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_supported_version = HAMMER_VOL_VERSION_TWO;
  51 int hammer_debug_io;
  52 int hammer_debug_general;
  53 int hammer_debug_debug = 1;             /* medium-error panics */
  54 int hammer_debug_inode;
  55 int hammer_debug_locks;
  56 int hammer_debug_btree;
  57 int hammer_debug_tid;
  58 int hammer_debug_recover;               /* -1 will disable, +1 will force */
  59 int hammer_debug_recover_faults;
  60 int hammer_cluster_enable = 1;          /* enable read clustering by default */
  61 int hammer_count_fsyncs;
  62 int hammer_count_inodes;
  63 int hammer_count_iqueued;
  64 int hammer_count_reclaiming;
  65 int hammer_count_records;
  66 int hammer_count_record_datas;
  67 int hammer_count_volumes;
  68 int hammer_count_buffers;
  69 int hammer_count_nodes;
  70 int64_t hammer_count_extra_space_used;
  71 int64_t hammer_stats_btree_lookups;
  72 int64_t hammer_stats_btree_searches;
  73 int64_t hammer_stats_btree_inserts;
  74 int64_t hammer_stats_btree_deletes;
  75 int64_t hammer_stats_btree_elements;
  76 int64_t hammer_stats_btree_splits;
  77 int64_t hammer_stats_btree_iterations;
  78 int64_t hammer_stats_btree_root_iterations;
  79 int64_t hammer_stats_record_iterations;
  80
  81 int64_t hammer_stats_file_read;
  82 int64_t hammer_stats_file_write;
  83 int64_t hammer_stats_file_iopsr;
  84 int64_t hammer_stats_file_iopsw;
  85 int64_t hammer_stats_disk_read;
  86 int64_t hammer_stats_disk_write;
  87 int64_t hammer_stats_inode_flushes;
  88 int64_t hammer_stats_commits;
  89
  90 int hammer_count_dirtybufspace;         /* global */
  91 int hammer_count_refedbufs;             /* global */
  92 int hammer_count_reservations;
  93 int hammer_count_io_running_read;
  94 int hammer_count_io_running_write;
  95 int hammer_count_io_locked;
  96 int hammer_limit_dirtybufspace;         /* per-mount */
  97 int hammer_limit_recs;                  /* as a whole XXX */
  98 int hammer_limit_inode_recs = 1024;     /* per inode */
  99 int hammer_autoflush = 2000;            /* auto flush */
 100 int hammer_bio_count;
 101 int hammer_verify_zone;
 102 int hammer_verify_data = 1;
 103 int hammer_write_mode;
 104 int hammer_yield_check = 16;
 105 int64_t hammer_contention_count;
 106 int64_t hammer_zone_limit;
 107
 108 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
 109 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
 110            &hammer_supported_version, 0, "");
 111 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
 112            &hammer_debug_general, 0, "");
 113 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
 114            &hammer_debug_io, 0, "");
 115 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
 116            &hammer_debug_debug, 0, "");
 117 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 118            &hammer_debug_inode, 0, "");
 119 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 120            &hammer_debug_locks, 0, "");
 121 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 122            &hammer_debug_btree, 0, "");
 123 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 124            &hammer_debug_tid, 0, "");
 125 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 126            &hammer_debug_recover, 0, "");
 127 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 128            &hammer_debug_recover_faults, 0, "");
 129 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
 130            &hammer_cluster_enable, 0, "");
 131
 132 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 133            &hammer_limit_dirtybufspace, 0, "");
 134 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 135            &hammer_limit_recs, 0, "");
 136 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
 137            &hammer_limit_inode_recs, 0, "");
 138
 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 140            &hammer_count_fsyncs, 0, "");
 141 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 142            &hammer_count_inodes, 0, "");
 143 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 144            &hammer_count_iqueued, 0, "");
 145 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 146            &hammer_count_reclaiming, 0, "");
 147 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 148            &hammer_count_records, 0, "");
 149 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 150            &hammer_count_record_datas, 0, "");
 151 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 152            &hammer_count_volumes, 0, "");
 153 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 154            &hammer_count_buffers, 0, "");
 155 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 156            &hammer_count_nodes, 0, "");
 157 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
 158            &hammer_count_extra_space_used, 0, "");
 159
 160 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 161            &hammer_stats_btree_searches, 0, "");
 162 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 163            &hammer_stats_btree_lookups, 0, "");
 164 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 165            &hammer_stats_btree_inserts, 0, "");
 166 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 167            &hammer_stats_btree_deletes, 0, "");
 168 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 169            &hammer_stats_btree_elements, 0, "");
