sys/vfs/hammer2/hammer2_vfsops.c

   1 /*-
   2  * Copyright (c) 2011-2013 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  * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  *
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in
  16  *    the documentation and/or other materials provided with the
  17  *    distribution.
  18  * 3. Neither the name of The DragonFly Project nor the names of its
  19  *    contributors may be used to endorse or promote products derived
  20  *    from this software without specific, prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  25  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  26  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  27  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  28  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  29  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  30  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  31  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  32  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  33  * SUCH DAMAGE.
  34  */
  35 #include <sys/param.h>
  36 #include <sys/systm.h>
  37 #include <sys/kernel.h>
  38 #include <sys/nlookup.h>
  39 #include <sys/vnode.h>
  40 #include <sys/mount.h>
  41 #include <sys/fcntl.h>
  42 #include <sys/buf.h>
  43 #include <sys/uuid.h>
  44 #include <sys/vfsops.h>
  45 #include <sys/sysctl.h>
  46 #include <sys/socket.h>
  47 #include <sys/objcache.h>
  48
  49 #include <sys/proc.h>
  50 #include <sys/namei.h>
  51 #include <sys/mountctl.h>
  52 #include <sys/dirent.h>
  53 #include <sys/uio.h>
  54
  55 #include <sys/mutex.h>
  56 #include <sys/mutex2.h>
  57
  58 #include "hammer2.h"
  59 #include "hammer2_disk.h"
  60 #include "hammer2_mount.h"
  61
  62 #include "hammer2.h"
  63 #include "hammer2_lz4.h"
  64
  65 #include "zlib/hammer2_zlib.h"
  66
  67 #define REPORT_REFS_ERRORS 1    /* XXX remove me */
  68
  69 MALLOC_DEFINE(M_OBJCACHE, "objcache", "Object Cache");
  70
  71 struct hammer2_sync_info {
  72         hammer2_trans_t trans;
  73         int error;
  74         int waitfor;
  75 };
  76
  77 TAILQ_HEAD(hammer2_mntlist, hammer2_mount);
  78 static struct hammer2_mntlist hammer2_mntlist;
  79 static struct lock hammer2_mntlk;
  80
  81 int hammer2_debug;
  82 int hammer2_cluster_enable = 1;
  83 int hammer2_hardlink_enable = 1;
  84 int hammer2_flush_pipe = 100;
  85 long hammer2_limit_dirty_chains;
  86 long hammer2_iod_file_read;
  87 long hammer2_iod_meta_read;
  88 long hammer2_iod_indr_read;
  89 long hammer2_iod_fmap_read;
  90 long hammer2_iod_volu_read;
  91 long hammer2_iod_file_write;
  92 long hammer2_iod_meta_write;
  93 long hammer2_iod_indr_write;
  94 long hammer2_iod_fmap_write;
  95 long hammer2_iod_volu_write;
  96 long hammer2_ioa_file_read;
  97 long hammer2_ioa_meta_read;
  98 long hammer2_ioa_indr_read;
  99 long hammer2_ioa_fmap_read;
 100 long hammer2_ioa_volu_read;
 101 long hammer2_ioa_fmap_write;
 102 long hammer2_ioa_file_write;
 103 long hammer2_ioa_meta_write;
 104 long hammer2_ioa_indr_write;
 105 long hammer2_ioa_volu_write;
 106
 107 MALLOC_DECLARE(C_BUFFER);
 108 MALLOC_DEFINE(C_BUFFER, "compbuffer", "Buffer used for compression.");
 109
 110 MALLOC_DECLARE(D_BUFFER);
 111 MALLOC_DEFINE(D_BUFFER, "decompbuffer", "Buffer used for decompression.");
 112
 113 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
 114
 115 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
 116            &hammer2_debug, 0, "");
 117 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
 118            &hammer2_cluster_enable, 0, "");
 119 SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
 120            &hammer2_hardlink_enable, 0, "");
 121 SYSCTL_INT(_vfs_hammer2, OID_AUTO, flush_pipe, CTLFLAG_RW,
 122            &hammer2_flush_pipe, 0, "");
 123 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, limit_dirty_chains, CTLFLAG_RW,
 124            &hammer2_limit_dirty_chains, 0, "");
 125
 126 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
 127            &hammer2_iod_file_read, 0, "");
 128 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
 129            &hammer2_iod_meta_read, 0, "");
 130 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
 131            &hammer2_iod_indr_read, 0, "");
 132 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_read, CTLFLAG_RW,
 133            &hammer2_iod_fmap_read, 0, "");
 134 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_read, CTLFLAG_RW,
 135            &hammer2_iod_volu_read, 0, "");
 136
 137 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
 138            &hammer2_iod_file_write, 0, "");
 139 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
 140            &hammer2_iod_meta_write, 0, "");
 141 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
 142            &hammer2_iod_indr_write, 0, "");
 143 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_write, CTLFLAG_RW,
 144            &hammer2_iod_fmap_write, 0, "");
 145 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
 146            &hammer2_iod_volu_write, 0, "");
 147
 148 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
 149            &hammer2_ioa_file_read, 0, "");
 150 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
 151            &hammer2_ioa_meta_read, 0, "");
 152 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
 153            &hammer2_ioa_indr_read, 0, "");
 154 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_read, CTLFLAG_RW,
 155            &hammer2_ioa_fmap_read, 0, "");
 156 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_read, CTLFLAG_RW,
 157            &hammer2_ioa_volu_read, 0, "");
 158
 159 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
 160            &hammer2_ioa_file_write, 0, "");
 161 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
 162            &hammer2_ioa_meta_write, 0, "");
 163 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
 164            &hammer2_ioa_indr_write, 0, "");
 165 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_write, CTLFLAG_RW,
 166            &hammer2_ioa_fmap_write, 0, "");
 167 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
 168            &hammer2_ioa_volu_write, 0, "");
 169
 170 static int hammer2_vfs_init(struct vfsconf *conf);
 171 static int hammer2_vfs_uninit(struct vfsconf *vfsp);
 172 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
 173                                 struct ucred *cred);
 174 static int hammer2_remount(hammer2_mount_t *, struct mount *, char *,
 175                                 struct vnode *, struct ucred *);
 176 static int hammer2_recovery(hammer2_mount_t *hmp);
 177 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
 178 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
 179 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
 180                                 struct ucred *cred);
 181 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 182                                 struct ucred *cred);
 183 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
 184                                 ino_t ino, struct vnode **vpp);
 185 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 186                                 struct fid *fhp, struct vnode **vpp);
 187 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 188 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 189                                 int *exflagsp, struct ucred **credanonp);
 190
 191 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
 192 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
 193
 194 static void hammer2_write_thread(void *arg);
 195
 196 static void hammer2_vfs_unmount_hmp1(struct mount *mp, hammer2_mount_t *hmp);
 197 static void hammer2_vfs_unmount_hmp2(struct mount *mp, hammer2_mount_t *hmp);
 198
 199 /*
 200  * Functions for compression in threads,
 201  * from hammer2_vnops.c
 202  */
 203 static void hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
 204                                 hammer2_inode_t *ip,
 205                                 hammer2_inode_data_t *ipdata,
 206                                 hammer2_chain_t **parentp,
 207                                 hammer2_key_t lbase, int ioflag, int pblksize,
 208                                 int *errorp);
 209 static void hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
 210                                 hammer2_inode_t *ip,
 211                                 hammer2_inode_data_t *ipdata,
 212                                 hammer2_chain_t **parentp,
 213                                 hammer2_key_t lbase, int ioflag,
 214                                 int pblksize, int *errorp, int comp_algo);
 215 static void hammer2_zero_check_and_write(struct buf *bp,
 216                                 hammer2_trans_t *trans, hammer2_inode_t *ip,
 217                                 hammer2_inode_data_t *ipdata,
 218                                 hammer2_chain_t **parentp,
 219                                 hammer2_key_t lbase,
 220                                 int ioflag, int pblksize, int *errorp);
 221 static int test_block_zeros(const char *buf, size_t bytes);
 222 static void zero_write(struct buf *bp, hammer2_trans_t *trans,
 223                                 hammer2_inode_t *ip,
 224                                 hammer2_inode_data_t *ipdata,
 225                                 hammer2_chain_t **parentp,
 226                                 hammer2_key_t lbase,
 227                                 int *errorp);
 228 static void hammer2_write_bp(hammer2_chain_t *chain, struct buf *bp,
 229                                 int ioflag, int pblksize, int *errorp);
 230
 231 static int hammer2_rcvdmsg(kdmsg_msg_t *msg);
 232 static void hammer2_autodmsg(kdmsg_msg_t *msg);
 233
 234
 235 /*
 236  * HAMMER2 vfs operations.
