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 1.68 2008/07/27 21:34:04 mneumann 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_bio_count;
  98 int hammer_verify_zone;
  99 int hammer_verify_data = 1;
 100 int hammer_write_mode;
 101 int64_t hammer_contention_count;
 102 int64_t hammer_zone_limit;
 103
 104 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
 105 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
 106            &hammer_debug_general, 0, "");
 107 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
 108            &hammer_debug_io, 0, "");
 109 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
 110            &hammer_debug_debug, 0, "");
 111 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 112            &hammer_debug_inode, 0, "");
 113 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 114            &hammer_debug_locks, 0, "");
 115 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 116            &hammer_debug_btree, 0, "");
 117 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 118            &hammer_debug_tid, 0, "");
 119 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 120            &hammer_debug_recover, 0, "");
 121 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 122            &hammer_debug_recover_faults, 0, "");
 123 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
 124            &hammer_cluster_enable, 0, "");
 125
 126 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 127            &hammer_limit_dirtybufspace, 0, "");
 128 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 129            &hammer_limit_recs, 0, "");
 130 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_iqueued, CTLFLAG_RW,
 131            &hammer_limit_iqueued, 0, "");
 132
 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 134            &hammer_count_fsyncs, 0, "");
 135 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 136            &hammer_count_inodes, 0, "");
 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 138            &hammer_count_iqueued, 0, "");
 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 140            &hammer_count_reclaiming, 0, "");
 141 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 142            &hammer_count_records, 0, "");
 143 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 144            &hammer_count_record_datas, 0, "");
 145 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 146            &hammer_count_volumes, 0, "");
 147 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 148            &hammer_count_buffers, 0, "");
 149 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 150            &hammer_count_nodes, 0, "");
 151 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
 152            &hammer_count_extra_space_used, 0, "");
 153
 154 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 155            &hammer_stats_btree_searches, 0, "");
 156 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 157            &hammer_stats_btree_lookups, 0, "");
 158 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 159            &hammer_stats_btree_inserts, 0, "");
 160 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 161            &hammer_stats_btree_deletes, 0, "");
 162 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 163            &hammer_stats_btree_elements, 0, "");
 164 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 165            &hammer_stats_btree_splits, 0, "");
 166 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 167            &hammer_stats_btree_iterations, 0, "");
 168 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 169            &hammer_stats_record_iterations, 0, "");
 170
 171 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
 172            &hammer_stats_file_read, 0, "");
 173 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
 174            &hammer_stats_file_write, 0, "");
 175 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
 176            &hammer_stats_file_iopsr, 0, "");
 177 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
 178            &hammer_stats_file_iopsw, 0, "");
 179 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
 180            &hammer_stats_disk_read, 0, "");
 181 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
 182            &hammer_stats_disk_write, 0, "");
 183 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
 184            &hammer_stats_inode_flushes, 0, "");
 185 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
 186            &hammer_stats_commits, 0, "");
 187
 188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 189            &hammer_count_dirtybufspace, 0, "");
 190 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 191            &hammer_count_refedbufs, 0, "");
 192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 193            &hammer_count_reservations, 0, "");
 194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 195            &hammer_count_io_running_read, 0, "");
 196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 197            &hammer_count_io_locked, 0, "");
