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 1.74 2008/11/13 02:18:43 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 = 1;             /* medium-error panics */
  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_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
  79 int64_t hammer_stats_file_read;
  80 int64_t hammer_stats_file_write;
  81 int64_t hammer_stats_file_iopsr;
  82 int64_t hammer_stats_file_iopsw;
  83 int64_t hammer_stats_disk_read;
  84 int64_t hammer_stats_disk_write;
  85 int64_t hammer_stats_inode_flushes;
  86 int64_t hammer_stats_commits;
  87
  88 int hammer_count_dirtybufspace;         /* global */
  89 int hammer_count_refedbufs;             /* global */
  90 int hammer_count_reservations;
  91 int hammer_count_io_running_read;
  92 int hammer_count_io_running_write;
  93 int hammer_count_io_locked;
  94 int hammer_limit_dirtybufspace;         /* per-mount */
  95 int hammer_limit_recs;                  /* as a whole XXX */
  96 int hammer_autoflush = 2000;            /* auto flush */
  97 int hammer_bio_count;
  98 int hammer_verify_zone;
  99 int hammer_verify_data = 1;
 100 int hammer_write_mode;
 101 int64_t hammer_contention_count;
 102 int64_t hammer_zone_limit;
 103
 104 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
 105 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
 106            &hammer_debug_general, 0, "");
 107 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
 108            &hammer_debug_io, 0, "");
 109 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
 110            &hammer_debug_debug, 0, "");
 111 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 112            &hammer_debug_inode, 0, "");
 113 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 114            &hammer_debug_locks, 0, "");
 115 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 116            &hammer_debug_btree, 0, "");
 117 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 118            &hammer_debug_tid, 0, "");
 119 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 120            &hammer_debug_recover, 0, "");
 121 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 122            &hammer_debug_recover_faults, 0, "");
 123 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
 124            &hammer_cluster_enable, 0, "");
 125
 126 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 127            &hammer_limit_dirtybufspace, 0, "");
 128 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 129            &hammer_limit_recs, 0, "");
 130
 131 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 132            &hammer_count_fsyncs, 0, "");
 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 134            &hammer_count_inodes, 0, "");
 135 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 136            &hammer_count_iqueued, 0, "");
 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 138            &hammer_count_reclaiming, 0, "");
 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 140            &hammer_count_records, 0, "");
 141 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 142            &hammer_count_record_datas, 0, "");
 143 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 144            &hammer_count_volumes, 0, "");
 145 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 146            &hammer_count_buffers, 0, "");
 147 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 148            &hammer_count_nodes, 0, "");
 149 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
 150            &hammer_count_extra_space_used, 0, "");
 151
 152 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 153            &hammer_stats_btree_searches, 0, "");
 154 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 155            &hammer_stats_btree_lookups, 0, "");
 156 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 157            &hammer_stats_btree_inserts, 0, "");
 158 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 159            &hammer_stats_btree_deletes, 0, "");
 160 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 161            &hammer_stats_btree_elements, 0, "");
 162 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 163            &hammer_stats_btree_splits, 0, "");
 164 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 165            &hammer_stats_btree_iterations, 0, "");
 166 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 167            &hammer_stats_record_iterations, 0, "");
 168
 169 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
 170            &hammer_stats_file_read, 0, "");
 171 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
 172            &hammer_stats_file_write, 0, "");
 173 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
 174            &hammer_stats_file_iopsr, 0, "");
 175 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
 176            &hammer_stats_file_iopsw, 0, "");
 177 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
 178            &hammer_stats_disk_read, 0, "");
 179 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
 180            &hammer_stats_disk_write, 0, "");
 181 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
 182            &hammer_stats_inode_flushes, 0, "");
 183 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
 184            &hammer_stats_commits, 0, "");
 185
 186 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 187            &hammer_count_dirtybufspace, 0, "");