 170 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 171            &hammer_stats_btree_splits, 0, "");
 172 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 173            &hammer_stats_btree_iterations, 0, "");
 174 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
 175            &hammer_stats_btree_root_iterations, 0, "");
 176 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 177            &hammer_stats_record_iterations, 0, "");
 178
 179 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
 180            &hammer_stats_file_read, 0, "");
 181 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
 182            &hammer_stats_file_write, 0, "");
 183 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
 184            &hammer_stats_file_iopsr, 0, "");
 185 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
 186            &hammer_stats_file_iopsw, 0, "");
 187 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
 188            &hammer_stats_disk_read, 0, "");
 189 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
 190            &hammer_stats_disk_write, 0, "");
 191 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
 192            &hammer_stats_inode_flushes, 0, "");
 193 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
 194            &hammer_stats_commits, 0, "");
 195
 196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 197            &hammer_count_dirtybufspace, 0, "");
 198 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 199            &hammer_count_refedbufs, 0, "");
 200 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 201            &hammer_count_reservations, 0, "");
 202 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 203            &hammer_count_io_running_read, 0, "");
 204 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 205            &hammer_count_io_locked, 0, "");
 206 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 207            &hammer_count_io_running_write, 0, "");
 208 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 209            &hammer_zone_limit, 0, "");
 210 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 211            &hammer_contention_count, 0, "");
 212 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
 213            &hammer_autoflush, 0, "");
 214 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 215            &hammer_verify_zone, 0, "");
 216 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
 217            &hammer_verify_data, 0, "");
 218 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 219            &hammer_write_mode, 0, "");
 220 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
 221            &hammer_yield_check, 0, "");
 222
 223 KTR_INFO_MASTER(hammer);
 224
 225 /*
 226  * VFS ABI
 227  */
 228 static void     hammer_free_hmp(struct mount *mp);
 229
 230 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 231                                 struct ucred *cred);
 232 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 233 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 234 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 235                                 struct ucred *cred);
 236 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 237                                 struct ucred *cred);
 238 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 239 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 240                                 ino_t ino, struct vnode **vpp);
 241 static int      hammer_vfs_init(struct vfsconf *conf);
 242 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 243                                 struct fid *fhp, struct vnode **vpp);
 244 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 245 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 246                                 int *exflagsp, struct ucred **credanonp);
 247
 248
 249 static struct vfsops hammer_vfsops = {
 250         .vfs_mount      = hammer_vfs_mount,
 251         .vfs_unmount    = hammer_vfs_unmount,
 252         .vfs_root       = hammer_vfs_root,
 253         .vfs_statfs     = hammer_vfs_statfs,
 254         .vfs_statvfs    = hammer_vfs_statvfs,
 255         .vfs_sync       = hammer_vfs_sync,
 256         .vfs_vget       = hammer_vfs_vget,
 257         .vfs_init       = hammer_vfs_init,
 258         .vfs_vptofh     = hammer_vfs_vptofh,
 259         .vfs_fhtovp     = hammer_vfs_fhtovp,
 260         .vfs_checkexp   = hammer_vfs_checkexp
 261 };
 262
 263 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
 264
 265 VFS_SET(hammer_vfsops, hammer, 0);
 266 MODULE_VERSION(hammer, 1);
 267
 268 static int
 269 hammer_vfs_init(struct vfsconf *conf)
 270 {
 271         int n;
 272
 273         if (hammer_limit_recs == 0) {
 274                 hammer_limit_recs = nbuf * 25;
 275                 n = kmalloc_limit(M_HAMMER) / 512;
 276                 if (hammer_limit_recs > n)
 277                         hammer_limit_recs = n;
 278         }
 279         if (hammer_limit_dirtybufspace == 0) {
 280                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 281                 if (hammer_limit_dirtybufspace < 100)
 282                         hammer_limit_dirtybufspace = 100;
 283         }
 284         return(0);
 285 }
 286
 287 static int
 288 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 289                  struct ucred *cred)
 290 {
