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_supported_version = HAMMER_VOL_VERSION_TWO;
  51 int hammer_debug_io;
  52 int hammer_debug_general;
  53 int hammer_debug_debug = 1;             /* medium-error panics */
  54 int hammer_debug_inode;
  55 int hammer_debug_locks;
  56 int hammer_debug_btree;
  57 int hammer_debug_tid;
  58 int hammer_debug_recover;               /* -1 will disable, +1 will force */
  59 int hammer_debug_recover_faults;
  60 int hammer_cluster_enable = 1;          /* enable read clustering by default */
  61 int hammer_count_fsyncs;
  62 int hammer_count_inodes;
  63 int hammer_count_iqueued;
  64 int hammer_count_reclaiming;
  65 int hammer_count_records;
  66 int hammer_count_record_datas;
  67 int hammer_count_volumes;
  68 int hammer_count_buffers;
  69 int hammer_count_nodes;
  70 int64_t hammer_count_extra_space_used;
  71 int64_t hammer_stats_btree_lookups;
  72 int64_t hammer_stats_btree_searches;
  73 int64_t hammer_stats_btree_inserts;
  74 int64_t hammer_stats_btree_deletes;
  75 int64_t hammer_stats_btree_elements;
  76 int64_t hammer_stats_btree_splits;
  77 int64_t hammer_stats_btree_iterations;
  78 int64_t hammer_stats_btree_root_iterations;
  79 int64_t hammer_stats_record_iterations;
  80
  81 int64_t hammer_stats_file_read;
  82 int64_t hammer_stats_file_write;
  83 int64_t hammer_stats_file_iopsr;
  84 int64_t hammer_stats_file_iopsw;
  85 int64_t hammer_stats_disk_read;
  86 int64_t hammer_stats_disk_write;
  87 int64_t hammer_stats_inode_flushes;
  88 int64_t hammer_stats_commits;
  89
  90 int hammer_count_dirtybufspace;         /* global */
  91 int hammer_count_refedbufs;             /* global */
  92 int hammer_count_reservations;
  93 int hammer_count_io_running_read;
  94 int hammer_count_io_running_write;
  95 int hammer_count_io_locked;
  96 int hammer_limit_dirtybufspace;         /* per-mount */
  97 int hammer_limit_recs;                  /* as a whole XXX */
  98 int hammer_autoflush = 2000;            /* auto flush */
  99 int hammer_bio_count;
 100 int hammer_verify_zone;
 101 int hammer_verify_data = 1;
 102 int hammer_write_mode;
 103 int64_t hammer_contention_count;
 104 int64_t hammer_zone_limit;
 105
 106 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
 107 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
 108            &hammer_supported_version, 0, "");
 109 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
 110            &hammer_debug_general, 0, "");
 111 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
 112            &hammer_debug_io, 0, "");
 113 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
 114            &hammer_debug_debug, 0, "");
 115 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 116            &hammer_debug_inode, 0, "");
 117 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 118            &hammer_debug_locks, 0, "");
 119 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 120            &hammer_debug_btree, 0, "");
 121 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 122            &hammer_debug_tid, 0, "");
 123 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 124            &hammer_debug_recover, 0, "");
 125 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 126            &hammer_debug_recover_faults, 0, "");
 127 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
 128            &hammer_cluster_enable, 0, "");
 129
 130 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 131            &hammer_limit_dirtybufspace, 0, "");
 132 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 133            &hammer_limit_recs, 0, "");
 134
 135 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 136            &hammer_count_fsyncs, 0, "");
 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 138            &hammer_count_inodes, 0, "");
 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 140            &hammer_count_iqueued, 0, "");
 141 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 142            &hammer_count_reclaiming, 0, "");
 143 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 144            &hammer_count_records, 0, "");
 145 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 146            &hammer_count_record_datas, 0, "");
 147 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 148            &hammer_count_volumes, 0, "");
 149 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 150            &hammer_count_buffers, 0, "");
 151 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 152            &hammer_count_nodes, 0, "");
 153 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
 154            &hammer_count_extra_space_used, 0, "");
 155
 156 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 157            &hammer_stats_btree_searches, 0, "");
 158 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 159            &hammer_stats_btree_lookups, 0, "");
 160 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 161            &hammer_stats_btree_inserts, 0, "");
 162 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 163            &hammer_stats_btree_deletes, 0, "");
 164 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 165            &hammer_stats_btree_elements, 0, "");
 166 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 167            &hammer_stats_btree_splits, 0, "");
