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 1.70 2008/07/31 04:42:04 dillon Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/vnode.h>
  41 #include <sys/mount.h>
  42 #include <sys/malloc.h>
  43 #include <sys/nlookup.h>
  44 #include <sys/fcntl.h>
  45 #include <sys/sysctl.h>
  46 #include <sys/buf.h>
  47 #include <sys/buf2.h>
  48 #include "hammer.h"
  49
  50 int hammer_debug_io;
  51 int hammer_debug_general;
  52 int hammer_debug_debug = 1;             /* medium-error panics */
  53 int hammer_debug_inode;
  54 int hammer_debug_locks;
  55 int hammer_debug_btree;
  56 int hammer_debug_tid;
  57 int hammer_debug_recover;               /* -1 will disable, +1 will force */
  58 int hammer_debug_recover_faults;
  59 int hammer_cluster_enable = 1;          /* enable read clustering by default */
  60 int hammer_count_fsyncs;
  61 int hammer_count_inodes;
  62 int hammer_count_iqueued;
  63 int hammer_count_reclaiming;
  64 int hammer_count_records;
  65 int hammer_count_record_datas;
  66 int hammer_count_volumes;
  67 int hammer_count_buffers;
  68 int hammer_count_nodes;
  69 int64_t hammer_count_extra_space_used;
  70 int64_t hammer_stats_btree_lookups;
  71 int64_t hammer_stats_btree_searches;
  72 int64_t hammer_stats_btree_inserts;
  73 int64_t hammer_stats_btree_deletes;
  74 int64_t hammer_stats_btree_elements;
  75 int64_t hammer_stats_btree_splits;
  76 int64_t hammer_stats_btree_iterations;
  77 int64_t hammer_stats_record_iterations;
  78
  79 int64_t hammer_stats_file_read;
  80 int64_t hammer_stats_file_write;
  81 int64_t hammer_stats_file_iopsr;
  82 int64_t hammer_stats_file_iopsw;
  83 int64_t hammer_stats_disk_read;
  84 int64_t hammer_stats_disk_write;
  85 int64_t hammer_stats_inode_flushes;
  86 int64_t hammer_stats_commits;
  87
  88 int hammer_count_dirtybufspace;         /* global */
  89 int hammer_count_refedbufs;             /* global */
  90 int hammer_count_reservations;
  91 int hammer_count_io_running_read;
  92 int hammer_count_io_running_write;
  93 int hammer_count_io_locked;
  94 int hammer_limit_dirtybufspace;         /* per-mount */
  95 int hammer_limit_recs;                  /* as a whole XXX */
  96 int hammer_limit_iqueued;               /* per-mount */
  97 int hammer_bio_count;
  98 int hammer_verify_zone;
  99 int hammer_verify_data = 1;
 100 int hammer_write_mode;
 101 int64_t hammer_contention_count;
 102 int64_t hammer_zone_limit;
 103
 104 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
 105 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
 106            &hammer_debug_general, 0, "");
 107 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
 108            &hammer_debug_io, 0, "");
 109 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
 110            &hammer_debug_debug, 0, "");
 111 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 112            &hammer_debug_inode, 0, "");
 113 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 114            &hammer_debug_locks, 0, "");
 115 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 116            &hammer_debug_btree, 0, "");
 117 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 118            &hammer_debug_tid, 0, "");
 119 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 120            &hammer_debug_recover, 0, "");
 121 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 122            &hammer_debug_recover_faults, 0, "");
 123 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
 124            &hammer_cluster_enable, 0, "");
 125
 126 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 127            &hammer_limit_dirtybufspace, 0, "");
 128 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 129            &hammer_limit_recs, 0, "");
 130 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_iqueued, CTLFLAG_RW,
 131            &hammer_limit_iqueued, 0, "");
 132
 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 134            &hammer_count_fsyncs, 0, "");
 135 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 136            &hammer_count_inodes, 0, "");
 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 138            &hammer_count_iqueued, 0, "");
 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 140            &hammer_count_reclaiming, 0, "");
 141 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 142            &hammer_count_records, 0, "");
 143 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 144            &hammer_count_record_datas, 0, "");
 145 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 146            &hammer_count_volumes, 0, "");
 147 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 148            &hammer_count_buffers, 0, "");
 149 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 150            &hammer_count_nodes, 0, "");
 151 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
 152            &hammer_count_extra_space_used, 0, "");
 153
 154 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 155            &hammer_stats_btree_searches, 0, "");
 156 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 157            &hammer_stats_btree_lookups, 0, "");