 291         struct hammer_mount_info info;
 292         hammer_mount_t hmp;
 293         hammer_volume_t rootvol;
 294         struct vnode *rootvp;
 295         struct vnode *devvp = NULL;
 296         const char *upath;      /* volume name in userspace */
 297         char *path;             /* volume name in system space */
 298         int error;
 299         int i;
 300         int master_id;
 301         int maxinodes;
 302
 303         /*
 304          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
 305          */
 306         if (mntpt == NULL) {
 307                 if ((error = bdevvp(rootdev, &devvp))) {
 308                         kprintf("hammer_mountroot: can't find devvp\n");
 309                         return (error);
 310                 }
 311                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 312                 bzero(&info, sizeof(info));
 313                 info.asof = 0;
 314                 info.hflags = 0;
 315                 info.nvolumes = 1;
 316         } else {
 317                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 318                         return (error);
 319         }
 320
 321         /*
 322          * updating or new mount
 323          */
 324         if (mp->mnt_flag & MNT_UPDATE) {
 325                 hmp = (void *)mp->mnt_data;
 326                 KKASSERT(hmp != NULL);
 327         } else {
 328                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 329                         return (EINVAL);
 330                 hmp = NULL;
 331         }
 332
 333         /*
 334          * master-id validation.  The master id may not be changed by a
 335          * mount update.
 336          */
 337         if (info.hflags & HMNT_MASTERID) {
 338                 if (hmp && hmp->master_id != info.master_id) {
 339                         kprintf("hammer: cannot change master id "
 340                                 "with mount update\n");
 341                         return(EINVAL);
 342                 }
 343                 master_id = info.master_id;
 344                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 345                         return (EINVAL);
 346         } else {
 347                 if (hmp)
 348                         master_id = hmp->master_id;
 349                 else
 350                         master_id = 0;
 351         }
 352
 353         /*
 354          * Interal mount data structure
 355          */
 356         if (hmp == NULL) {
 357                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 358                 mp->mnt_data = (qaddr_t)hmp;
 359                 hmp->mp = mp;
 360                 /*TAILQ_INIT(&hmp->recycle_list);*/
 361
 362                 /*
 363                  * Make sure kmalloc type limits are set appropriately.  If root
 364                  * increases the vnode limit you may have to do a dummy remount
 365                  * to adjust the HAMMER inode limit.
 366                  */
 367                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
 368                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
 369
 370                 maxinodes = desiredvnodes + desiredvnodes / 5 +
 371                             HAMMER_RECLAIM_WAIT;
 372                 kmalloc_raise_limit(hmp->m_inodes,
 373                                     maxinodes * sizeof(struct hammer_inode));
 374
 375                 hmp->root_btree_beg.localization = 0x00000000U;
 376                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 377                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 378                 hmp->root_btree_beg.create_tid = 1;
 379                 hmp->root_btree_beg.delete_tid = 1;
 380                 hmp->root_btree_beg.rec_type = 0;
 381                 hmp->root_btree_beg.obj_type = 0;
 382
 383                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 384                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 385                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 386                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 387                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 388                 hmp->root_btree_end.rec_type = 0xFFFFU;
 389                 hmp->root_btree_end.obj_type = 0;
 390
 391                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 392                 hmp->krate.count = -16; /* initial burst */
 393
 394                 hmp->sync_lock.refs = 1;
 395                 hmp->free_lock.refs = 1;
 396                 hmp->undo_lock.refs = 1;
 397                 hmp->blkmap_lock.refs = 1;
 398
 399                 TAILQ_INIT(&hmp->delay_list);
 400                 TAILQ_INIT(&hmp->flush_group_list);
 401                 TAILQ_INIT(&hmp->objid_cache_list);
 402                 TAILQ_INIT(&hmp->undo_lru_list);
 403                 TAILQ_INIT(&hmp->reclaim_list);
 404         }
 405         hmp->hflags &= ~HMNT_USERFLAGS;
 406         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 407
 408         hmp->master_id = master_id;
 409
 410         if (info.asof) {
 411                 mp->mnt_flag |= MNT_RDONLY;
 412                 hmp->asof = info.asof;
 413         } else {
 414                 hmp->asof = HAMMER_MAX_TID;
 415         }
 416
 417         /*
 418          * Re-open read-write if originally read-only, or vise-versa.