 188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 189            &hammer_count_refedbufs, 0, "");
 190 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 191            &hammer_count_reservations, 0, "");
 192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 193            &hammer_count_io_running_read, 0, "");
 194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 195            &hammer_count_io_locked, 0, "");
 196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 197            &hammer_count_io_running_write, 0, "");
 198 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 199            &hammer_zone_limit, 0, "");
 200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 201            &hammer_contention_count, 0, "");
 202 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
 203            &hammer_autoflush, 0, "");
 204 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 205            &hammer_verify_zone, 0, "");
 206 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
 207            &hammer_verify_data, 0, "");
 208 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 209            &hammer_write_mode, 0, "");
 210
 211 KTR_INFO_MASTER(hammer);
 212
 213 /*
 214  * VFS ABI
 215  */
 216 static void     hammer_free_hmp(struct mount *mp);
 217
 218 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 219                                 struct ucred *cred);
 220 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 221 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 222 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 223                                 struct ucred *cred);
 224 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 225                                 struct ucred *cred);
 226 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 227 static int      hammer_vfs_vget(struct mount *mp, ino_t ino,
 228                                 struct vnode **vpp);
 229 static int      hammer_vfs_init(struct vfsconf *conf);
 230 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 231                                 struct fid *fhp, struct vnode **vpp);
 232 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 233 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 234                                 int *exflagsp, struct ucred **credanonp);
 235
 236
 237 static struct vfsops hammer_vfsops = {
 238         .vfs_mount      = hammer_vfs_mount,
 239         .vfs_unmount    = hammer_vfs_unmount,
 240         .vfs_root       = hammer_vfs_root,
 241         .vfs_statfs     = hammer_vfs_statfs,
 242         .vfs_statvfs    = hammer_vfs_statvfs,
 243         .vfs_sync       = hammer_vfs_sync,
 244         .vfs_vget       = hammer_vfs_vget,
 245         .vfs_init       = hammer_vfs_init,
 246         .vfs_vptofh     = hammer_vfs_vptofh,
 247         .vfs_fhtovp     = hammer_vfs_fhtovp,
 248         .vfs_checkexp   = hammer_vfs_checkexp
 249 };
 250
 251 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
 252
 253 VFS_SET(hammer_vfsops, hammer, 0);
 254 MODULE_VERSION(hammer, 1);
 255
 256 static int
 257 hammer_vfs_init(struct vfsconf *conf)
 258 {
 259         int n;
 260
 261         if (hammer_limit_recs == 0) {
 262                 hammer_limit_recs = nbuf * 25;
 263                 n = kmalloc_limit(M_HAMMER) / 512;
 264                 if (hammer_limit_recs > n)
 265                         hammer_limit_recs = n;
 266         }
 267         if (hammer_limit_dirtybufspace == 0) {
 268                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 269                 if (hammer_limit_dirtybufspace < 100)
 270                         hammer_limit_dirtybufspace = 100;
 271         }
 272         return(0);
 273 }
 274
 275 static int
 276 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 277                  struct ucred *cred)
 278 {
 279         struct hammer_mount_info info;
 280         hammer_mount_t hmp;
 281         hammer_volume_t rootvol;
 282         struct vnode *rootvp;
 283         struct vnode *devvp = NULL;
 284         const char *upath;      /* volume name in userspace */
 285         char *path;             /* volume name in system space */
 286         int error;
 287         int i;
 288         int master_id;
 289         int maxinodes;
 290
 291         /*
 292          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
 293          */
 294         if (mntpt == NULL) {
 295                 if ((error = bdevvp(rootdev, &devvp))) {
 296                         kprintf("hammer_mountroot: can't find devvp\n");
 297                         return (error);
 298                 }
 299                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 300                 bzero(&info, sizeof(info));
 301                 info.asof = 0;
 302                 info.hflags = 0;
 303                 info.nvolumes = 1;
 304         } else {
 305                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 306                         return (error);
 307         }
 308
 309         /*
 310          * updating or new mount
 311          */
 312         if (mp->mnt_flag & MNT_UPDATE) {
 313                 hmp = (void *)mp->mnt_data;
 314                 KKASSERT(hmp != NULL);
 315         } else {
 316                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 317                         return (EINVAL);
 318                 hmp = NULL;
 319         }
 320
 321         /*
 322          * master-id validation.  The master id may not be changed by a
 323          * mount update.