 237  */
 238 static struct vfsops hammer2_vfsops = {
 239         .vfs_init       = hammer2_vfs_init,
 240         .vfs_uninit = hammer2_vfs_uninit,
 241         .vfs_sync       = hammer2_vfs_sync,
 242         .vfs_mount      = hammer2_vfs_mount,
 243         .vfs_unmount    = hammer2_vfs_unmount,
 244         .vfs_root       = hammer2_vfs_root,
 245         .vfs_statfs     = hammer2_vfs_statfs,
 246         .vfs_statvfs    = hammer2_vfs_statvfs,
 247         .vfs_vget       = hammer2_vfs_vget,
 248         .vfs_vptofh     = hammer2_vfs_vptofh,
 249         .vfs_fhtovp     = hammer2_vfs_fhtovp,
 250         .vfs_checkexp   = hammer2_vfs_checkexp
 251 };
 252
 253 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
 254
 255 VFS_SET(hammer2_vfsops, hammer2, 0);
 256 MODULE_VERSION(hammer2, 1);
 257
 258 static
 259 int
 260 hammer2_vfs_init(struct vfsconf *conf)
 261 {
 262         static struct objcache_malloc_args margs_read;
 263         static struct objcache_malloc_args margs_write;
 264
 265         int error;
 266
 267         error = 0;
 268
 269         if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
 270                 error = EINVAL;
 271         if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
 272                 error = EINVAL;
 273         if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
 274                 error = EINVAL;
 275
 276         if (error)
 277                 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
 278
 279         margs_read.objsize = 65536;
 280         margs_read.mtype = D_BUFFER;
 281
 282         margs_write.objsize = 32768;
 283         margs_write.mtype = C_BUFFER;
 284
 285         cache_buffer_read = objcache_create(margs_read.mtype->ks_shortdesc,
 286                                 0, 1, NULL, NULL, NULL, objcache_malloc_alloc,
 287                                 objcache_malloc_free, &margs_read);
 288         cache_buffer_write = objcache_create(margs_write.mtype->ks_shortdesc,
 289                                 0, 1, NULL, NULL, NULL, objcache_malloc_alloc,
 290                                 objcache_malloc_free, &margs_write);
 291
 292         lockinit(&hammer2_mntlk, "mntlk", 0, 0);
 293         TAILQ_INIT(&hammer2_mntlist);
 294
 295         hammer2_limit_dirty_chains = desiredvnodes / 10;
 296
 297         return (error);
 298 }
 299
 300 static
 301 int
 302 hammer2_vfs_uninit(struct vfsconf *vfsp __unused)
 303 {
 304         objcache_destroy(cache_buffer_read);
 305         objcache_destroy(cache_buffer_write);
 306         return 0;
 307 }
 308
 309 /*
 310  * Mount or remount HAMMER2 fileystem from physical media
 311  *
 312  *      mountroot
 313  *              mp              mount point structure
 314  *              path            NULL
 315  *              data            <unused>
 316  *              cred            <unused>
 317  *
 318  *      mount
 319  *              mp              mount point structure
 320  *              path            path to mount point
 321  *              data            pointer to argument structure in user space
 322  *                      volume  volume path (device@LABEL form)
 323  *                      hflags  user mount flags
 324  *              cred            user credentials
 325  *
 326  * RETURNS:     0       Success
 327  *              !0      error number
 328  */
 329 static
 330 int
 331 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
 332                   struct ucred *cred)
 333 {
 334         struct hammer2_mount_info info;
 335         hammer2_pfsmount_t *pmp;
 336         hammer2_mount_t *hmp;
 337         hammer2_key_t key_next;
 338         hammer2_key_t key_dummy;
 339         hammer2_key_t lhc;
 340         struct vnode *devvp;
 341         struct nlookupdata nd;
 342         hammer2_chain_t *parent;
 343         hammer2_chain_t *schain;
 344         hammer2_chain_t *rchain;
 345         struct file *fp;
 346         char devstr[MNAMELEN];
 347         size_t size;
 348         size_t done;
 349         char *dev;
 350         char *label;
 351         int ronly = 1;
 352         int error;
 353         int cache_index;
 354         int i;
 355
 356         hmp = NULL;
 357         pmp = NULL;
 358         dev = NULL;
 359         label = NULL;
 360         devvp = NULL;
 361         cache_index = -1;
 362
 363         kprintf("hammer2_mount\n");
 364
 365         if (path == NULL) {
 366                 /*
 367                  * Root mount
 368                  */
 369                 bzero(&info, sizeof(info));
 370                 info.cluster_fd = -1;
 371                 return (EOPNOTSUPP);
 372         } else {
 373                 /*
 374                  * Non-root mount or updating a mount
 375                  */
 376                 error = copyin(data, &info, sizeof(info));
 377                 if (error)
 378                         return (error);
 379
 380                 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
 381                 if (error)
 382                         return (error);
 383
 384                 /* Extract device and label */
 385                 dev = devstr;
 386                 label = strchr(devstr, '@');
 387                 if (label == NULL ||
 388                     ((label + 1) - dev) > done) {
 389                         return (EINVAL);
 390                 }
 391                 *label = '\0';
 392                 label++;
 393                 if (*label == '\0')
 394                         return (EINVAL);
 395
 396                 if (mp->mnt_flag & MNT_UPDATE) {
 397                         /* Update mount */
 398                         /* HAMMER2 implements NFS export via mountctl */
 399                         pmp = MPTOPMP(mp);
 400                         for (i = 0; i < pmp->cluster.nchains; ++i) {
 401                                 hmp = pmp->cluster.chains[i]->hmp;
 402                                 devvp = hmp->devvp;
 403                                 error = hammer2_remount(hmp, mp, path,
 404                                                         devvp, cred);
 405                                 if (error)
 406                                         break;
 407                         }
 408                         return error;
 409                 }
 410         }
 411
 412         /*
 413          * PFS mount
 414          *
 415          * Lookup name and verify it refers to a block device.