 198 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 199            &hammer_count_io_running_write, 0, "");
 200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 201            &hammer_zone_limit, 0, "");
 202 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 203            &hammer_contention_count, 0, "");
 204 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 205            &hammer_verify_zone, 0, "");
 206 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
 207            &hammer_verify_data, 0, "");
 208 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 209            &hammer_write_mode, 0, "");
 210
 211 KTR_INFO_MASTER(hammer);
 212
 213 /*
 214  * VFS ABI
 215  */
 216 static void     hammer_free_hmp(struct mount *mp);
 217
 218 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 219                                 struct ucred *cred);
 220 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 221 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 222 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 223                                 struct ucred *cred);
 224 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 225                                 struct ucred *cred);
 226 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 227 static int      hammer_vfs_vget(struct mount *mp, ino_t ino,
 228                                 struct vnode **vpp);
 229 static int      hammer_vfs_init(struct vfsconf *conf);
 230 static int      hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp,
 231                                 struct vnode **vpp);
 232 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 233 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 234                                 int *exflagsp, struct ucred **credanonp);
 235
 236
 237 static struct vfsops hammer_vfsops = {
 238         .vfs_mount      = hammer_vfs_mount,
 239         .vfs_unmount    = hammer_vfs_unmount,
 240         .vfs_root       = hammer_vfs_root,
 241         .vfs_statfs     = hammer_vfs_statfs,
 242         .vfs_statvfs    = hammer_vfs_statvfs,
 243         .vfs_sync       = hammer_vfs_sync,
 244         .vfs_vget       = hammer_vfs_vget,
 245         .vfs_init       = hammer_vfs_init,
 246         .vfs_vptofh     = hammer_vfs_vptofh,
 247         .vfs_fhtovp     = hammer_vfs_fhtovp,
 248         .vfs_checkexp   = hammer_vfs_checkexp
 249 };
 250
 251 MALLOC_DEFINE(M_HAMMER, "hammer-mount", "hammer mount");
 252
 253 VFS_SET(hammer_vfsops, hammer, 0);
 254 MODULE_VERSION(hammer, 1);
 255
 256 static int
 257 hammer_vfs_init(struct vfsconf *conf)
 258 {
 259         int n;
 260
 261         if (hammer_limit_recs == 0) {
 262                 hammer_limit_recs = nbuf * 25;
 263                 n = kmalloc_limit(M_HAMMER) / 512;
 264                 if (hammer_limit_recs > n)
 265                         hammer_limit_recs = n;
 266         }
 267         if (hammer_limit_dirtybufspace == 0) {
 268                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 269                 if (hammer_limit_dirtybufspace < 100)
 270                         hammer_limit_dirtybufspace = 100;
 271         }
 272         if (hammer_limit_iqueued == 0)
 273                 hammer_limit_iqueued = desiredvnodes / 5;
 274         return(0);
 275 }
 276
 277 static int
 278 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 279                  struct ucred *cred)
 280 {
 281         struct hammer_mount_info info;
 282         hammer_mount_t hmp;
 283         hammer_volume_t rootvol;
 284         struct vnode *rootvp;
 285         struct vnode *devvp = NULL;
 286         const char *upath;      /* volume name in userspace */
 287         char *path;             /* volume name in system space */
 288         int error;
 289         int i;
 290         int master_id;
 291         if (mntpt == NULL) {
 292                 /*
 293                  * Root mount
 294                  */
 295                 if ((error = bdevvp(rootdev, &devvp))) {
 296                         kprintf("hammer_mountroot: can't find devvp\n");
 297                         return (error);
 298                 }
 299                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 300                 bzero(&info, sizeof(info));
 301                 info.asof = 0;
 302                 info.hflags = 0;
 303                 info.nvolumes = 1;
 304         } else {
 305                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 306                         return (error);
 307         }
 308
 309         /*
 310          * updating or new mount
 311          */
 312         if (mp->mnt_flag & MNT_UPDATE) {
 313                 hmp = (void *)mp->mnt_data;
 314                 KKASSERT(hmp != NULL);
 315         } else {
 316                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 317                         return (EINVAL);
 318                 hmp = NULL;
 319         }
 320
 321         /*
 322          * master-id validation.  The master id may not be changed by a
 323          * mount update.
