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 int hammer_yield_check = 16;
 104 int64_t hammer_contention_count;
 105 int64_t hammer_zone_limit;
 106
 107 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
 108 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
 109            &hammer_supported_version, 0, "");
 110 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
 111            &hammer_debug_general, 0, "");
 112 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
 113            &hammer_debug_io, 0, "");
 114 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
 115            &hammer_debug_debug, 0, "");
 116 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 117            &hammer_debug_inode, 0, "");
 118 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 119            &hammer_debug_locks, 0, "");
 120 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 121            &hammer_debug_btree, 0, "");
 122 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 123            &hammer_debug_tid, 0, "");
 124 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 125            &hammer_debug_recover, 0, "");
 126 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 127            &hammer_debug_recover_faults, 0, "");
 128 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
 129            &hammer_cluster_enable, 0, "");
 130
 131 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 132            &hammer_limit_dirtybufspace, 0, "");
 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 134            &hammer_limit_recs, 0, "");
 135
 136 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 137            &hammer_count_fsyncs, 0, "");
 138 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 139            &hammer_count_inodes, 0, "");
 140 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 141            &hammer_count_iqueued, 0, "");
 142 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 143            &hammer_count_reclaiming, 0, "");
 144 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 145            &hammer_count_records, 0, "");
 146 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 147            &hammer_count_record_datas, 0, "");
 148 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 149            &hammer_count_volumes, 0, "");
 150 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 151            &hammer_count_buffers, 0, "");
 152 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 153            &hammer_count_nodes, 0, "");
 154 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
 155            &hammer_count_extra_space_used, 0, "");
 156
 157 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 158            &hammer_stats_btree_searches, 0, "");
 159 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 160            &hammer_stats_btree_lookups, 0, "");
 161 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 162            &hammer_stats_btree_inserts, 0, "");
 163 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 164            &hammer_stats_btree_deletes, 0, "");
 165 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 166            &hammer_stats_btree_elements, 0, "");
 167 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 168            &hammer_stats_btree_splits, 0, "");
 169 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 170            &hammer_stats_btree_iterations, 0, "");
 171 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
 172            &hammer_stats_btree_root_iterations, 0, "");
 173 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 174            &hammer_stats_record_iterations, 0, "");
 175
 176 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
 177            &hammer_stats_file_read, 0, "");
 178 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
 179            &hammer_stats_file_write, 0, "");
 180 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
 181            &hammer_stats_file_iopsr, 0, "");
 182 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
 183            &hammer_stats_file_iopsw, 0, "");
 184 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
 185            &hammer_stats_disk_read, 0, "");
 186 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
 187            &hammer_stats_disk_write, 0, "");
 188 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
 189            &hammer_stats_inode_flushes, 0, "");
 190 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
 191            &hammer_stats_commits, 0, "");
 192
 193 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 194            &hammer_count_dirtybufspace, 0, "");
 195 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 196            &hammer_count_refedbufs, 0, "");
 197 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 198            &hammer_count_reservations, 0, "");
 199 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 200            &hammer_count_io_running_read, 0, "");
 201 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 202            &hammer_count_io_locked, 0, "");
 203 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 204            &hammer_count_io_running_write, 0, "");
 205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 206            &hammer_zone_limit, 0, "");
 207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 208            &hammer_contention_count, 0, "");
 209 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
 210            &hammer_autoflush, 0, "");
 211 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 212            &hammer_verify_zone, 0, "");
 213 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
 214            &hammer_verify_data, 0, "");
 215 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 216            &hammer_write_mode, 0, "");
