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_limit_inode_recs = 1024;     /* per inode */
  99 int hammer_autoflush = 2000;            /* auto flush */
 100 int hammer_bio_count;
 101 int hammer_verify_zone;
 102 int hammer_verify_data = 1;
 103 int hammer_write_mode;
 104 int hammer_yield_check = 16;
 105 int hammer_fsync_mode;
 106 int64_t hammer_contention_count;
 107 int64_t hammer_zone_limit;
 108
 109 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
 110 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
 111            &hammer_supported_version, 0, "");
 112 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
 113            &hammer_debug_general, 0, "");
 114 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
 115            &hammer_debug_io, 0, "");
 116 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
 117            &hammer_debug_debug, 0, "");
 118 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 119            &hammer_debug_inode, 0, "");
 120 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 121            &hammer_debug_locks, 0, "");
 122 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 123            &hammer_debug_btree, 0, "");
 124 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 125            &hammer_debug_tid, 0, "");
 126 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 127            &hammer_debug_recover, 0, "");
 128 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 129            &hammer_debug_recover_faults, 0, "");
 130 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
 131            &hammer_cluster_enable, 0, "");
 132
 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 134            &hammer_limit_dirtybufspace, 0, "");
 135 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 136            &hammer_limit_recs, 0, "");
 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
 138            &hammer_limit_inode_recs, 0, "");
 139
 140 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 141            &hammer_count_fsyncs, 0, "");
 142 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 143            &hammer_count_inodes, 0, "");
 144 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 145            &hammer_count_iqueued, 0, "");
 146 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 147            &hammer_count_reclaiming, 0, "");
 148 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 149            &hammer_count_records, 0, "");
 150 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 151            &hammer_count_record_datas, 0, "");
 152 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 153            &hammer_count_volumes, 0, "");
 154 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 155            &hammer_count_buffers, 0, "");
 156 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 157            &hammer_count_nodes, 0, "");
 158 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
 159            &hammer_count_extra_space_used, 0, "");
 160
 161 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 162            &hammer_stats_btree_searches, 0, "");
 163 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 164            &hammer_stats_btree_lookups, 0, "");
 165 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 166            &hammer_stats_btree_inserts, 0, "");
 167 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 168            &hammer_stats_btree_deletes, 0, "");
 169 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 170            &hammer_stats_btree_elements, 0, "");
 171 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 172            &hammer_stats_btree_splits, 0, "");
 173 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 174            &hammer_stats_btree_iterations, 0, "");
 175 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
 176            &hammer_stats_btree_root_iterations, 0, "");
 177 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 178            &hammer_stats_record_iterations, 0, "");
 179
 180 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
 181            &hammer_stats_file_read, 0, "");
 182 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
 183            &hammer_stats_file_write, 0, "");
 184 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
 185            &hammer_stats_file_iopsr, 0, "");
 186 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
 187            &hammer_stats_file_iopsw, 0, "");
 188 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
 189            &hammer_stats_disk_read, 0, "");
 190 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
 191            &hammer_stats_disk_write, 0, "");
 192 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
 193            &hammer_stats_inode_flushes, 0, "");
 194 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
 195            &hammer_stats_commits, 0, "");
 196
 197 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 198            &hammer_count_dirtybufspace, 0, "");
 199 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 200            &hammer_count_refedbufs, 0, "");
 201 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 202            &hammer_count_reservations, 0, "");
 203 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 204            &hammer_count_io_running_read, 0, "");
 205 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 206            &hammer_count_io_locked, 0, "");
 207 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 208            &hammer_count_io_running_write, 0, "");
 209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 210            &hammer_zone_limit, 0, "");
 211 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 212            &hammer_contention_count, 0, "");
 213 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
 214            &hammer_autoflush, 0, "");
 215 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 216            &hammer_verify_zone, 0, "");
 217 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