 168 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 169            &hammer_stats_btree_iterations, 0, "");
 170 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
 171            &hammer_stats_btree_root_iterations, 0, "");
 172 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 173            &hammer_stats_record_iterations, 0, "");
 174
 175 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
 176            &hammer_stats_file_read, 0, "");
 177 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
 178            &hammer_stats_file_write, 0, "");
 179 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
 180            &hammer_stats_file_iopsr, 0, "");
 181 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
 182            &hammer_stats_file_iopsw, 0, "");
 183 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
 184            &hammer_stats_disk_read, 0, "");
 185 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
 186            &hammer_stats_disk_write, 0, "");
 187 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
 188            &hammer_stats_inode_flushes, 0, "");
 189 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
 190            &hammer_stats_commits, 0, "");
 191
 192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 193            &hammer_count_dirtybufspace, 0, "");
 194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 195            &hammer_count_refedbufs, 0, "");
 196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 197            &hammer_count_reservations, 0, "");
 198 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 199            &hammer_count_io_running_read, 0, "");
 200 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 201            &hammer_count_io_locked, 0, "");
 202 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 203            &hammer_count_io_running_write, 0, "");
 204 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 205            &hammer_zone_limit, 0, "");
 206 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 207            &hammer_contention_count, 0, "");
 208 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
 209            &hammer_autoflush, 0, "");
 210 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 211            &hammer_verify_zone, 0, "");
 212 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
 213            &hammer_verify_data, 0, "");
 214 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 215            &hammer_write_mode, 0, "");
 216
 217 KTR_INFO_MASTER(hammer);
 218
 219 /*
 220  * VFS ABI
 221  */
 222 static void     hammer_free_hmp(struct mount *mp);
 223
 224 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 225                                 struct ucred *cred);
 226 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 227 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 228 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 229                                 struct ucred *cred);
 230 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 231                                 struct ucred *cred);
 232 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 233 static int      hammer_vfs_vget(struct mount *mp, ino_t ino,
 234                                 struct vnode **vpp);
 235 static int      hammer_vfs_init(struct vfsconf *conf);
 236 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 237                                 struct fid *fhp, struct vnode **vpp);
 238 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 239 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 240                                 int *exflagsp, struct ucred **credanonp);
 241
 242
 243 static struct vfsops hammer_vfsops = {
 244         .vfs_mount      = hammer_vfs_mount,
 245         .vfs_unmount    = hammer_vfs_unmount,
 246         .vfs_root       = hammer_vfs_root,
 247         .vfs_statfs     = hammer_vfs_statfs,
 248         .vfs_statvfs    = hammer_vfs_statvfs,
 249         .vfs_sync       = hammer_vfs_sync,
 250         .vfs_vget       = hammer_vfs_vget,
 251         .vfs_init       = hammer_vfs_init,
 252         .vfs_vptofh     = hammer_vfs_vptofh,
 253         .vfs_fhtovp     = hammer_vfs_fhtovp,
 254         .vfs_checkexp   = hammer_vfs_checkexp
 255 };
 256
 257 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
 258
 259 VFS_SET(hammer_vfsops, hammer, 0);
 260 MODULE_VERSION(hammer, 1);
 261
 262 static int
 263 hammer_vfs_init(struct vfsconf *conf)
 264 {
 265         int n;
 266
 267         if (hammer_limit_recs == 0) {
 268                 hammer_limit_recs = nbuf * 25;
 269                 n = kmalloc_limit(M_HAMMER) / 512;
 270                 if (hammer_limit_recs > n)
 271                         hammer_limit_recs = n;
 272         }
 273         if (hammer_limit_dirtybufspace == 0) {
 274                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 275                 if (hammer_limit_dirtybufspace < 100)
 276                         hammer_limit_dirtybufspace = 100;
 277         }
 278         return(0);
 279 }
 280
 281 static int
 282 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 283                  struct ucred *cred)
 284 {
 285         struct hammer_mount_info info;
 286         hammer_mount_t hmp;
 287         hammer_volume_t rootvol;