 324          */
 325         if (info.hflags & HMNT_MASTERID) {
 326                 if (hmp && hmp->master_id != info.master_id) {
 327                         kprintf("hammer: cannot change master id "
 328                                 "with mount update\n");
 329                         return(EINVAL);
 330                 }
 331                 master_id = info.master_id;
 332                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 333                         return (EINVAL);
 334         } else {
 335                 if (hmp)
 336                         master_id = hmp->master_id;
 337                 else
 338                         master_id = 0;
 339         }
 340
 341         /*
 342          * Interal mount data structure
 343          */
 344         if (hmp == NULL) {
 345                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 346                 mp->mnt_data = (qaddr_t)hmp;
 347                 hmp->mp = mp;
 348                 hmp->namekey_iterator = mycpu->gd_time_seconds;
 349                 /*TAILQ_INIT(&hmp->recycle_list);*/
 350
 351                 hmp->root_btree_beg.localization = 0x00000000U;
 352                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 353                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 354                 hmp->root_btree_beg.create_tid = 1;
 355                 hmp->root_btree_beg.delete_tid = 1;
 356                 hmp->root_btree_beg.rec_type = 0;
 357                 hmp->root_btree_beg.obj_type = 0;
 358
 359                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 360                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 361                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 362                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 363                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 364                 hmp->root_btree_end.rec_type = 0xFFFFU;
 365                 hmp->root_btree_end.obj_type = 0;
 366
 367                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 368                 hmp->krate.count = -16; /* initial burst */
 369
 370                 hmp->sync_lock.refs = 1;
 371                 hmp->free_lock.refs = 1;
 372                 hmp->undo_lock.refs = 1;
 373                 hmp->blkmap_lock.refs = 1;
 374
 375                 TAILQ_INIT(&hmp->delay_list);
 376                 TAILQ_INIT(&hmp->flush_group_list);
 377                 TAILQ_INIT(&hmp->objid_cache_list);
 378                 TAILQ_INIT(&hmp->undo_lru_list);
 379                 TAILQ_INIT(&hmp->reclaim_list);
 380         }
 381         hmp->hflags &= ~HMNT_USERFLAGS;
 382         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 383
 384         hmp->master_id = master_id;
 385
 386         if (info.asof) {
 387                 mp->mnt_flag |= MNT_RDONLY;
 388                 hmp->asof = info.asof;
 389         } else {
 390                 hmp->asof = HAMMER_MAX_TID;
 391         }
 392
 393         /*
 394          * Re-open read-write if originally read-only, or vise-versa.
 395          */
 396         if (mp->mnt_flag & MNT_UPDATE) {
 397                 error = 0;
 398                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 399                         kprintf("HAMMER read-only -> read-write\n");
 400                         hmp->ronly = 0;
 401                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 402                                 hammer_adjust_volume_mode, NULL);
 403                         rootvol = hammer_get_root_volume(hmp, &error);
 404                         if (rootvol) {
 405                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 406                                 bcopy(rootvol->ondisk->vol0_blockmap,
 407                                       hmp->blockmap,
 408                                       sizeof(hmp->blockmap));
 409                                 hammer_rel_volume(rootvol, 0);
 410                         }
 411                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 412                                 hammer_reload_inode, NULL);
 413                         /* kernel clears MNT_RDONLY */
 414                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 415                         kprintf("HAMMER read-write -> read-only\n");
 416                         hmp->ronly = 1; /* messy */
 417                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 418                                 hammer_reload_inode, NULL);
 419                         hmp->ronly = 0;
 420                         hammer_flusher_sync(hmp);
 421                         hammer_flusher_sync(hmp);
 422                         hammer_flusher_sync(hmp);
 423                         hmp->ronly = 1;
 424                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 425                                 hammer_adjust_volume_mode, NULL);
 426                 }
 427                 return(error);
 428         }
 429
 430         RB_INIT(&hmp->rb_vols_root);
 431         RB_INIT(&hmp->rb_inos_root);
 432         RB_INIT(&hmp->rb_nods_root);
 433         RB_INIT(&hmp->rb_undo_root);
 434         RB_INIT(&hmp->rb_resv_root);
 435         RB_INIT(&hmp->rb_bufs_root);
 436         RB_INIT(&hmp->rb_pfsm_root);
 437
 438         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 439
 440         TAILQ_INIT(&hmp->volu_list);
 441         TAILQ_INIT(&hmp->undo_list);
 442         TAILQ_INIT(&hmp->data_list);
 443         TAILQ_INIT(&hmp->meta_list);
 444         TAILQ_INIT(&hmp->lose_list);
 445
 446         /*
 447          * Load volumes
 448          */
 449         path = objcache_get(namei_oc, M_WAITOK);
 450         hmp->nvolumes = -1;
 451         for (i = 0; i < info.nvolumes; ++i) {
 452                 if (mntpt == NULL) {
 453                         /*
 454                          * Root mount.