 158 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 159            &hammer_stats_btree_inserts, 0, "");
 160 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 161            &hammer_stats_btree_deletes, 0, "");
 162 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 163            &hammer_stats_btree_elements, 0, "");
 164 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 165            &hammer_stats_btree_splits, 0, "");
 166 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 167            &hammer_stats_btree_iterations, 0, "");
 168 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 169            &hammer_stats_record_iterations, 0, "");
 170
 171 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
 172            &hammer_stats_file_read, 0, "");
 173 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
 174            &hammer_stats_file_write, 0, "");
 175 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
 176            &hammer_stats_file_iopsr, 0, "");
 177 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
 178            &hammer_stats_file_iopsw, 0, "");
 179 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
 180            &hammer_stats_disk_read, 0, "");
 181 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
 182            &hammer_stats_disk_write, 0, "");
 183 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
 184            &hammer_stats_inode_flushes, 0, "");
 185 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
 186            &hammer_stats_commits, 0, "");
 187
 188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 189            &hammer_count_dirtybufspace, 0, "");
 190 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 191            &hammer_count_refedbufs, 0, "");
 192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 193            &hammer_count_reservations, 0, "");
 194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 195            &hammer_count_io_running_read, 0, "");
 196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 197            &hammer_count_io_locked, 0, "");
 198 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 199            &hammer_count_io_running_write, 0, "");
 200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 201            &hammer_zone_limit, 0, "");
 202 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 203            &hammer_contention_count, 0, "");
 204 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 205            &hammer_verify_zone, 0, "");
 206 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
 207            &hammer_verify_data, 0, "");
 208 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 209            &hammer_write_mode, 0, "");
 210
 211 KTR_INFO_MASTER(hammer);
 212
 213 /*
 214  * VFS ABI
 215  */
 216 static void     hammer_free_hmp(struct mount *mp);
 217
 218 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 219                                 struct ucred *cred);
 220 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 221 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 222 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 223                                 struct ucred *cred);
 224 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 225                                 struct ucred *cred);
 226 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 227 static int      hammer_vfs_vget(struct mount *mp, ino_t ino,
 228                                 struct vnode **vpp);
 229 static int      hammer_vfs_init(struct vfsconf *conf);
 230 static int      hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp,
 231                                 struct vnode **vpp);
 232 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 233 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 234                                 int *exflagsp, struct ucred **credanonp);
 235
 236
 237 static struct vfsops hammer_vfsops = {
 238         .vfs_mount      = hammer_vfs_mount,
 239         .vfs_unmount    = hammer_vfs_unmount,
 240         .vfs_root       = hammer_vfs_root,
 241         .vfs_statfs     = hammer_vfs_statfs,
 242         .vfs_statvfs    = hammer_vfs_statvfs,
 243         .vfs_sync       = hammer_vfs_sync,
 244         .vfs_vget       = hammer_vfs_vget,
 245         .vfs_init       = hammer_vfs_init,
 246         .vfs_vptofh     = hammer_vfs_vptofh,
 247         .vfs_fhtovp     = hammer_vfs_fhtovp,
 248         .vfs_checkexp   = hammer_vfs_checkexp
 249 };
 250
 251 MALLOC_DEFINE(M_HAMMER, "hammer-general", "hammer general");
 252 MALLOC_DEFINE(M_HAMMER_INO, "hammer-inodes", "hammer inodes");
 253
 254 VFS_SET(hammer_vfsops, hammer, 0);
 255 MODULE_VERSION(hammer, 1);
 256
 257 static int
 258 hammer_vfs_init(struct vfsconf *conf)
 259 {
 260         int n;
 261
 262         if (hammer_limit_recs == 0) {
 263                 hammer_limit_recs = nbuf * 25;
 264                 n = kmalloc_limit(M_HAMMER) / 512;
 265                 if (hammer_limit_recs > n)
 266                         hammer_limit_recs = n;
 267         }
 268         if (hammer_limit_dirtybufspace == 0) {