 419          */
 420         if (mp->mnt_flag & MNT_UPDATE) {
 421                 error = 0;
 422                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 423                         kprintf("HAMMER read-only -> read-write\n");
 424                         hmp->ronly = 0;
 425                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 426                                 hammer_adjust_volume_mode, NULL);
 427                         rootvol = hammer_get_root_volume(hmp, &error);
 428                         if (rootvol) {
 429                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 430                                 bcopy(rootvol->ondisk->vol0_blockmap,
 431                                       hmp->blockmap,
 432                                       sizeof(hmp->blockmap));
 433                                 hammer_rel_volume(rootvol, 0);
 434                         }
 435                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 436                                 hammer_reload_inode, NULL);
 437                         /* kernel clears MNT_RDONLY */
 438                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 439                         kprintf("HAMMER read-write -> read-only\n");
 440                         hmp->ronly = 1; /* messy */
 441                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 442                                 hammer_reload_inode, NULL);
 443                         hmp->ronly = 0;
 444                         hammer_flusher_sync(hmp);
 445                         hammer_flusher_sync(hmp);
 446                         hammer_flusher_sync(hmp);
 447                         hmp->ronly = 1;
 448                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 449                                 hammer_adjust_volume_mode, NULL);
 450                 }
 451                 return(error);
 452         }
 453
 454         RB_INIT(&hmp->rb_vols_root);
 455         RB_INIT(&hmp->rb_inos_root);
 456         RB_INIT(&hmp->rb_nods_root);
 457         RB_INIT(&hmp->rb_undo_root);
 458         RB_INIT(&hmp->rb_resv_root);
 459         RB_INIT(&hmp->rb_bufs_root);
 460         RB_INIT(&hmp->rb_pfsm_root);
 461
 462         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 463
 464         TAILQ_INIT(&hmp->volu_list);
 465         TAILQ_INIT(&hmp->undo_list);
 466         TAILQ_INIT(&hmp->data_list);
 467         TAILQ_INIT(&hmp->meta_list);
 468         TAILQ_INIT(&hmp->lose_list);
 469
 470         /*
 471          * Load volumes
 472          */
 473         path = objcache_get(namei_oc, M_WAITOK);
 474         hmp->nvolumes = -1;
 475         for (i = 0; i < info.nvolumes; ++i) {
 476                 if (mntpt == NULL) {
 477                         /*
 478                          * Root mount.
 479                          * Only one volume; and no need for copyin.