 217 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
 218            &hammer_yield_check, 0, "");
 219
 220 KTR_INFO_MASTER(hammer);
 221
 222 /*
 223  * VFS ABI
 224  */
 225 static void     hammer_free_hmp(struct mount *mp);
 226
 227 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 228                                 struct ucred *cred);
 229 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 230 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 231 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 232                                 struct ucred *cred);
 233 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 234                                 struct ucred *cred);
 235 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 236 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 237                                 ino_t ino, struct vnode **vpp);
 238 static int      hammer_vfs_init(struct vfsconf *conf);
 239 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 240                                 struct fid *fhp, struct vnode **vpp);
 241 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 242 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 243                                 int *exflagsp, struct ucred **credanonp);
 244
 245
 246 static struct vfsops hammer_vfsops = {
 247         .vfs_mount      = hammer_vfs_mount,
 248         .vfs_unmount    = hammer_vfs_unmount,
 249         .vfs_root       = hammer_vfs_root,
 250         .vfs_statfs     = hammer_vfs_statfs,
 251         .vfs_statvfs    = hammer_vfs_statvfs,
 252         .vfs_sync       = hammer_vfs_sync,
 253         .vfs_vget       = hammer_vfs_vget,
 254         .vfs_init       = hammer_vfs_init,
 255         .vfs_vptofh     = hammer_vfs_vptofh,
 256         .vfs_fhtovp     = hammer_vfs_fhtovp,
 257         .vfs_checkexp   = hammer_vfs_checkexp
 258 };
 259
 260 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
 261
 262 VFS_SET(hammer_vfsops, hammer, 0);
 263 MODULE_VERSION(hammer, 1);
 264
 265 static int
 266 hammer_vfs_init(struct vfsconf *conf)
 267 {
 268         int n;
 269
 270         if (hammer_limit_recs == 0) {
 271                 hammer_limit_recs = nbuf * 25;
 272                 n = kmalloc_limit(M_HAMMER) / 512;
 273                 if (hammer_limit_recs > n)
 274                         hammer_limit_recs = n;
 275         }
 276         if (hammer_limit_dirtybufspace == 0) {
 277                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 278                 if (hammer_limit_dirtybufspace < 100)
 279                         hammer_limit_dirtybufspace = 100;
 280         }
 281         return(0);
 282 }
 283
 284 static int
 285 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 286                  struct ucred *cred)
 287 {
 288         struct hammer_mount_info info;
 289         hammer_mount_t hmp;
 290         hammer_volume_t rootvol;
 291         struct vnode *rootvp;
 292         struct vnode *devvp = NULL;
 293         const char *upath;      /* volume name in userspace */
 294         char *path;             /* volume name in system space */
 295         int error;
 296         int i;
 297         int master_id;
 298         int maxinodes;
 299
 300         /*
 301          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
 302          */
 303         if (mntpt == NULL) {
 304                 if ((error = bdevvp(rootdev, &devvp))) {
 305                         kprintf("hammer_mountroot: can't find devvp\n");
 306                         return (error);
 307                 }
 308                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 309                 bzero(&info, sizeof(info));
 310                 info.asof = 0;
 311                 info.hflags = 0;
 312                 info.nvolumes = 1;
 313         } else {
 314                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 315                         return (error);
 316         }
 317
 318         /*
 319          * updating or new mount
 320          */
 321         if (mp->mnt_flag & MNT_UPDATE) {
 322                 hmp = (void *)mp->mnt_data;
 323                 KKASSERT(hmp != NULL);
 324         } else {
 325                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 326                         return (EINVAL);
 327                 hmp = NULL;
 328         }
 329
 330         /*
 331          * master-id validation.  The master id may not be changed by a
 332          * mount update.
 333          */
 334         if (info.hflags & HMNT_MASTERID) {
 335                 if (hmp && hmp->master_id != info.master_id) {
 336                         kprintf("hammer: cannot change master id "
 337                                 "with mount update\n");
 338                         return(EINVAL);
 339                 }
 340                 master_id = info.master_id;
 341                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 342                         return (EINVAL);
 343         } else {
 344                 if (hmp)
 345                         master_id = hmp->master_id;
 346                 else
 347                         master_id = 0;
 348         }
 349
 350         /*
 351          * Interal mount data structure
 352          */
 353         if (hmp == NULL) {
 354                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 355                 mp->mnt_data = (qaddr_t)hmp;
 356                 hmp->mp = mp;
 357                 /*TAILQ_INIT(&hmp->recycle_list);*/
 358
 359                 /*
 360                  * Make sure kmalloc type limits are set appropriately.  If root
 361                  * increases the vnode limit you may have to do a dummy remount
 362                  * to adjust the HAMMER inode limit.
