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_error_panic;                 /* panic on error levels */
  61 int hammer_cluster_enable = 1;          /* enable read clustering by default */
  62 int hammer_count_fsyncs;
  63 int hammer_count_inodes;
  64 int hammer_count_iqueued;
  65 int hammer_count_reclaiming;
  66 int hammer_count_records;
  67 int hammer_count_record_datas;
  68 int hammer_count_volumes;
  69 int hammer_count_buffers;
  70 int hammer_count_nodes;
  71 int64_t hammer_count_extra_space_used;
  72 int64_t hammer_stats_btree_lookups;
  73 int64_t hammer_stats_btree_searches;
  74 int64_t hammer_stats_btree_inserts;
  75 int64_t hammer_stats_btree_deletes;
  76 int64_t hammer_stats_btree_elements;
  77 int64_t hammer_stats_btree_splits;
  78 int64_t hammer_stats_btree_iterations;
  79 int64_t hammer_stats_btree_root_iterations;
  80 int64_t hammer_stats_record_iterations;
  81
  82 int64_t hammer_stats_file_read;
  83 int64_t hammer_stats_file_write;
  84 int64_t hammer_stats_file_iopsr;
  85 int64_t hammer_stats_file_iopsw;
  86 int64_t hammer_stats_disk_read;
  87 int64_t hammer_stats_disk_write;
  88 int64_t hammer_stats_inode_flushes;
  89 int64_t hammer_stats_commits;
  90
  91 int hammer_count_dirtybufspace;         /* global */
  92 int hammer_count_refedbufs;             /* global */
  93 int hammer_count_reservations;
  94 int hammer_count_io_running_read;
  95 int hammer_count_io_running_write;
  96 int hammer_count_io_locked;
  97 int hammer_limit_dirtybufspace;         /* per-mount */
  98 int hammer_limit_recs;                  /* as a whole XXX */
  99 int hammer_limit_inode_recs = 1024;     /* per inode */
 100 int hammer_autoflush = 2000;            /* auto flush */
 101 int hammer_bio_count;
 102 int hammer_verify_zone;
 103 int hammer_verify_data = 1;
 104 int hammer_write_mode;
 105 int hammer_yield_check = 16;
 106 int hammer_fsync_mode;
 107 int64_t hammer_contention_count;
 108 int64_t hammer_zone_limit;
 109
 110 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
 111 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
 112            &hammer_supported_version, 0, "");
 113 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
 114            &hammer_debug_general, 0, "");
 115 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
 116            &hammer_debug_io, 0, "");
 117 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
 118            &hammer_debug_debug, 0, "");
 119 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 120            &hammer_debug_inode, 0, "");
 121 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 122            &hammer_debug_locks, 0, "");
 123 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 124            &hammer_debug_btree, 0, "");
 125 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 126            &hammer_debug_tid, 0, "");
 127 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 128            &hammer_debug_recover, 0, "");
 129 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 130            &hammer_debug_recover_faults, 0, "");
 131 SYSCTL_INT(_vfs_hammer, OID_AUTO, error_panic, CTLFLAG_RW,
 132            &hammer_error_panic, 0, "");
 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
 134            &hammer_cluster_enable, 0, "");
 135
 136 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 137            &hammer_limit_dirtybufspace, 0, "");
 138 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 139            &hammer_limit_recs, 0, "");
 140 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
 141            &hammer_limit_inode_recs, 0, "");
 142
 143 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 144            &hammer_count_fsyncs, 0, "");
 145 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 146            &hammer_count_inodes, 0, "");
 147 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 148            &hammer_count_iqueued, 0, "");
 149 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 150            &hammer_count_reclaiming, 0, "");
 151 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 152            &hammer_count_records, 0, "");
 153 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 154            &hammer_count_record_datas, 0, "");
 155 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 156            &hammer_count_volumes, 0, "");
 157 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 158            &hammer_count_buffers, 0, "");
 159 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 160            &hammer_count_nodes, 0, "");
 161 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
 162            &hammer_count_extra_space_used, 0, "");
 163
 164 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 165            &hammer_stats_btree_searches, 0, "");
 166 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 167            &hammer_stats_btree_lookups, 0, "");
 168 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 169            &hammer_stats_btree_inserts, 0, "");
 170 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 171            &hammer_stats_btree_deletes, 0, "");
 172 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 173            &hammer_stats_btree_elements, 0, "");
 174 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 175            &hammer_stats_btree_splits, 0, "");