 324          */
 325         if (info.hflags & HMNT_MASTERID) {
 326                 if (hmp && hmp->master_id != info.master_id) {
 327                         kprintf("hammer: cannot change master id "
 328                                 "with mount update\n");
 329                         return(EINVAL);
 330                 }
 331                 master_id = info.master_id;
 332                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 333                         return (EINVAL);
 334         } else {
 335                 if (hmp)
 336                         master_id = hmp->master_id;
 337                 else
 338                         master_id = 0;
 339         }
 340
 341         /*
 342          * Interal mount data structure
 343          */
 344         if (hmp == NULL) {
 345                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 346                 mp->mnt_data = (qaddr_t)hmp;
 347                 hmp->mp = mp;
 348                 /*TAILQ_INIT(&hmp->recycle_list);*/
 349
 350                 /*
 351                  * Make sure kmalloc type limits are set appropriately.  If root
 352                  * increases the vnode limit you may have to do a dummy remount
 353                  * to adjust the HAMMER inode limit.
 354                  */
 355                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
 356                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
 357
 358                 maxinodes = desiredvnodes + desiredvnodes / 5 +
 359                             HAMMER_RECLAIM_WAIT;
 360                 kmalloc_raise_limit(hmp->m_inodes,
 361                                     maxinodes * sizeof(struct hammer_inode));
 362
 363                 hmp->root_btree_beg.localization = 0x00000000U;
 364                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 365                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 366                 hmp->root_btree_beg.create_tid = 1;
 367                 hmp->root_btree_beg.delete_tid = 1;
 368                 hmp->root_btree_beg.rec_type = 0;
 369                 hmp->root_btree_beg.obj_type = 0;
 370
 371                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 372                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 373                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 374                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 375                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 376                 hmp->root_btree_end.rec_type = 0xFFFFU;
 377                 hmp->root_btree_end.obj_type = 0;
 378
 379                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 380                 hmp->krate.count = -16; /* initial burst */
 381
 382                 hmp->sync_lock.refs = 1;
 383                 hmp->free_lock.refs = 1;
 384                 hmp->undo_lock.refs = 1;
 385                 hmp->blkmap_lock.refs = 1;
 386
 387                 TAILQ_INIT(&hmp->delay_list);
 388                 TAILQ_INIT(&hmp->flush_group_list);
 389                 TAILQ_INIT(&hmp->objid_cache_list);
 390                 TAILQ_INIT(&hmp->undo_lru_list);
 391                 TAILQ_INIT(&hmp->reclaim_list);
 392         }
 393         hmp->hflags &= ~HMNT_USERFLAGS;
 394         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 395
 396         hmp->master_id = master_id;
 397
 398         if (info.asof) {
 399                 mp->mnt_flag |= MNT_RDONLY;
 400                 hmp->asof = info.asof;
 401         } else {
 402                 hmp->asof = HAMMER_MAX_TID;
 403         }
 404
 405         /*
 406          * Re-open read-write if originally read-only, or vise-versa.
 407          */
 408         if (mp->mnt_flag & MNT_UPDATE) {
 409                 error = 0;
 410                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 411                         kprintf("HAMMER read-only -> read-write\n");
 412                         hmp->ronly = 0;
 413                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 414                                 hammer_adjust_volume_mode, NULL);
 415                         rootvol = hammer_get_root_volume(hmp, &error);
 416                         if (rootvol) {
 417                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 418                                 bcopy(rootvol->ondisk->vol0_blockmap,
 419                                       hmp->blockmap,
 420                                       sizeof(hmp->blockmap));
 421                                 hammer_rel_volume(rootvol, 0);
 422                         }
 423                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 424                                 hammer_reload_inode, NULL);
 425                         /* kernel clears MNT_RDONLY */
 426                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 427                         kprintf("HAMMER read-write -> read-only\n");
 428                         hmp->ronly = 1; /* messy */
 429                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 430                                 hammer_reload_inode, NULL);
 431                         hmp->ronly = 0;
 432                         hammer_flusher_sync(hmp);
 433                         hammer_flusher_sync(hmp);
 434                         hammer_flusher_sync(hmp);
 435                         hmp->ronly = 1;
 436                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 437                                 hammer_adjust_volume_mode, NULL);
 438                 }
 439                 return(error);
 440         }
 441
 442         RB_INIT(&hmp->rb_vols_root);
 443         RB_INIT(&hmp->rb_inos_root);
 444         RB_INIT(&hmp->rb_nods_root);
 445         RB_INIT(&hmp->rb_undo_root);
 446         RB_INIT(&hmp->rb_resv_root);
 447         RB_INIT(&hmp->rb_bufs_root);
 448         RB_INIT(&hmp->rb_pfsm_root);
 449
 450         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 451
 452         TAILQ_INIT(&hmp->volu_list);
 453         TAILQ_INIT(&hmp->undo_list);
 454         TAILQ_INIT(&hmp->data_list);
 455         TAILQ_INIT(&hmp->meta_list);
 456         TAILQ_INIT(&hmp->lose_list);
 457
 458         /*
 459          * Load volumes
 460          */
 461         path = objcache_get(namei_oc, M_WAITOK);
 462         hmp->nvolumes = -1;
 463         for (i = 0; i < info.nvolumes; ++i) {
 464                 if (mntpt == NULL) {
 465                         /*
 466                          * Root mount.