 416          */
 417         error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
 418         if (error == 0)
 419                 error = nlookup(&nd);
 420         if (error == 0)
 421                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
 422         nlookup_done(&nd);
 423
 424         if (error == 0) {
 425                 if (vn_isdisk(devvp, &error))
 426                         error = vfs_mountedon(devvp);
 427         }
 428
 429         /*
 430          * Determine if the device has already been mounted.  After this
 431          * check hmp will be non-NULL if we are doing the second or more
 432          * hammer2 mounts from the same device.
 433          */
 434         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
 435         TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
 436                 if (hmp->devvp == devvp)
 437                         break;
 438         }
 439
 440         /*
 441          * Open the device if this isn't a secondary mount and construct
 442          * the H2 device mount (hmp).
 443          */
 444         if (hmp == NULL) {
 445                 if (error == 0 && vcount(devvp) > 0)
 446                         error = EBUSY;
 447
 448                 /*
 449                  * Now open the device
 450                  */
 451                 if (error == 0) {
 452                         ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 453                         vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
 454                         error = vinvalbuf(devvp, V_SAVE, 0, 0);
 455                         if (error == 0) {
 456                                 error = VOP_OPEN(devvp,
 457                                                  ronly ? FREAD : FREAD | FWRITE,
 458                                                  FSCRED, NULL);
 459                         }
 460                         vn_unlock(devvp);
 461                 }
 462                 if (error && devvp) {
 463                         vrele(devvp);
 464                         devvp = NULL;
 465                 }
 466                 if (error) {
 467                         lockmgr(&hammer2_mntlk, LK_RELEASE);
 468                         return error;
 469                 }
 470                 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
 471                 hmp->ronly = ronly;
 472                 hmp->devvp = devvp;
 473                 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
 474                 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
 475                 RB_INIT(&hmp->iotree);
 476
 477                 lockinit(&hmp->alloclk, "h2alloc", 0, 0);
 478                 lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
 479                 TAILQ_INIT(&hmp->transq);
 480
 481                 /*
 482                  * vchain setup. vchain.data is embedded.
 483                  * vchain.refs is initialized and will never drop to 0.
 484                  *
 485                  * NOTE! voldata is not yet loaded.
 486                  */
 487                 hmp->vchain.hmp = hmp;
 488                 hmp->vchain.refs = 1;
 489                 hmp->vchain.data = (void *)&hmp->voldata;
 490                 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
 491                 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
 492                 hmp->vchain.delete_tid = HAMMER2_MAX_TID;
 493
 494                 hammer2_chain_core_alloc(NULL, &hmp->vchain, NULL);
 495                 /* hmp->vchain.u.xxx is left NULL */
 496
 497                 /*
 498                  * fchain setup.  fchain.data is embedded.
 499                  * fchain.refs is initialized and will never drop to 0.
 500                  *
 501                  * The data is not used but needs to be initialized to
 502                  * pass assertion muster.  We use this chain primarily
 503                  * as a placeholder for the freemap's top-level RBTREE
 504                  * so it does not interfere with the volume's topology
 505                  * RBTREE.
 506                  */
 507                 hmp->fchain.hmp = hmp;
 508                 hmp->fchain.refs = 1;
 509                 hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
 510                 hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
 511                 hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
 512                 hmp->fchain.bref.methods =
 513                         HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
 514                         HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
 515                 hmp->fchain.delete_tid = HAMMER2_MAX_TID;
 516
 517                 hammer2_chain_core_alloc(NULL, &hmp->fchain, NULL);
 518                 /* hmp->fchain.u.xxx is left NULL */
 519
 520                 /*
 521                  * Install the volume header and initialize fields from
 522                  * voldata.
 523                  */
 524                 error = hammer2_install_volume_header(hmp);
 525                 if (error) {
 526                         ++hmp->pmp_count;
 527                         hammer2_vfs_unmount_hmp1(mp, hmp);
 528                         hammer2_vfs_unmount_hmp2(mp, hmp);
 529                         hammer2_vfs_unmount(mp, MNT_FORCE);
 530                         return error;
 531                 }
 532
 533                 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
 534                 hmp->vchain.modify_tid = hmp->voldata.mirror_tid;
 535                 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
 536                 hmp->fchain.modify_tid = hmp->voldata.freemap_tid;
 537
 538                 /*
 539                  * First locate the super-root inode, which is key 0
 540                  * relative to the volume header's blockset.
 541                  *
 542                  * Then locate the root inode by scanning the directory keyspace
 543                  * represented by the label.
 544                  */
 545                 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
 546                 schain = hammer2_chain_lookup(&parent, &key_dummy,
 547                                       HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY,
 548                                       &cache_index, 0);
 549                 hammer2_chain_lookup_done(parent);
 550                 if (schain == NULL) {
 551                         kprintf("hammer2_mount: invalid super-root\n");
 552                         ++hmp->pmp_count;
 553                         hammer2_vfs_unmount_hmp1(mp, hmp);
 554                         hammer2_vfs_unmount_hmp2(mp, hmp);
 555                         hammer2_vfs_unmount(mp, MNT_FORCE);
 556                         return EINVAL;
 557                 }
 558
 559                 /*
 560                  * NOTE: inode_get sucks up schain's lock.
 561                  */
 562                 atomic_set_int(&schain->flags, HAMMER2_CHAIN_PFSROOT);
 563                 hmp->sroot = hammer2_inode_get(NULL, NULL, schain);
 564                 hammer2_inode_ref(hmp->sroot);
 565                 hammer2_inode_unlock_ex(hmp->sroot, schain);
 566                 schain = NULL;
 567                 /* leave hmp->sroot with one ref */
 568
 569                 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
 570                         error = hammer2_recovery(hmp);
 571                         /* XXX do something with error */
 572                 }
 573         }
 574
 575         /*
 576          * Block device opened successfully, finish initializing the
 577          * mount structure.
 578          *
 579          * From this point on we have to call hammer2_unmount() on failure.
 580          */
 581         pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
 582
 583         kmalloc_create(&pmp->minode, "HAMMER2-inodes");
 584         kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
 585
 586         spin_init(&pmp->inum_spin);
 587         RB_INIT(&pmp->inum_tree);
 588
 589         kdmsg_iocom_init(&pmp->iocom, pmp,
 590                          KDMSG_IOCOMF_AUTOCONN |
 591                          KDMSG_IOCOMF_AUTOSPAN |
 592                          KDMSG_IOCOMF_AUTOCIRC,
 593                          pmp->mmsg, hammer2_rcvdmsg);
 594
 595         ccms_domain_init(&pmp->ccms_dom);
 596         ++hmp->pmp_count;
 597         lockmgr(&hammer2_mntlk, LK_RELEASE);
 598         kprintf("hammer2_mount hmp=%p pmp=%p pmpcnt=%d\n",
 599                 hmp, pmp, hmp->pmp_count);
 600
 601         mp->mnt_flag = MNT_LOCAL;
 602         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;   /* all entry pts are SMP */
 603         mp->mnt_kern_flag |= MNTK_THR_SYNC;     /* new vsyncscan semantics */
 604
 605         /*
 606          * required mount structure initializations
 607          */
 608         mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
 609         mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
 610
 611         mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
 612         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
 613
 614         /*
 615          * Optional fields
 616          */
 617         mp->mnt_iosize_max = MAXPHYS;
 618         mp->mnt_data = (qaddr_t)pmp;
 619         pmp->mp = mp;
 620
 621         /*
 622          * Lookup mount point under the media-localized super-root.