 176 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 177            &hammer_stats_btree_iterations, 0, "");
 178 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
 179            &hammer_stats_btree_root_iterations, 0, "");
 180 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 181            &hammer_stats_record_iterations, 0, "");
 182
 183 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
 184            &hammer_stats_file_read, 0, "");
 185 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
 186            &hammer_stats_file_write, 0, "");
 187 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
 188            &hammer_stats_file_iopsr, 0, "");
 189 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
 190            &hammer_stats_file_iopsw, 0, "");
 191 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
 192            &hammer_stats_disk_read, 0, "");
 193 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
 194            &hammer_stats_disk_write, 0, "");
 195 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
 196            &hammer_stats_inode_flushes, 0, "");
 197 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
 198            &hammer_stats_commits, 0, "");
 199
 200 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 201            &hammer_count_dirtybufspace, 0, "");
 202 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 203            &hammer_count_refedbufs, 0, "");
 204 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 205            &hammer_count_reservations, 0, "");
 206 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 207            &hammer_count_io_running_read, 0, "");
 208 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 209            &hammer_count_io_locked, 0, "");
 210 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 211            &hammer_count_io_running_write, 0, "");
 212 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 213            &hammer_zone_limit, 0, "");
 214 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 215            &hammer_contention_count, 0, "");
 216 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
 217            &hammer_autoflush, 0, "");
 218 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 219            &hammer_verify_zone, 0, "");
 220 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
 221            &hammer_verify_data, 0, "");
 222 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 223            &hammer_write_mode, 0, "");
 224 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
 225            &hammer_yield_check, 0, "");
 226 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
 227            &hammer_fsync_mode, 0, "");
 228
 229 KTR_INFO_MASTER(hammer);
 230
 231 /*
 232  * VFS ABI
 233  */
 234 static void     hammer_free_hmp(struct mount *mp);
 235
 236 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 237                                 struct ucred *cred);
 238 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 239 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 240 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 241                                 struct ucred *cred);
 242 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 243                                 struct ucred *cred);
 244 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 245 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 246                                 ino_t ino, struct vnode **vpp);
 247 static int      hammer_vfs_init(struct vfsconf *conf);
 248 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 249                                 struct fid *fhp, struct vnode **vpp);
 250 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 251 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 252                                 int *exflagsp, struct ucred **credanonp);
 253
 254
 255 static struct vfsops hammer_vfsops = {
 256         .vfs_mount      = hammer_vfs_mount,
 257         .vfs_unmount    = hammer_vfs_unmount,
 258         .vfs_root       = hammer_vfs_root,
 259         .vfs_statfs     = hammer_vfs_statfs,
 260         .vfs_statvfs    = hammer_vfs_statvfs,
 261         .vfs_sync       = hammer_vfs_sync,
 262         .vfs_vget       = hammer_vfs_vget,
 263         .vfs_init       = hammer_vfs_init,
 264         .vfs_vptofh     = hammer_vfs_vptofh,
 265         .vfs_fhtovp     = hammer_vfs_fhtovp,
 266         .vfs_checkexp   = hammer_vfs_checkexp
 267 };
 268
 269 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
 270
 271 VFS_SET(hammer_vfsops, hammer, 0);
 272 MODULE_VERSION(hammer, 1);
 273
 274 static int
 275 hammer_vfs_init(struct vfsconf *conf)
 276 {
 277         int n;
 278
 279         if (hammer_limit_recs == 0) {
 280                 hammer_limit_recs = nbuf * 25;
 281                 n = kmalloc_limit(M_HAMMER) / 512;
 282                 if (hammer_limit_recs > n)
 283                         hammer_limit_recs = n;
 284         }
 285         if (hammer_limit_dirtybufspace == 0) {
 286                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 287                 if (hammer_limit_dirtybufspace < 100)
 288                         hammer_limit_dirtybufspace = 100;
 289         }
 290         return(0);
 291 }
 292
 293 static int
 294 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 295                  struct ucred *cred)
 296 {
 297         struct hammer_mount_info info;