 467                          * Only one volume; and no need for copyin.
 468                          */
 469                         KKASSERT(info.nvolumes == 1);
 470                         ksnprintf(path, MAXPATHLEN, "/dev/%s",
 471                                   mp->mnt_stat.f_mntfromname);
 472                         error = 0;
 473                 } else {
 474                         error = copyin(&info.volumes[i], &upath,
 475                                        sizeof(char *));
 476                         if (error == 0)
 477                                 error = copyinstr(upath, path,
 478                                                   MAXPATHLEN, NULL);
 479                 }
 480                 if (error == 0)
 481                         error = hammer_install_volume(hmp, path, devvp);
 482                 if (error)
 483                         break;
 484         }
 485         objcache_put(namei_oc, path);
 486
 487         /*
 488          * Make sure we found a root volume
 489          */
 490         if (error == 0 && hmp->rootvol == NULL) {
 491                 kprintf("hammer_mount: No root volume found!\n");
 492                 error = EINVAL;
 493         }
 494
 495         /*
 496          * Check that all required volumes are available
 497          */
 498         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 499                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 500                 error = EINVAL;
 501         }
 502
 503         if (error) {
 504                 hammer_free_hmp(mp);
 505                 return (error);
 506         }
 507
 508         /*
 509          * No errors, setup enough of the mount point so we can lookup the
 510          * root vnode.
 511          */
 512         mp->mnt_iosize_max = MAXPHYS;
 513         mp->mnt_kern_flag |= MNTK_FSMID;
 514
 515         /*
 516          * note: f_iosize is used by vnode_pager_haspage() when constructing
 517          * its VOP_BMAP call.
 518          */
 519         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 520         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 521
 522         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 523         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 524
 525         mp->mnt_maxsymlinklen = 255;
 526         mp->mnt_flag |= MNT_LOCAL;
 527
 528         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 529         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 530         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 531
 532         /*
 533          * The root volume's ondisk pointer is only valid if we hold a
 534          * reference to it.
 535          */
 536         rootvol = hammer_get_root_volume(hmp, &error);
 537         if (error)
 538                 goto failed;
 539
 540         /*
 541          * Perform any necessary UNDO operations.  The recovery code does
 542          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 543          * and then re-copy it again after recovery is complete.
 544          *
 545          * If this is a read-only mount the UNDO information is retained
 546          * in memory in the form of dirty buffer cache buffers, and not
 547          * written back to the media.
 548          */
 549         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 550               sizeof(hmp->blockmap));
 551
 552         /*
 553          * Check filesystem version
 554          */
 555         hmp->version = rootvol->ondisk->vol_version;
 556         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
 557             hmp->version > HAMMER_VOL_VERSION_MAX) {
 558                 kprintf("HAMMER: mount unsupported fs version %d\n",
 559                         hmp->version);
 560                 error = ERANGE;
 561                 goto done;
 562         }
 563
 564         /*
 565          * The undo_rec_limit limits the size of flush groups to avoid
 566          * blowing out the UNDO FIFO.  This calculation is typically in
 567          * the tens of thousands and is designed primarily when small
 568          * HAMMER filesystems are created.
 569          */
 570         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 571         if (hammer_debug_general & 0x0001)
 572                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 573
 574         error = hammer_recover(hmp, rootvol);
 575         if (error) {
 576                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 577                 goto done;
 578         }
 579
 580         /*
 581          * Finish setup now that we have a good root volume.