 623          */
 624         parent = hammer2_inode_lock_ex(hmp->sroot);
 625         lhc = hammer2_dirhash(label, strlen(label));
 626         rchain = hammer2_chain_lookup(&parent, &key_next,
 627                                       lhc, lhc + HAMMER2_DIRHASH_LOMASK,
 628                                       &cache_index, 0);
 629         while (rchain) {
 630                 if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
 631                     strcmp(label, rchain->data->ipdata.filename) == 0) {
 632                         break;
 633                 }
 634                 rchain = hammer2_chain_next(&parent, rchain, &key_next,
 635                                             key_next,
 636                                             lhc + HAMMER2_DIRHASH_LOMASK,
 637                                             &cache_index, 0);
 638         }
 639         hammer2_inode_unlock_ex(hmp->sroot, parent);
 640
 641         if (rchain == NULL) {
 642                 kprintf("hammer2_mount: PFS label not found\n");
 643                 hammer2_vfs_unmount_hmp1(mp, hmp);
 644                 hammer2_vfs_unmount_hmp2(mp, hmp);
 645                 hammer2_vfs_unmount(mp, MNT_FORCE);
 646                 return EINVAL;
 647         }
 648         if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
 649                 hammer2_chain_unlock(rchain);
 650                 kprintf("hammer2_mount: PFS label already mounted!\n");
 651                 hammer2_vfs_unmount_hmp1(mp, hmp);
 652                 hammer2_vfs_unmount_hmp2(mp, hmp);
 653                 hammer2_vfs_unmount(mp, MNT_FORCE);
 654                 return EBUSY;
 655         }
 656 #if 0
 657         if (rchain->flags & HAMMER2_CHAIN_RECYCLE) {
 658                 kprintf("hammer2_mount: PFS label currently recycling\n");
 659                 hammer2_vfs_unmount_hmp1(mp, hmp);
 660                 hammer2_vfs_unmount_hmp2(mp, hmp);
 661                 hammer2_vfs_unmount(mp, MNT_FORCE);
 662                 return EBUSY;
 663         }
 664 #endif
 665         /*
 666          * After this point hammer2_vfs_unmount() has visibility on hmp
 667          * and manual hmp1/hmp2 calls are not needed on fatal errors.
 668          */
 669
 670         atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
 671
 672         /*
 673          * NOTE: *_get() integrates chain's lock into the inode lock.
 674          */
 675         hammer2_chain_ref(rchain);              /* for pmp->rchain */
 676         pmp->cluster.nchains = 1;
 677         pmp->cluster.chains[0] = rchain;
 678         pmp->iroot = hammer2_inode_get(pmp, NULL, rchain);
 679         hammer2_inode_ref(pmp->iroot);          /* ref for pmp->iroot */
 680
 681         KKASSERT(rchain->pmp == NULL);          /* tracking pmp for rchain */
 682         rchain->pmp = pmp;
 683
 684         hammer2_inode_unlock_ex(pmp->iroot, rchain);
 685
 686         kprintf("iroot %p\n", pmp->iroot);
 687
 688         /*
 689          * The logical file buffer bio write thread handles things
 690          * like physical block assignment and compression.
 691          */
 692         mtx_init(&pmp->wthread_mtx);
 693         bioq_init(&pmp->wthread_bioq);
 694         pmp->wthread_destroy = 0;
 695         lwkt_create(hammer2_write_thread, pmp,
 696                     &pmp->wthread_td, NULL, 0, -1, "hwrite-%s", label);
 697
 698         /*
 699          * Ref the cluster management messaging descriptor.  The mount
 700          * program deals with the other end of the communications pipe.
 701          */
 702         fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
 703         if (fp == NULL) {
 704                 kprintf("hammer2_mount: bad cluster_fd!\n");
 705                 hammer2_vfs_unmount(mp, MNT_FORCE);
 706                 return EBADF;
 707         }
 708         hammer2_cluster_reconnect(pmp, fp);
 709
 710         /*
 711          * Finish setup
 712          */
 713         vfs_getnewfsid(mp);
 714         vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
 715         vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
 716         vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
 717
 718         copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
 719         bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
 720         bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
 721         copyinstr(path, mp->mnt_stat.f_mntonname,
 722                   sizeof(mp->mnt_stat.f_mntonname) - 1,
 723                   &size);
 724
 725         /*
 726          * Initial statfs to prime mnt_stat.
 727          */
 728         hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
 729
 730         return 0;
 731 }
 732
 733 /*
 734  * Handle bioq for strategy write
 735  */
 736 static
 737 void
 738 hammer2_write_thread(void *arg)
 739 {
 740         hammer2_pfsmount_t *pmp;
 741         struct bio *bio;
 742         struct buf *bp;
 743         hammer2_trans_t trans;
 744         struct vnode *vp;
 745         hammer2_inode_t *ip;
 746         hammer2_chain_t *parent;
 747         hammer2_chain_t **parentp;
 748         hammer2_inode_data_t *ipdata;
 749         hammer2_key_t lbase;
 750         int lblksize;
 751         int pblksize;
 752         int error;
 753
 754         pmp = arg;
 755
 756         mtx_lock(&pmp->wthread_mtx);
 757         while (pmp->wthread_destroy == 0) {
 758                 if (bioq_first(&pmp->wthread_bioq) == NULL) {
 759                         mtxsleep(&pmp->wthread_bioq, &pmp->wthread_mtx,
 760                                  0, "h2bioqw", 0);
 761                 }
 762                 parent = NULL;
 763                 parentp = &parent;
 764
 765                 hammer2_trans_init(&trans, pmp, NULL, HAMMER2_TRANS_BUFCACHE);
 766
 767                 while ((bio = bioq_takefirst(&pmp->wthread_bioq)) != NULL) {
 768                         /*
 769                          * dummy bio for synchronization.  The transaction
 770                          * must be reinitialized.
 771                          */
 772                         if (bio->bio_buf == NULL) {
 773                                 bio->bio_flags |= BIO_DONE;
 774                                 wakeup(bio);
 775                                 hammer2_trans_done(&trans);
 776                                 hammer2_trans_init(&trans, pmp, NULL,
 777                                                    HAMMER2_TRANS_BUFCACHE);
 778                                 continue;
 779                         }
 780
 781                         /*
 782                          * else normal bio processing
 783                          */
 784                         mtx_unlock(&pmp->wthread_mtx);
 785
 786                         hammer2_lwinprog_drop(pmp);
 787
 788                         error = 0;
 789                         bp = bio->bio_buf;
 790                         vp = bp->b_vp;
 791                         ip = VTOI(vp);
 792
 793                         /*
 794                          * Inode is modified, flush size and mtime changes
 795                          * to ensure that the file size remains consistent
 796                          * with the buffers being flushed.