 218            &hammer_verify_data, 0, "");
 219 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 220            &hammer_write_mode, 0, "");
 221 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
 222            &hammer_yield_check, 0, "");
 223 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
 224            &hammer_fsync_mode, 0, "");
 225
 226 KTR_INFO_MASTER(hammer);
 227
 228 /*
 229  * VFS ABI
 230  */
 231 static void     hammer_free_hmp(struct mount *mp);
 232
 233 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 234                                 struct ucred *cred);
 235 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 236 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 237 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 238                                 struct ucred *cred);
 239 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 240                                 struct ucred *cred);
 241 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 242 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 243                                 ino_t ino, struct vnode **vpp);
 244 static int      hammer_vfs_init(struct vfsconf *conf);
 245 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 246                                 struct fid *fhp, struct vnode **vpp);
 247 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 248 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 249                                 int *exflagsp, struct ucred **credanonp);
 250
 251
 252 static struct vfsops hammer_vfsops = {
 253         .vfs_mount      = hammer_vfs_mount,
 254         .vfs_unmount    = hammer_vfs_unmount,
 255         .vfs_root       = hammer_vfs_root,
 256         .vfs_statfs     = hammer_vfs_statfs,
 257         .vfs_statvfs    = hammer_vfs_statvfs,
 258         .vfs_sync       = hammer_vfs_sync,
 259         .vfs_vget       = hammer_vfs_vget,
 260         .vfs_init       = hammer_vfs_init,
 261         .vfs_vptofh     = hammer_vfs_vptofh,
 262         .vfs_fhtovp     = hammer_vfs_fhtovp,
 263         .vfs_checkexp   = hammer_vfs_checkexp
 264 };
 265
 266 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
 267
 268 VFS_SET(hammer_vfsops, hammer, 0);
 269 MODULE_VERSION(hammer, 1);
 270
 271 static int
 272 hammer_vfs_init(struct vfsconf *conf)
 273 {
 274         int n;
 275
 276         if (hammer_limit_recs == 0) {
 277                 hammer_limit_recs = nbuf * 25;
 278                 n = kmalloc_limit(M_HAMMER) / 512;
 279                 if (hammer_limit_recs > n)
 280                         hammer_limit_recs = n;
 281         }
 282         if (hammer_limit_dirtybufspace == 0) {
 283                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 284                 if (hammer_limit_dirtybufspace < 100)
 285                         hammer_limit_dirtybufspace = 100;
 286         }
 287         return(0);
 288 }
 289
 290 static int
 291 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 292                  struct ucred *cred)
 293 {
 294         struct hammer_mount_info info;
 295         hammer_mount_t hmp;
 296         hammer_volume_t rootvol;
 297         struct vnode *rootvp;
 298         struct vnode *devvp = NULL;
 299         const char *upath;      /* volume name in userspace */
 300         char *path;             /* volume name in system space */
 301         int error;
 302         int i;
 303         int master_id;
 304         int maxinodes;
 305
 306         /*
 307          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
 308          */
 309         if (mntpt == NULL) {
 310                 if ((error = bdevvp(rootdev, &devvp))) {
 311                         kprintf("hammer_mountroot: can't find devvp\n");
 312                         return (error);
 313                 }
 314                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 315                 bzero(&info, sizeof(info));
 316                 info.asof = 0;
 317                 info.hflags = 0;
 318                 info.nvolumes = 1;
 319         } else {
 320                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 321                         return (error);
 322         }
 323
 324         /*
 325          * updating or new mount
 326          */
 327         if (mp->mnt_flag & MNT_UPDATE) {
 328                 hmp = (void *)mp->mnt_data;
 329                 KKASSERT(hmp != NULL);
 330         } else {
 331                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 332                         return (EINVAL);
 333                 hmp = NULL;
 334         }
 335
 336         /*
 337          * master-id validation.  The master id may not be changed by a
 338          * mount update.
 339          */
 340         if (info.hflags & HMNT_MASTERID) {
 341                 if (hmp && hmp->master_id != info.master_id) {
 342                         kprintf("hammer: cannot change master id "
 343                                 "with mount update\n");
 344                         return(EINVAL);
 345                 }
 346                 master_id = info.master_id;
 347                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 348                         return (EINVAL);
 349         } else {
 350                 if (hmp)
 351                         master_id = hmp->master_id;
 352                 else
 353                         master_id = 0;
 354         }
 355
 356         /*
 357          * Interal mount data structure
 358          */
 359         if (hmp == NULL) {
 360                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 361                 mp->mnt_data = (qaddr_t)hmp;
 362                 hmp->mp = mp;
 363                 /*TAILQ_INIT(&hmp->recycle_list);*/
 364
 365                 /*
 366                  * Make sure kmalloc type limits are set appropriately.  If root
 367                  * increases the vnode limit you may have to do a dummy remount
 368                  * to adjust the HAMMER inode limit.