 288         struct vnode *rootvp;
 289         struct vnode *devvp = NULL;
 290         const char *upath;      /* volume name in userspace */
 291         char *path;             /* volume name in system space */
 292         int error;
 293         int i;
 294         int master_id;
 295         int maxinodes;
 296
 297         /*
 298          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
 299          */
 300         if (mntpt == NULL) {
 301                 if ((error = bdevvp(rootdev, &devvp))) {
 302                         kprintf("hammer_mountroot: can't find devvp\n");
 303                         return (error);
 304                 }
 305                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 306                 bzero(&info, sizeof(info));
 307                 info.asof = 0;
 308                 info.hflags = 0;
 309                 info.nvolumes = 1;
 310         } else {
 311                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 312                         return (error);
 313         }
 314
 315         /*
 316          * updating or new mount
 317          */
 318         if (mp->mnt_flag & MNT_UPDATE) {
 319                 hmp = (void *)mp->mnt_data;
 320                 KKASSERT(hmp != NULL);
 321         } else {
 322                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 323                         return (EINVAL);
 324                 hmp = NULL;
 325         }
 326
 327         /*
 328          * master-id validation.  The master id may not be changed by a
 329          * mount update.
 330          */
 331         if (info.hflags & HMNT_MASTERID) {
 332                 if (hmp && hmp->master_id != info.master_id) {
 333                         kprintf("hammer: cannot change master id "
 334                                 "with mount update\n");
 335                         return(EINVAL);
 336                 }
 337                 master_id = info.master_id;
 338                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 339                         return (EINVAL);
 340         } else {
 341                 if (hmp)
 342                         master_id = hmp->master_id;
 343                 else
 344                         master_id = 0;
 345         }
 346
 347         /*
 348          * Interal mount data structure
 349          */
 350         if (hmp == NULL) {
 351                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 352                 mp->mnt_data = (qaddr_t)hmp;
 353                 hmp->mp = mp;
 354                 /*TAILQ_INIT(&hmp->recycle_list);*/
 355
 356                 /*
 357                  * Make sure kmalloc type limits are set appropriately.  If root
 358                  * increases the vnode limit you may have to do a dummy remount
 359                  * to adjust the HAMMER inode limit.
 360                  */
 361                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
 362                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
 363
 364                 maxinodes = desiredvnodes + desiredvnodes / 5 +
 365                             HAMMER_RECLAIM_WAIT;
 366                 kmalloc_raise_limit(hmp->m_inodes,
 367                                     maxinodes * sizeof(struct hammer_inode));
 368
 369                 hmp->root_btree_beg.localization = 0x00000000U;
 370                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 371                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 372                 hmp->root_btree_beg.create_tid = 1;
 373                 hmp->root_btree_beg.delete_tid = 1;
 374                 hmp->root_btree_beg.rec_type = 0;
 375                 hmp->root_btree_beg.obj_type = 0;
 376
 377                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 378                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 379                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 380                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 381                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 382                 hmp->root_btree_end.rec_type = 0xFFFFU;
 383                 hmp->root_btree_end.obj_type = 0;
 384
 385                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 386                 hmp->krate.count = -16; /* initial burst */
 387
 388                 hmp->sync_lock.refs = 1;
 389                 hmp->free_lock.refs = 1;
 390                 hmp->undo_lock.refs = 1;
 391                 hmp->blkmap_lock.refs = 1;
 392
 393                 TAILQ_INIT(&hmp->delay_list);
 394                 TAILQ_INIT(&hmp->flush_group_list);
 395                 TAILQ_INIT(&hmp->objid_cache_list);
 396                 TAILQ_INIT(&hmp->undo_lru_list);
 397                 TAILQ_INIT(&hmp->reclaim_list);
 398         }
 399         hmp->hflags &= ~HMNT_USERFLAGS;
 400         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 401
 402         hmp->master_id = master_id;
 403
 404         if (info.asof) {
 405                 mp->mnt_flag |= MNT_RDONLY;
 406                 hmp->asof = info.asof;
 407         } else {
 408                 hmp->asof = HAMMER_MAX_TID;
 409         }
 410
 411         /*
 412          * Re-open read-write if originally read-only, or vise-versa.