 797                          */
 798                         parent = hammer2_inode_lock_ex(ip);
 799                         if (ip->flags & (HAMMER2_INODE_RESIZED |
 800                                          HAMMER2_INODE_MTIME)) {
 801                                 hammer2_inode_fsync(&trans, ip, parentp);
 802                         }
 803                         ipdata = hammer2_chain_modify_ip(&trans, ip,
 804                                                          parentp, 0);
 805                         lblksize = hammer2_calc_logical(ip, bio->bio_offset,
 806                                                         &lbase, NULL);
 807                         pblksize = hammer2_calc_physical(ip, lbase);
 808                         hammer2_write_file_core(bp, &trans, ip, ipdata,
 809                                                 parentp,
 810                                                 lbase, IO_ASYNC,
 811                                                 pblksize, &error);
 812                         hammer2_inode_unlock_ex(ip, parent);
 813                         if (error) {
 814                                 kprintf("hammer2: error in buffer write\n");
 815                                 bp->b_flags |= B_ERROR;
 816                                 bp->b_error = EIO;
 817                         }
 818                         biodone(bio);
 819                         mtx_lock(&pmp->wthread_mtx);
 820                 }
 821                 hammer2_trans_done(&trans);
 822         }
 823         pmp->wthread_destroy = -1;
 824         wakeup(&pmp->wthread_destroy);
 825
 826         mtx_unlock(&pmp->wthread_mtx);
 827 }
 828
 829 void
 830 hammer2_bioq_sync(hammer2_pfsmount_t *pmp)
 831 {
 832         struct bio sync_bio;
 833
 834         bzero(&sync_bio, sizeof(sync_bio));     /* dummy with no bio_buf */
 835         mtx_lock(&pmp->wthread_mtx);
 836         if (pmp->wthread_destroy == 0) {
 837                 if (TAILQ_EMPTY(&pmp->wthread_bioq.queue)) {
 838                        bioq_insert_tail(&pmp->wthread_bioq, &sync_bio);
 839                        wakeup(&pmp->wthread_bioq);
 840                 } else {
 841                        bioq_insert_tail(&pmp->wthread_bioq, &sync_bio);
 842                 }
 843                 while ((sync_bio.bio_flags & BIO_DONE) == 0)
 844                         mtxsleep(&sync_bio, &pmp->wthread_mtx, 0, "h2bioq", 0);
 845         }
 846         mtx_unlock(&pmp->wthread_mtx);
 847 }
 848
 849 /*
 850  * Return a chain suitable for I/O, creating the chain if necessary
 851  * and assigning its physical block.
 852  */
 853 static
 854 hammer2_chain_t *
 855 hammer2_assign_physical(hammer2_trans_t *trans,
 856                         hammer2_inode_t *ip, hammer2_chain_t **parentp,
 857                         hammer2_key_t lbase, int pblksize, int *errorp)
 858 {
 859         hammer2_chain_t *parent;
 860         hammer2_chain_t *chain;
 861         hammer2_off_t pbase;
 862         hammer2_key_t key_dummy;
 863         int pradix = hammer2_getradix(pblksize);
 864         int cache_index = -1;
 865
 866         /*
 867          * Locate the chain associated with lbase, return a locked chain.
 868          * However, do not instantiate any data reference (which utilizes a
 869          * device buffer) because we will be using direct IO via the
 870          * logical buffer cache buffer.
 871          */
 872         *errorp = 0;
 873         KKASSERT(pblksize >= HAMMER2_MIN_ALLOC);
 874 retry:
 875         parent = *parentp;
 876         hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS); /* extra lock */
 877         chain = hammer2_chain_lookup(&parent, &key_dummy,
 878                                      lbase, lbase,
 879                                      &cache_index, HAMMER2_LOOKUP_NODATA);
 880
 881         if (chain == NULL) {
 882                 /*
 883                  * We found a hole, create a new chain entry.
 884                  *
 885                  * NOTE: DATA chains are created without device backing
 886                  *       store (nor do we want any).
 887                  */
 888                 *errorp = hammer2_chain_create(trans, &parent, &chain,
 889                                                lbase, HAMMER2_PBUFRADIX,
 890                                                HAMMER2_BREF_TYPE_DATA,
 891                                                pblksize);
 892                 if (chain == NULL) {
 893                         hammer2_chain_lookup_done(parent);
 894                         panic("hammer2_chain_create: par=%p error=%d\n",
 895                                 parent, *errorp);
 896                         goto retry;
 897                 }
 898
 899                 pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
 900                 /*ip->delta_dcount += pblksize;*/
 901         } else {
 902                 switch (chain->bref.type) {
 903                 case HAMMER2_BREF_TYPE_INODE:
 904                         /*
 905                          * The data is embedded in the inode.  The
 906                          * caller is responsible for marking the inode
 907                          * modified and copying the data to the embedded
 908                          * area.
 909                          */
 910                         pbase = NOOFFSET;
 911                         break;
 912                 case HAMMER2_BREF_TYPE_DATA:
 913                         if (chain->bytes != pblksize) {
 914                                 hammer2_chain_resize(trans, ip,
 915                                                      parent, &chain,
 916                                                      pradix,
 917                                                      HAMMER2_MODIFY_OPTDATA);
 918                         }
 919                         hammer2_chain_modify(trans, &chain,
 920                                              HAMMER2_MODIFY_OPTDATA);
 921                         pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
 922                         break;
 923                 default:
 924                         panic("hammer2_assign_physical: bad type");
 925                         /* NOT REACHED */
 926                         pbase = NOOFFSET;
 927                         break;
 928                 }
 929         }
 930
 931         /*
 932          * Cleanup.  If chain wound up being the inode (i.e. DIRECTDATA),
 933          * we might have to replace *parentp.
 934          */
 935         hammer2_chain_lookup_done(parent);
 936         if (chain) {
 937                 if (*parentp != chain &&
 938                     (*parentp)->core == chain->core) {
 939                         parent = *parentp;
 940                         *parentp = chain;               /* eats lock */
 941                         hammer2_chain_unlock(parent);
 942                         hammer2_chain_lock(chain, 0);   /* need another */
 943                 }
 944                 /* else chain already locked for return */
 945         }
 946         return (chain);
 947 }
 948
 949 /*
 950  * From hammer2_vnops.c.
 951  * The core write function which determines which path to take
 952  * depending on compression settings.
 953  */
 954 static
 955 void
 956 hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
 957                         hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
 958                         hammer2_chain_t **parentp,
 959                         hammer2_key_t lbase, int ioflag, int pblksize,
 960                         int *errorp)
 961 {
 962         hammer2_chain_t *chain;
 963
 964         switch(HAMMER2_DEC_COMP(ipdata->comp_algo)) {
 965         case HAMMER2_COMP_NONE:
 966                 /*
 967                  * We have to assign physical storage to the buffer
 968                  * we intend to dirty or write now to avoid deadlocks
 969                  * in the strategy code later.
 970                  *
 971                  * This can return NOOFFSET for inode-embedded data.
 972                  * The strategy code will take care of it in that case.
 973                  */
 974                 chain = hammer2_assign_physical(trans, ip, parentp,
 975                                                 lbase, pblksize,
 976                                                 errorp);
 977                 hammer2_write_bp(chain, bp, ioflag, pblksize, errorp);
 978                 if (chain)
 979                         hammer2_chain_unlock(chain);
 980                 break;
 981         case HAMMER2_COMP_AUTOZERO:
 982                 /*
 983                  * Check for zero-fill only
 984                  */
 985                 hammer2_zero_check_and_write(bp, trans, ip,
 986                                     ipdata, parentp, lbase,
 987                                     ioflag, pblksize, errorp);
 988                 break;
 989         case HAMMER2_COMP_LZ4:
 990         case HAMMER2_COMP_ZLIB:
 991         default:
 992                 /*
 993                  * Check for zero-fill and attempt compression.