 369                  */
 370                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
 371                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
 372
 373                 maxinodes = desiredvnodes + desiredvnodes / 5 +
 374                             HAMMER_RECLAIM_WAIT;
 375                 kmalloc_raise_limit(hmp->m_inodes,
 376                                     maxinodes * sizeof(struct hammer_inode));
 377
 378                 hmp->root_btree_beg.localization = 0x00000000U;
 379                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 380                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 381                 hmp->root_btree_beg.create_tid = 1;
 382                 hmp->root_btree_beg.delete_tid = 1;
 383                 hmp->root_btree_beg.rec_type = 0;
 384                 hmp->root_btree_beg.obj_type = 0;
 385
 386                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 387                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 388                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 389                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 390                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 391                 hmp->root_btree_end.rec_type = 0xFFFFU;
 392                 hmp->root_btree_end.obj_type = 0;
 393
 394                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 395                 hmp->krate.count = -16; /* initial burst */
 396
 397                 hmp->sync_lock.refs = 1;
 398                 hmp->free_lock.refs = 1;
 399                 hmp->undo_lock.refs = 1;
 400                 hmp->blkmap_lock.refs = 1;
 401
 402                 TAILQ_INIT(&hmp->delay_list);
 403                 TAILQ_INIT(&hmp->flush_group_list);
 404                 TAILQ_INIT(&hmp->objid_cache_list);
 405                 TAILQ_INIT(&hmp->undo_lru_list);
 406                 TAILQ_INIT(&hmp->reclaim_list);
 407         }
 408         hmp->hflags &= ~HMNT_USERFLAGS;
 409         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 410
 411         hmp->master_id = master_id;
 412
 413         if (info.asof) {
 414                 mp->mnt_flag |= MNT_RDONLY;
 415                 hmp->asof = info.asof;
 416         } else {
 417                 hmp->asof = HAMMER_MAX_TID;
 418         }
 419
 420         /*
 421          * Re-open read-write if originally read-only, or vise-versa.
 422          */
 423         if (mp->mnt_flag & MNT_UPDATE) {
 424                 error = 0;
 425                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 426                         kprintf("HAMMER read-only -> read-write\n");
 427                         hmp->ronly = 0;
 428                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 429                                 hammer_adjust_volume_mode, NULL);
 430                         rootvol = hammer_get_root_volume(hmp, &error);
 431                         if (rootvol) {
 432                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 433                                 bcopy(rootvol->ondisk->vol0_blockmap,
 434                                       hmp->blockmap,
 435                                       sizeof(hmp->blockmap));
 436                                 hammer_rel_volume(rootvol, 0);
 437                         }
 438                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 439                                 hammer_reload_inode, NULL);
 440                         /* kernel clears MNT_RDONLY */
 441                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 442                         kprintf("HAMMER read-write -> read-only\n");
 443                         hmp->ronly = 1; /* messy */
 444                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 445                                 hammer_reload_inode, NULL);
 446                         hmp->ronly = 0;
 447                         hammer_flusher_sync(hmp);
 448                         hammer_flusher_sync(hmp);
 449                         hammer_flusher_sync(hmp);
 450                         hmp->ronly = 1;
 451                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 452                                 hammer_adjust_volume_mode, NULL);
 453                 }
 454                 return(error);
 455         }
 456
 457         RB_INIT(&hmp->rb_vols_root);
 458         RB_INIT(&hmp->rb_inos_root);
 459         RB_INIT(&hmp->rb_nods_root);
 460         RB_INIT(&hmp->rb_undo_root);
 461         RB_INIT(&hmp->rb_resv_root);
 462         RB_INIT(&hmp->rb_bufs_root);
 463         RB_INIT(&hmp->rb_pfsm_root);
 464
 465         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 466
 467         TAILQ_INIT(&hmp->volu_list);
 468         TAILQ_INIT(&hmp->undo_list);
 469         TAILQ_INIT(&hmp->data_list);
 470         TAILQ_INIT(&hmp->meta_list);
 471         TAILQ_INIT(&hmp->lose_list);
 472
 473         /*
 474          * Load volumes
 475          */
 476         path = objcache_get(namei_oc, M_WAITOK);
 477         hmp->nvolumes = -1;
 478         for (i = 0; i < info.nvolumes; ++i) {
 479                 if (mntpt == NULL) {
 480                         /*
 481                          * Root mount.