 582          *
 583          * The top 16 bits of fsid.val[1] is a pfs id.
 584          */
 585         ksnprintf(mp->mnt_stat.f_mntfromname,
 586                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 587                   rootvol->ondisk->vol_name);
 588         mp->mnt_stat.f_fsid.val[0] =
 589                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 590         mp->mnt_stat.f_fsid.val[1] =
 591                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 592         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 593
 594         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 595         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 596                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 597
 598         /*
 599          * Certain often-modified fields in the root volume are cached in
 600          * the hammer_mount structure so we do not have to generate lots
 601          * of little UNDO structures for them.
 602          *
 603          * Recopy after recovery.  This also has the side effect of
 604          * setting our cached undo FIFO's first_offset, which serves to
 605          * placemark the FIFO start for the NEXT flush cycle while the
 606          * on-disk first_offset represents the LAST flush cycle.
 607          */
 608         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 609         hmp->flush_tid1 = hmp->next_tid;
 610         hmp->flush_tid2 = hmp->next_tid;
 611         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 612               sizeof(hmp->blockmap));
 613         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 614
 615         hammer_flusher_create(hmp);
 616
 617         /*
 618          * Locate the root directory using the root cluster's B-Tree as a
 619          * starting point.  The root directory uses an obj_id of 1.
 620          *
 621          * FUTURE: Leave the root directory cached referenced but unlocked
 622          * in hmp->rootvp (need to flush it on unmount).
 623          */
 624         error = hammer_vfs_vget(mp, 1, &rootvp);
 625         if (error)
 626                 goto done;
 627         vput(rootvp);
 628         /*vn_unlock(hmp->rootvp);*/
 629
 630 done:
 631         hammer_rel_volume(rootvol, 0);
 632 failed:
 633         /*
 634          * Cleanup and return.
 635          */
 636         if (error)
 637                 hammer_free_hmp(mp);
 638         return (error);
 639 }
 640
 641 static int
 642 hammer_vfs_unmount(struct mount *mp, int mntflags)
 643 {
 644 #if 0
 645         struct hammer_mount *hmp = (void *)mp->mnt_data;
 646 #endif
 647         int flags;
 648         int error;
 649
 650         /*
 651          * Clean out the vnodes
 652          */
 653         flags = 0;
 654         if (mntflags & MNT_FORCE)
 655                 flags |= FORCECLOSE;
 656         if ((error = vflush(mp, 0, flags)) != 0)
 657                 return (error);
 658
 659         /*
 660          * Clean up the internal mount structure and related entities.  This
 661          * may issue I/O.
 662          */
 663         hammer_free_hmp(mp);
 664         return(0);
 665 }
 666
 667 /*
 668  * Clean up the internal mount structure and disassociate it from the mount.
 669  * This may issue I/O.
 670  */
 671 static void
 672 hammer_free_hmp(struct mount *mp)
 673 {
 674         struct hammer_mount *hmp = (void *)mp->mnt_data;
 675         hammer_flush_group_t flg;
 676         int count;
 677
 678         /*
 679          * Flush anything dirty.  This won't even run if the
 680          * filesystem errored-out.
 681          */
 682         count = 0;
 683         while (hammer_flusher_haswork(hmp)) {
 684                 hammer_flusher_sync(hmp);
 685                 ++count;
 686                 if (count >= 5) {
 687                         if (count == 5)
 688                                 kprintf("HAMMER: umount flushing.");
 689                         else
 690                                 kprintf(".");
 691                         tsleep(hmp, 0, "hmrufl", hz);
 692                 }
 693                 if (count == 30) {
 694                         kprintf("giving up\n");
 695                         break;
 696                 }
 697         }
 698         if (count >= 5 && count < 30)
 699                 kprintf("\n");
 700
 701         /*
 702          * If the mount had a critical error we have to destroy any
 703          * remaining inodes before we can finish cleaning up the flusher.
 704          */
 705         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 706                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 707                         hammer_destroy_inode_callback, NULL);
 708         }
 709
 710         /*
 711          * There shouldn't be any inodes left now and any left over
 712          * flush groups should now be empty.