 363                  */
 364                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
 365                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
 366
 367                 maxinodes = desiredvnodes + desiredvnodes / 5 +
 368                             HAMMER_RECLAIM_WAIT;
 369                 kmalloc_raise_limit(hmp->m_inodes,
 370                                     maxinodes * sizeof(struct hammer_inode));
 371
 372                 hmp->root_btree_beg.localization = 0x00000000U;
 373                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 374                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 375                 hmp->root_btree_beg.create_tid = 1;
 376                 hmp->root_btree_beg.delete_tid = 1;
 377                 hmp->root_btree_beg.rec_type = 0;
 378                 hmp->root_btree_beg.obj_type = 0;
 379
 380                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 381                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 382                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 383                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 384                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 385                 hmp->root_btree_end.rec_type = 0xFFFFU;
 386                 hmp->root_btree_end.obj_type = 0;
 387
 388                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 389                 hmp->krate.count = -16; /* initial burst */
 390
 391                 hmp->sync_lock.refs = 1;
 392                 hmp->free_lock.refs = 1;
 393                 hmp->undo_lock.refs = 1;
 394                 hmp->blkmap_lock.refs = 1;
 395
 396                 TAILQ_INIT(&hmp->delay_list);
 397                 TAILQ_INIT(&hmp->flush_group_list);
 398                 TAILQ_INIT(&hmp->objid_cache_list);
 399                 TAILQ_INIT(&hmp->undo_lru_list);
 400                 TAILQ_INIT(&hmp->reclaim_list);
 401         }
 402         hmp->hflags &= ~HMNT_USERFLAGS;
 403         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 404
 405         hmp->master_id = master_id;
 406
 407         if (info.asof) {
 408                 mp->mnt_flag |= MNT_RDONLY;
 409                 hmp->asof = info.asof;
 410         } else {
 411                 hmp->asof = HAMMER_MAX_TID;
 412         }
 413
 414         /*
 415          * Re-open read-write if originally read-only, or vise-versa.
 416          */
 417         if (mp->mnt_flag & MNT_UPDATE) {
 418                 error = 0;
 419                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 420                         kprintf("HAMMER read-only -> read-write\n");
 421                         hmp->ronly = 0;
 422                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 423                                 hammer_adjust_volume_mode, NULL);
 424                         rootvol = hammer_get_root_volume(hmp, &error);
 425                         if (rootvol) {
 426                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 427                                 bcopy(rootvol->ondisk->vol0_blockmap,
 428                                       hmp->blockmap,
 429                                       sizeof(hmp->blockmap));
 430                                 hammer_rel_volume(rootvol, 0);
 431                         }
 432                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 433                                 hammer_reload_inode, NULL);
 434                         /* kernel clears MNT_RDONLY */
 435                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 436                         kprintf("HAMMER read-write -> read-only\n");
 437                         hmp->ronly = 1; /* messy */
 438                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 439                                 hammer_reload_inode, NULL);
 440                         hmp->ronly = 0;
 441                         hammer_flusher_sync(hmp);
 442                         hammer_flusher_sync(hmp);
 443                         hammer_flusher_sync(hmp);
 444                         hmp->ronly = 1;
 445                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 446                                 hammer_adjust_volume_mode, NULL);
 447                 }
 448                 return(error);
 449         }
 450
 451         RB_INIT(&hmp->rb_vols_root);
 452         RB_INIT(&hmp->rb_inos_root);
 453         RB_INIT(&hmp->rb_nods_root);
 454         RB_INIT(&hmp->rb_undo_root);
 455         RB_INIT(&hmp->rb_resv_root);
 456         RB_INIT(&hmp->rb_bufs_root);
 457         RB_INIT(&hmp->rb_pfsm_root);
 458
 459         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 460
 461         TAILQ_INIT(&hmp->volu_list);
 462         TAILQ_INIT(&hmp->undo_list);
 463         TAILQ_INIT(&hmp->data_list);
 464         TAILQ_INIT(&hmp->meta_list);
 465         TAILQ_INIT(&hmp->lose_list);
 466
 467         /*
 468          * Load volumes
 469          */
 470         path = objcache_get(namei_oc, M_WAITOK);
 471         hmp->nvolumes = -1;
 472         for (i = 0; i < info.nvolumes; ++i) {
 473                 if (mntpt == NULL) {
 474                         /*
 475                          * Root mount.