 994                  */
 995                 hammer2_compress_and_write(bp, trans, ip,
 996                                            ipdata, parentp,
 997                                            lbase, ioflag,
 998                                            pblksize, errorp,
 999                                            ipdata->comp_algo);
1000                 break;
1001         }
1002         ipdata = &ip->chain->data->ipdata;      /* reload */
1003 }
1004
1005 /*
1006  * From hammer2_vnops.c
1007  * Generic function that will perform the compression in compression
1008  * write path. The compression algorithm is determined by the settings
1009  * obtained from inode.
1010  */
1011 static
1012 void
1013 hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
1014         hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
1015         hammer2_chain_t **parentp,
1016         hammer2_key_t lbase, int ioflag, int pblksize,
1017         int *errorp, int comp_algo)
1018 {
1019         hammer2_chain_t *chain;
1020         int comp_size;
1021         int comp_block_size;
1022         char *comp_buffer;
1023
1024         if (test_block_zeros(bp->b_data, pblksize)) {
1025                 zero_write(bp, trans, ip, ipdata, parentp, lbase, errorp);
1026                 return;
1027         }
1028
1029         comp_size = 0;
1030         comp_buffer = NULL;
1031
1032         KKASSERT(pblksize / 2 <= 32768);
1033
1034         if (ip->comp_heuristic < 8 || (ip->comp_heuristic & 7) == 0) {
1035                 z_stream strm_compress;
1036                 int comp_level;
1037                 int ret;
1038
1039                 switch(HAMMER2_DEC_COMP(comp_algo)) {
1040                 case HAMMER2_COMP_LZ4:
1041                         comp_buffer = objcache_get(cache_buffer_write,
1042                                                    M_INTWAIT);
1043                         comp_size = LZ4_compress_limitedOutput(
1044                                         bp->b_data,
1045                                         &comp_buffer[sizeof(int)],
1046                                         pblksize,
1047                                         pblksize / 2 - sizeof(int));
1048                         /*
1049                          * We need to prefix with the size, LZ4
1050                          * doesn't do it for us.  Add the related
1051                          * overhead.
1052                          */
1053                         *(int *)comp_buffer = comp_size;
1054                         if (comp_size)
1055                                 comp_size += sizeof(int);
1056                         break;
1057                 case HAMMER2_COMP_ZLIB:
1058                         comp_level = HAMMER2_DEC_LEVEL(comp_algo);
1059                         if (comp_level == 0)
1060                                 comp_level = 6; /* default zlib compression */
1061                         else if (comp_level < 6)
1062                                 comp_level = 6;
1063                         else if (comp_level > 9)
1064                                 comp_level = 9;
1065                         ret = deflateInit(&strm_compress, comp_level);
1066                         if (ret != Z_OK) {
1067                                 kprintf("HAMMER2 ZLIB: fatal error "
1068                                         "on deflateInit.\n");
1069                         }
1070
1071                         comp_buffer = objcache_get(cache_buffer_write,
1072                                                    M_INTWAIT);
1073                         strm_compress.next_in = bp->b_data;
1074                         strm_compress.avail_in = pblksize;
1075                         strm_compress.next_out = comp_buffer;
1076                         strm_compress.avail_out = pblksize / 2;
1077                         ret = deflate(&strm_compress, Z_FINISH);
1078                         if (ret == Z_STREAM_END) {
1079                                 comp_size = pblksize / 2 -
1080                                             strm_compress.avail_out;
1081                         } else {
1082                                 comp_size = 0;
1083                         }
1084                         ret = deflateEnd(&strm_compress);
1085                         break;
1086                 default:
1087                         kprintf("Error: Unknown compression method.\n");
1088                         kprintf("Comp_method = %d.\n", comp_algo);
1089                         break;
1090                 }
1091         }
1092
1093         if (comp_size == 0) {
1094                 /*
1095                  * compression failed or turned off
1096                  */
1097                 comp_block_size = pblksize;     /* safety */
1098                 if (++ip->comp_heuristic > 128)
1099                         ip->comp_heuristic = 8;
1100         } else {
1101                 /*
1102                  * compression succeeded
1103                  */
1104                 ip->comp_heuristic = 0;
1105                 if (comp_size <= 1024) {
1106                         comp_block_size = 1024;
1107                 } else if (comp_size <= 2048) {
1108                         comp_block_size = 2048;
1109                 } else if (comp_size <= 4096) {
1110                         comp_block_size = 4096;
1111                 } else if (comp_size <= 8192) {
1112                         comp_block_size = 8192;
1113                 } else if (comp_size <= 16384) {
1114                         comp_block_size = 16384;
1115                 } else if (comp_size <= 32768) {
1116                         comp_block_size = 32768;
1117                 } else {
1118                         panic("hammer2: WRITE PATH: "
1119                               "Weird comp_size value.");
1120                         /* NOT REACHED */
1121                         comp_block_size = pblksize;
1122                 }
1123         }
1124
1125         chain = hammer2_assign_physical(trans, ip, parentp,
1126                                         lbase, comp_block_size,
1127                                         errorp);
1128         ipdata = &ip->chain->data->ipdata;      /* RELOAD */
1129
1130         if (*errorp) {
1131                 kprintf("WRITE PATH: An error occurred while "
1132                         "assigning physical space.\n");
1133                 KKASSERT(chain == NULL);
1134         } else {
1135                 /* Get device offset */
1136                 hammer2_io_t *dio;
1137                 char *bdata;
1138                 int temp_check;
1139
1140                 KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1141
1142                 switch(chain->bref.type) {
1143                 case HAMMER2_BREF_TYPE_INODE:
1144                         KKASSERT(chain->data->ipdata.op_flags &
1145                                  HAMMER2_OPFLAG_DIRECTDATA);
1146                         KKASSERT(bp->b_loffset == 0);
1147                         bcopy(bp->b_data, chain->data->ipdata.u.data,
1148                               HAMMER2_EMBEDDED_BYTES);
1149                         break;
1150                 case HAMMER2_BREF_TYPE_DATA:
1151                         temp_check = HAMMER2_DEC_CHECK(chain->bref.methods);
1152
1153                         /*
1154                          * Optimize out the read-before-write
1155                          * if possible.
1156                          */
1157                         *errorp = hammer2_io_newnz(chain->hmp,
1158                                                    chain->bref.data_off,
1159                                                    chain->bytes,
1160                                                    &dio);
1161                         if (*errorp) {
1162                                 hammer2_io_brelse(&dio);
1163                                 kprintf("hammer2: WRITE PATH: "
1164                                         "dbp bread error\n");
1165                                 break;
1166                         }
1167                         bdata = hammer2_io_data(dio, chain->bref.data_off);
1168
1169                         /*
1170                          * When loading the block make sure we don't
1171                          * leave garbage after the compressed data.