 713          */
 714         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 715         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 716                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 717                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
 718                 if (flg->refs) {
 719                         kprintf("HAMMER: Warning, flush_group %p was "
 720                                 "not empty on umount!\n", flg);
 721                 }
 722                 kfree(flg, hmp->m_misc);
 723         }
 724
 725         /*
 726          * We can finally destroy the flusher
 727          */
 728         hammer_flusher_destroy(hmp);
 729
 730         /*
 731          * We may have held recovered buffers due to a read-only mount.
 732          * These must be discarded.
 733          */
 734         if (hmp->ronly)
 735                 hammer_recover_flush_buffers(hmp, NULL, -1);
 736
 737         /*
 738          * Unload buffers and then volumes
 739          */
 740         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 741                 hammer_unload_buffer, NULL);
 742         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 743                 hammer_unload_volume, NULL);
 744
 745         mp->mnt_data = NULL;
 746         mp->mnt_flag &= ~MNT_LOCAL;
 747         hmp->mp = NULL;
 748         hammer_destroy_objid_cache(hmp);
 749         kmalloc_destroy(&hmp->m_misc);
 750         kmalloc_destroy(&hmp->m_inodes);
 751         kfree(hmp, M_HAMMER);
 752 }
 753
 754 /*
 755  * Report critical errors.  ip may be NULL.
 756  */
 757 void
 758 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 759                       int error, const char *msg)
 760 {
 761         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 762         krateprintf(&hmp->krate,
 763                 "HAMMER(%s): Critical error inode=%lld %s\n",
 764                 hmp->mp->mnt_stat.f_mntfromname,
 765                 (ip ? ip->obj_id : -1), msg);
 766         if (hmp->ronly == 0) {
 767                 hmp->ronly = 2;         /* special errored read-only mode */
 768                 hmp->mp->mnt_flag |= MNT_RDONLY;
 769                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 770                         hmp->mp->mnt_stat.f_mntfromname);
 771         }
 772         hmp->error = error;
 773 }
 774
 775
 776 /*
 777  * Obtain a vnode for the specified inode number.  An exclusively locked
 778  * vnode is returned.
 779  */
 780 int
 781 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 782 {
 783         struct hammer_transaction trans;
 784         struct hammer_mount *hmp = (void *)mp->mnt_data;
 785         struct hammer_inode *ip;
 786         int error;
 787
 788         hammer_simple_transaction(&trans, hmp);
 789
 790         /*
 791          * Lookup the requested HAMMER inode.  The structure must be
 792          * left unlocked while we manipulate the related vnode to avoid
 793          * a deadlock.
 794          */
 795         ip = hammer_get_inode(&trans, NULL, ino,
 796                               hmp->asof, HAMMER_DEF_LOCALIZATION,
 797                               0, &error);
 798         if (ip == NULL) {
 799                 *vpp = NULL;
 800                 hammer_done_transaction(&trans);
 801                 return(error);
 802         }
 803         error = hammer_get_vnode(ip, vpp);
 804         hammer_rel_inode(ip, 0);
 805         hammer_done_transaction(&trans);
 806         return (error);
 807 }
 808
 809 /*
 810  * Return the root vnode for the filesystem.
 811  *
 812  * HAMMER stores the root vnode in the hammer_mount structure so
 813  * getting it is easy.
