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 1.63.2.7 2008/09/25 01:42:52 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         if (mntpt == NULL) {
 296                 /*
 297                  * Root mount
 298                  */
 299                 if ((error = bdevvp(rootdev, &devvp))) {
 300                         kprintf("hammer_mountroot: can't find devvp\n");
 301                         return (error);
 302                 }
 303                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 304                 bzero(&info, sizeof(info));
 305                 info.asof = 0;
 306                 info.hflags = 0;
 307                 info.nvolumes = 1;
 308         } else {
 309                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 310                         return (error);
 311         }
 312
 313         /*
 314          * updating or new mount
 315          */
 316         if (mp->mnt_flag & MNT_UPDATE) {
 317                 hmp = (void *)mp->mnt_data;
 318                 KKASSERT(hmp != NULL);
 319         } else {
 320                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 321                         return (EINVAL);
 322                 hmp = NULL;
 323         }
 324
 325         /*
 326          * master-id validation.  The master id may not be changed by a
 327          * mount update.
 328          */
 329         if (info.hflags & HMNT_MASTERID) {
 330                 if (hmp && hmp->master_id != info.master_id) {
 331                         kprintf("hammer: cannot change master id "
 332                                 "with mount update\n");
 333                         return(EINVAL);
 334                 }
 335                 master_id = info.master_id;
 336                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 337                         return (EINVAL);
 338         } else {
 339                 if (hmp)
 340                         master_id = hmp->master_id;
 341                 else
 342                         master_id = 0;
 343         }
 344
 345         /*
 346          * Interal mount data structure
 347          */
 348         if (hmp == NULL) {
 349                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 350                 mp->mnt_data = (qaddr_t)hmp;
 351                 hmp->mp = mp;
 352                 hmp->namekey_iterator = mycpu->gd_time_seconds;
 353                 /*TAILQ_INIT(&hmp->recycle_list);*/
 354
 355                 hmp->root_btree_beg.localization = 0x00000000U;
 356                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 357                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 358                 hmp->root_btree_beg.create_tid = 1;
 359                 hmp->root_btree_beg.delete_tid = 1;
 360                 hmp->root_btree_beg.rec_type = 0;
 361                 hmp->root_btree_beg.obj_type = 0;
 362
 363                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 364                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 365                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 366                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 367                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 368                 hmp->root_btree_end.rec_type = 0xFFFFU;
 369                 hmp->root_btree_end.obj_type = 0;
 370
 371                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 372                 hmp->krate.count = -16; /* initial burst */
 373
 374                 hmp->sync_lock.refs = 1;
 375                 hmp->free_lock.refs = 1;
 376                 hmp->undo_lock.refs = 1;
 377                 hmp->blkmap_lock.refs = 1;
 378
 379                 TAILQ_INIT(&hmp->delay_list);
 380                 TAILQ_INIT(&hmp->flush_group_list);
 381                 TAILQ_INIT(&hmp->objid_cache_list);
 382                 TAILQ_INIT(&hmp->undo_lru_list);
 383                 TAILQ_INIT(&hmp->reclaim_list);
 384         }
 385         hmp->hflags &= ~HMNT_USERFLAGS;
 386         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 387
 388         hmp->master_id = master_id;
 389
 390         if (info.asof) {
 391                 mp->mnt_flag |= MNT_RDONLY;
 392                 hmp->asof = info.asof;
 393         } else {
 394                 hmp->asof = HAMMER_MAX_TID;
 395         }
 396
 397         /*
 398          * Re-open read-write if originally read-only, or vise-versa.
