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 1.63.2.4 2008/07/26 05:37:20 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-mount", "hammer mount");
 252
 253 VFS_SET(hammer_vfsops, hammer, 0);
 254 MODULE_VERSION(hammer, 1);
 255
 256 static int
 257 hammer_vfs_init(struct vfsconf *conf)
 258 {
 259         int n;
 260
 261         if (hammer_limit_recs == 0) {
 262                 hammer_limit_recs = nbuf * 25;
 263                 n = kmalloc_limit(M_HAMMER) / 512;
 264                 if (hammer_limit_recs > n)
 265                         hammer_limit_recs = n;
 266         }
 267         if (hammer_limit_dirtybufspace == 0) {
 268                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 269                 if (hammer_limit_dirtybufspace < 100)
 270                         hammer_limit_dirtybufspace = 100;
 271         }
 272         if (hammer_limit_iqueued == 0)
 273                 hammer_limit_iqueued = desiredvnodes / 5;
 274         return(0);
 275 }
 276
 277 static int
 278 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 279                  struct ucred *cred)
 280 {
 281         struct hammer_mount_info info;
 282         hammer_mount_t hmp;
 283         hammer_volume_t rootvol;
 284         struct vnode *rootvp;
 285         const char *upath;      /* volume name in userspace */
 286         char *path;             /* volume name in system space */
 287         int error;
 288         int i;
 289         int master_id;
 290
 291         if ((error = copyin(data, &info, sizeof(info))) != 0)
 292                 return (error);
 293
 294         /*
 295          * updating or new mount
 296          */
 297         if (mp->mnt_flag & MNT_UPDATE) {
 298                 hmp = (void *)mp->mnt_data;
 299                 KKASSERT(hmp != NULL);
 300         } else {
 301                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 302                         return (EINVAL);
 303                 hmp = NULL;
 304         }
 305
 306         /*
 307          * master-id validation.  The master id may not be changed by a
 308          * mount update.
 309          */
 310         if (info.hflags & HMNT_MASTERID) {
 311                 if (hmp && hmp->master_id != info.master_id) {
 312                         kprintf("hammer: cannot change master id "
 313                                 "with mount update\n");
 314                         return(EINVAL);
 315                 }
 316                 master_id = info.master_id;
 317                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 318                         return (EINVAL);
 319         } else {
 320                 if (hmp)
 321                         master_id = hmp->master_id;
 322                 else
 323                         master_id = 0;
 324         }
 325
 326         /*
 327          * Interal mount data structure
 328          */
 329         if (hmp == NULL) {
 330                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 331                 mp->mnt_data = (qaddr_t)hmp;
 332                 hmp->mp = mp;
 333                 hmp->namekey_iterator = mycpu->gd_time_seconds;
 334                 /*TAILQ_INIT(&hmp->recycle_list);*/
 335
 336                 hmp->root_btree_beg.localization = 0x00000000U;
 337                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 338                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 339                 hmp->root_btree_beg.create_tid = 1;
 340                 hmp->root_btree_beg.delete_tid = 1;
 341                 hmp->root_btree_beg.rec_type = 0;
 342                 hmp->root_btree_beg.obj_type = 0;
 343
 344                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 345                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 346                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 347                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 348                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 349                 hmp->root_btree_end.rec_type = 0xFFFFU;
 350                 hmp->root_btree_end.obj_type = 0;
 351
 352                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 353                 hmp->krate.count = -16; /* initial burst */
 354
 355                 hmp->sync_lock.refs = 1;
 356                 hmp->free_lock.refs = 1;
 357                 hmp->undo_lock.refs = 1;
 358                 hmp->blkmap_lock.refs = 1;
 359
 360                 TAILQ_INIT(&hmp->delay_list);
 361                 TAILQ_INIT(&hmp->flush_group_list);
 362                 TAILQ_INIT(&hmp->objid_cache_list);
 363                 TAILQ_INIT(&hmp->undo_lru_list);
 364                 TAILQ_INIT(&hmp->reclaim_list);
 365         }
 366         hmp->hflags &= ~HMNT_USERFLAGS;
 367         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 368
 369         hmp->master_id = master_id;
 370
 371         if (info.asof) {
 372                 mp->mnt_flag |= MNT_RDONLY;
 373                 hmp->asof = info.asof;
 374         } else {
 375                 hmp->asof = HAMMER_MAX_TID;
 376         }
 377
 378         /*
 379          * Re-open read-write if originally read-only, or vise-versa.