 455                          * Only one volume; and no need for copyin.
 456                          */
 457                         KKASSERT(info.nvolumes == 1);
 458                         ksnprintf(path, MAXPATHLEN, "/dev/%s",
 459                                   mp->mnt_stat.f_mntfromname);
 460                         error = 0;
 461                 } else {
 462                         error = copyin(&info.volumes[i], &upath,
 463                                        sizeof(char *));
 464                         if (error == 0)
 465                                 error = copyinstr(upath, path,
 466                                                   MAXPATHLEN, NULL);
 467                 }
 468                 if (error == 0)
 469                         error = hammer_install_volume(hmp, path, devvp);
 470                 if (error)
 471                         break;
 472         }
 473         objcache_put(namei_oc, path);
 474
 475         /*
 476          * Make sure we found a root volume
 477          */
 478         if (error == 0 && hmp->rootvol == NULL) {
 479                 kprintf("hammer_mount: No root volume found!\n");
 480                 error = EINVAL;
 481         }
 482
 483         /*
 484          * Check that all required volumes are available
 485          */
 486         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 487                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 488                 error = EINVAL;
 489         }
 490
 491         if (error) {
 492                 hammer_free_hmp(mp);
 493                 return (error);
 494         }
 495
 496         /*
 497          * No errors, setup enough of the mount point so we can lookup the
 498          * root vnode.
 499          */
 500         mp->mnt_iosize_max = MAXPHYS;
 501         mp->mnt_kern_flag |= MNTK_FSMID;
 502
 503         /*
 504          * note: f_iosize is used by vnode_pager_haspage() when constructing
 505          * its VOP_BMAP call.
 506          */
 507         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 508         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 509
 510         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 511         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 512
 513         mp->mnt_maxsymlinklen = 255;
 514         mp->mnt_flag |= MNT_LOCAL;
 515
 516         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 517         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 518         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 519
 520         /*
 521          * The root volume's ondisk pointer is only valid if we hold a
 522          * reference to it.
 523          */
 524         rootvol = hammer_get_root_volume(hmp, &error);
 525         if (error)
 526                 goto failed;
 527
 528         /*
 529          * Perform any necessary UNDO operations.  The recovery code does
 530          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 531          * and then re-copy it again after recovery is complete.
 532          *
 533          * If this is a read-only mount the UNDO information is retained
 534          * in memory in the form of dirty buffer cache buffers, and not
 535          * written back to the media.
 536          */
 537         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 538               sizeof(hmp->blockmap));
 539
 540         /*
 541          * The undo_rec_limit limits the size of flush groups to avoid
 542          * blowing out the UNDO FIFO.  This calculation is typically in
 543          * the tens of thousands and is designed primarily when small
 544          * HAMMER filesystems are created.
 545          */
 546         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 547         if (hammer_debug_general & 0x0001)
 548                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 549
 550         error = hammer_recover(hmp, rootvol);
 551         if (error) {
 552                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 553                 goto done;
 554         }
 555
 556         /*
 557          * Finish setup now that we have a good root volume.
 558          *
 559          * The top 16 bits of fsid.val[1] is a pfs id.
 560          */
 561         ksnprintf(mp->mnt_stat.f_mntfromname,
 562                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 563                   rootvol->ondisk->vol_name);
 564         mp->mnt_stat.f_fsid.val[0] =
 565                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 566         mp->mnt_stat.f_fsid.val[1] =
 567                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 568         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 569
 570         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 571         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 572                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 573
 574         /*
 575          * Certain often-modified fields in the root volume are cached in
 576          * the hammer_mount structure so we do not have to generate lots
 577          * of little UNDO structures for them.