 399          */
 400         if (mp->mnt_flag & MNT_UPDATE) {
 401                 error = 0;
 402                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 403                         kprintf("HAMMER read-only -> read-write\n");
 404                         hmp->ronly = 0;
 405                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 406                                 hammer_adjust_volume_mode, NULL);
 407                         rootvol = hammer_get_root_volume(hmp, &error);
 408                         if (rootvol) {
 409                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 410                                 bcopy(rootvol->ondisk->vol0_blockmap,
 411                                       hmp->blockmap,
 412                                       sizeof(hmp->blockmap));
 413                                 hammer_rel_volume(rootvol, 0);
 414                         }
 415                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 416                                 hammer_reload_inode, NULL);
 417                         /* kernel clears MNT_RDONLY */
 418                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 419                         kprintf("HAMMER read-write -> read-only\n");
 420                         hmp->ronly = 1; /* messy */
 421                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 422                                 hammer_reload_inode, NULL);
 423                         hmp->ronly = 0;
 424                         hammer_flusher_sync(hmp);
 425                         hammer_flusher_sync(hmp);
 426                         hammer_flusher_sync(hmp);
 427                         hmp->ronly = 1;
 428                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 429                                 hammer_adjust_volume_mode, NULL);
 430                 }
 431                 return(error);
 432         }
 433
 434         RB_INIT(&hmp->rb_vols_root);
 435         RB_INIT(&hmp->rb_inos_root);
 436         RB_INIT(&hmp->rb_nods_root);
 437         RB_INIT(&hmp->rb_undo_root);
 438         RB_INIT(&hmp->rb_resv_root);
 439         RB_INIT(&hmp->rb_bufs_root);
 440         RB_INIT(&hmp->rb_pfsm_root);
 441
 442         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 443
 444         TAILQ_INIT(&hmp->volu_list);
 445         TAILQ_INIT(&hmp->undo_list);
 446         TAILQ_INIT(&hmp->data_list);
 447         TAILQ_INIT(&hmp->meta_list);
 448         TAILQ_INIT(&hmp->lose_list);
 449
 450         /*
 451          * Load volumes
 452          */
 453         path = objcache_get(namei_oc, M_WAITOK);
 454         hmp->nvolumes = -1;
 455         for (i = 0; i < info.nvolumes; ++i) {
 456                 if (mntpt == NULL) {
 457                         /*
 458                          * Root mount.
 459                          * Only one volume; and no need for copyin.
 460                          */
 461                         KKASSERT(info.nvolumes == 1);
 462                         ksnprintf(path, MAXPATHLEN, "/dev/%s",
 463                                   mp->mnt_stat.f_mntfromname);
 464                         error = 0;
 465                 } else {
 466                         error = copyin(&info.volumes[i], &upath,
 467                                        sizeof(char *));
 468                         if (error == 0)
 469                                 error = copyinstr(upath, path,
 470                                                   MAXPATHLEN, NULL);
 471                 }
 472                 if (error == 0)
 473                         error = hammer_install_volume(hmp, path, devvp);
 474                 if (error)
 475                         break;
 476         }
 477         objcache_put(namei_oc, path);
 478
 479         /*
 480          * Make sure we found a root volume
 481          */
 482         if (error == 0 && hmp->rootvol == NULL) {
 483                 kprintf("hammer_mount: No root volume found!\n");
 484                 error = EINVAL;
 485         }
 486
 487         /*
 488          * Check that all required volumes are available
 489          */
 490         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 491                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 492                 error = EINVAL;
 493         }
 494
 495         if (error) {
 496                 hammer_free_hmp(mp);
 497                 return (error);
 498         }
 499
 500         /*
 501          * No errors, setup enough of the mount point so we can lookup the
 502          * root vnode.
 503          */
 504         mp->mnt_iosize_max = MAXPHYS;
 505         mp->mnt_kern_flag |= MNTK_FSMID;
 506
 507         /*
 508          * note: f_iosize is used by vnode_pager_haspage() when constructing
 509          * its VOP_BMAP call.
 510          */
 511         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 512         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 513
 514         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 515         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 516
 517         mp->mnt_maxsymlinklen = 255;
 518         mp->mnt_flag |= MNT_LOCAL;
 519
 520         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 521         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 522         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 523
 524         /*
 525          * The root volume's ondisk pointer is only valid if we hold a
 526          * reference to it.
 527          */
 528         rootvol = hammer_get_root_volume(hmp, &error);
 529         if (error)
 530                 goto failed;
 531
 532         /*
 533          * Perform any necessary UNDO operations.  The recovery code does
 534          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 535          * and then re-copy it again after recovery is complete.
 536          *
 537          * If this is a read-only mount the UNDO information is retained
 538          * in memory in the form of dirty buffer cache buffers, and not
 539          * written back to the media.
 540          */
 541         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 542               sizeof(hmp->blockmap));
 543
 544         /*
 545          * The undo_rec_limit limits the size of flush groups to avoid
 546          * blowing out the UNDO FIFO.  This calculation is typically in
 547          * the tens of thousands and is designed primarily when small
 548          * HAMMER filesystems are created.