 269                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 270                 if (hammer_limit_dirtybufspace < 100)
 271                         hammer_limit_dirtybufspace = 100;
 272         }
 273         if (hammer_limit_iqueued == 0)
 274                 hammer_limit_iqueued = desiredvnodes / 5;
 275         return(0);
 276 }
 277
 278 static int
 279 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 280                  struct ucred *cred)
 281 {
 282         struct hammer_mount_info info;
 283         hammer_mount_t hmp;
 284         hammer_volume_t rootvol;
 285         struct vnode *rootvp;
 286         struct vnode *devvp = NULL;
 287         const char *upath;      /* volume name in userspace */
 288         char *path;             /* volume name in system space */
 289         int error;
 290         int i;
 291         int master_id;
 292         if (mntpt == NULL) {
 293                 /*
 294                  * Root mount
 295                  */
 296                 if ((error = bdevvp(rootdev, &devvp))) {
 297                         kprintf("hammer_mountroot: can't find devvp\n");
 298                         return (error);
 299                 }
 300                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 301                 bzero(&info, sizeof(info));
 302                 info.asof = 0;
 303                 info.hflags = 0;
 304                 info.nvolumes = 1;
 305         } else {
 306                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 307                         return (error);
 308         }
 309
 310         /*
 311          * updating or new mount
 312          */
 313         if (mp->mnt_flag & MNT_UPDATE) {
 314                 hmp = (void *)mp->mnt_data;
 315                 KKASSERT(hmp != NULL);
 316         } else {
 317                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 318                         return (EINVAL);
 319                 hmp = NULL;
 320         }
 321
 322         /*
 323          * master-id validation.  The master id may not be changed by a
 324          * mount update.
 325          */
 326         if (info.hflags & HMNT_MASTERID) {
 327                 if (hmp && hmp->master_id != info.master_id) {
 328                         kprintf("hammer: cannot change master id "
 329                                 "with mount update\n");
 330                         return(EINVAL);
 331                 }
 332                 master_id = info.master_id;
 333                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 334                         return (EINVAL);
 335         } else {
 336                 if (hmp)
 337                         master_id = hmp->master_id;
 338                 else
 339                         master_id = 0;
 340         }
 341
 342         /*
 343          * Interal mount data structure
 344          */
 345         if (hmp == NULL) {
 346                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 347                 mp->mnt_data = (qaddr_t)hmp;
 348                 hmp->mp = mp;
 349                 hmp->namekey_iterator = mycpu->gd_time_seconds;
 350                 /*TAILQ_INIT(&hmp->recycle_list);*/
 351
 352                 hmp->root_btree_beg.localization = 0x00000000U;
 353                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 354                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 355                 hmp->root_btree_beg.create_tid = 1;
 356                 hmp->root_btree_beg.delete_tid = 1;
 357                 hmp->root_btree_beg.rec_type = 0;
 358                 hmp->root_btree_beg.obj_type = 0;
 359
 360                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 361                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 362                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 363                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 364                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 365                 hmp->root_btree_end.rec_type = 0xFFFFU;
 366                 hmp->root_btree_end.obj_type = 0;
 367
 368                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 369                 hmp->krate.count = -16; /* initial burst */
 370
 371                 hmp->sync_lock.refs = 1;
 372                 hmp->free_lock.refs = 1;
 373                 hmp->undo_lock.refs = 1;
 374                 hmp->blkmap_lock.refs = 1;
 375
 376                 TAILQ_INIT(&hmp->delay_list);
 377                 TAILQ_INIT(&hmp->flush_group_list);
 378                 TAILQ_INIT(&hmp->objid_cache_list);
 379                 TAILQ_INIT(&hmp->undo_lru_list);
 380                 TAILQ_INIT(&hmp->reclaim_list);
 381         }
 382         hmp->hflags &= ~HMNT_USERFLAGS;
 383         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 384
 385         hmp->master_id = master_id;
 386
 387         if (info.asof) {
 388                 mp->mnt_flag |= MNT_RDONLY;
 389                 hmp->asof = info.asof;
 390         } else {
 391                 hmp->asof = HAMMER_MAX_TID;
 392         }
 393
 394         /*
 395          * Re-open read-write if originally read-only, or vise-versa.