 298         hammer_mount_t hmp;
 299         hammer_volume_t rootvol;
 300         struct vnode *rootvp;
 301         struct vnode *devvp = NULL;
 302         const char *upath;      /* volume name in userspace */
 303         char *path;             /* volume name in system space */
 304         int error;
 305         int i;
 306         int master_id;
 307         int maxinodes;
 308
 309         /*
 310          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
 311          */
 312         if (mntpt == NULL) {
 313                 if ((error = bdevvp(rootdev, &devvp))) {
 314                         kprintf("hammer_mountroot: can't find devvp\n");
 315                         return (error);
 316                 }
 317                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 318                 bzero(&info, sizeof(info));
 319                 info.asof = 0;
 320                 info.hflags = 0;
 321                 info.nvolumes = 1;
 322         } else {
 323                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 324                         return (error);
 325         }
 326
 327         /*
 328          * updating or new mount
 329          */
 330         if (mp->mnt_flag & MNT_UPDATE) {
 331                 hmp = (void *)mp->mnt_data;
 332                 KKASSERT(hmp != NULL);
 333         } else {
 334                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 335                         return (EINVAL);
 336                 hmp = NULL;
 337         }
 338
 339         /*
 340          * master-id validation.  The master id may not be changed by a
 341          * mount update.
 342          */
 343         if (info.hflags & HMNT_MASTERID) {
 344                 if (hmp && hmp->master_id != info.master_id) {
 345                         kprintf("hammer: cannot change master id "
 346                                 "with mount update\n");
 347                         return(EINVAL);
 348                 }
 349                 master_id = info.master_id;
 350                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 351                         return (EINVAL);
 352         } else {
 353                 if (hmp)
 354                         master_id = hmp->master_id;
 355                 else
 356                         master_id = 0;
 357         }
 358
 359         /*
 360          * Interal mount data structure
 361          */
 362         if (hmp == NULL) {
 363                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 364                 mp->mnt_data = (qaddr_t)hmp;
 365                 hmp->mp = mp;
 366                 /*TAILQ_INIT(&hmp->recycle_list);*/
 367
 368                 /*
 369                  * Make sure kmalloc type limits are set appropriately.  If root
 370                  * increases the vnode limit you may have to do a dummy remount
 371                  * to adjust the HAMMER inode limit.
 372                  */
 373                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
 374                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
 375
 376                 maxinodes = desiredvnodes + desiredvnodes / 5 +
 377                             HAMMER_RECLAIM_WAIT;
 378                 kmalloc_raise_limit(hmp->m_inodes,
 379                                     maxinodes * sizeof(struct hammer_inode));
 380
 381                 hmp->root_btree_beg.localization = 0x00000000U;
 382                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 383                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 384                 hmp->root_btree_beg.create_tid = 1;
 385                 hmp->root_btree_beg.delete_tid = 1;
 386                 hmp->root_btree_beg.rec_type = 0;
 387                 hmp->root_btree_beg.obj_type = 0;
 388
 389                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 390                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 391                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 392                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 393                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 394                 hmp->root_btree_end.rec_type = 0xFFFFU;
 395                 hmp->root_btree_end.obj_type = 0;
 396
 397                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 398                 hmp->krate.count = -16; /* initial burst */
 399
 400                 hmp->sync_lock.refs = 1;
 401                 hmp->free_lock.refs = 1;
 402                 hmp->undo_lock.refs = 1;
 403                 hmp->blkmap_lock.refs = 1;
 404
 405                 TAILQ_INIT(&hmp->delay_list);
 406                 TAILQ_INIT(&hmp->flush_group_list);
 407                 TAILQ_INIT(&hmp->objid_cache_list);
 408                 TAILQ_INIT(&hmp->undo_lru_list);
 409                 TAILQ_INIT(&hmp->reclaim_list);
 410         }
 411         hmp->hflags &= ~HMNT_USERFLAGS;
 412         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 413
 414         hmp->master_id = master_id;
 415
 416         if (info.asof) {
 417                 mp->mnt_flag |= MNT_RDONLY;
 418                 hmp->asof = info.asof;
 419         } else {
 420                 hmp->asof = HAMMER_MAX_TID;
 421         }
 422
 423         /*
 424          * Re-open read-write if originally read-only, or vise-versa.