 549          */
 550         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 551         if (hammer_debug_general & 0x0001)
 552                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 553
 554         error = hammer_recover(hmp, rootvol);
 555         if (error) {
 556                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 557                 goto done;
 558         }
 559
 560         /*
 561          * Finish setup now that we have a good root volume.
 562          *
 563          * The top 16 bits of fsid.val[1] is a pfs id.
 564          */
 565         ksnprintf(mp->mnt_stat.f_mntfromname,
 566                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 567                   rootvol->ondisk->vol_name);
 568         mp->mnt_stat.f_fsid.val[0] =
 569                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 570         mp->mnt_stat.f_fsid.val[1] =
 571                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 572         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 573
 574         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 575         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 576                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 577
 578         /*
 579          * Certain often-modified fields in the root volume are cached in
 580          * the hammer_mount structure so we do not have to generate lots
 581          * of little UNDO structures for them.
 582          *
 583          * Recopy after recovery.  This also has the side effect of
 584          * setting our cached undo FIFO's first_offset, which serves to
 585          * placemark the FIFO start for the NEXT flush cycle while the
 586          * on-disk first_offset represents the LAST flush cycle.
 587          */
 588         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 589         hmp->flush_tid1 = hmp->next_tid;
 590         hmp->flush_tid2 = hmp->next_tid;
 591         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 592               sizeof(hmp->blockmap));
 593         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 594
 595         hammer_flusher_create(hmp);
 596
 597         /*
 598          * Locate the root directory using the root cluster's B-Tree as a
 599          * starting point.  The root directory uses an obj_id of 1.
 600          *
 601          * FUTURE: Leave the root directory cached referenced but unlocked
 602          * in hmp->rootvp (need to flush it on unmount).
 603          */
 604         error = hammer_vfs_vget(mp, 1, &rootvp);
 605         if (error)
 606                 goto done;
 607         vput(rootvp);
 608         /*vn_unlock(hmp->rootvp);*/
 609
 610 done:
 611         hammer_rel_volume(rootvol, 0);
 612 failed:
 613         /*
 614          * Cleanup and return.
 615          */
 616         if (error)
 617                 hammer_free_hmp(mp);
 618         return (error);
 619 }
 620
 621 static int
 622 hammer_vfs_unmount(struct mount *mp, int mntflags)
 623 {
 624 #if 0
 625         struct hammer_mount *hmp = (void *)mp->mnt_data;
 626 #endif
 627         int flags;
 628         int error;
 629
 630         /*
 631          * Clean out the vnodes
 632          */
 633         flags = 0;
 634         if (mntflags & MNT_FORCE)
 635                 flags |= FORCECLOSE;
 636         if ((error = vflush(mp, 0, flags)) != 0)
 637                 return (error);
 638
 639         /*
 640          * Clean up the internal mount structure and related entities.  This
 641          * may issue I/O.
 642          */
 643         hammer_free_hmp(mp);
 644         return(0);
 645 }
 646
 647 /*
 648  * Clean up the internal mount structure and disassociate it from the mount.
 649  * This may issue I/O.
 650  */
 651 static void
 652 hammer_free_hmp(struct mount *mp)
 653 {
 654         struct hammer_mount *hmp = (void *)mp->mnt_data;
 655         hammer_flush_group_t flg;
 656         int count;
 657
 658         /*
 659          * Flush anything dirty.  This won't even run if the
 660          * filesystem errored-out.
 661          */
 662         count = 0;
 663         while (hammer_flusher_haswork(hmp)) {
 664                 hammer_flusher_sync(hmp);
 665                 ++count;
 666                 if (count >= 5) {
 667                         if (count == 5)
 668                                 kprintf("HAMMER: umount flushing.");
 669                         else
 670                                 kprintf(".");
 671                         tsleep(hmp, 0, "hmrufl", hz);
 672                 }
 673                 if (count == 30) {
 674                         kprintf("giving up\n");
 675                         break;
 676                 }
 677         }
 678         if (count >= 5 && count < 30)
 679                 kprintf("\n");
 680
 681         /*
 682          * If the mount had a critical error we have to destroy any
 683          * remaining inodes before we can finish cleaning up the flusher.