 578          *
 579          * Recopy after recovery.  This also has the side effect of
 580          * setting our cached undo FIFO's first_offset, which serves to
 581          * placemark the FIFO start for the NEXT flush cycle while the
 582          * on-disk first_offset represents the LAST flush cycle.
 583          */
 584         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 585         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 586               sizeof(hmp->blockmap));
 587         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 588
 589         hammer_flusher_create(hmp);
 590
 591         /*
 592          * Locate the root directory using the root cluster's B-Tree as a
 593          * starting point.  The root directory uses an obj_id of 1.
 594          *
 595          * FUTURE: Leave the root directory cached referenced but unlocked
 596          * in hmp->rootvp (need to flush it on unmount).
 597          */
 598         error = hammer_vfs_vget(mp, 1, &rootvp);
 599         if (error)
 600                 goto done;
 601         vput(rootvp);
 602         /*vn_unlock(hmp->rootvp);*/
 603
 604 done:
 605         hammer_rel_volume(rootvol, 0);
 606 failed:
 607         /*
 608          * Cleanup and return.
 609          */
 610         if (error)
 611                 hammer_free_hmp(mp);
 612         return (error);
 613 }
 614
 615 static int
 616 hammer_vfs_unmount(struct mount *mp, int mntflags)
 617 {
 618 #if 0
 619         struct hammer_mount *hmp = (void *)mp->mnt_data;
 620 #endif
 621         int flags;
 622         int error;
 623
 624         /*
 625          * Clean out the vnodes
 626          */
 627         flags = 0;
 628         if (mntflags & MNT_FORCE)
 629                 flags |= FORCECLOSE;
 630         if ((error = vflush(mp, 0, flags)) != 0)
 631                 return (error);
 632
 633         /*
 634          * Clean up the internal mount structure and related entities.  This
 635          * may issue I/O.
 636          */
 637         hammer_free_hmp(mp);
 638         return(0);
 639 }
 640
 641 /*
 642  * Clean up the internal mount structure and disassociate it from the mount.
 643  * This may issue I/O.
 644  */
 645 static void
 646 hammer_free_hmp(struct mount *mp)
 647 {
 648         struct hammer_mount *hmp = (void *)mp->mnt_data;
 649         hammer_flush_group_t flg;
 650         int count;
 651
 652         /*
 653          * Flush anything dirty.  This won't even run if the
 654          * filesystem errored-out.
 655          */
 656         count = 0;
 657         while (hammer_flusher_haswork(hmp)) {
 658                 hammer_flusher_sync(hmp);
 659                 ++count;
 660                 if (count >= 5) {
 661                         if (count == 5)
 662                                 kprintf("HAMMER: umount flushing.");
 663                         else
 664                                 kprintf(".");
 665                         tsleep(hmp, 0, "hmrufl", hz);
 666                 }
 667                 if (count == 30) {
 668                         kprintf("giving up\n");
 669                         break;
 670                 }
 671         }
 672         if (count >= 5 && count < 30)
 673                 kprintf("\n");
 674
 675         /*
 676          * If the mount had a critical error we have to destroy any
 677          * remaining inodes before we can finish cleaning up the flusher.
 678          */
 679         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 680                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 681                         hammer_destroy_inode_callback, NULL);
 682         }
 683
 684         /*
 685          * There shouldn't be any inodes left now and any left over
 686          * flush groups should now be empty.
 687          */
 688         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 689         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 690                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 691                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
 692                 if (flg->refs) {
 693                         kprintf("HAMMER: Warning, flush_group %p was "
 694                                 "not empty on umount!\n", flg);
 695                 }
 696                 kfree(flg, M_HAMMER);
 697         }
 698
 699         /*
 700          * We can finally destroy the flusher
 701          */
 702         hammer_flusher_destroy(hmp);
 703
 704         /*
 705          * We may have held recovered buffers due to a read-only mount.