 425          */
 426         if (mp->mnt_flag & MNT_UPDATE) {
 427                 error = 0;
 428                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 429                         kprintf("HAMMER read-only -> read-write\n");
 430                         hmp->ronly = 0;
 431                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 432                                 hammer_adjust_volume_mode, NULL);
 433                         rootvol = hammer_get_root_volume(hmp, &error);
 434                         if (rootvol) {
 435                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 436                                 bcopy(rootvol->ondisk->vol0_blockmap,
 437                                       hmp->blockmap,
 438                                       sizeof(hmp->blockmap));
 439                                 hammer_rel_volume(rootvol, 0);
 440                         }
 441                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 442                                 hammer_reload_inode, NULL);
 443                         /* kernel clears MNT_RDONLY */
 444                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 445                         kprintf("HAMMER read-write -> read-only\n");
 446                         hmp->ronly = 1; /* messy */
 447                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 448                                 hammer_reload_inode, NULL);
 449                         hmp->ronly = 0;
 450                         hammer_flusher_sync(hmp);
 451                         hammer_flusher_sync(hmp);
 452                         hammer_flusher_sync(hmp);
 453                         hmp->ronly = 1;
 454                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 455                                 hammer_adjust_volume_mode, NULL);
 456                 }
 457                 return(error);
 458         }
 459
 460         RB_INIT(&hmp->rb_vols_root);
 461         RB_INIT(&hmp->rb_inos_root);
 462         RB_INIT(&hmp->rb_nods_root);
 463         RB_INIT(&hmp->rb_undo_root);
 464         RB_INIT(&hmp->rb_resv_root);
 465         RB_INIT(&hmp->rb_bufs_root);
 466         RB_INIT(&hmp->rb_pfsm_root);
 467
 468         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 469
 470         TAILQ_INIT(&hmp->volu_list);
 471         TAILQ_INIT(&hmp->undo_list);
 472         TAILQ_INIT(&hmp->data_list);
 473         TAILQ_INIT(&hmp->meta_list);
 474         TAILQ_INIT(&hmp->lose_list);
 475
 476         /*
 477          * Load volumes
 478          */
 479         path = objcache_get(namei_oc, M_WAITOK);
 480         hmp->nvolumes = -1;
 481         for (i = 0; i < info.nvolumes; ++i) {
 482                 if (mntpt == NULL) {
 483                         /*
 484                          * Root mount.
 485                          * Only one volume; and no need for copyin.
 486                          */
 487                         KKASSERT(info.nvolumes == 1);
 488                         ksnprintf(path, MAXPATHLEN, "/dev/%s",
 489                                   mp->mnt_stat.f_mntfromname);
 490                         error = 0;
 491                 } else {
 492                         error = copyin(&info.volumes[i], &upath,
 493                                        sizeof(char *));
 494                         if (error == 0)
 495                                 error = copyinstr(upath, path,
 496                                                   MAXPATHLEN, NULL);
 497                 }
 498                 if (error == 0)
 499                         error = hammer_install_volume(hmp, path, devvp);
 500                 if (error)
 501                         break;
 502         }
 503         objcache_put(namei_oc, path);
 504
 505         /*
 506          * Make sure we found a root volume
 507          */
 508         if (error == 0 && hmp->rootvol == NULL) {
 509                 kprintf("hammer_mount: No root volume found!\n");
 510                 error = EINVAL;
 511         }
 512
 513         /*
 514          * Check that all required volumes are available
 515          */
 516         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 517                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 518                 error = EINVAL;
 519         }
 520
 521         if (error) {
 522                 hammer_free_hmp(mp);
 523                 return (error);
 524         }
 525
 526         /*
 527          * No errors, setup enough of the mount point so we can lookup the
 528          * root vnode.