1172                          */
1173                         if (comp_size) {
1174                                 chain->bref.methods =
1175                                         HAMMER2_ENC_COMP(comp_algo) +
1176                                         HAMMER2_ENC_CHECK(temp_check);
1177                                 bcopy(comp_buffer, bdata, comp_size);
1178                                 if (comp_size != comp_block_size) {
1179                                         bzero(bdata + comp_size,
1180                                               comp_block_size - comp_size);
1181                                 }
1182                         } else {
1183                                 chain->bref.methods =
1184                                         HAMMER2_ENC_COMP(
1185                                                 HAMMER2_COMP_NONE) +
1186                                         HAMMER2_ENC_CHECK(temp_check);
1187                                 bcopy(bp->b_data, bdata, pblksize);
1188                         }
1189
1190                         /*
1191                          * Device buffer is now valid, chain is no
1192                          * longer in the initial state.
1193                          */
1194                         atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
1195
1196                         /* Now write the related bdp. */
1197                         if (ioflag & IO_SYNC) {
1198                                 /*
1199                                  * Synchronous I/O requested.
1200                                  */
1201                                 hammer2_io_bwrite(&dio);
1202                         /*
1203                         } else if ((ioflag & IO_DIRECT) &&
1204                                    loff + n == pblksize) {
1205                                 hammer2_io_bdwrite(&dio);
1206                         */
1207                         } else if (ioflag & IO_ASYNC) {
1208                                 hammer2_io_bawrite(&dio);
1209                         } else {
1210                                 hammer2_io_bdwrite(&dio);
1211                         }
1212                         break;
1213                 default:
1214                         panic("hammer2_write_bp: bad chain type %d\n",
1215                                 chain->bref.type);
1216                         /* NOT REACHED */
1217                         break;
1218                 }
1219
1220                 hammer2_chain_unlock(chain);
1221         }
1222         if (comp_buffer)
1223                 objcache_put(cache_buffer_write, comp_buffer);
1224 }
1225
1226 /*
1227  * Function that performs zero-checking and writing without compression,
1228  * it corresponds to default zero-checking path.
1229  */
1230 static
1231 void
1232 hammer2_zero_check_and_write(struct buf *bp, hammer2_trans_t *trans,
1233         hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
1234         hammer2_chain_t **parentp,
1235         hammer2_key_t lbase, int ioflag, int pblksize, int *errorp)
1236 {
1237         hammer2_chain_t *chain;
1238
1239         if (test_block_zeros(bp->b_data, pblksize)) {
1240                 zero_write(bp, trans, ip, ipdata, parentp, lbase, errorp);
1241         } else {
1242                 chain = hammer2_assign_physical(trans, ip, parentp,
1243                                                 lbase, pblksize, errorp);
1244                 hammer2_write_bp(chain, bp, ioflag, pblksize, errorp);
1245                 if (chain)
1246                         hammer2_chain_unlock(chain);
1247         }
1248 }
1249
1250 /*
1251  * A function to test whether a block of data contains only zeros,
1252  * returns TRUE (non-zero) if the block is all zeros.
1253  */
1254 static
1255 int
1256 test_block_zeros(const char *buf, size_t bytes)
1257 {
1258         size_t i;
1259
1260         for (i = 0; i < bytes; i += sizeof(long)) {
1261                 if (*(const long *)(buf + i) != 0)
1262                         return (0);
1263         }
1264         return (1);
1265 }
1266
1267 /*
1268  * Function to "write" a block that contains only zeros.
1269  */
1270 static
1271 void
1272 zero_write(struct buf *bp, hammer2_trans_t *trans, hammer2_inode_t *ip,
1273         hammer2_inode_data_t *ipdata, hammer2_chain_t **parentp,
1274         hammer2_key_t lbase, int *errorp __unused)
1275 {
1276         hammer2_chain_t *parent;
1277         hammer2_chain_t *chain;
1278         hammer2_key_t key_dummy;
1279         int cache_index = -1;
1280
1281         parent = hammer2_chain_lookup_init(*parentp, 0);
1282
1283         chain = hammer2_chain_lookup(&parent, &key_dummy, lbase, lbase,
1284                                      &cache_index, HAMMER2_LOOKUP_NODATA);
1285         if (chain) {
1286                 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
1287                         bzero(chain->data->ipdata.u.data,
1288                               HAMMER2_EMBEDDED_BYTES);
1289                 } else {
1290                         hammer2_chain_delete(trans, chain, 0);
1291                 }
1292                 hammer2_chain_unlock(chain);
1293         }
1294         hammer2_chain_lookup_done(parent);
1295 }
1296
1297 /*
1298  * Function to write the data as it is, without performing any sort of
1299  * compression. This function is used in path without compression and
1300  * default zero-checking path.
1301  */
1302 static
1303 void
1304 hammer2_write_bp(hammer2_chain_t *chain, struct buf *bp, int ioflag,
1305                                 int pblksize, int *errorp)
1306 {
1307         hammer2_io_t *dio;
1308         char *bdata;
1309         int error;
1310         int temp_check = HAMMER2_DEC_CHECK(chain->bref.methods);
1311
1312         KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1313
1314         switch(chain->bref.type) {
1315         case HAMMER2_BREF_TYPE_INODE:
1316                 KKASSERT(chain->data->ipdata.op_flags &
1317                          HAMMER2_OPFLAG_DIRECTDATA);
1318                 KKASSERT(bp->b_loffset == 0);
1319                 bcopy(bp->b_data, chain->data->ipdata.u.data,
1320                       HAMMER2_EMBEDDED_BYTES);
1321                 error = 0;
1322                 break;
1323         case HAMMER2_BREF_TYPE_DATA:
1324                 error = hammer2_io_newnz(chain->hmp, chain->bref.data_off,
1325                                          chain->bytes, &dio);
1326                 if (error) {
1327                         hammer2_io_bqrelse(&dio);
1328                         kprintf("hammer2: WRITE PATH: dbp bread error\n");
1329                         break;
1330                 }
1331                 bdata = hammer2_io_data(dio, chain->bref.data_off);
1332
1333                 chain->bref.methods = HAMMER2_ENC_COMP(HAMMER2_COMP_NONE) +
1334                                       HAMMER2_ENC_CHECK(temp_check);
1335                 bcopy(bp->b_data, bdata, chain->bytes);
1336
1337                 /*
1338                  * Device buffer is now valid, chain is no
1339                  * longer in the initial state.
1340                  */
1341                 atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
1342
1343                 if (ioflag & IO_SYNC) {
1344                         /*
1345                          * Synchronous I/O requested.
1346                          */
1347                         hammer2_io_bwrite(&dio);
1348                 /*
1349                 } else if ((ioflag & IO_DIRECT) && loff + n == pblksize) {
1350                         hammer2_io_bdwrite(&dio);
1351                 */
1352                 } else if (ioflag & IO_ASYNC) {
1353                         hammer2_io_bawrite(&dio);
1354                 } else {
1355                         hammer2_io_bdwrite(&dio);
1356                 }
1357                 break;
1358         default:
1359                 panic("hammer2_write_bp: bad chain type %d\n",
1360                       chain->bref.type);
1361                 /* NOT REACHED */
1362                 error = 0;
1363                 break;
1364         }
1365         *errorp = error;
1366 }
1367
1368 static
1369 int
1370 hammer2_remount(hammer2_mount_t *hmp, struct mount *mp, char *path,
1371                 struct vnode *devvp, struct ucred *cred)
1372 {
1373         int error;
1374
1375         if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
1376                 error = hammer2_recovery(hmp);
1377         } else {
1378                 error = 0;
1379         }
1380         return error;
1381 }
1382
1383 static
1384 int
1385 hammer2_vfs_unmount(struct mount *mp, int mntflags)
1386 {
1387         hammer2_pfsmount_t *pmp;
1388         hammer2_mount_t *hmp;
1389         hammer2_chain_t *rchain;
1390         int flags;
1391         int error = 0;
1392         int i;
1393
1394         pmp = MPTOPMP(mp);
1395
1396         if (pmp == NULL)
1397                 return(0);
1398
1399         ccms_domain_uninit(&pmp->ccms_dom);
1400         kdmsg_iocom_uninit(&pmp->iocom);        /* XXX chain dependency */
1401
1402         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
1403
1404         /*
1405          * If mount initialization proceeded far enough we must flush
1406          * its vnodes.