 380          */
 381         if (mp->mnt_flag & MNT_UPDATE) {
 382                 error = 0;
 383                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 384                         kprintf("HAMMER read-only -> read-write\n");
 385                         hmp->ronly = 0;
 386                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 387                                 hammer_adjust_volume_mode, NULL);
 388                         rootvol = hammer_get_root_volume(hmp, &error);
 389                         if (rootvol) {
 390                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 391                                 bcopy(rootvol->ondisk->vol0_blockmap,
 392                                       hmp->blockmap,
 393                                       sizeof(hmp->blockmap));
 394                                 hammer_rel_volume(rootvol, 0);
 395                         }
 396                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 397                                 hammer_reload_inode, NULL);
 398                         /* kernel clears MNT_RDONLY */
 399                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 400                         kprintf("HAMMER read-write -> read-only\n");
 401                         hmp->ronly = 1; /* messy */
 402                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 403                                 hammer_reload_inode, NULL);
 404                         hmp->ronly = 0;
 405                         hammer_flusher_sync(hmp);
 406                         hammer_flusher_sync(hmp);
 407                         hammer_flusher_sync(hmp);
 408                         hmp->ronly = 1;
 409                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 410                                 hammer_adjust_volume_mode, NULL);
 411                 }
 412                 return(error);
 413         }
 414
 415         RB_INIT(&hmp->rb_vols_root);
 416         RB_INIT(&hmp->rb_inos_root);
 417         RB_INIT(&hmp->rb_nods_root);
 418         RB_INIT(&hmp->rb_undo_root);
 419         RB_INIT(&hmp->rb_resv_root);
 420         RB_INIT(&hmp->rb_bufs_root);
 421         RB_INIT(&hmp->rb_pfsm_root);
 422
 423         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 424
 425         TAILQ_INIT(&hmp->volu_list);
 426         TAILQ_INIT(&hmp->undo_list);
 427         TAILQ_INIT(&hmp->data_list);
 428         TAILQ_INIT(&hmp->meta_list);
 429         TAILQ_INIT(&hmp->lose_list);
 430
 431         /*
 432          * Load volumes
 433          */
 434         path = objcache_get(namei_oc, M_WAITOK);
 435         hmp->nvolumes = -1;
 436         for (i = 0; i < info.nvolumes; ++i) {
 437                 error = copyin(&info.volumes[i], &upath, sizeof(char *));
 438                 if (error == 0)
 439                         error = copyinstr(upath, path, MAXPATHLEN, NULL);
 440                 if (error == 0)
 441                         error = hammer_install_volume(hmp, path);
 442                 if (error)
 443                         break;
 444         }
 445         objcache_put(namei_oc, path);
 446
 447         /*
 448          * Make sure we found a root volume
 449          */
 450         if (error == 0 && hmp->rootvol == NULL) {
 451                 kprintf("hammer_mount: No root volume found!\n");
 452                 error = EINVAL;
 453         }
 454
 455         /*
 456          * Check that all required volumes are available
 457          */
 458         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 459                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 460                 error = EINVAL;
 461         }
 462
 463         if (error) {
 464                 hammer_free_hmp(mp);
 465                 return (error);
 466         }
 467
 468         /*
 469          * No errors, setup enough of the mount point so we can lookup the
 470          * root vnode.
 471          */
 472         mp->mnt_iosize_max = MAXPHYS;
 473         mp->mnt_kern_flag |= MNTK_FSMID;
 474
 475         /*
 476          * note: f_iosize is used by vnode_pager_haspage() when constructing
 477          * its VOP_BMAP call.
 478          */
 479         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 480         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 481
 482         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 483         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 484
 485         mp->mnt_maxsymlinklen = 255;
 486         mp->mnt_flag |= MNT_LOCAL;
 487
 488         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 489         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 490         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 491
 492         /*
 493          * The root volume's ondisk pointer is only valid if we hold a
 494          * reference to it.
 495          */
 496         rootvol = hammer_get_root_volume(hmp, &error);
 497         if (error)
 498                 goto failed;
 499
 500         /*
 501          * Perform any necessary UNDO operations.  The recovery code does
 502          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 503          * and then re-copy it again after recovery is complete.