 684          */
 685         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 686                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 687                         hammer_destroy_inode_callback, NULL);
 688         }
 689
 690         /*
 691          * There shouldn't be any inodes left now and any left over
 692          * flush groups should now be empty.
 693          */
 694         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 695         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 696                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 697                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
 698                 if (flg->refs) {
 699                         kprintf("HAMMER: Warning, flush_group %p was "
 700                                 "not empty on umount!\n", flg);
 701                 }
 702                 kfree(flg, M_HAMMER);
 703         }
 704
 705         /*
 706          * We can finally destroy the flusher
 707          */
 708         hammer_flusher_destroy(hmp);
 709
 710         /*
 711          * We may have held recovered buffers due to a read-only mount.
 712          * These must be discarded.
 713          */
 714         if (hmp->ronly)
 715                 hammer_recover_flush_buffers(hmp, NULL, -1);
 716
 717         /*
 718          * Unload buffers and then volumes
 719          */
 720         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 721                 hammer_unload_buffer, NULL);
 722         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 723                 hammer_unload_volume, NULL);
 724
 725         mp->mnt_data = NULL;
 726         mp->mnt_flag &= ~MNT_LOCAL;
 727         hmp->mp = NULL;
 728         hammer_destroy_objid_cache(hmp);
 729         kfree(hmp, M_HAMMER);
 730 }
 731
 732 /*
 733  * Report critical errors.  ip may be NULL.
 734  */
 735 void
 736 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 737                       int error, const char *msg)
 738 {
 739         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 740         krateprintf(&hmp->krate,
 741                 "HAMMER(%s): Critical error inode=%lld %s\n",
 742                 hmp->mp->mnt_stat.f_mntfromname,
 743                 (ip ? ip->obj_id : -1), msg);
 744         if (hmp->ronly == 0) {
 745                 hmp->ronly = 2;         /* special errored read-only mode */
 746                 hmp->mp->mnt_flag |= MNT_RDONLY;
 747                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 748                         hmp->mp->mnt_stat.f_mntfromname);
 749         }
 750         hmp->error = error;
 751 }
 752
 753
 754 /*
 755  * Obtain a vnode for the specified inode number.  An exclusively locked
 756  * vnode is returned.
 757  */
 758 int
 759 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 760 {
 761         struct hammer_transaction trans;
 762         struct hammer_mount *hmp = (void *)mp->mnt_data;
 763         struct hammer_inode *ip;
 764         int error;
 765
 766         hammer_simple_transaction(&trans, hmp);
 767
 768         /*
 769          * Lookup the requested HAMMER inode.  The structure must be
 770          * left unlocked while we manipulate the related vnode to avoid
 771          * a deadlock.
 772          */
 773         ip = hammer_get_inode(&trans, NULL, ino,
 774                               hmp->asof, HAMMER_DEF_LOCALIZATION,
 775                               0, &error);
 776         if (ip == NULL) {
 777                 *vpp = NULL;
 778                 hammer_done_transaction(&trans);
 779                 return(error);
 780         }
 781         error = hammer_get_vnode(ip, vpp);
 782         hammer_rel_inode(ip, 0);
 783         hammer_done_transaction(&trans);
 784         return (error);
 785 }
 786
 787 /*
 788  * Return the root vnode for the filesystem.
 789  *
 790  * HAMMER stores the root vnode in the hammer_mount structure so
 791  * getting it is easy.