 706          * These must be discarded.
 707          */
 708         if (hmp->ronly)
 709                 hammer_recover_flush_buffers(hmp, NULL, -1);
 710
 711         /*
 712          * Unload buffers and then volumes
 713          */
 714         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 715                 hammer_unload_buffer, NULL);
 716         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 717                 hammer_unload_volume, NULL);
 718
 719         mp->mnt_data = NULL;
 720         mp->mnt_flag &= ~MNT_LOCAL;
 721         hmp->mp = NULL;
 722         hammer_destroy_objid_cache(hmp);
 723         kfree(hmp, M_HAMMER);
 724 }
 725
 726 /*
 727  * Report critical errors.  ip may be NULL.
 728  */
 729 void
 730 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 731                       int error, const char *msg)
 732 {
 733         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 734         krateprintf(&hmp->krate,
 735                 "HAMMER(%s): Critical error inode=%lld %s\n",
 736                 hmp->mp->mnt_stat.f_mntfromname,
 737                 (ip ? ip->obj_id : -1), msg);
 738         if (hmp->ronly == 0) {
 739                 hmp->ronly = 2;         /* special errored read-only mode */
 740                 hmp->mp->mnt_flag |= MNT_RDONLY;
 741                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 742                         hmp->mp->mnt_stat.f_mntfromname);
 743         }
 744         hmp->error = error;
 745 }
 746
 747
 748 /*
 749  * Obtain a vnode for the specified inode number.  An exclusively locked
 750  * vnode is returned.
 751  */
 752 int
 753 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 754 {
 755         struct hammer_transaction trans;
 756         struct hammer_mount *hmp = (void *)mp->mnt_data;
 757         struct hammer_inode *ip;
 758         int error;
 759
 760         hammer_simple_transaction(&trans, hmp);
 761
 762         /*
 763          * Lookup the requested HAMMER inode.  The structure must be
 764          * left unlocked while we manipulate the related vnode to avoid
 765          * a deadlock.
 766          */
 767         ip = hammer_get_inode(&trans, NULL, ino,
 768                               hmp->asof, HAMMER_DEF_LOCALIZATION,
 769                               0, &error);
 770         if (ip == NULL) {
 771                 *vpp = NULL;
 772                 hammer_done_transaction(&trans);
 773                 return(error);
 774         }
 775         error = hammer_get_vnode(ip, vpp);
 776         hammer_rel_inode(ip, 0);
 777         hammer_done_transaction(&trans);
 778         return (error);
 779 }
 780
 781 /*
 782  * Return the root vnode for the filesystem.
 783  *
 784  * HAMMER stores the root vnode in the hammer_mount structure so
 785  * getting it is easy.
 786  */
 787 static int
 788 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 789 {
 790 #if 0
 791         struct hammer_mount *hmp = (void *)mp->mnt_data;
 792 #endif
 793         int error;
 794
 795         error = hammer_vfs_vget(mp, 1, vpp);
 796         return (error);
 797 }
 798
 799 static int
 800 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 801 {
 802         struct hammer_mount *hmp = (void *)mp->mnt_data;
 803         hammer_volume_t volume;
 804         hammer_volume_ondisk_t ondisk;
 805         int error;
 806         int64_t bfree;
 807
 808         volume = hammer_get_root_volume(hmp, &error);
 809         if (error)
 810                 return(error);
 811         ondisk = volume->ondisk;
 812
 813         /*
 814          * Basic stats
 815          */
 816         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 817         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 818         hammer_rel_volume(volume, 0);
 819
 820         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
 821         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 822         if (mp->mnt_stat.f_files < 0)
 823                 mp->mnt_stat.f_files = 0;
 824
 825         *sbp = mp->mnt_stat;
 826         return(0);
 827 }
 828
 829 static int
 830 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 831 {
 832         struct hammer_mount *hmp = (void *)mp->mnt_data;
 833         hammer_volume_t volume;
 834         hammer_volume_ondisk_t ondisk;
 835         int error;
 836         int64_t bfree;
 837
 838         volume = hammer_get_root_volume(hmp, &error);
 839         if (error)
 840                 return(error);
 841         ondisk = volume->ondisk;
 842
 843         /*
 844          * Basic stats
 845          */
 846         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 847         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 848         hammer_rel_volume(volume, 0);
 849
 850         mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
 851         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 852         if (mp->mnt_vstat.f_files < 0)
 853                 mp->mnt_vstat.f_files = 0;
 854         *sbp = mp->mnt_vstat;
 855         return(0);
 856 }
 857
 858 /*
 859  * Sync the filesystem.  Currently we have to run it twice, the second
 860  * one will advance the undo start index to the end index, so if a crash
 861  * occurs no undos will be run on mount.