 504          *
 505          * If this is a read-only mount the UNDO information is retained
 506          * in memory in the form of dirty buffer cache buffers, and not
 507          * written back to the media.
 508          */
 509         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 510               sizeof(hmp->blockmap));
 511
 512         /*
 513          * The undo_rec_limit limits the size of flush groups to avoid
 514          * blowing out the UNDO FIFO.  This calculation is typically in
 515          * the tens of thousands and is designed primarily when small
 516          * HAMMER filesystems are created.
 517          */
 518         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 519         if (hammer_debug_general & 0x0001)
 520                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 521
 522         error = hammer_recover(hmp, rootvol);
 523         if (error) {
 524                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 525                 goto done;
 526         }
 527
 528         /*
 529          * Finish setup now that we have a good root volume.
 530          *
 531          * The top 16 bits of fsid.val[1] is a pfs id.
 532          */
 533         ksnprintf(mp->mnt_stat.f_mntfromname,
 534                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 535                   rootvol->ondisk->vol_name);
 536         mp->mnt_stat.f_fsid.val[0] =
 537                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 538         mp->mnt_stat.f_fsid.val[1] =
 539                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 540         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 541
 542         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 543         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 544                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 545
 546         /*
 547          * Certain often-modified fields in the root volume are cached in
 548          * the hammer_mount structure so we do not have to generate lots
 549          * of little UNDO structures for them.
 550          *
 551          * Recopy after recovery.  This also has the side effect of
 552          * setting our cached undo FIFO's first_offset, which serves to
 553          * placemark the FIFO start for the NEXT flush cycle while the
 554          * on-disk first_offset represents the LAST flush cycle.
 555          */
 556         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 557         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 558               sizeof(hmp->blockmap));
 559         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 560
 561         hammer_flusher_create(hmp);
 562
 563         /*
 564          * Locate the root directory using the root cluster's B-Tree as a
 565          * starting point.  The root directory uses an obj_id of 1.
 566          *
 567          * FUTURE: Leave the root directory cached referenced but unlocked
 568          * in hmp->rootvp (need to flush it on unmount).
 569          */
 570         error = hammer_vfs_vget(mp, 1, &rootvp);
 571         if (error)
 572                 goto done;
 573         vput(rootvp);
 574         /*vn_unlock(hmp->rootvp);*/
 575
 576 done:
 577         hammer_rel_volume(rootvol, 0);
 578 failed:
 579         /*
 580          * Cleanup and return.
 581          */
 582         if (error)
 583                 hammer_free_hmp(mp);
 584         return (error);
 585 }
 586
 587 static int
 588 hammer_vfs_unmount(struct mount *mp, int mntflags)
 589 {
 590 #if 0
 591         struct hammer_mount *hmp = (void *)mp->mnt_data;
 592 #endif
 593         int flags;
 594         int error;
 595
 596         /*
 597          * Clean out the vnodes
 598          */
 599         flags = 0;
 600         if (mntflags & MNT_FORCE)
 601                 flags |= FORCECLOSE;
 602         if ((error = vflush(mp, 0, flags)) != 0)
 603                 return (error);
 604
 605         /*
 606          * Clean up the internal mount structure and related entities.  This
 607          * may issue I/O.
 608          */
 609         hammer_free_hmp(mp);
 610         return(0);
 611 }
 612
 613 /*
 614  * Clean up the internal mount structure and disassociate it from the mount.
 615  * This may issue I/O.
 616  */
 617 static void
 618 hammer_free_hmp(struct mount *mp)
 619 {
 620         struct hammer_mount *hmp = (void *)mp->mnt_data;
 621         hammer_flush_group_t flg;
 622         int count;
 623
 624         /*
 625          * Flush anything dirty.  This won't even run if the
 626          * filesystem errored-out.
 627          */
 628         count = 0;
 629         while (hammer_flusher_haswork(hmp)) {
 630                 hammer_flusher_sync(hmp);
 631                 ++count;
 632                 if (count >= 5) {
 633                         if (count == 5)
 634                                 kprintf("HAMMER: umount flushing.");
 635                         else
 636                                 kprintf(".");
 637                         tsleep(hmp, 0, "hmrufl", hz);
 638                 }
 639                 if (count == 30) {
 640                         kprintf("giving up\n");
 641                         break;
 642                 }
 643         }
 644         if (count >= 5 && count < 30)
 645                 kprintf("\n");
 646
 647         /*
 648          * If the mount had a critical error we have to destroy any
 649          * remaining inodes before we can finish cleaning up the flusher.