 814  */
 815 static int
 816 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 817 {
 818 #if 0
 819         struct hammer_mount *hmp = (void *)mp->mnt_data;
 820 #endif
 821         int error;
 822
 823         error = hammer_vfs_vget(mp, 1, vpp);
 824         return (error);
 825 }
 826
 827 static int
 828 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 829 {
 830         struct hammer_mount *hmp = (void *)mp->mnt_data;
 831         hammer_volume_t volume;
 832         hammer_volume_ondisk_t ondisk;
 833         int error;
 834         int64_t bfree;
 835
 836         volume = hammer_get_root_volume(hmp, &error);
 837         if (error)
 838                 return(error);
 839         ondisk = volume->ondisk;
 840
 841         /*
 842          * Basic stats
 843          */
 844         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 845         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 846         hammer_rel_volume(volume, 0);
 847
 848         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
 849         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 850         if (mp->mnt_stat.f_files < 0)
 851                 mp->mnt_stat.f_files = 0;
 852
 853         *sbp = mp->mnt_stat;
 854         return(0);
 855 }
 856
 857 static int
 858 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 859 {
 860         struct hammer_mount *hmp = (void *)mp->mnt_data;
 861         hammer_volume_t volume;
 862         hammer_volume_ondisk_t ondisk;
 863         int error;
 864         int64_t bfree;
 865
 866         volume = hammer_get_root_volume(hmp, &error);
 867         if (error)
 868                 return(error);
 869         ondisk = volume->ondisk;
 870
 871         /*
 872          * Basic stats
 873          */
 874         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 875         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 876         hammer_rel_volume(volume, 0);
 877
 878         mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
 879         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 880         if (mp->mnt_vstat.f_files < 0)
 881                 mp->mnt_vstat.f_files = 0;
 882         *sbp = mp->mnt_vstat;
 883         return(0);
 884 }
 885
 886 /*
 887  * Sync the filesystem.  Currently we have to run it twice, the second
 888  * one will advance the undo start index to the end index, so if a crash
 889  * occurs no undos will be run on mount.
 890  *
 891  * We do not sync the filesystem if we are called from a panic.  If we did
 892  * we might end up blowing up a sync that was already in progress.
 893  */
 894 static int
 895 hammer_vfs_sync(struct mount *mp, int waitfor)
 896 {
 897         struct hammer_mount *hmp = (void *)mp->mnt_data;
 898         int error;
 899
 900         if (panicstr == NULL) {
 901                 error = hammer_sync_hmp(hmp, waitfor);
 902         } else {
 903                 error = EIO;
 904         }
 905         return (error);
 906 }
 907
 908 /*
 909  * Convert a vnode to a file handle.
 910  */
 911 static int
 912 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 913 {
 914         hammer_inode_t ip;
 915
 916         KKASSERT(MAXFIDSZ >= 16);
 917         ip = VTOI(vp);
 918         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 919         fhp->fid_ext = ip->obj_localization >> 16;
 920         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 921         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 922         return(0);
 923 }
 924
 925
 926 /*
 927  * Convert a file handle back to a vnode.
 928  *
 929  * Use rootvp to enforce PFS isolation when a PFS is exported via a
 930  * null mount.
 931  */
 932 static int
 933 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 934                   struct fid *fhp, struct vnode **vpp)
 935 {
 936         struct hammer_transaction trans;
 937         struct hammer_inode *ip;
 938         struct hammer_inode_info info;
 939         int error;
 940         u_int32_t localization;
 941
 942         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 943         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 944         if (rootvp)
 945                 localization = VTOI(rootvp)->obj_localization;
 946         else
 947                 localization = (u_int32_t)fhp->fid_ext << 16;
 948
 949         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 950
 951         /*
 952          * Get/allocate the hammer_inode structure.  The structure must be
 953          * unlocked while we manipulate the related vnode to avoid a
 954          * deadlock.
 955          */
 956         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 957                               info.obj_asof, localization, 0, &error);
 958         if (ip == NULL) {
 959                 *vpp = NULL;
 960                 return(error);
 961         }
 962         error = hammer_get_vnode(ip, vpp);
 963         hammer_rel_inode(ip, 0);
 964         hammer_done_transaction(&trans);
 965         return (error);
 966 }
 967
 968 static int
 969 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 970                     int *exflagsp, struct ucred **credanonp)
 971 {
 972         hammer_mount_t hmp = (void *)mp->mnt_data;
 973         struct netcred *np;
 974         int error;
 975
 976         np = vfs_export_lookup(mp, &hmp->export, nam);
 977         if (np) {
 978                 *exflagsp = np->netc_exflags;
 979                 *credanonp = &np->netc_anon;
 980                 error = 0;
 981         } else {
 982                 error = EACCES;
 983         }
 984         return (error);
 985
 986 }
 987
 988 int
 989 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
 990 {
 991         hammer_mount_t hmp = (void *)mp->mnt_data;
 992         int error;
 993
 994         switch(op) {
 995         case MOUNTCTL_SET_EXPORT:
 996                 error = vfs_export(mp, &hmp->export, export);
 997                 break;
 998         default:
 999                 error = EOPNOTSUPP;
1000                 break;
1001         }
1002         return(error);
1003 }
1004