 396          */
 397         if (mp->mnt_flag & MNT_UPDATE) {
 398                 error = 0;
 399                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 400                         kprintf("HAMMER read-only -> read-write\n");
 401                         hmp->ronly = 0;
 402                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 403                                 hammer_adjust_volume_mode, NULL);
 404                         rootvol = hammer_get_root_volume(hmp, &error);
 405                         if (rootvol) {
 406                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 407                                 bcopy(rootvol->ondisk->vol0_blockmap,
 408                                       hmp->blockmap,
 409                                       sizeof(hmp->blockmap));
 410                                 hammer_rel_volume(rootvol, 0);
 411                         }
 412                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 413                                 hammer_reload_inode, NULL);
 414                         /* kernel clears MNT_RDONLY */
 415                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 416                         kprintf("HAMMER read-write -> read-only\n");
 417                         hmp->ronly = 1; /* messy */
 418                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 419                                 hammer_reload_inode, NULL);
 420                         hmp->ronly = 0;
 421                         hammer_flusher_sync(hmp);
 422                         hammer_flusher_sync(hmp);
 423                         hammer_flusher_sync(hmp);
 424                         hmp->ronly = 1;
 425                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 426                                 hammer_adjust_volume_mode, NULL);
 427                 }
 428                 return(error);
 429         }
 430
 431         RB_INIT(&hmp->rb_vols_root);
 432         RB_INIT(&hmp->rb_inos_root);
 433         RB_INIT(&hmp->rb_nods_root);
 434         RB_INIT(&hmp->rb_undo_root);
 435         RB_INIT(&hmp->rb_resv_root);
 436         RB_INIT(&hmp->rb_bufs_root);
 437         RB_INIT(&hmp->rb_pfsm_root);
 438
 439         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 440
 441         TAILQ_INIT(&hmp->volu_list);
 442         TAILQ_INIT(&hmp->undo_list);
 443         TAILQ_INIT(&hmp->data_list);
 444         TAILQ_INIT(&hmp->meta_list);
 445         TAILQ_INIT(&hmp->lose_list);
 446
 447         /*
 448          * Load volumes
 449          */
 450         path = objcache_get(namei_oc, M_WAITOK);
 451         hmp->nvolumes = -1;
 452         for (i = 0; i < info.nvolumes; ++i) {
 453                 if (mntpt == NULL) {
 454                         /*
 455                          * Root mount.
 456                          * Only one volume; and no need for copyin.
 457                          */
 458                         KKASSERT(info.nvolumes == 1);
 459                         ksnprintf(path, MAXPATHLEN, "/dev/%s",
 460                                   mp->mnt_stat.f_mntfromname);
 461                         error = 0;
 462                 } else {
 463                         error = copyin(&info.volumes[i], &upath,
 464                                        sizeof(char *));
 465                         if (error == 0)
 466                                 error = copyinstr(upath, path,
 467                                                   MAXPATHLEN, NULL);
 468                 }
 469                 if (error == 0)
 470                         error = hammer_install_volume(hmp, path, devvp);
 471                 if (error)
 472                         break;
 473         }
 474         objcache_put(namei_oc, path);
 475
 476         /*
 477          * Make sure we found a root volume
 478          */
 479         if (error == 0 && hmp->rootvol == NULL) {
 480                 kprintf("hammer_mount: No root volume found!\n");
 481                 error = EINVAL;
 482         }
 483
 484         /*
 485          * Check that all required volumes are available
 486          */
 487         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 488                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 489                 error = EINVAL;
 490         }
 491
 492         if (error) {
 493                 hammer_free_hmp(mp);
 494                 return (error);
 495         }
 496
 497         /*
 498          * No errors, setup enough of the mount point so we can lookup the
 499          * root vnode.
 500          */
 501         mp->mnt_iosize_max = MAXPHYS;
 502         mp->mnt_kern_flag |= MNTK_FSMID;
 503
 504         /*
 505          * note: f_iosize is used by vnode_pager_haspage() when constructing
 506          * its VOP_BMAP call.
 507          */
 508         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 509         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 510
 511         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 512         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 513
 514         mp->mnt_maxsymlinklen = 255;
 515         mp->mnt_flag |= MNT_LOCAL;
 516
 517         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 518         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 519         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 520
 521         /*
 522          * The root volume's ondisk pointer is only valid if we hold a
 523          * reference to it.
 524          */
 525         rootvol = hammer_get_root_volume(hmp, &error);
 526         if (error)
 527                 goto failed;
 528
 529         /*
 530          * Perform any necessary UNDO operations.  The recovery code does
 531          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 532          * and then re-copy it again after recovery is complete.