 529          */
 530         mp->mnt_iosize_max = MAXPHYS;
 531         mp->mnt_kern_flag |= MNTK_FSMID;
 532
 533         /*
 534          * note: f_iosize is used by vnode_pager_haspage() when constructing
 535          * its VOP_BMAP call.
 536          */
 537         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 538         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 539
 540         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 541         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 542
 543         mp->mnt_maxsymlinklen = 255;
 544         mp->mnt_flag |= MNT_LOCAL;
 545
 546         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 547         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 548         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 549
 550         /*
 551          * The root volume's ondisk pointer is only valid if we hold a
 552          * reference to it.
 553          */
 554         rootvol = hammer_get_root_volume(hmp, &error);
 555         if (error)
 556                 goto failed;
 557
 558         /*
 559          * Perform any necessary UNDO operations.  The recovery code does
 560          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 561          * and then re-copy it again after recovery is complete.
 562          *
 563          * If this is a read-only mount the UNDO information is retained
 564          * in memory in the form of dirty buffer cache buffers, and not
 565          * written back to the media.
 566          */
 567         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 568               sizeof(hmp->blockmap));
 569
 570         /*
 571          * Check filesystem version
 572          */
 573         hmp->version = rootvol->ondisk->vol_version;
 574         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
 575             hmp->version > HAMMER_VOL_VERSION_MAX) {
 576                 kprintf("HAMMER: mount unsupported fs version %d\n",
 577                         hmp->version);
 578                 error = ERANGE;
 579                 goto done;
 580         }
 581
 582         /*
 583          * The undo_rec_limit limits the size of flush groups to avoid
 584          * blowing out the UNDO FIFO.  This calculation is typically in
 585          * the tens of thousands and is designed primarily when small
 586          * HAMMER filesystems are created.
 587          */
 588         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 589         if (hammer_debug_general & 0x0001)
 590                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 591
 592         error = hammer_recover(hmp, rootvol);
 593         if (error) {
 594                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 595                 goto done;
 596         }
 597
 598         /*
 599          * Finish setup now that we have a good root volume.
 600          *
 601          * The top 16 bits of fsid.val[1] is a pfs id.
 602          */
 603         ksnprintf(mp->mnt_stat.f_mntfromname,
 604                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 605                   rootvol->ondisk->vol_name);
 606         mp->mnt_stat.f_fsid.val[0] =
 607                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 608         mp->mnt_stat.f_fsid.val[1] =
 609                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 610         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 611
 612         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 613         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 614                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 615
 616         /*
 617          * Certain often-modified fields in the root volume are cached in
 618          * the hammer_mount structure so we do not have to generate lots
 619          * of little UNDO structures for them.
 620          *
 621          * Recopy after recovery.  This also has the side effect of
 622          * setting our cached undo FIFO's first_offset, which serves to
 623          * placemark the FIFO start for the NEXT flush cycle while the
 624          * on-disk first_offset represents the LAST flush cycle.
 625          */
 626         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 627         hmp->flush_tid1 = hmp->next_tid;
 628         hmp->flush_tid2 = hmp->next_tid;
 629         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 630               sizeof(hmp->blockmap));
 631         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 632
 633         hammer_flusher_create(hmp);
 634
 635         /*
 636          * Locate the root directory using the root cluster's B-Tree as a
 637          * starting point.  The root directory uses an obj_id of 1.