 482                          * Only one volume; and no need for copyin.
 483                          */
 484                         KKASSERT(info.nvolumes == 1);
 485                         ksnprintf(path, MAXPATHLEN, "/dev/%s",
 486                                   mp->mnt_stat.f_mntfromname);
 487                         error = 0;
 488                 } else {
 489                         error = copyin(&info.volumes[i], &upath,
 490                                        sizeof(char *));
 491                         if (error == 0)
 492                                 error = copyinstr(upath, path,
 493                                                   MAXPATHLEN, NULL);
 494                 }
 495                 if (error == 0)
 496                         error = hammer_install_volume(hmp, path, devvp);
 497                 if (error)
 498                         break;
 499         }
 500         objcache_put(namei_oc, path);
 501
 502         /*
 503          * Make sure we found a root volume
 504          */
 505         if (error == 0 && hmp->rootvol == NULL) {
 506                 kprintf("hammer_mount: No root volume found!\n");
 507                 error = EINVAL;
 508         }
 509
 510         /*
 511          * Check that all required volumes are available
 512          */
 513         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 514                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 515                 error = EINVAL;
 516         }
 517
 518         if (error) {
 519                 hammer_free_hmp(mp);
 520                 return (error);
 521         }
 522
 523         /*
 524          * No errors, setup enough of the mount point so we can lookup the
 525          * root vnode.
 526          */
 527         mp->mnt_iosize_max = MAXPHYS;
 528         mp->mnt_kern_flag |= MNTK_FSMID;
 529
 530         /*
 531          * note: f_iosize is used by vnode_pager_haspage() when constructing
 532          * its VOP_BMAP call.
 533          */
 534         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 535         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 536
 537         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 538         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 539
 540         mp->mnt_maxsymlinklen = 255;
 541         mp->mnt_flag |= MNT_LOCAL;
 542
 543         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 544         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 545         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 546
 547         /*
 548          * The root volume's ondisk pointer is only valid if we hold a
 549          * reference to it.
 550          */
 551         rootvol = hammer_get_root_volume(hmp, &error);
 552         if (error)
 553                 goto failed;
 554
 555         /*
 556          * Perform any necessary UNDO operations.  The recovery code does
 557          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 558          * and then re-copy it again after recovery is complete.
 559          *
 560          * If this is a read-only mount the UNDO information is retained
 561          * in memory in the form of dirty buffer cache buffers, and not
 562          * written back to the media.
 563          */
 564         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 565               sizeof(hmp->blockmap));
 566
 567         /*
 568          * Check filesystem version
 569          */
 570         hmp->version = rootvol->ondisk->vol_version;
 571         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
 572             hmp->version > HAMMER_VOL_VERSION_MAX) {
 573                 kprintf("HAMMER: mount unsupported fs version %d\n",
 574                         hmp->version);
 575                 error = ERANGE;
 576                 goto done;
 577         }
 578
 579         /*
 580          * The undo_rec_limit limits the size of flush groups to avoid
 581          * blowing out the UNDO FIFO.  This calculation is typically in
 582          * the tens of thousands and is designed primarily when small
 583          * HAMMER filesystems are created.
 584          */
 585         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 586         if (hammer_debug_general & 0x0001)
 587                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 588
 589         error = hammer_recover(hmp, rootvol);
 590         if (error) {
 591                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 592                 goto done;
 593         }
 594
 595         /*
 596          * Finish setup now that we have a good root volume.
 597          *
 598          * The top 16 bits of fsid.val[1] is a pfs id.
 599          */
 600         ksnprintf(mp->mnt_stat.f_mntfromname,
 601                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 602                   rootvol->ondisk->vol_name);
 603         mp->mnt_stat.f_fsid.val[0] =
 604                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 605         mp->mnt_stat.f_fsid.val[1] =
 606                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 607         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 608
 609         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 610         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 611                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 612
 613         /*
 614          * Certain often-modified fields in the root volume are cached in
 615          * the hammer_mount structure so we do not have to generate lots
 616          * of little UNDO structures for them.
