691         volume = hammer_get_root_volume(hmp, &error);
 692         if (error)
 693                 return(error);
 694         ondisk = volume->ondisk;
 695
 696         /*
 697          * Basic stats
 698          */
 699         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 700         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 701         hammer_rel_volume(volume, 0);
 702
 703         mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
 704         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 705         if (mp->mnt_vstat.f_files < 0)
 706                 mp->mnt_vstat.f_files = 0;
 707         *sbp = mp->mnt_vstat;
 708         return(0);
 709 }
 710
 711 /*
 712  * Sync the filesystem.  Currently we have to run it twice, the second
 713  * one will advance the undo start index to the end index, so if a crash
 714  * occurs no undos will be run on mount.
 715  *
 716  * We do not sync the filesystem if we are called from a panic.  If we did
 717  * we might end up blowing up a sync that was already in progress.
 718  */
 719 static int
 720 hammer_vfs_sync(struct mount *mp, int waitfor)
 721 {
 722         struct hammer_mount *hmp = (void *)mp->mnt_data;
 723         int error;
 724
 725         if (panicstr == NULL) {
 726                 error = hammer_sync_hmp(hmp, waitfor);
 727                 if (error == 0)
 728                         error = hammer_sync_hmp(hmp, waitfor);
 729         } else {
 730                 error = EIO;
 731         }
 732         return (error);
 733 }
 734
 735 /*
 736  * Convert a vnode to a file handle.
 737  */
 738 static int
 739 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 740 {
 741         hammer_inode_t ip;
 742
 743         KKASSERT(MAXFIDSZ >= 16);
 744         ip = VTOI(vp);
 745         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 746         fhp->fid_ext = ip->obj_localization >> 16;
 747         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 748         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 749         return(0);
 750 }
 751
 752
 753 /*
 754  * Convert a file handle back to a vnode.
 755  */
 756 static int
 757 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
 758 {
 759         struct hammer_transaction trans;
 760         struct hammer_inode *ip;
 761         struct hammer_inode_info info;
 762         int error;
 763         u_int32_t localization;
 764
 765         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 766         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 767         localization = (u_int32_t)fhp->fid_ext << 16;
 768
 769         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 770
 771         /*
 772          * Get/allocate the hammer_inode structure.  The structure must be
 773          * unlocked while we manipulate the related vnode to avoid a
 774          * deadlock.
 775          */
 776         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 777                               info.obj_asof, localization, 0, &error);
 778         if (ip == NULL) {
 779                 *vpp = NULL;
 780                 return(error);
 781         }
 782         error = hammer_get_vnode(ip, vpp);
 783         hammer_rel_inode(ip, 0);
 784         hammer_done_transaction(&trans);
 785         return (error);
 786 }
 787
 788 static int
 789 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 790                     int *exflagsp, struct ucred **credanonp)
 791 {
 792         hammer_mount_t hmp = (void *)mp->mnt_data;
 793         struct netcred *np;
 794         int error;
 795
 796         np = vfs_export_lookup(mp, &hmp->export, nam);
 797         if (np) {
 798                 *exflagsp = np->netc_exflags;
 799                 *credanonp = &np->netc_anon;
 800                 error = 0;
 801         } else {
 802                 error = EACCES;
 803         }
 804         return (error);
 805
 806 }
 807
 808 int
 809 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
 810 {
 811         hammer_mount_t hmp = (void *)mp->mnt_data;
 812         int error;
 813
 814         switch(op) {
 815         case MOUNTCTL_SET_EXPORT:
 816                 error = vfs_export(mp, &hmp->export, export);
 817                 break;
 818         default:
 819                 error = EOPNOTSUPP;
 820                 break;
 821         }
 822         return(error);
 823 }
 824