 638          *
 639          * FUTURE: Leave the root directory cached referenced but unlocked
 640          * in hmp->rootvp (need to flush it on unmount).
 641          */
 642         error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
 643         if (error)
 644                 goto done;
 645         vput(rootvp);
 646         /*vn_unlock(hmp->rootvp);*/
 647
 648 done:
 649         hammer_rel_volume(rootvol, 0);
 650 failed:
 651         /*
 652          * Cleanup and return.
 653          */
 654         if (error)
 655                 hammer_free_hmp(mp);
 656         return (error);
 657 }
 658
 659 static int
 660 hammer_vfs_unmount(struct mount *mp, int mntflags)
 661 {
 662 #if 0
 663         struct hammer_mount *hmp = (void *)mp->mnt_data;
 664 #endif
 665         int flags;
 666         int error;
 667
 668         /*
 669          * Clean out the vnodes
 670          */
 671         flags = 0;
 672         if (mntflags & MNT_FORCE)
 673                 flags |= FORCECLOSE;
 674         if ((error = vflush(mp, 0, flags)) != 0)
 675                 return (error);
 676
 677         /*
 678          * Clean up the internal mount structure and related entities.  This
 679          * may issue I/O.
 680          */
 681         hammer_free_hmp(mp);
 682         return(0);
 683 }
 684
 685 /*
 686  * Clean up the internal mount structure and disassociate it from the mount.
 687  * This may issue I/O.
 688  */
 689 static void
 690 hammer_free_hmp(struct mount *mp)
 691 {
 692         struct hammer_mount *hmp = (void *)mp->mnt_data;
 693         hammer_flush_group_t flg;
 694         int count;
 695         int dummy;
 696
 697         /*
 698          * Flush anything dirty.  This won't even run if the
 699          * filesystem errored-out.
 700          */
 701         count = 0;
 702         while (hammer_flusher_haswork(hmp)) {
 703                 hammer_flusher_sync(hmp);
 704                 ++count;
 705                 if (count >= 5) {
 706                         if (count == 5)
 707                                 kprintf("HAMMER: umount flushing.");
 708                         else
 709                                 kprintf(".");
 710                         tsleep(&dummy, 0, "hmrufl", hz);
 711                 }
 712                 if (count == 30) {
 713                         kprintf("giving up\n");
 714                         break;
 715                 }
 716         }
 717         if (count >= 5 && count < 30)
 718                 kprintf("\n");
 719
 720         /*
 721          * If the mount had a critical error we have to destroy any
 722          * remaining inodes before we can finish cleaning up the flusher.
 723          */
 724         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 725                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 726                         hammer_destroy_inode_callback, NULL);
 727         }
 728
 729         /*
 730          * There shouldn't be any inodes left now and any left over
 731          * flush groups should now be empty.
 732          */
 733         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 734         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 735                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 736                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
 737                 if (flg->refs) {
 738                         kprintf("HAMMER: Warning, flush_group %p was "
 739                                 "not empty on umount!\n", flg);
 740                 }
 741                 kfree(flg, hmp->m_misc);
 742         }
 743
 744         /*
 745          * We can finally destroy the flusher
 746          */
 747         hammer_flusher_destroy(hmp);
 748
 749         /*
 750          * We may have held recovered buffers due to a read-only mount.
 751          * These must be discarded.
 752          */
 753         if (hmp->ronly)
 754                 hammer_recover_flush_buffers(hmp, NULL, -1);
 755
 756         /*
 757          * Unload buffers and then volumes
 758          */
 759         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 760                 hammer_unload_buffer, NULL);
 761         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 762                 hammer_unload_volume, NULL);
 763
 764         mp->mnt_data = NULL;
 765         mp->mnt_flag &= ~MNT_LOCAL;
 766         hmp->mp = NULL;
 767         hammer_destroy_objid_cache(hmp);
 768         kmalloc_destroy(&hmp->m_misc);
 769         kmalloc_destroy(&hmp->m_inodes);
 770         kfree(hmp, M_HAMMER);
 771 }
 772
 773 /*
 774  * Report critical errors.  ip may be NULL.