 413          */
 414         if (mp->mnt_flag & MNT_UPDATE) {
 415                 error = 0;
 416                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 417                         kprintf("HAMMER read-only -> read-write\n");
 418                         hmp->ronly = 0;
 419                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 420                                 hammer_adjust_volume_mode, NULL);
 421                         rootvol = hammer_get_root_volume(hmp, &error);
 422                         if (rootvol) {
 423                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 424                                 bcopy(rootvol->ondisk->vol0_blockmap,
 425                                       hmp->blockmap,
 426                                       sizeof(hmp->blockmap));
 427                                 hammer_rel_volume(rootvol, 0);
 428                         }
 429                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 430                                 hammer_reload_inode, NULL);
 431                         /* kernel clears MNT_RDONLY */
 432                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 433                         kprintf("HAMMER read-write -> read-only\n");
 434                         hmp->ronly = 1; /* messy */
 435                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 436                                 hammer_reload_inode, NULL);
 437                         hmp->ronly = 0;
 438                         hammer_flusher_sync(hmp);
 439                         hammer_flusher_sync(hmp);
 440                         hammer_flusher_sync(hmp);
 441                         hmp->ronly = 1;
 442                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 443                                 hammer_adjust_volume_mode, NULL);
 444                 }
 445                 return(error);
 446         }
 447
 448         RB_INIT(&hmp->rb_vols_root);
 449         RB_INIT(&hmp->rb_inos_root);
 450         RB_INIT(&hmp->rb_nods_root);
 451         RB_INIT(&hmp->rb_undo_root);
 452         RB_INIT(&hmp->rb_resv_root);
 453         RB_INIT(&hmp->rb_bufs_root);
 454         RB_INIT(&hmp->rb_pfsm_root);
 455
 456         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 457
 458         TAILQ_INIT(&hmp->volu_list);
 459         TAILQ_INIT(&hmp->undo_list);
 460         TAILQ_INIT(&hmp->data_list);
 461         TAILQ_INIT(&hmp->meta_list);
 462         TAILQ_INIT(&hmp->lose_list);
 463
 464         /*
 465          * Load volumes
 466          */
 467         path = objcache_get(namei_oc, M_WAITOK);
 468         hmp->nvolumes = -1;
 469         for (i = 0; i < info.nvolumes; ++i) {
 470                 if (mntpt == NULL) {
 471                         /*
 472                          * Root mount.
 473                          * Only one volume; and no need for copyin.