 533          *
 534          * If this is a read-only mount the UNDO information is retained
 535          * in memory in the form of dirty buffer cache buffers, and not
 536          * written back to the media.
 537          */
 538         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 539               sizeof(hmp->blockmap));
 540
 541         /*
 542          * The undo_rec_limit limits the size of flush groups to avoid
 543          * blowing out the UNDO FIFO.  This calculation is typically in
 544          * the tens of thousands and is designed primarily when small
 545          * HAMMER filesystems are created.
 546          */
 547         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 548         if (hammer_debug_general & 0x0001)
 549                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 550
 551         error = hammer_recover(hmp, rootvol);
 552         if (error) {
 553                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 554                 goto done;
 555         }
 556
 557         /*
 558          * Finish setup now that we have a good root volume.
 559          *
 560          * The top 16 bits of fsid.val[1] is a pfs id.
 561          */
 562         ksnprintf(mp->mnt_stat.f_mntfromname,
 563                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 564                   rootvol->ondisk->vol_name);
 565         mp->mnt_stat.f_fsid.val[0] =
 566                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 567         mp->mnt_stat.f_fsid.val[1] =
 568                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 569         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 570
 571         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 572         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 573                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 574
 575         /*
 576          * Certain often-modified fields in the root volume are cached in
 577          * the hammer_mount structure so we do not have to generate lots
 578          * of little UNDO structures for them.
 579          *
 580          * Recopy after recovery.  This also has the side effect of
 581          * setting our cached undo FIFO's first_offset, which serves to
 582          * placemark the FIFO start for the NEXT flush cycle while the
 583          * on-disk first_offset represents the LAST flush cycle.
 584          */
 585         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 586         hmp->flush_tid1 = hmp->next_tid;
 587         hmp->flush_tid2 = hmp->next_tid;
 588         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 589               sizeof(hmp->blockmap));
 590         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 591
 592         hammer_flusher_create(hmp);
 593
 594         /*
 595          * Locate the root directory using the root cluster's B-Tree as a
 596          * starting point.  The root directory uses an obj_id of 1.
 597          *
 598          * FUTURE: Leave the root directory cached referenced but unlocked
 599          * in hmp->rootvp (need to flush it on unmount).
 600          */
 601         error = hammer_vfs_vget(mp, 1, &rootvp);
 602         if (error)
 603                 goto done;
 604         vput(rootvp);
 605         /*vn_unlock(hmp->rootvp);*/
 606
 607 done:
 608         hammer_rel_volume(rootvol, 0);
 609 failed:
 610         /*
 611          * Cleanup and return.
 612          */
 613         if (error)
 614                 hammer_free_hmp(mp);
 615         return (error);
 616 }
 617
 618 static int
 619 hammer_vfs_unmount(struct mount *mp, int mntflags)
 620 {
 621 #if 0
 622         struct hammer_mount *hmp = (void *)mp->mnt_data;
 623 #endif
 624         int flags;
 625         int error;
 626
 627         /*
 628          * Clean out the vnodes
 629          */
 630         flags = 0;
 631         if (mntflags & MNT_FORCE)
 632                 flags |= FORCECLOSE;
 633         if ((error = vflush(mp, 0, flags)) != 0)
 634                 return (error);
 635
 636         /*
 637          * Clean up the internal mount structure and related entities.  This
 638          * may issue I/O.
 639          */
 640         hammer_free_hmp(mp);
 641         return(0);
 642 }
 643
 644 /*
 645  * Clean up the internal mount structure and disassociate it from the mount.
 646  * This may issue I/O.
 647  */
 648 static void
 649 hammer_free_hmp(struct mount *mp)
 650 {
 651         struct hammer_mount *hmp = (void *)mp->mnt_data;
 652         hammer_flush_group_t flg;
 653         int count;
 654
 655         /*
 656          * Flush anything dirty.  This won't even run if the
 657          * filesystem errored-out.
 658          */
 659         count = 0;
 660         while (hammer_flusher_haswork(hmp)) {
 661                 hammer_flusher_sync(hmp);
 662                 ++count;
 663                 if (count >= 5) {
 664                         if (count == 5)
 665                                 kprintf("HAMMER: umount flushing.");
 666                         else
 667                                 kprintf(".");
 668                         tsleep(hmp, 0, "hmrufl", hz);
 669                 }
 670                 if (count == 30) {
 671                         kprintf("giving up\n");
 672                         break;
 673                 }
 674         }
 675         if (count >= 5 && count < 30)
 676                 kprintf("\n");
 677
 678         /*
 679          * If the mount had a critical error we have to destroy any
 680          * remaining inodes before we can finish cleaning up the flusher.