 775  */
 776 void
 777 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 778                       int error, const char *msg)
 779 {
 780         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 781
 782         krateprintf(&hmp->krate,
 783                     "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
 784                     hmp->mp->mnt_stat.f_mntfromname,
 785                     (intmax_t)(ip ? ip->obj_id : -1),
 786                     error, msg);
 787
 788         if (hmp->ronly == 0) {
 789                 hmp->ronly = 2;         /* special errored read-only mode */
 790                 hmp->mp->mnt_flag |= MNT_RDONLY;
 791                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 792                         hmp->mp->mnt_stat.f_mntfromname);
 793         }
 794         hmp->error = error;
 795         if (hammer_error_panic > 2)
 796                 Debugger("Entering debugger");
 797 }
 798
 799
 800 /*
 801  * Obtain a vnode for the specified inode number.  An exclusively locked
 802  * vnode is returned.
 803  */
 804 int
 805 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 806                 ino_t ino, struct vnode **vpp)
 807 {
 808         struct hammer_transaction trans;
 809         struct hammer_mount *hmp = (void *)mp->mnt_data;
 810         struct hammer_inode *ip;
 811         int error;
 812         u_int32_t localization;
 813
 814         hammer_simple_transaction(&trans, hmp);
 815
 816         /*
 817          * If a directory vnode is supplied (mainly NFS) then we can acquire
 818          * the PFS domain from it.  Otherwise we would only be able to vget
 819          * inodes in the root PFS.
 820          */
 821         if (dvp) {
 822                 localization = HAMMER_DEF_LOCALIZATION +
 823                                 VTOI(dvp)->obj_localization;
 824         } else {
 825                 localization = HAMMER_DEF_LOCALIZATION;
 826         }
 827
 828         /*
 829          * Lookup the requested HAMMER inode.  The structure must be
 830          * left unlocked while we manipulate the related vnode to avoid
 831          * a deadlock.
 832          */
 833         ip = hammer_get_inode(&trans, NULL, ino,
 834                               hmp->asof, localization,
 835                               0, &error);
 836         if (ip == NULL) {
 837                 *vpp = NULL;
 838                 hammer_done_transaction(&trans);
 839                 return(error);
 840         }
 841         error = hammer_get_vnode(ip, vpp);
 842         hammer_rel_inode(ip, 0);
 843         hammer_done_transaction(&trans);
 844         return (error);
 845 }
 846
 847 /*
 848  * Return the root vnode for the filesystem.
 849  *
 850  * HAMMER stores the root vnode in the hammer_mount structure so
 851  * getting it is easy.
 852  */
 853 static int
 854 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 855 {
 856 #if 0
 857         struct hammer_mount *hmp = (void *)mp->mnt_data;
 858 #endif
 859         int error;
 860
 861         error = hammer_vfs_vget(mp, NULL, 1, vpp);
 862         return (error);
 863 }
 864
 865 static int
 866 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 867 {
 868         struct hammer_mount *hmp = (void *)mp->mnt_data;
 869         hammer_volume_t volume;
 870         hammer_volume_ondisk_t ondisk;
 871         int error;
 872         int64_t bfree;
 873         int64_t breserved;
 874
 875         volume = hammer_get_root_volume(hmp, &error);
 876         if (error)
 877                 return(error);
 878         ondisk = volume->ondisk;
 879
 880         /*
 881          * Basic stats
 882          */
 883         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
 884         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 885         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 886         hammer_rel_volume(volume, 0);
 887
 888         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
 889         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 890         if (mp->mnt_stat.f_files < 0)
 891                 mp->mnt_stat.f_files = 0;
 892
 893         *sbp = mp->mnt_stat;
 894         return(0);
 895 }
 896
 897 static int
 898 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 899 {
 900         struct hammer_mount *hmp = (void *)mp->mnt_data;
 901         hammer_volume_t volume;
 902         hammer_volume_ondisk_t ondisk;
 903         int error;
 904         int64_t bfree;
 905         int64_t breserved;
 906
 907         volume = hammer_get_root_volume(hmp, &error);
 908         if (error)
 909                 return(error);
 910         ondisk = volume->ondisk;
 911
 912         /*
 913          * Basic stats
 914          */
 915         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
 916         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 917         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 918         hammer_rel_volume(volume, 0);
 919
 920         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
 921         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 922         if (mp->mnt_vstat.f_files < 0)
 923                 mp->mnt_vstat.f_files = 0;
 924         *sbp = mp->mnt_vstat;
 925         return(0);
 926 }
 927
 928 /*
 929  * Sync the filesystem.  Currently we have to run it twice, the second
 930  * one will advance the undo start index to the end index, so if a crash
 931  * occurs no undos will be run on mount.