 474                          */
 475                         KKASSERT(info.nvolumes == 1);
 476                         ksnprintf(path, MAXPATHLEN, "/dev/%s",
 477                                   mp->mnt_stat.f_mntfromname);
 478                         error = 0;
 479                 } else {
 480                         error = copyin(&info.volumes[i], &upath,
 481                                        sizeof(char *));
 482                         if (error == 0)
 483                                 error = copyinstr(upath, path,
 484                                                   MAXPATHLEN, NULL);
 485                 }
 486                 if (error == 0)
 487                         error = hammer_install_volume(hmp, path, devvp);
 488                 if (error)
 489                         break;
 490         }
 491         objcache_put(namei_oc, path);
 492
 493         /*
 494          * Make sure we found a root volume
 495          */
 496         if (error == 0 && hmp->rootvol == NULL) {
 497                 kprintf("hammer_mount: No root volume found!\n");
 498                 error = EINVAL;
 499         }
 500
 501         /*
 502          * Check that all required volumes are available
 503          */
 504         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 505                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 506                 error = EINVAL;
 507         }
 508
 509         if (error) {
 510                 hammer_free_hmp(mp);
 511                 return (error);
 512         }
 513
 514         /*
 515          * No errors, setup enough of the mount point so we can lookup the
 516          * root vnode.
 517          */
 518         mp->mnt_iosize_max = MAXPHYS;
 519         mp->mnt_kern_flag |= MNTK_FSMID;
 520
 521         /*
 522          * note: f_iosize is used by vnode_pager_haspage() when constructing
 523          * its VOP_BMAP call.
 524          */
 525         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 526         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 527
 528         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 529         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 530
 531         mp->mnt_maxsymlinklen = 255;
 532         mp->mnt_flag |= MNT_LOCAL;
 533
 534         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 535         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 536         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 537
 538         /*
 539          * The root volume's ondisk pointer is only valid if we hold a
 540          * reference to it.
 541          */
 542         rootvol = hammer_get_root_volume(hmp, &error);
 543         if (error)
 544                 goto failed;
 545
 546         /*
 547          * Perform any necessary UNDO operations.  The recovery code does
 548          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 549          * and then re-copy it again after recovery is complete.
 550          *
 551          * If this is a read-only mount the UNDO information is retained
 552          * in memory in the form of dirty buffer cache buffers, and not
 553          * written back to the media.
 554          */
 555         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 556               sizeof(hmp->blockmap));
 557
 558         /*
 559          * Check filesystem version
 560          */
 561         hmp->version = rootvol->ondisk->vol_version;
 562         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
 563             hmp->version > HAMMER_VOL_VERSION_MAX) {
 564                 kprintf("HAMMER: mount unsupported fs version %d\n",
 565                         hmp->version);
 566                 error = ERANGE;
 567                 goto done;
 568         }
 569
 570         /*
 571          * The undo_rec_limit limits the size of flush groups to avoid
 572          * blowing out the UNDO FIFO.  This calculation is typically in
 573          * the tens of thousands and is designed primarily when small
 574          * HAMMER filesystems are created.
 575          */
 576         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 577         if (hammer_debug_general & 0x0001)
 578                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 579
 580         error = hammer_recover(hmp, rootvol);
 581         if (error) {
 582                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 583                 goto done;
 584         }
 585
 586         /*
 587          * Finish setup now that we have a good root volume.
 588          *
 589          * The top 16 bits of fsid.val[1] is a pfs id.
 590          */
 591         ksnprintf(mp->mnt_stat.f_mntfromname,
 592                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 593                   rootvol->ondisk->vol_name);
 594         mp->mnt_stat.f_fsid.val[0] =
 595                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 596         mp->mnt_stat.f_fsid.val[1] =
 597                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 598         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 599
 600         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 601         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 602                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 603
 604         /*
 605          * Certain often-modified fields in the root volume are cached in
 606          * the hammer_mount structure so we do not have to generate lots
 607          * of little UNDO structures for them.
 608          *
 609          * Recopy after recovery.  This also has the side effect of
 610          * setting our cached undo FIFO's first_offset, which serves to
 611          * placemark the FIFO start for the NEXT flush cycle while the
 612          * on-disk first_offset represents the LAST flush cycle.