 792  */
 793 static int
 794 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 795 {
 796 #if 0
 797         struct hammer_mount *hmp = (void *)mp->mnt_data;
 798 #endif
 799         int error;
 800
 801         error = hammer_vfs_vget(mp, 1, vpp);
 802         return (error);
 803 }
 804
 805 static int
 806 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 807 {
 808         struct hammer_mount *hmp = (void *)mp->mnt_data;
 809         hammer_volume_t volume;
 810         hammer_volume_ondisk_t ondisk;
 811         int error;
 812         int64_t bfree;
 813
 814         volume = hammer_get_root_volume(hmp, &error);
 815         if (error)
 816                 return(error);
 817         ondisk = volume->ondisk;
 818
 819         /*
 820          * Basic stats
 821          */
 822         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 823         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 824         hammer_rel_volume(volume, 0);
 825
 826         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
 827         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 828         if (mp->mnt_stat.f_files < 0)
 829                 mp->mnt_stat.f_files = 0;
 830
 831         *sbp = mp->mnt_stat;
 832         return(0);
 833 }
 834
 835 static int
 836 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 837 {
 838         struct hammer_mount *hmp = (void *)mp->mnt_data;
 839         hammer_volume_t volume;
 840         hammer_volume_ondisk_t ondisk;
 841         int error;
 842         int64_t bfree;
 843
 844         volume = hammer_get_root_volume(hmp, &error);
 845         if (error)
 846                 return(error);
 847         ondisk = volume->ondisk;
 848
 849         /*
 850          * Basic stats
 851          */
 852         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 853         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 854         hammer_rel_volume(volume, 0);
 855
 856         mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
 857         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 858         if (mp->mnt_vstat.f_files < 0)
 859                 mp->mnt_vstat.f_files = 0;
 860         *sbp = mp->mnt_vstat;
 861         return(0);
 862 }
 863
 864 /*
 865  * Sync the filesystem.  Currently we have to run it twice, the second
 866  * one will advance the undo start index to the end index, so if a crash
 867  * occurs no undos will be run on mount.
 868  *
 869  * We do not sync the filesystem if we are called from a panic.  If we did
 870  * we might end up blowing up a sync that was already in progress.
 871  */
 872 static int
 873 hammer_vfs_sync(struct mount *mp, int waitfor)
 874 {
 875         struct hammer_mount *hmp = (void *)mp->mnt_data;
 876         int error;
 877
 878         if (panicstr == NULL) {
 879                 error = hammer_sync_hmp(hmp, waitfor);
 880         } else {
 881                 error = EIO;
 882         }
 883         return (error);
 884 }
 885
 886 /*
 887  * Convert a vnode to a file handle.
 888  */
 889 static int
 890 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 891 {
 892         hammer_inode_t ip;
 893
 894         KKASSERT(MAXFIDSZ >= 16);
 895         ip = VTOI(vp);
 896         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 897         fhp->fid_ext = ip->obj_localization >> 16;
 898         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 899         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 900         return(0);
 901 }
 902
 903
 904 /*
 905  * Convert a file handle back to a vnode.
 906  *
 907  * Use rootvp to enforce PFS isolation when a PFS is exported via a
 908  * null mount.
 909  */
 910 static int
 911 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 912                   struct fid *fhp, struct vnode **vpp)
 913 {
 914         struct hammer_transaction trans;
 915         struct hammer_inode *ip;
 916         struct hammer_inode_info info;
 917         int error;
 918         u_int32_t localization;
 919
 920         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 921         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 922         if (rootvp)
 923                 localization = VTOI(rootvp)->obj_localization;
 924         else
 925                 localization = (u_int32_t)fhp->fid_ext << 16;
 926
 927         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 928
 929         /*
 930          * Get/allocate the hammer_inode structure.  The structure must be
 931          * unlocked while we manipulate the related vnode to avoid a
 932          * deadlock.
 933          */
 934         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 935                               info.obj_asof, localization, 0, &error);
 936         if (ip == NULL) {
 937                 *vpp = NULL;
 938                 return(error);
 939         }
 940         error = hammer_get_vnode(ip, vpp);
 941         hammer_rel_inode(ip, 0);
 942         hammer_done_transaction(&trans);
 943         return (error);
 944 }
 945
 946 static int
 947 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 948                     int *exflagsp, struct ucred **credanonp)
 949 {
 950         hammer_mount_t hmp = (void *)mp->mnt_data;
 951         struct netcred *np;
 952         int error;
 953
 954         np = vfs_export_lookup(mp, &hmp->export, nam);
 955         if (np) {
 956                 *exflagsp = np->netc_exflags;
 957                 *credanonp = &np->netc_anon;
 958                 error = 0;
 959         } else {
 960                 error = EACCES;
 961         }
 962         return (error);
 963
 964 }
 965
 966 int
 967 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
 968 {
 969         hammer_mount_t hmp = (void *)mp->mnt_data;
 970         int error;
 971
 972         switch(op) {
 973         case MOUNTCTL_SET_EXPORT:
 974                 error = vfs_export(mp, &hmp->export, export);
 975                 break;
 976         default:
 977                 error = EOPNOTSUPP;
 978                 break;
 979         }
 980         return(error);
 981 }
 982