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