 476                          * Only one volume; and no need for copyin.
 477                          */
 478                         KKASSERT(info.nvolumes == 1);
 479                         ksnprintf(path, MAXPATHLEN, "/dev/%s",
 480                                   mp->mnt_stat.f_mntfromname);
 481                         error = 0;
 482                 } else {
 483                         error = copyin(&info.volumes[i], &upath,
 484                                        sizeof(char *));
 485                         if (error == 0)
 486                                 error = copyinstr(upath, path,
 487                                                   MAXPATHLEN, NULL);
 488                 }
 489                 if (error == 0)
 490                         error = hammer_install_volume(hmp, path, devvp);
 491                 if (error)
 492                         break;
 493         }
 494         objcache_put(namei_oc, path);
 495
 496         /*
 497          * Make sure we found a root volume
 498          */
 499         if (error == 0 && hmp->rootvol == NULL) {
 500                 kprintf("hammer_mount: No root volume found!\n");
 501                 error = EINVAL;
 502         }
 503
 504         /*
 505          * Check that all required volumes are available
 506          */
 507         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 508                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 509                 error = EINVAL;
 510         }
 511
 512         if (error) {
 513                 hammer_free_hmp(mp);
 514                 return (error);
 515         }
 516
 517         /*
 518          * No errors, setup enough of the mount point so we can lookup the
 519          * root vnode.
 520          */
 521         mp->mnt_iosize_max = MAXPHYS;
 522         mp->mnt_kern_flag |= MNTK_FSMID;
 523
 524         /*
 525          * note: f_iosize is used by vnode_pager_haspage() when constructing
 526          * its VOP_BMAP call.
 527          */
 528         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 529         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 530
 531         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 532         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 533
 534         mp->mnt_maxsymlinklen = 255;
 535         mp->mnt_flag |= MNT_LOCAL;
 536
 537         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 538         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 539         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 540
 541         /*
 542          * The root volume's ondisk pointer is only valid if we hold a
 543          * reference to it.
 544          */
 545         rootvol = hammer_get_root_volume(hmp, &error);
 546         if (error)
 547                 goto failed;
 548
 549         /*
 550          * Perform any necessary UNDO operations.  The recovery code does
 551          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 552          * and then re-copy it again after recovery is complete.
 553          *
 554          * If this is a read-only mount the UNDO information is retained
 555          * in memory in the form of dirty buffer cache buffers, and not
 556          * written back to the media.
 557          */
 558         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 559               sizeof(hmp->blockmap));
 560
 561         /*
 562          * Check filesystem version
 563          */
 564         hmp->version = rootvol->ondisk->vol_version;
 565         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
 566             hmp->version > HAMMER_VOL_VERSION_MAX) {
 567                 kprintf("HAMMER: mount unsupported fs version %d\n",
 568                         hmp->version);
 569                 error = ERANGE;
 570                 goto done;
 571         }
 572
 573         /*
 574          * The undo_rec_limit limits the size of flush groups to avoid
 575          * blowing out the UNDO FIFO.  This calculation is typically in
 576          * the tens of thousands and is designed primarily when small
 577          * HAMMER filesystems are created.
 578          */
 579         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 580         if (hammer_debug_general & 0x0001)
 581                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 582
 583         error = hammer_recover(hmp, rootvol);
 584         if (error) {
 585                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 586                 goto done;
 587         }
 588
 589         /*
 590          * Finish setup now that we have a good root volume.
 591          *
 592          * The top 16 bits of fsid.val[1] is a pfs id.
 593          */
 594         ksnprintf(mp->mnt_stat.f_mntfromname,
 595                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 596                   rootvol->ondisk->vol_name);
 597         mp->mnt_stat.f_fsid.val[0] =
 598                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 599         mp->mnt_stat.f_fsid.val[1] =
 600                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 601         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 602
 603         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 604         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 605                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 606
 607         /*
 608          * Certain often-modified fields in the root volume are cached in
 609          * the hammer_mount structure so we do not have to generate lots
 610          * of little UNDO structures for them.