 613          */
 614         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 615         hmp->flush_tid1 = hmp->next_tid;
 616         hmp->flush_tid2 = hmp->next_tid;
 617         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 618               sizeof(hmp->blockmap));
 619         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 620
 621         hammer_flusher_create(hmp);
 622
 623         /*
 624          * Locate the root directory using the root cluster's B-Tree as a
 625          * starting point.  The root directory uses an obj_id of 1.
 626          *
 627          * FUTURE: Leave the root directory cached referenced but unlocked
 628          * in hmp->rootvp (need to flush it on unmount).
 629          */
 630         error = hammer_vfs_vget(mp, 1, &rootvp);
 631         if (error)
 632                 goto done;
 633         vput(rootvp);
 634         /*vn_unlock(hmp->rootvp);*/
 635
 636 done:
 637         hammer_rel_volume(rootvol, 0);
 638 failed:
 639         /*
 640          * Cleanup and return.
 641          */
 642         if (error)
 643                 hammer_free_hmp(mp);
 644         return (error);
 645 }
 646
 647 static int
 648 hammer_vfs_unmount(struct mount *mp, int mntflags)
 649 {
 650 #if 0
 651         struct hammer_mount *hmp = (void *)mp->mnt_data;
 652 #endif
 653         int flags;
 654         int error;
 655
 656         /*
 657          * Clean out the vnodes
 658          */
 659         flags = 0;
 660         if (mntflags & MNT_FORCE)
 661                 flags |= FORCECLOSE;
 662         if ((error = vflush(mp, 0, flags)) != 0)
 663                 return (error);
 664
 665         /*
 666          * Clean up the internal mount structure and related entities.  This
 667          * may issue I/O.
 668          */
 669         hammer_free_hmp(mp);
 670         return(0);
 671 }
 672
 673 /*
 674  * Clean up the internal mount structure and disassociate it from the mount.
 675  * This may issue I/O.
 676  */
 677 static void
 678 hammer_free_hmp(struct mount *mp)
 679 {
 680         struct hammer_mount *hmp = (void *)mp->mnt_data;
 681         hammer_flush_group_t flg;
 682         int count;
 683         int dummy;
 684
 685         /*
 686          * Flush anything dirty.  This won't even run if the
 687          * filesystem errored-out.
 688          */
 689         count = 0;
 690         while (hammer_flusher_haswork(hmp)) {
 691                 hammer_flusher_sync(hmp);
 692                 ++count;
 693                 if (count >= 5) {
 694                         if (count == 5)
 695                                 kprintf("HAMMER: umount flushing.");
 696                         else
 697                                 kprintf(".");
 698                         tsleep(&dummy, 0, "hmrufl", hz);
 699                 }
 700                 if (count == 30) {
 701                         kprintf("giving up\n");
 702                         break;
 703                 }
 704         }
 705         if (count >= 5 && count < 30)
 706                 kprintf("\n");
 707
 708         /*
 709          * If the mount had a critical error we have to destroy any
 710          * remaining inodes before we can finish cleaning up the flusher.
 711          */
 712         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 713                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 714                         hammer_destroy_inode_callback, NULL);
 715         }
 716
 717         /*
 718          * There shouldn't be any inodes left now and any left over
 719          * flush groups should now be empty.
 720          */
 721         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 722         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 723                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 724                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
 725                 if (flg->refs) {
 726                         kprintf("HAMMER: Warning, flush_group %p was "
 727                                 "not empty on umount!\n", flg);
 728                 }
 729                 kfree(flg, hmp->m_misc);
 730         }
 731
 732         /*
 733          * We can finally destroy the flusher
 734          */
 735         hammer_flusher_destroy(hmp);
 736
 737         /*
 738          * We may have held recovered buffers due to a read-only mount.
 739          * These must be discarded.