 617          *
 618          * Recopy after recovery.  This also has the side effect of
 619          * setting our cached undo FIFO's first_offset, which serves to
 620          * placemark the FIFO start for the NEXT flush cycle while the
 621          * on-disk first_offset represents the LAST flush cycle.
 622          */
 623         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 624         hmp->flush_tid1 = hmp->next_tid;
 625         hmp->flush_tid2 = hmp->next_tid;
 626         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 627               sizeof(hmp->blockmap));
 628         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 629
 630         hammer_flusher_create(hmp);
 631
 632         /*
 633          * Locate the root directory using the root cluster's B-Tree as a
 634          * starting point.  The root directory uses an obj_id of 1.
 635          *
 636          * FUTURE: Leave the root directory cached referenced but unlocked
 637          * in hmp->rootvp (need to flush it on unmount).
 638          */
 639         error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
 640         if (error)
 641                 goto done;
 642         vput(rootvp);
 643         /*vn_unlock(hmp->rootvp);*/
 644
 645 done:
 646         hammer_rel_volume(rootvol, 0);
 647 failed:
 648         /*
 649          * Cleanup and return.
 650          */
 651         if (error)
 652                 hammer_free_hmp(mp);
 653         return (error);
 654 }
 655
 656 static int
 657 hammer_vfs_unmount(struct mount *mp, int mntflags)
 658 {
 659 #if 0
 660         struct hammer_mount *hmp = (void *)mp->mnt_data;
 661 #endif
 662         int flags;
 663         int error;
 664
 665         /*
 666          * Clean out the vnodes
 667          */
 668         flags = 0;
 669         if (mntflags & MNT_FORCE)
 670                 flags |= FORCECLOSE;
 671         if ((error = vflush(mp, 0, flags)) != 0)
 672                 return (error);
 673
 674         /*
 675          * Clean up the internal mount structure and related entities.  This
 676          * may issue I/O.
 677          */
 678         hammer_free_hmp(mp);
 679         return(0);
 680 }
 681
 682 /*
 683  * Clean up the internal mount structure and disassociate it from the mount.
 684  * This may issue I/O.
 685  */
 686 static void
 687 hammer_free_hmp(struct mount *mp)
 688 {
 689         struct hammer_mount *hmp = (void *)mp->mnt_data;
 690         hammer_flush_group_t flg;
 691         int count;
 692         int dummy;
 693
 694         /*
 695          * Flush anything dirty.  This won't even run if the
 696          * filesystem errored-out.
 697          */
 698         count = 0;
 699         while (hammer_flusher_haswork(hmp)) {
 700                 hammer_flusher_sync(hmp);
 701                 ++count;
 702                 if (count >= 5) {
 703                         if (count == 5)
 704                                 kprintf("HAMMER: umount flushing.");
 705                         else
 706                                 kprintf(".");
 707                         tsleep(&dummy, 0, "hmrufl", hz);
 708                 }
 709                 if (count == 30) {
 710                         kprintf("giving up\n");
 711                         break;
 712                 }
 713         }
 714         if (count >= 5 && count < 30)
 715                 kprintf("\n");
 716
 717         /*
 718          * If the mount had a critical error we have to destroy any
 719          * remaining inodes before we can finish cleaning up the flusher.
 720          */
 721         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 722                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 723                         hammer_destroy_inode_callback, NULL);
 724         }
 725
 726         /*
 727          * There shouldn't be any inodes left now and any left over
 728          * flush groups should now be empty.
 729          */
 730         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 731         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 732                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 733                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
 734                 if (flg->refs) {
 735                         kprintf("HAMMER: Warning, flush_group %p was "
 736                                 "not empty on umount!\n", flg);
 737                 }
 738                 kfree(flg, hmp->m_misc);
 739         }
 740
 741         /*
 742          * We can finally destroy the flusher
 743          */
 744         hammer_flusher_destroy(hmp);
 745
 746         /*
 747          * We may have held recovered buffers due to a read-only mount.
 748          * These must be discarded.