 932  *
 933  * We do not sync the filesystem if we are called from a panic.  If we did
 934  * we might end up blowing up a sync that was already in progress.
 935  */
 936 static int
 937 hammer_vfs_sync(struct mount *mp, int waitfor)
 938 {
 939         struct hammer_mount *hmp = (void *)mp->mnt_data;
 940         int error;
 941
 942         if (panicstr == NULL) {
 943                 error = hammer_sync_hmp(hmp, waitfor);
 944         } else {
 945                 error = EIO;
 946         }
 947         return (error);
 948 }
 949
 950 /*
 951  * Convert a vnode to a file handle.
 952  */
 953 static int
 954 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 955 {
 956         hammer_inode_t ip;
 957
 958         KKASSERT(MAXFIDSZ >= 16);
 959         ip = VTOI(vp);
 960         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 961         fhp->fid_ext = ip->obj_localization >> 16;
 962         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 963         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 964         return(0);
 965 }
 966
 967
 968 /*
 969  * Convert a file handle back to a vnode.
 970  *
 971  * Use rootvp to enforce PFS isolation when a PFS is exported via a
 972  * null mount.
 973  */
 974 static int
 975 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 976                   struct fid *fhp, struct vnode **vpp)
 977 {
 978         struct hammer_transaction trans;
 979         struct hammer_inode *ip;
 980         struct hammer_inode_info info;
 981         int error;
 982         u_int32_t localization;
 983
 984         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 985         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 986         if (rootvp)
 987                 localization = VTOI(rootvp)->obj_localization;
 988         else
 989                 localization = (u_int32_t)fhp->fid_ext << 16;
 990
 991         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 992
 993         /*
 994          * Get/allocate the hammer_inode structure.  The structure must be
 995          * unlocked while we manipulate the related vnode to avoid a
 996          * deadlock.
 997          */
 998         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 999                               info.obj_asof, localization, 0, &error);
1000         if (ip == NULL) {
1001                 *vpp = NULL;
1002                 return(error);
1003         }
1004         error = hammer_get_vnode(ip, vpp);
1005         hammer_rel_inode(ip, 0);
1006         hammer_done_transaction(&trans);
1007         return (error);
1008 }
1009
1010 static int
1011 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1012                     int *exflagsp, struct ucred **credanonp)
1013 {
1014         hammer_mount_t hmp = (void *)mp->mnt_data;
1015         struct netcred *np;
1016         int error;
1017
1018         np = vfs_export_lookup(mp, &hmp->export, nam);
1019         if (np) {
1020                 *exflagsp = np->netc_exflags;
1021                 *credanonp = &np->netc_anon;
1022                 error = 0;
1023         } else {
1024                 error = EACCES;
1025         }
1026         return (error);
1027
1028 }
1029
1030 int
1031 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1032 {
1033         hammer_mount_t hmp = (void *)mp->mnt_data;
1034         int error;
1035
1036         switch(op) {
1037         case MOUNTCTL_SET_EXPORT:
1038                 error = vfs_export(mp, &hmp->export, export);
1039                 break;
1040         default:
1041                 error = EOPNOTSUPP;
1042                 break;
1043         }
1044         return(error);
1045 }
1046