 681          */
 682         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 683                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 684                         hammer_destroy_inode_callback, NULL);
 685         }
 686
 687         /*
 688          * There shouldn't be any inodes left now and any left over
 689          * flush groups should now be empty.
 690          */
 691         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 692         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 693                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 694                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
 695                 if (flg->refs) {
 696                         kprintf("HAMMER: Warning, flush_group %p was "
 697                                 "not empty on umount!\n", flg);
 698                 }
 699                 kfree(flg, M_HAMMER);
 700         }
 701
 702         /*
 703          * We can finally destroy the flusher
 704          */
 705         hammer_flusher_destroy(hmp);
 706
 707         /*
 708          * We may have held recovered buffers due to a read-only mount.
 709          * These must be discarded.
 710          */
 711         if (hmp->ronly)
 712                 hammer_recover_flush_buffers(hmp, NULL, -1);
 713
 714         /*
 715          * Unload buffers and then volumes
 716          */
 717         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 718                 hammer_unload_buffer, NULL);
 719         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 720                 hammer_unload_volume, NULL);
 721
 722         mp->mnt_data = NULL;
 723         mp->mnt_flag &= ~MNT_LOCAL;
 724         hmp->mp = NULL;
 725         hammer_destroy_objid_cache(hmp);
 726         kfree(hmp, M_HAMMER);
 727 }
 728
 729 /*
 730  * Report critical errors.  ip may be NULL.
 731  */
 732 void
 733 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 734                       int error, const char *msg)
 735 {
 736         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 737         krateprintf(&hmp->krate,
 738                 "HAMMER(%s): Critical error inode=%lld %s\n",
 739                 hmp->mp->mnt_stat.f_mntfromname,
 740                 (ip ? ip->obj_id : -1), msg);
 741         if (hmp->ronly == 0) {
 742                 hmp->ronly = 2;         /* special errored read-only mode */
 743                 hmp->mp->mnt_flag |= MNT_RDONLY;
 744                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 745                         hmp->mp->mnt_stat.f_mntfromname);
 746         }
 747         hmp->error = error;
 748 }
 749
 750
 751 /*
 752  * Obtain a vnode for the specified inode number.  An exclusively locked
 753  * vnode is returned.
 754  */
 755 int
 756 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 757 {
 758         struct hammer_transaction trans;
 759         struct hammer_mount *hmp = (void *)mp->mnt_data;
 760         struct hammer_inode *ip;
 761         int error;
 762
 763         hammer_simple_transaction(&trans, hmp);
 764
 765         /*
 766          * Lookup the requested HAMMER inode.  The structure must be
 767          * left unlocked while we manipulate the related vnode to avoid
 768          * a deadlock.
 769          */
 770         ip = hammer_get_inode(&trans, NULL, ino,
 771                               hmp->asof, HAMMER_DEF_LOCALIZATION,
 772                               0, &error);
 773         if (ip == NULL) {
 774                 *vpp = NULL;
 775                 hammer_done_transaction(&trans);
 776                 return(error);
 777         }
 778         error = hammer_get_vnode(ip, vpp);
 779         hammer_rel_inode(ip, 0);
 780         hammer_done_transaction(&trans);
 781         return (error);
 782 }
 783
 784 /*
 785  * Return the root vnode for the filesystem.
 786  *
 787  * HAMMER stores the root vnode in the hammer_mount structure so
 788  * getting it is easy.