 480                          */
 481                         KKASSERT(info.nvolumes == 1);
 482                         ksnprintf(path, MAXPATHLEN, "/dev/%s",
 483                                   mp->mnt_stat.f_mntfromname);
 484                         error = 0;
 485                 } else {
 486                         error = copyin(&info.volumes[i], &upath,
 487                                        sizeof(char *));
 488                         if (error == 0)
 489                                 error = copyinstr(upath, path,
 490                                                   MAXPATHLEN, NULL);
 491                 }
 492                 if (error == 0)
 493                         error = hammer_install_volume(hmp, path, devvp);
 494                 if (error)
 495                         break;
 496         }
 497         objcache_put(namei_oc, path);
 498
 499         /*
 500          * Make sure we found a root volume
 501          */
 502         if (error == 0 && hmp->rootvol == NULL) {
 503                 kprintf("hammer_mount: No root volume found!\n");
 504                 error = EINVAL;
 505         }
 506
 507         /*
 508          * Check that all required volumes are available
 509          */
 510         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 511                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 512                 error = EINVAL;
 513         }
 514
 515         if (error) {
 516                 hammer_free_hmp(mp);
 517                 return (error);
 518         }
 519
 520         /*
 521          * No errors, setup enough of the mount point so we can lookup the
 522          * root vnode.
 523          */
 524         mp->mnt_iosize_max = MAXPHYS;
 525         mp->mnt_kern_flag |= MNTK_FSMID;
 526
 527         /*
 528          * note: f_iosize is used by vnode_pager_haspage() when constructing
 529          * its VOP_BMAP call.
 530          */
 531         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 532         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 533
 534         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 535         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 536
 537         mp->mnt_maxsymlinklen = 255;
 538         mp->mnt_flag |= MNT_LOCAL;
 539
 540         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 541         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 542         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 543
 544         /*
 545          * The root volume's ondisk pointer is only valid if we hold a
 546          * reference to it.
 547          */
 548         rootvol = hammer_get_root_volume(hmp, &error);
 549         if (error)
 550                 goto failed;
 551
 552         /*
 553          * Perform any necessary UNDO operations.  The recovery code does
 554          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 555          * and then re-copy it again after recovery is complete.
 556          *
 557          * If this is a read-only mount the UNDO information is retained
 558          * in memory in the form of dirty buffer cache buffers, and not
 559          * written back to the media.
 560          */
 561         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 562               sizeof(hmp->blockmap));
 563
 564         /*
 565          * Check filesystem version
 566          */
 567         hmp->version = rootvol->ondisk->vol_version;
 568         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
 569             hmp->version > HAMMER_VOL_VERSION_MAX) {
 570                 kprintf("HAMMER: mount unsupported fs version %d\n",
 571                         hmp->version);
 572                 error = ERANGE;
 573                 goto done;
 574         }
 575
 576         /*
 577          * The undo_rec_limit limits the size of flush groups to avoid
 578          * blowing out the UNDO FIFO.  This calculation is typically in
 579          * the tens of thousands and is designed primarily when small
 580          * HAMMER filesystems are created.
 581          */
 582         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 583         if (hammer_debug_general & 0x0001)
 584                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 585
 586         error = hammer_recover(hmp, rootvol);
 587         if (error) {
 588                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 589                 goto done;
 590         }
 591
 592         /*
 593          * Finish setup now that we have a good root volume.
 594          *
 595          * The top 16 bits of fsid.val[1] is a pfs id.
 596          */
 597         ksnprintf(mp->mnt_stat.f_mntfromname,
 598                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 599                   rootvol->ondisk->vol_name);
 600         mp->mnt_stat.f_fsid.val[0] =
 601                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 602         mp->mnt_stat.f_fsid.val[1] =
 603                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 604         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 605
 606         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 607         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 608                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 609
 610         /*
 611          * Certain often-modified fields in the root volume are cached in
 612          * the hammer_mount structure so we do not have to generate lots
 613          * of little UNDO structures for them.
 614          *
 615          * Recopy after recovery.  This also has the side effect of
 616          * setting our cached undo FIFO's first_offset, which serves to
 617          * placemark the FIFO start for the NEXT flush cycle while the
 618          * on-disk first_offset represents the LAST flush cycle.
 619          */
 620         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 621         hmp->flush_tid1 = hmp->next_tid;
 622         hmp->flush_tid2 = hmp->next_tid;
 623         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 624               sizeof(hmp->blockmap));
 625         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 626
 627         hammer_flusher_create(hmp);
 628
 629         /*
 630          * Locate the root directory using the root cluster's B-Tree as a
 631          * starting point.  The root directory uses an obj_id of 1.