 862  *
 863  * We do not sync the filesystem if we are called from a panic.  If we did
 864  * we might end up blowing up a sync that was already in progress.
 865  */
 866 static int
 867 hammer_vfs_sync(struct mount *mp, int waitfor)
 868 {
 869         struct hammer_mount *hmp = (void *)mp->mnt_data;
 870         int error;
 871
 872         if (panicstr == NULL) {
 873                 error = hammer_sync_hmp(hmp, waitfor);
 874                 if (error == 0)
 875                         error = hammer_sync_hmp(hmp, waitfor);
 876         } else {
 877                 error = EIO;
 878         }
 879         return (error);
 880 }
 881
 882 /*
 883  * Convert a vnode to a file handle.
 884  */
 885 static int
 886 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 887 {
 888         hammer_inode_t ip;
 889
 890         KKASSERT(MAXFIDSZ >= 16);
 891         ip = VTOI(vp);
 892         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 893         fhp->fid_ext = ip->obj_localization >> 16;
 894         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 895         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 896         return(0);
 897 }
 898
 899
 900 /*
 901  * Convert a file handle back to a vnode.
 902  */
 903 static int
 904 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
 905 {
 906         struct hammer_transaction trans;
 907         struct hammer_inode *ip;
 908         struct hammer_inode_info info;
 909         int error;
 910         u_int32_t localization;
 911
 912         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 913         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 914         localization = (u_int32_t)fhp->fid_ext << 16;
 915
 916         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 917
 918         /*
 919          * Get/allocate the hammer_inode structure.  The structure must be
 920          * unlocked while we manipulate the related vnode to avoid a
 921          * deadlock.
 922          */
 923         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 924                               info.obj_asof, localization, 0, &error);
 925         if (ip == NULL) {
 926                 *vpp = NULL;
 927                 return(error);
 928         }
 929         error = hammer_get_vnode(ip, vpp);
 930         hammer_rel_inode(ip, 0);
 931         hammer_done_transaction(&trans);
 932         return (error);
 933 }
 934
 935 static int
 936 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 937                     int *exflagsp, struct ucred **credanonp)
 938 {
 939         hammer_mount_t hmp = (void *)mp->mnt_data;
 940         struct netcred *np;
 941         int error;
 942
 943         np = vfs_export_lookup(mp, &hmp->export, nam);
 944         if (np) {
 945                 *exflagsp = np->netc_exflags;
 946                 *credanonp = &np->netc_anon;
 947                 error = 0;
 948         } else {
 949                 error = EACCES;
 950         }
 951         return (error);
 952
 953 }
 954
 955 int
 956 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
 957 {
 958         hammer_mount_t hmp = (void *)mp->mnt_data;
 959         int error;
 960
 961         switch(op) {
 962         case MOUNTCTL_SET_EXPORT:
 963                 error = vfs_export(mp, &hmp->export, export);
 964                 break;
 965         default:
 966                 error = EOPNOTSUPP;
 967                 break;
 968         }
 969         return(error);
 970 }
 971