 740          */
 741         if (hmp->ronly)
 742                 hammer_recover_flush_buffers(hmp, NULL, -1);
 743
 744         /*
 745          * Unload buffers and then volumes
 746          */
 747         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 748                 hammer_unload_buffer, NULL);
 749         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 750                 hammer_unload_volume, NULL);
 751
 752         mp->mnt_data = NULL;
 753         mp->mnt_flag &= ~MNT_LOCAL;
 754         hmp->mp = NULL;
 755         hammer_destroy_objid_cache(hmp);
 756         kmalloc_destroy(&hmp->m_misc);
 757         kmalloc_destroy(&hmp->m_inodes);
 758         kfree(hmp, M_HAMMER);
 759 }
 760
 761 /*
 762  * Report critical errors.  ip may be NULL.
 763  */
 764 void
 765 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 766                       int error, const char *msg)
 767 {
 768         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 769         krateprintf(&hmp->krate,
 770                 "HAMMER(%s): Critical error inode=%lld %s\n",
 771                 hmp->mp->mnt_stat.f_mntfromname,
 772                 (long long)(ip ? ip->obj_id : -1), msg);
 773         if (hmp->ronly == 0) {
 774                 hmp->ronly = 2;         /* special errored read-only mode */
 775                 hmp->mp->mnt_flag |= MNT_RDONLY;
 776                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 777                         hmp->mp->mnt_stat.f_mntfromname);
 778         }
 779         hmp->error = error;
 780 }
 781
 782
 783 /*
 784  * Obtain a vnode for the specified inode number.  An exclusively locked
 785  * vnode is returned.
 786  */
 787 int
 788 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 789 {
 790         struct hammer_transaction trans;
 791         struct hammer_mount *hmp = (void *)mp->mnt_data;
 792         struct hammer_inode *ip;
 793         int error;
 794
 795         hammer_simple_transaction(&trans, hmp);
 796
 797         /*
 798          * Lookup the requested HAMMER inode.  The structure must be
 799          * left unlocked while we manipulate the related vnode to avoid
 800          * a deadlock.
 801          */
 802         ip = hammer_get_inode(&trans, NULL, ino,
 803                               hmp->asof, HAMMER_DEF_LOCALIZATION,
 804                               0, &error);
 805         if (ip == NULL) {
 806                 *vpp = NULL;
 807                 hammer_done_transaction(&trans);
 808                 return(error);
 809         }
 810         error = hammer_get_vnode(ip, vpp);
 811         hammer_rel_inode(ip, 0);
 812         hammer_done_transaction(&trans);
 813         return (error);
 814 }
 815
 816 /*
 817  * Return the root vnode for the filesystem.
 818  *
 819  * HAMMER stores the root vnode in the hammer_mount structure so
 820  * getting it is easy.
 821  */
 822 static int
 823 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 824 {
 825 #if 0
 826         struct hammer_mount *hmp = (void *)mp->mnt_data;
 827 #endif
 828         int error;
 829
 830         error = hammer_vfs_vget(mp, 1, vpp);
 831         return (error);
 832 }
 833
 834 static int
 835 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 836 {
 837         struct hammer_mount *hmp = (void *)mp->mnt_data;
 838         hammer_volume_t volume;
 839         hammer_volume_ondisk_t ondisk;
 840         int error;
 841         int64_t bfree;
 842         int64_t breserved;
 843
 844         volume = hammer_get_root_volume(hmp, &error);
 845         if (error)
 846                 return(error);
 847         ondisk = volume->ondisk;
 848
 849         /*
 850          * Basic stats
 851          */
 852         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
 853         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 854         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 855         hammer_rel_volume(volume, 0);
 856
 857         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
 858         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 859         if (mp->mnt_stat.f_files < 0)
 860                 mp->mnt_stat.f_files = 0;
 861
 862         *sbp = mp->mnt_stat;
 863         return(0);
 864 }
 865
 866 static int
 867 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 868 {
 869         struct hammer_mount *hmp = (void *)mp->mnt_data;
 870         hammer_volume_t volume;
 871         hammer_volume_ondisk_t ondisk;
 872         int error;
 873         int64_t bfree;
 874         int64_t breserved;
 875
 876         volume = hammer_get_root_volume(hmp, &error);
 877         if (error)
 878                 return(error);
 879         ondisk = volume->ondisk;
 880
 881         /*
 882          * Basic stats
 883          */
 884         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
 885         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 886         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 887         hammer_rel_volume(volume, 0);
 888
 889         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
 890         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 891         if (mp->mnt_vstat.f_files < 0)
 892                 mp->mnt_vstat.f_files = 0;
 893         *sbp = mp->mnt_vstat;
 894         return(0);
 895 }
 896
 897 /*
 898  * Sync the filesystem.  Currently we have to run it twice, the second
 899  * one will advance the undo start index to the end index, so if a crash
 900  * occurs no undos will be run on mount.