 789  */
 790 static int
 791 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 792 {
 793 #if 0
 794         struct hammer_mount *hmp = (void *)mp->mnt_data;
 795 #endif
 796         int error;
 797
 798         error = hammer_vfs_vget(mp, 1, vpp);
 799         return (error);
 800 }
 801
 802 static int
 803 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 804 {
 805         struct hammer_mount *hmp = (void *)mp->mnt_data;
 806         hammer_volume_t volume;
 807         hammer_volume_ondisk_t ondisk;
 808         int error;
 809         int64_t bfree;
 810
 811         volume = hammer_get_root_volume(hmp, &error);
 812         if (error)
 813                 return(error);
 814         ondisk = volume->ondisk;
 815
 816         /*
 817          * Basic stats
 818          */
 819         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 820         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 821         hammer_rel_volume(volume, 0);
 822
 823         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
 824         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 825         if (mp->mnt_stat.f_files < 0)
 826                 mp->mnt_stat.f_files = 0;
 827
 828         *sbp = mp->mnt_stat;
 829         return(0);
 830 }
 831
 832 static int
 833 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 834 {
 835         struct hammer_mount *hmp = (void *)mp->mnt_data;
 836         hammer_volume_t volume;
 837         hammer_volume_ondisk_t ondisk;
 838         int error;
 839         int64_t bfree;
 840
 841         volume = hammer_get_root_volume(hmp, &error);
 842         if (error)
 843                 return(error);
 844         ondisk = volume->ondisk;
 845
 846         /*
 847          * Basic stats
 848          */
 849         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 850         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 851         hammer_rel_volume(volume, 0);
 852
 853         mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
 854         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 855         if (mp->mnt_vstat.f_files < 0)
 856                 mp->mnt_vstat.f_files = 0;
 857         *sbp = mp->mnt_vstat;
 858         return(0);
 859 }
 860
 861 /*
 862  * Sync the filesystem.  Currently we have to run it twice, the second
 863  * one will advance the undo start index to the end index, so if a crash
 864  * occurs no undos will be run on mount.
 865  *
 866  * We do not sync the filesystem if we are called from a panic.  If we did
 867  * we might end up blowing up a sync that was already in progress.
 868  */
 869 static int
 870 hammer_vfs_sync(struct mount *mp, int waitfor)
 871 {
 872         struct hammer_mount *hmp = (void *)mp->mnt_data;
 873         int error;
 874
 875         if (panicstr == NULL) {
 876                 error = hammer_sync_hmp(hmp, waitfor);
 877         } else {
 878                 error = EIO;
 879         }
 880         return (error);
 881 }
 882
 883 /*
 884  * Convert a vnode to a file handle.
 885  */
 886 static int
 887 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 888 {
 889         hammer_inode_t ip;
 890
 891         KKASSERT(MAXFIDSZ >= 16);
 892         ip = VTOI(vp);
 893         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 894         fhp->fid_ext = ip->obj_localization >> 16;
 895         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 896         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 897         return(0);
 898 }
 899
 900
 901 /*
 902  * Convert a file handle back to a vnode.
 903  */
 904 static int
 905 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
 906 {
 907         struct hammer_transaction trans;
 908         struct hammer_inode *ip;
 909         struct hammer_inode_info info;
 910         int error;
 911         u_int32_t localization;
 912
 913         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 914         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 915         localization = (u_int32_t)fhp->fid_ext << 16;
 916
 917         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 918
 919         /*
 920          * Get/allocate the hammer_inode structure.  The structure must be
 921          * unlocked while we manipulate the related vnode to avoid a
 922          * deadlock.
 923          */
 924         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 925                               info.obj_asof, localization, 0, &error);
 926         if (ip == NULL) {
 927                 *vpp = NULL;
 928                 return(error);
 929         }
 930         error = hammer_get_vnode(ip, vpp);
 931         hammer_rel_inode(ip, 0);
 932         hammer_done_transaction(&trans);
 933         return (error);
 934 }
 935
 936 static int
 937 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 938                     int *exflagsp, struct ucred **credanonp)
 939 {
 940         hammer_mount_t hmp = (void *)mp->mnt_data;
 941         struct netcred *np;
 942         int error;
 943
 944         np = vfs_export_lookup(mp, &hmp->export, nam);
 945         if (np) {
 946                 *exflagsp = np->netc_exflags;
 947                 *credanonp = &np->netc_anon;
 948                 error = 0;
 949         } else {
 950                 error = EACCES;
 951         }
 952         return (error);
 953
 954 }
 955
 956 int
 957 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
 958 {
 959         hammer_mount_t hmp = (void *)mp->mnt_data;
 960         int error;
 961
 962         switch(op) {
 963         case MOUNTCTL_SET_EXPORT:
 964                 error = vfs_export(mp, &hmp->export, export);
 965                 break;
 966         default:
 967                 error = EOPNOTSUPP;
 968                 break;
 969         }
 970         return(error);
 971 }
 972