 749          */
 750         if (hmp->ronly)
 751                 hammer_recover_flush_buffers(hmp, NULL, -1);
 752
 753         /*
 754          * Unload buffers and then volumes
 755          */
 756         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 757                 hammer_unload_buffer, NULL);
 758         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 759                 hammer_unload_volume, NULL);
 760
 761         mp->mnt_data = NULL;
 762         mp->mnt_flag &= ~MNT_LOCAL;
 763         hmp->mp = NULL;
 764         hammer_destroy_objid_cache(hmp);
 765         kmalloc_destroy(&hmp->m_misc);
 766         kmalloc_destroy(&hmp->m_inodes);
 767         kfree(hmp, M_HAMMER);
 768 }
 769
 770 /*
 771  * Report critical errors.  ip may be NULL.
 772  */
 773 void
 774 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 775                       int error, const char *msg)
 776 {
 777         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 778         krateprintf(&hmp->krate,
 779                 "HAMMER(%s): Critical error inode=%lld %s\n",
 780                 hmp->mp->mnt_stat.f_mntfromname,
 781                 (long long)(ip ? ip->obj_id : -1), msg);
 782         if (hmp->ronly == 0) {
 783                 hmp->ronly = 2;         /* special errored read-only mode */
 784                 hmp->mp->mnt_flag |= MNT_RDONLY;
 785                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 786                         hmp->mp->mnt_stat.f_mntfromname);
 787         }
 788         hmp->error = error;
 789 }
 790
 791
 792 /*
 793  * Obtain a vnode for the specified inode number.  An exclusively locked
 794  * vnode is returned.
 795  */
 796 int
 797 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 798                 ino_t ino, struct vnode **vpp)
 799 {
 800         struct hammer_transaction trans;
 801         struct hammer_mount *hmp = (void *)mp->mnt_data;
 802         struct hammer_inode *ip;
 803         int error;
 804         u_int32_t localization;
 805
 806         hammer_simple_transaction(&trans, hmp);
 807
 808         /*
 809          * If a directory vnode is supplied (mainly NFS) then we can acquire
 810          * the PFS domain from it.  Otherwise we would only be able to vget
 811          * inodes in the root PFS.
 812          */
 813         if (dvp) {
 814                 localization = HAMMER_DEF_LOCALIZATION +
 815                                 VTOI(dvp)->obj_localization;
 816         } else {
 817                 localization = HAMMER_DEF_LOCALIZATION;
 818         }
 819
 820         /*
 821          * Lookup the requested HAMMER inode.  The structure must be
 822          * left unlocked while we manipulate the related vnode to avoid
 823          * a deadlock.
 824          */
 825         ip = hammer_get_inode(&trans, NULL, ino,
 826                               hmp->asof, localization,
 827                               0, &error);
 828         if (ip == NULL) {
 829                 *vpp = NULL;
 830                 hammer_done_transaction(&trans);
 831                 return(error);
 832         }
 833         error = hammer_get_vnode(ip, vpp);
 834         hammer_rel_inode(ip, 0);
 835         hammer_done_transaction(&trans);
 836         return (error);
 837 }
 838
 839 /*
 840  * Return the root vnode for the filesystem.
 841  *
 842  * HAMMER stores the root vnode in the hammer_mount structure so
 843  * getting it is easy.