 650          */
 651         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 652                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 653                         hammer_destroy_inode_callback, NULL);
 654         }
 655
 656         /*
 657          * There shouldn't be any inodes left now and any left over
 658          * flush groups should now be empty.
 659          */
 660         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 661         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 662                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 663                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
 664                 if (flg->refs) {
 665                         kprintf("HAMMER: Warning, flush_group %p was "
 666                                 "not empty on umount!\n", flg);
 667                 }
 668                 kfree(flg, M_HAMMER);
 669         }
 670
 671         /*
 672          * We can finally destroy the flusher
 673          */
 674         hammer_flusher_destroy(hmp);
 675
 676         /*
 677          * We may have held recovered buffers due to a read-only mount.
 678          * These must be discarded.
 679          */
 680         if (hmp->ronly)
 681                 hammer_recover_flush_buffers(hmp, NULL, -1);
 682
 683         /*
 684          * Unload buffers and then volumes
 685          */
 686         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 687                 hammer_unload_buffer, NULL);
 688         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 689                 hammer_unload_volume, NULL);
 690
 691         mp->mnt_data = NULL;
 692         mp->mnt_flag &= ~MNT_LOCAL;
 693         hmp->mp = NULL;
 694         hammer_destroy_objid_cache(hmp);
 695         kfree(hmp, M_HAMMER);
 696 }
 697
 698 /*
 699  * Report critical errors.  ip may be NULL.
 700  */
 701 void
 702 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 703                       int error, const char *msg)
 704 {
 705         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 706         krateprintf(&hmp->krate,
 707                 "HAMMER(%s): Critical error inode=%lld %s\n",
 708                 hmp->mp->mnt_stat.f_mntfromname,
 709                 (ip ? ip->obj_id : -1), msg);
 710         if (hmp->ronly == 0) {
 711                 hmp->ronly = 2;         /* special errored read-only mode */
 712                 hmp->mp->mnt_flag |= MNT_RDONLY;
 713                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 714                         hmp->mp->mnt_stat.f_mntfromname);
 715         }
 716         hmp->error = error;
 717 }
 718
 719
 720 /*
 721  * Obtain a vnode for the specified inode number.  An exclusively locked
 722  * vnode is returned.
 723  */
 724 int
 725 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 726 {
 727         struct hammer_transaction trans;
 728         struct hammer_mount *hmp = (void *)mp->mnt_data;
 729         struct hammer_inode *ip;
 730         int error;
 731
 732         hammer_simple_transaction(&trans, hmp);
 733
 734         /*
 735          * Lookup the requested HAMMER inode.  The structure must be
 736          * left unlocked while we manipulate the related vnode to avoid
 737          * a deadlock.
 738          */
 739         ip = hammer_get_inode(&trans, NULL, ino,
 740                               hmp->asof, HAMMER_DEF_LOCALIZATION,
 741                               0, &error);
 742         if (ip == NULL) {
 743                 *vpp = NULL;
 744                 hammer_done_transaction(&trans);
 745                 return(error);
 746         }
 747         error = hammer_get_vnode(ip, vpp);
 748         hammer_rel_inode(ip, 0);
 749         hammer_done_transaction(&trans);
 750         return (error);
 751 }
 752
 753 /*
 754  * Return the root vnode for the filesystem.
 755  *
 756  * HAMMER stores the root vnode in the hammer_mount structure so
 757  * getting it is easy.