 632          *
 633          * FUTURE: Leave the root directory cached referenced but unlocked
 634          * in hmp->rootvp (need to flush it on unmount).
 635          */
 636         error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
 637         if (error)
 638                 goto done;
 639         vput(rootvp);
 640         /*vn_unlock(hmp->rootvp);*/
 641
 642 done:
 643         hammer_rel_volume(rootvol, 0);
 644 failed:
 645         /*
 646          * Cleanup and return.
 647          */
 648         if (error)
 649                 hammer_free_hmp(mp);
 650         return (error);
 651 }
 652
 653 static int
 654 hammer_vfs_unmount(struct mount *mp, int mntflags)
 655 {
 656 #if 0
 657         struct hammer_mount *hmp = (void *)mp->mnt_data;
 658 #endif
 659         int flags;
 660         int error;
 661
 662         /*
 663          * Clean out the vnodes
 664          */
 665         flags = 0;
 666         if (mntflags & MNT_FORCE)
 667                 flags |= FORCECLOSE;
 668         if ((error = vflush(mp, 0, flags)) != 0)
 669                 return (error);
 670
 671         /*
 672          * Clean up the internal mount structure and related entities.  This
 673          * may issue I/O.
 674          */
 675         hammer_free_hmp(mp);
 676         return(0);
 677 }
 678
 679 /*
 680  * Clean up the internal mount structure and disassociate it from the mount.
 681  * This may issue I/O.
 682  */
 683 static void
 684 hammer_free_hmp(struct mount *mp)
 685 {
 686         struct hammer_mount *hmp = (void *)mp->mnt_data;
 687         hammer_flush_group_t flg;
 688         int count;
 689         int dummy;
 690
 691         /*
 692          * Flush anything dirty.  This won't even run if the
 693          * filesystem errored-out.
 694          */
 695         count = 0;
 696         while (hammer_flusher_haswork(hmp)) {
 697                 hammer_flusher_sync(hmp);
 698                 ++count;
 699                 if (count >= 5) {
 700                         if (count == 5)
 701                                 kprintf("HAMMER: umount flushing.");
 702                         else
 703                                 kprintf(".");
 704                         tsleep(&dummy, 0, "hmrufl", hz);
 705                 }
 706                 if (count == 30) {
 707                         kprintf("giving up\n");
 708                         break;
 709                 }
 710         }
 711         if (count >= 5 && count < 30)
 712                 kprintf("\n");
 713
 714         /*
 715          * If the mount had a critical error we have to destroy any
 716          * remaining inodes before we can finish cleaning up the flusher.
 717          */
 718         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 719                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 720                         hammer_destroy_inode_callback, NULL);
 721         }
 722
 723         /*
 724          * There shouldn't be any inodes left now and any left over
 725          * flush groups should now be empty.
 726          */
 727         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 728         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 729                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 730                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
 731                 if (flg->refs) {
 732                         kprintf("HAMMER: Warning, flush_group %p was "
 733                                 "not empty on umount!\n", flg);
 734                 }
 735                 kfree(flg, hmp->m_misc);
 736         }
 737
 738         /*
 739          * We can finally destroy the flusher
 740          */
 741         hammer_flusher_destroy(hmp);
 742
 743         /*
 744          * We may have held recovered buffers due to a read-only mount.
 745          * These must be discarded.
 746          */
 747         if (hmp->ronly)
 748                 hammer_recover_flush_buffers(hmp, NULL, -1);
 749
 750         /*
 751          * Unload buffers and then volumes
 752          */
 753         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 754                 hammer_unload_buffer, NULL);
 755         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 756                 hammer_unload_volume, NULL);
 757
 758         mp->mnt_data = NULL;
 759         mp->mnt_flag &= ~MNT_LOCAL;
 760         hmp->mp = NULL;
 761         hammer_destroy_objid_cache(hmp);
 762         kmalloc_destroy(&hmp->m_misc);
 763         kmalloc_destroy(&hmp->m_inodes);
 764         kfree(hmp, M_HAMMER);
 765 }
 766
 767 /*
 768  * Report critical errors.  ip may be NULL.