 844  */
 845 static int
 846 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 847 {
 848 #if 0
 849         struct hammer_mount *hmp = (void *)mp->mnt_data;
 850 #endif
 851         int error;
 852
 853         error = hammer_vfs_vget(mp, NULL, 1, vpp);
 854         return (error);
 855 }
 856
 857 static int
 858 hammer_vfs_statfs(struct mount *mp, struct statfs *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         int64_t breserved;
 866
 867         volume = hammer_get_root_volume(hmp, &error);
 868         if (error)
 869                 return(error);
 870         ondisk = volume->ondisk;
 871
 872         /*
 873          * Basic stats
 874          */
 875         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
 876         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 877         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 878         hammer_rel_volume(volume, 0);
 879
 880         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
 881         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 882         if (mp->mnt_stat.f_files < 0)
 883                 mp->mnt_stat.f_files = 0;
 884
 885         *sbp = mp->mnt_stat;
 886         return(0);
 887 }
 888
 889 static int
 890 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 891 {
 892         struct hammer_mount *hmp = (void *)mp->mnt_data;
 893         hammer_volume_t volume;
 894         hammer_volume_ondisk_t ondisk;
 895         int error;
 896         int64_t bfree;
 897         int64_t breserved;
 898
 899         volume = hammer_get_root_volume(hmp, &error);
 900         if (error)
 901                 return(error);
 902         ondisk = volume->ondisk;
 903
 904         /*
 905          * Basic stats
 906          */
 907         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
 908         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 909         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 910         hammer_rel_volume(volume, 0);
 911
 912         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
 913         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 914         if (mp->mnt_vstat.f_files < 0)
 915                 mp->mnt_vstat.f_files = 0;
 916         *sbp = mp->mnt_vstat;
 917         return(0);
 918 }
 919
 920 /*
 921  * Sync the filesystem.  Currently we have to run it twice, the second
 922  * one will advance the undo start index to the end index, so if a crash
 923  * occurs no undos will be run on mount.
 924  *
 925  * We do not sync the filesystem if we are called from a panic.  If we did
 926  * we might end up blowing up a sync that was already in progress.
 927  */
 928 static int
 929 hammer_vfs_sync(struct mount *mp, int waitfor)
 930 {
 931         struct hammer_mount *hmp = (void *)mp->mnt_data;
 932         int error;
 933
 934         if (panicstr == NULL) {
 935                 error = hammer_sync_hmp(hmp, waitfor);
 936         } else {
 937                 error = EIO;
 938         }
 939         return (error);
 940 }
 941
 942 /*
 943  * Convert a vnode to a file handle.
 944  */
 945 static int
 946 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 947 {
 948         hammer_inode_t ip;
 949
 950         KKASSERT(MAXFIDSZ >= 16);
 951         ip = VTOI(vp);
 952         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 953         fhp->fid_ext = ip->obj_localization >> 16;
 954         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 955         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 956         return(0);
 957 }
 958
 959
 960 /*
 961  * Convert a file handle back to a vnode.
 962  *
 963  * Use rootvp to enforce PFS isolation when a PFS is exported via a
 964  * null mount.
 965  */
 966 static int
 967 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 968                   struct fid *fhp, struct vnode **vpp)
 969 {
 970         struct hammer_transaction trans;
 971         struct hammer_inode *ip;
 972         struct hammer_inode_info info;
 973         int error;
 974         u_int32_t localization;
 975
 976         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 977         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 978         if (rootvp)
 979                 localization = VTOI(rootvp)->obj_localization;
 980         else
 981                 localization = (u_int32_t)fhp->fid_ext << 16;
 982
 983         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 984
 985         /*
 986          * Get/allocate the hammer_inode structure.  The structure must be
 987          * unlocked while we manipulate the related vnode to avoid a
 988          * deadlock.
 989          */
 990         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 991                               info.obj_asof, localization, 0, &error);
 992         if (ip == NULL) {
 993                 *vpp = NULL;
 994                 return(error);
 995         }
 996         error = hammer_get_vnode(ip, vpp);
 997         hammer_rel_inode(ip, 0);
 998         hammer_done_transaction(&trans);
 999         return (error);
1000 }
1001
1002 static int
1003 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1004                     int *exflagsp, struct ucred **credanonp)
1005 {
1006         hammer_mount_t hmp = (void *)mp->mnt_data;
1007         struct netcred *np;
1008         int error;
1009
1010         np = vfs_export_lookup(mp, &hmp->export, nam);
1011         if (np) {
1012                 *exflagsp = np->netc_exflags;
1013                 *credanonp = &np->netc_anon;
1014                 error = 0;
1015         } else {
1016                 error = EACCES;
1017         }
1018         return (error);
1019
1020 }
1021
1022 int
1023 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1024 {
1025         hammer_mount_t hmp = (void *)mp->mnt_data;
1026         int error;
1027
1028         switch(op) {
1029         case MOUNTCTL_SET_EXPORT:
1030                 error = vfs_export(mp, &hmp->export, export);
1031                 break;
1032         default:
1033                 error = EOPNOTSUPP;
1034                 break;
1035         }
1036         return(error);
1037 }
1038