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