 769  */
 770 void
 771 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 772                       int error, const char *msg)
 773 {
 774         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 775         krateprintf(&hmp->krate,
 776                 "HAMMER(%s): Critical error inode=%lld %s\n",
 777                 hmp->mp->mnt_stat.f_mntfromname,
 778                 (long long)(ip ? ip->obj_id : -1), msg);
 779         if (hmp->ronly == 0) {
 780                 hmp->ronly = 2;         /* special errored read-only mode */
 781                 hmp->mp->mnt_flag |= MNT_RDONLY;
 782                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 783                         hmp->mp->mnt_stat.f_mntfromname);
 784         }
 785         hmp->error = error;
 786 }
 787
 788
 789 /*
 790  * Obtain a vnode for the specified inode number.  An exclusively locked
 791  * vnode is returned.
 792  */
 793 int
 794 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 795                 ino_t ino, struct vnode **vpp)
 796 {
 797         struct hammer_transaction trans;
 798         struct hammer_mount *hmp = (void *)mp->mnt_data;
 799         struct hammer_inode *ip;
 800         int error;
 801         u_int32_t localization;
 802
 803         hammer_simple_transaction(&trans, hmp);
 804
 805         /*
 806          * If a directory vnode is supplied (mainly NFS) then we can acquire
 807          * the PFS domain from it.  Otherwise we would only be able to vget
 808          * inodes in the root PFS.
 809          */
 810         if (dvp) {
 811                 localization = HAMMER_DEF_LOCALIZATION +
 812                                 VTOI(dvp)->obj_localization;
 813         } else {
 814                 localization = HAMMER_DEF_LOCALIZATION;
 815         }
 816
 817         /*
 818          * Lookup the requested HAMMER inode.  The structure must be
 819          * left unlocked while we manipulate the related vnode to avoid
 820          * a deadlock.
 821          */
 822         ip = hammer_get_inode(&trans, NULL, ino,
 823                               hmp->asof, localization,
 824                               0, &error);
 825         if (ip == NULL) {
 826                 *vpp = NULL;
 827                 hammer_done_transaction(&trans);
 828                 return(error);
 829         }
 830         error = hammer_get_vnode(ip, vpp);
 831         hammer_rel_inode(ip, 0);
 832         hammer_done_transaction(&trans);
 833         return (error);
 834 }
 835
 836 /*
 837  * Return the root vnode for the filesystem.
 838  *
 839  * HAMMER stores the root vnode in the hammer_mount structure so
 840  * getting it is easy.
 841  */
 842 static int
 843 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 844 {
 845 #if 0
 846         struct hammer_mount *hmp = (void *)mp->mnt_data;
 847 #endif
 848         int error;
 849
 850         error = hammer_vfs_vget(mp, NULL, 1, vpp);
 851         return (error);
 852 }
 853
 854 static int
 855 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 856 {
 857         struct hammer_mount *hmp = (void *)mp->mnt_data;
 858         hammer_volume_t volume;
 859         hammer_volume_ondisk_t ondisk;
 860         int error;
 861         int64_t bfree;
 862         int64_t breserved;
 863
 864         volume = hammer_get_root_volume(hmp, &error);
 865         if (error)
 866                 return(error);
 867         ondisk = volume->ondisk;
 868
 869         /*
 870          * Basic stats
 871          */
 872         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
 873         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 874         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 875         hammer_rel_volume(volume, 0);
 876
 877         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
 878         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 879         if (mp->mnt_stat.f_files < 0)
 880                 mp->mnt_stat.f_files = 0;
 881
 882         *sbp = mp->mnt_stat;
 883         return(0);
 884 }
 885
 886 static int
 887 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 888 {
 889         struct hammer_mount *hmp = (void *)mp->mnt_data;
 890         hammer_volume_t volume;
 891         hammer_volume_ondisk_t ondisk;
 892         int error;
 893         int64_t bfree;
 894         int64_t breserved;
 895
 896         volume = hammer_get_root_volume(hmp, &error);
 897         if (error)
 898                 return(error);
 899         ondisk = volume->ondisk;
 900
 901         /*
 902          * Basic stats
 903          */
 904         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
 905         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 906         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 907         hammer_rel_volume(volume, 0);
 908
 909         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
 910         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 911         if (mp->mnt_vstat.f_files < 0)
 912                 mp->mnt_vstat.f_files = 0;
 913         *sbp = mp->mnt_vstat;
 914         return(0);
 915 }
 916
 917 /*
 918  * Sync the filesystem.  Currently we have to run it twice, the second
 919  * one will advance the undo start index to the end index, so if a crash
 920  * occurs no undos will be run on mount.