 611          *
 612          * Recopy after recovery.  This also has the side effect of
 613          * setting our cached undo FIFO's first_offset, which serves to
 614          * placemark the FIFO start for the NEXT flush cycle while the
 615          * on-disk first_offset represents the LAST flush cycle.
 616          */
 617         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 618         hmp->flush_tid1 = hmp->next_tid;
 619         hmp->flush_tid2 = hmp->next_tid;
 620         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 621               sizeof(hmp->blockmap));
 622         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 623
 624         hammer_flusher_create(hmp);
 625
 626         /*
 627          * Locate the root directory using the root cluster's B-Tree as a
 628          * starting point.  The root directory uses an obj_id of 1.
 629          *
 630          * FUTURE: Leave the root directory cached referenced but unlocked
 631          * in hmp->rootvp (need to flush it on unmount).
 632          */
 633         error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
 634         if (error)
 635                 goto done;
 636         vput(rootvp);
 637         /*vn_unlock(hmp->rootvp);*/
 638
 639 done:
 640         hammer_rel_volume(rootvol, 0);
 641 failed:
 642         /*
 643          * Cleanup and return.
 644          */
 645         if (error)
 646                 hammer_free_hmp(mp);
 647         return (error);
 648 }
 649
 650 static int
 651 hammer_vfs_unmount(struct mount *mp, int mntflags)
 652 {
 653 #if 0
 654         struct hammer_mount *hmp = (void *)mp->mnt_data;
 655 #endif
 656         int flags;
 657         int error;
 658
 659         /*
 660          * Clean out the vnodes
 661          */
 662         flags = 0;
 663         if (mntflags & MNT_FORCE)
 664                 flags |= FORCECLOSE;
 665         if ((error = vflush(mp, 0, flags)) != 0)
 666                 return (error);
 667
 668         /*
 669          * Clean up the internal mount structure and related entities.  This
 670          * may issue I/O.
 671          */
 672         hammer_free_hmp(mp);
 673         return(0);
 674 }
 675
 676 /*
 677  * Clean up the internal mount structure and disassociate it from the mount.
 678  * This may issue I/O.
 679  */
 680 static void
 681 hammer_free_hmp(struct mount *mp)
 682 {
 683         struct hammer_mount *hmp = (void *)mp->mnt_data;
 684         hammer_flush_group_t flg;
 685         int count;
 686         int dummy;
 687
 688         /*
 689          * Flush anything dirty.  This won't even run if the
 690          * filesystem errored-out.
 691          */
 692         count = 0;
 693         while (hammer_flusher_haswork(hmp)) {
 694                 hammer_flusher_sync(hmp);
 695                 ++count;
 696                 if (count >= 5) {
 697                         if (count == 5)
 698                                 kprintf("HAMMER: umount flushing.");
 699                         else
 700                                 kprintf(".");
 701                         tsleep(&dummy, 0, "hmrufl", hz);
 702                 }
 703                 if (count == 30) {
 704                         kprintf("giving up\n");
 705                         break;
 706                 }
 707         }
 708         if (count >= 5 && count < 30)
 709                 kprintf("\n");
 710
 711         /*
 712          * If the mount had a critical error we have to destroy any
 713          * remaining inodes before we can finish cleaning up the flusher.
 714          */
 715         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 716                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 717                         hammer_destroy_inode_callback, NULL);
 718         }
 719
 720         /*
 721          * There shouldn't be any inodes left now and any left over
 722          * flush groups should now be empty.
 723          */
 724         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 725         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 726                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 727                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
 728                 if (flg->refs) {
 729                         kprintf("HAMMER: Warning, flush_group %p was "
 730                                 "not empty on umount!\n", flg);
 731                 }
 732                 kfree(flg, hmp->m_misc);
 733         }
 734
 735         /*
 736          * We can finally destroy the flusher
 737          */
 738         hammer_flusher_destroy(hmp);
 739
 740         /*
 741          * We may have held recovered buffers due to a read-only mount.
 742          * These must be discarded.