 901  *
 902  * We do not sync the filesystem if we are called from a panic.  If we did
 903  * we might end up blowing up a sync that was already in progress.
 904  */
 905 static int
 906 hammer_vfs_sync(struct mount *mp, int waitfor)
 907 {
 908         struct hammer_mount *hmp = (void *)mp->mnt_data;
 909         int error;
 910
 911         if (panicstr == NULL) {
 912                 error = hammer_sync_hmp(hmp, waitfor);
 913         } else {
 914                 error = EIO;
 915         }
 916         return (error);
 917 }
 918
 919 /*
 920  * Convert a vnode to a file handle.
 921  */
 922 static int
 923 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 924 {
 925         hammer_inode_t ip;
 926
 927         KKASSERT(MAXFIDSZ >= 16);
 928         ip = VTOI(vp);
 929         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 930         fhp->fid_ext = ip->obj_localization >> 16;
 931         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 932         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 933         return(0);
 934 }
 935
 936
 937 /*
 938  * Convert a file handle back to a vnode.
 939  *
 940  * Use rootvp to enforce PFS isolation when a PFS is exported via a
 941  * null mount.
 942  */
 943 static int
 944 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 945                   struct fid *fhp, struct vnode **vpp)
 946 {
 947         struct hammer_transaction trans;
 948         struct hammer_inode *ip;
 949         struct hammer_inode_info info;
 950         int error;
 951         u_int32_t localization;
 952
 953         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 954         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 955         if (rootvp)
 956                 localization = VTOI(rootvp)->obj_localization;
 957         else
 958                 localization = (u_int32_t)fhp->fid_ext << 16;
 959
 960         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 961
 962         /*
 963          * Get/allocate the hammer_inode structure.  The structure must be
 964          * unlocked while we manipulate the related vnode to avoid a
 965          * deadlock.
 966          */
 967         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 968                               info.obj_asof, localization, 0, &error);
 969         if (ip == NULL) {
 970                 *vpp = NULL;
 971                 return(error);
 972         }
 973         error = hammer_get_vnode(ip, vpp);
 974         hammer_rel_inode(ip, 0);
 975         hammer_done_transaction(&trans);
 976         return (error);
 977 }
 978
 979 static int
 980 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 981                     int *exflagsp, struct ucred **credanonp)
 982 {
 983         hammer_mount_t hmp = (void *)mp->mnt_data;
 984         struct netcred *np;
 985         int error;
 986
 987         np = vfs_export_lookup(mp, &hmp->export, nam);
 988         if (np) {
 989                 *exflagsp = np->netc_exflags;
 990                 *credanonp = &np->netc_anon;
 991                 error = 0;
 992         } else {
 993                 error = EACCES;
 994         }
 995         return (error);
 996
 997 }
 998
 999 int
1000 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1001 {
1002         hammer_mount_t hmp = (void *)mp->mnt_data;
1003         int error;
1004
1005         switch(op) {
1006         case MOUNTCTL_SET_EXPORT:
1007                 error = vfs_export(mp, &hmp->export, export);
1008                 break;
1009         default:
1010                 error = EOPNOTSUPP;
1011                 break;
1012         }
1013         return(error);
1014 }
1015