 921  *
 922  * We do not sync the filesystem if we are called from a panic.  If we did
 923  * we might end up blowing up a sync that was already in progress.
 924  */
 925 static int
 926 hammer_vfs_sync(struct mount *mp, int waitfor)
 927 {
 928         struct hammer_mount *hmp = (void *)mp->mnt_data;
 929         int error;
 930
 931         if (panicstr == NULL) {
 932                 error = hammer_sync_hmp(hmp, waitfor);
 933         } else {
 934                 error = EIO;
 935         }
 936         return (error);
 937 }
 938
 939 /*
 940  * Convert a vnode to a file handle.
 941  */
 942 static int
 943 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 944 {
 945         hammer_inode_t ip;
 946
 947         KKASSERT(MAXFIDSZ >= 16);
 948         ip = VTOI(vp);
 949         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 950         fhp->fid_ext = ip->obj_localization >> 16;
 951         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 952         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 953         return(0);
 954 }
 955
 956
 957 /*
 958  * Convert a file handle back to a vnode.
 959  *
 960  * Use rootvp to enforce PFS isolation when a PFS is exported via a
 961  * null mount.
 962  */
 963 static int
 964 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 965                   struct fid *fhp, struct vnode **vpp)
 966 {
 967         struct hammer_transaction trans;
 968         struct hammer_inode *ip;
 969         struct hammer_inode_info info;
 970         int error;
 971         u_int32_t localization;
 972
 973         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 974         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 975         if (rootvp)
 976                 localization = VTOI(rootvp)->obj_localization;
 977         else
 978                 localization = (u_int32_t)fhp->fid_ext << 16;
 979
 980         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 981
 982         /*
 983          * Get/allocate the hammer_inode structure.  The structure must be
 984          * unlocked while we manipulate the related vnode to avoid a
 985          * deadlock.
 986          */
 987         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 988                               info.obj_asof, localization, 0, &error);
 989         if (ip == NULL) {
 990                 *vpp = NULL;
 991                 return(error);
 992         }
 993         error = hammer_get_vnode(ip, vpp);
 994         hammer_rel_inode(ip, 0);
 995         hammer_done_transaction(&trans);
 996         return (error);
 997 }
 998
 999 static int
1000 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1001                     int *exflagsp, struct ucred **credanonp)
1002 {
1003         hammer_mount_t hmp = (void *)mp->mnt_data;
1004         struct netcred *np;
1005         int error;
1006
1007         np = vfs_export_lookup(mp, &hmp->export, nam);
1008         if (np) {
1009                 *exflagsp = np->netc_exflags;
1010                 *credanonp = &np->netc_anon;
1011                 error = 0;
1012         } else {
1013                 error = EACCES;
1014         }
1015         return (error);
1016
1017 }
1018
1019 int
1020 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1021 {
1022         hammer_mount_t hmp = (void *)mp->mnt_data;
1023         int error;
1024
1025         switch(op) {
1026         case MOUNTCTL_SET_EXPORT:
1027                 error = vfs_export(mp, &hmp->export, export);
1028                 break;
1029         default:
1030                 error = EOPNOTSUPP;
1031                 break;
1032         }
1033         return(error);
1034 }
1035