 743          */
 744         if (hmp->ronly)
 745                 hammer_recover_flush_buffers(hmp, NULL, -1);
 746
 747         /*
 748          * Unload buffers and then volumes
 749          */
 750         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 751                 hammer_unload_buffer, NULL);
 752         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 753                 hammer_unload_volume, NULL);
 754
 755         mp->mnt_data = NULL;
 756         mp->mnt_flag &= ~MNT_LOCAL;
 757         hmp->mp = NULL;
 758         hammer_destroy_objid_cache(hmp);
 759         kmalloc_destroy(&hmp->m_misc);
 760         kmalloc_destroy(&hmp->m_inodes);
 761         kfree(hmp, M_HAMMER);
 762 }
 763
 764 /*
 765  * Report critical errors.  ip may be NULL.
 766  */
 767 void
 768 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 769                       int error, const char *msg)
 770 {
 771         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 772         krateprintf(&hmp->krate,
 773                 "HAMMER(%s): Critical error inode=%lld %s\n",
 774                 hmp->mp->mnt_stat.f_mntfromname,
 775                 (long long)(ip ? ip->obj_id : -1), msg);
 776         if (hmp->ronly == 0) {
 777                 hmp->ronly = 2;         /* special errored read-only mode */
 778                 hmp->mp->mnt_flag |= MNT_RDONLY;
 779                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 780                         hmp->mp->mnt_stat.f_mntfromname);
 781         }
 782         hmp->error = error;
 783 }
 784
 785
 786 /*
 787  * Obtain a vnode for the specified inode number.  An exclusively locked
 788  * vnode is returned.
 789  */
 790 int
 791 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 792                 ino_t ino, struct vnode **vpp)
 793 {
 794         struct hammer_transaction trans;
 795         struct hammer_mount *hmp = (void *)mp->mnt_data;
 796         struct hammer_inode *ip;
 797         int error;
 798         u_int32_t localization;
 799
 800         hammer_simple_transaction(&trans, hmp);
 801
 802         /*
 803          * If a directory vnode is supplied (mainly NFS) then we can acquire
 804          * the PFS domain from it.  Otherwise we would only be able to vget
 805          * inodes in the root PFS.
 806          */
 807         if (dvp) {
 808                 localization = HAMMER_DEF_LOCALIZATION +
 809                                 VTOI(dvp)->obj_localization;
 810         } else {
 811                 localization = HAMMER_DEF_LOCALIZATION;
 812         }
 813
 814         /*
 815          * Lookup the requested HAMMER inode.  The structure must be
 816          * left unlocked while we manipulate the related vnode to avoid
 817          * a deadlock.
 818          */
 819         ip = hammer_get_inode(&trans, NULL, ino,
 820                               hmp->asof, localization,
 821                               0, &error);
 822         if (ip == NULL) {
 823                 *vpp = NULL;
 824                 hammer_done_transaction(&trans);
 825                 return(error);
 826         }
 827         error = hammer_get_vnode(ip, vpp);
 828         hammer_rel_inode(ip, 0);
 829         hammer_done_transaction(&trans);
 830         return (error);
 831 }
 832
 833 /*
 834  * Return the root vnode for the filesystem.
 835  *
 836  * HAMMER stores the root vnode in the hammer_mount structure so
 837  * getting it is easy.
 838  */
 839 static int
 840 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 841 {
 842 #if 0
 843         struct hammer_mount *hmp = (void *)mp->mnt_data;
 844 #endif
 845         int error;
 846
 847         error = hammer_vfs_vget(mp, NULL, 1, vpp);
 848         return (error);
 849 }
 850
 851 static int
 852 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 853 {
 854         struct hammer_mount *hmp = (void *)mp->mnt_data;
 855         hammer_volume_t volume;
 856         hammer_volume_ondisk_t ondisk;
 857         int error;
 858         int64_t bfree;
 859         int64_t breserved;
 860
 861         volume = hammer_get_root_volume(hmp, &error);
 862         if (error)
 863                 return(error);
 864         ondisk = volume->ondisk;
 865
 866         /*
 867          * Basic stats
 868          */
 869         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
 870         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 871         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 872         hammer_rel_volume(volume, 0);
 873
 874         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
 875         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 876         if (mp->mnt_stat.f_files < 0)
 877                 mp->mnt_stat.f_files = 0;
 878
 879         *sbp = mp->mnt_stat;
 880         return(0);
 881 }
 882
 883 static int
 884 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 885 {
 886         struct hammer_mount *hmp = (void *)mp->mnt_data;
 887         hammer_volume_t volume;
 888         hammer_volume_ondisk_t ondisk;
 889         int error;
 890         int64_t bfree;
 891         int64_t breserved;
 892
 893         volume = hammer_get_root_volume(hmp, &error);
 894         if (error)
 895                 return(error);
 896         ondisk = volume->ondisk;
 897
 898         /*
 899          * Basic stats
 900          */
 901         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
 902         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 903         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 904         hammer_rel_volume(volume, 0);
 905
 906         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
 907         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 908         if (mp->mnt_vstat.f_files < 0)
 909                 mp->mnt_vstat.f_files = 0;
 910         *sbp = mp->mnt_vstat;
 911         return(0);
 912 }
 913
 914 /*
 915  * Sync the filesystem.  Currently we have to run it twice, the second
 916  * one will advance the undo start index to the end index, so if a crash
 917  * occurs no undos will be run on mount.