 758  */
 759 static int
 760 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 761 {
 762 #if 0
 763         struct hammer_mount *hmp = (void *)mp->mnt_data;
 764 #endif
 765         int error;
 766
 767         error = hammer_vfs_vget(mp, 1, vpp);
 768         return (error);
 769 }
 770
 771 static int
 772 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 773 {
 774         struct hammer_mount *hmp = (void *)mp->mnt_data;
 775         hammer_volume_t volume;
 776         hammer_volume_ondisk_t ondisk;
 777         int error;
 778         int64_t bfree;
 779
 780         volume = hammer_get_root_volume(hmp, &error);
 781         if (error)
 782                 return(error);
 783         ondisk = volume->ondisk;
 784
 785         /*
 786          * Basic stats
 787          */
 788         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 789         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 790         hammer_rel_volume(volume, 0);
 791
 792         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
 793         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 794         if (mp->mnt_stat.f_files < 0)
 795                 mp->mnt_stat.f_files = 0;
 796
 797         *sbp = mp->mnt_stat;
 798         return(0);
 799 }
 800
 801 static int
 802 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 803 {
 804         struct hammer_mount *hmp = (void *)mp->mnt_data;
 805         hammer_volume_t volume;
 806         hammer_volume_ondisk_t ondisk;
 807         int error;
 808         int64_t bfree;
 809
 810         volume = hammer_get_root_volume(hmp, &error);
 811         if (error)
 812                 return(error);
 813         ondisk = volume->ondisk;
 814
 815         /*
 816          * Basic stats
 817          */
 818         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 819         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 820         hammer_rel_volume(volume, 0);
 821
 822         mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
 823         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 824         if (mp->mnt_vstat.f_files < 0)
 825                 mp->mnt_vstat.f_files = 0;
 826         *sbp = mp->mnt_vstat;
 827         return(0);
 828 }
 829
 830 /*
 831  * Sync the filesystem.  Currently we have to run it twice, the second
 832  * one will advance the undo start index to the end index, so if a crash
 833  * occurs no undos will be run on mount.
 834  *
 835  * We do not sync the filesystem if we are called from a panic.  If we did
 836  * we might end up blowing up a sync that was already in progress.
 837  */
 838 static int
 839 hammer_vfs_sync(struct mount *mp, int waitfor)
 840 {
 841         struct hammer_mount *hmp = (void *)mp->mnt_data;
 842         int error;
 843
 844         if (panicstr == NULL) {
 845                 error = hammer_sync_hmp(hmp, waitfor);
 846                 if (error == 0)
 847                         error = hammer_sync_hmp(hmp, waitfor);
 848         } else {
 849                 error = EIO;
 850         }
 851         return (error);
 852 }
 853
 854 /*
 855  * Convert a vnode to a file handle.
 856  */
 857 static int
 858 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 859 {
 860         hammer_inode_t ip;
 861
 862         KKASSERT(MAXFIDSZ >= 16);
 863         ip = VTOI(vp);
 864         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 865         fhp->fid_ext = ip->obj_localization >> 16;
 866         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 867         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 868         return(0);
 869 }
 870
 871
 872 /*
 873  * Convert a file handle back to a vnode.
 874  */
 875 static int
 876 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
 877 {
 878         struct hammer_transaction trans;
 879         struct hammer_inode *ip;
 880         struct hammer_inode_info info;
 881         int error;
 882         u_int32_t localization;
 883
 884         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 885         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 886         localization = (u_int32_t)fhp->fid_ext << 16;
 887
 888         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 889
 890         /*
 891          * Get/allocate the hammer_inode structure.  The structure must be
 892          * unlocked while we manipulate the related vnode to avoid a
 893          * deadlock.
 894          */
 895         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 896                               info.obj_asof, localization, 0, &error);
 897         if (ip == NULL) {
 898                 *vpp = NULL;
 899                 return(error);
 900         }
 901         error = hammer_get_vnode(ip, vpp);
 902         hammer_rel_inode(ip, 0);
 903         hammer_done_transaction(&trans);
 904         return (error);
 905 }
 906
 907 static int
 908 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 909                     int *exflagsp, struct ucred **credanonp)
 910 {
 911         hammer_mount_t hmp = (void *)mp->mnt_data;
 912         struct netcred *np;
 913         int error;
 914
 915         np = vfs_export_lookup(mp, &hmp->export, nam);
 916         if (np) {
 917                 *exflagsp = np->netc_exflags;
 918                 *credanonp = &np->netc_anon;
 919                 error = 0;
 920         } else {
 921                 error = EACCES;
 922         }
 923         return (error);
 924
 925 }
 926
 927 int
 928 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
 929 {
 930         hammer_mount_t hmp = (void *)mp->mnt_data;
 931         int error;
 932
 933         switch(op) {
 934         case MOUNTCTL_SET_EXPORT:
 935                 error = vfs_export(mp, &hmp->export, export);
 936                 break;
 937         default:
 938                 error = EOPNOTSUPP;
 939                 break;
 940         }
 941         return(error);
 942 }
 943