 918  *
 919  * We do not sync the filesystem if we are called from a panic.  If we did
 920  * we might end up blowing up a sync that was already in progress.
 921  */
 922 static int
 923 hammer_vfs_sync(struct mount *mp, int waitfor)
 924 {
 925         struct hammer_mount *hmp = (void *)mp->mnt_data;
 926         int error;
 927
 928         if (panicstr == NULL) {
 929                 error = hammer_sync_hmp(hmp, waitfor);
 930         } else {
 931                 error = EIO;
 932         }
 933         return (error);
 934 }
 935
 936 /*
 937  * Convert a vnode to a file handle.
 938  */
 939 static int
 940 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 941 {
 942         hammer_inode_t ip;
 943
 944         KKASSERT(MAXFIDSZ >= 16);
 945         ip = VTOI(vp);
 946         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 947         fhp->fid_ext = ip->obj_localization >> 16;
 948         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 949         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 950         return(0);
 951 }
 952
 953
 954 /*
 955  * Convert a file handle back to a vnode.
 956  *
 957  * Use rootvp to enforce PFS isolation when a PFS is exported via a
 958  * null mount.
 959  */
 960 static int
 961 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 962                   struct fid *fhp, struct vnode **vpp)
 963 {
 964         struct hammer_transaction trans;
 965         struct hammer_inode *ip;
 966         struct hammer_inode_info info;
 967         int error;
 968         u_int32_t localization;
 969
 970         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 971         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 972         if (rootvp)
 973                 localization = VTOI(rootvp)->obj_localization;
 974         else
 975                 localization = (u_int32_t)fhp->fid_ext << 16;
 976
 977         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 978
 979         /*
 980          * Get/allocate the hammer_inode structure.  The structure must be
 981          * unlocked while we manipulate the related vnode to avoid a
 982          * deadlock.
 983          */
 984         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 985                               info.obj_asof, localization, 0, &error);
 986         if (ip == NULL) {
 987                 *vpp = NULL;
 988                 return(error);
 989         }
 990         error = hammer_get_vnode(ip, vpp);
 991         hammer_rel_inode(ip, 0);
 992         hammer_done_transaction(&trans);
 993         return (error);
 994 }
 995
 996 static int
 997 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 998                     int *exflagsp, struct ucred **credanonp)
 999 {
1000         hammer_mount_t hmp = (void *)mp->mnt_data;
1001         struct netcred *np;
1002         int error;
1003
1004         np = vfs_export_lookup(mp, &hmp->export, nam);
1005         if (np) {
1006                 *exflagsp = np->netc_exflags;
1007                 *credanonp = &np->netc_anon;
1008                 error = 0;
1009         } else {
1010                 error = EACCES;
1011         }
1012         return (error);
1013
1014 }
1015
1016 int
1017 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1018 {
1019         hammer_mount_t hmp = (void *)mp->mnt_data;
1020         int error;
1021
1022         switch(op) {
1023         case MOUNTCTL_SET_EXPORT:
1024                 error = vfs_export(mp, &hmp->export, export);
1025                 break;
1026         default:
1027                 error = EOPNOTSUPP;
1028                 break;
1029         }
1030         return(error);
1031 }
1032