1407          */
1408         if (mntflags & MNT_FORCE)
1409                 flags = FORCECLOSE;
1410         else
1411                 flags = 0;
1412         if (pmp->iroot) {
1413                 error = vflush(mp, 0, flags);
1414                 if (error)
1415                         goto failed;
1416         }
1417
1418         if (pmp->wthread_td) {
1419                 mtx_lock(&pmp->wthread_mtx);
1420                 pmp->wthread_destroy = 1;
1421                 wakeup(&pmp->wthread_bioq);
1422                 while (pmp->wthread_destroy != -1) {
1423                         mtxsleep(&pmp->wthread_destroy,
1424                                 &pmp->wthread_mtx, 0,
1425                                 "umount-sleep", 0);
1426                 }
1427                 mtx_unlock(&pmp->wthread_mtx);
1428                 pmp->wthread_td = NULL;
1429         }
1430
1431         for (i = 0; i < pmp->cluster.nchains; ++i) {
1432                 hmp = pmp->cluster.chains[i]->hmp;
1433
1434                 hammer2_vfs_unmount_hmp1(mp, hmp);
1435
1436                 /*
1437                  * Cleanup the root and super-root chain elements
1438                  * (which should be clean).
1439                  */
1440                 if (pmp->iroot) {
1441 #if REPORT_REFS_ERRORS
1442                         if (pmp->iroot->refs != 1)
1443                                 kprintf("PMP->IROOT %p REFS WRONG %d\n",
1444                                         pmp->iroot, pmp->iroot->refs);
1445 #else
1446                         KKASSERT(pmp->iroot->refs == 1);
1447 #endif
1448                         /* ref for pmp->iroot */
1449                         hammer2_inode_drop(pmp->iroot);
1450                         pmp->iroot = NULL;
1451                 }
1452
1453                 rchain = pmp->cluster.chains[i];
1454                 if (rchain) {
1455                         atomic_clear_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
1456 #if REPORT_REFS_ERRORS
1457                         if (rchain->refs != 1)
1458                                 kprintf("PMP->RCHAIN %p REFS WRONG %d\n",
1459                                         rchain, rchain->refs);
1460 #else
1461                         KKASSERT(rchain->refs == 1);
1462 #endif
1463                         hammer2_chain_drop(rchain);
1464                         pmp->cluster.chains[i] = NULL;
1465                 }
1466
1467                 hammer2_vfs_unmount_hmp2(mp, hmp);
1468         }
1469
1470         pmp->mp = NULL;
1471         mp->mnt_data = NULL;
1472
1473         kmalloc_destroy(&pmp->mmsg);
1474         kmalloc_destroy(&pmp->minode);
1475
1476         kfree(pmp, M_HAMMER2);
1477         error = 0;
1478
1479 failed:
1480         lockmgr(&hammer2_mntlk, LK_RELEASE);
1481
1482         return (error);
1483 }
1484
1485 static
1486 void
1487 hammer2_vfs_unmount_hmp1(struct mount *mp, hammer2_mount_t *hmp)
1488 {
1489         hammer2_mount_exlock(hmp);
1490         --hmp->pmp_count;
1491
1492         kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
1493
1494         /*
1495          * Flush any left over chains.  The voldata lock is only used
1496          * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
1497          *
1498          * Flush twice to ensure that the freemap is completely
1499          * synchronized.  If we only do it once the next mount's
1500          * recovery scan will have to do some fixups (which isn't
1501          * bad, but we don't want it to have to do it except when
1502          * recovering from a crash).
1503          */
1504         hammer2_voldata_lock(hmp);
1505         if (((hmp->vchain.flags | hmp->fchain.flags) &
1506              HAMMER2_CHAIN_MODIFIED) ||
1507             hmp->vchain.core->update_hi > hmp->voldata.mirror_tid ||
1508             hmp->fchain.core->update_hi > hmp->voldata.freemap_tid) {
1509                 hammer2_voldata_unlock(hmp, 0);
1510                 hammer2_vfs_sync(mp, MNT_WAIT);
1511                 /*hammer2_vfs_sync(mp, MNT_WAIT);*/
1512         } else {
1513                 hammer2_voldata_unlock(hmp, 0);
1514         }
1515         if (hmp->pmp_count == 0) {
1516                 if (((hmp->vchain.flags | hmp->fchain.flags) &
1517                      HAMMER2_CHAIN_MODIFIED) ||
1518                     (hmp->vchain.core->update_hi >
1519                      hmp->voldata.mirror_tid) ||
1520                     (hmp->fchain.core->update_hi >
1521                      hmp->voldata.freemap_tid)) {
1522                         kprintf("hammer2_unmount: chains left over "
1523                                 "after final sync\n");
1524                         kprintf("    vchain %08x update_hi %jx/%jx\n",
1525                                 hmp->vchain.flags,
1526                                 hmp->voldata.mirror_tid,
1527                                 hmp->vchain.core->update_hi);
1528                         kprintf("    fchain %08x update_hi %jx/%jx\n",
1529                                 hmp->fchain.flags,
1530                                 hmp->voldata.freemap_tid,
1531                                 hmp->fchain.core->update_hi);
1532
1533                         if (hammer2_debug & 0x0010)
1534                                 Debugger("entered debugger");
1535                 }
1536         }
1537 }
1538
1539 static
1540 void
1541 hammer2_vfs_unmount_hmp2(struct mount *mp, hammer2_mount_t *hmp)
1542 {
1543         struct vnode *devvp;
1544         int dumpcnt;
1545         int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
1546
1547         /*
1548          * If no PFS's left drop the master hammer2_mount for the
1549          * device.
1550          */
1551         if (hmp->pmp_count == 0) {
1552                 if (hmp->sroot) {
1553                         hammer2_inode_drop(hmp->sroot);
1554                         hmp->sroot = NULL;
1555                 }
1556
1557                 /*
1558                  * Finish up with the device vnode
1559                  */
1560                 if ((devvp = hmp->devvp) != NULL) {
1561                         vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1562                         vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
1563                         hmp->devvp = NULL;
1564                         VOP_CLOSE(devvp,
1565                                   (ronly ? FREAD : FREAD|FWRITE));
1566                         vn_unlock(devvp);
1567                         vrele(devvp);
1568                         devvp = NULL;
1569                 }
1570
1571                 /*
1572                  * Final drop of embedded freemap root chain to
1573                  * clean up fchain.core (fchain structure is not
1574                  * flagged ALLOCATED so it is cleaned out and then
1575                  * left to rot).
1576                  */
1577                 hammer2_chain_drop(&hmp->fchain);
1578
1579                 /*
1580                  * Final drop of embedded volume root chain to clean
1581                  * up vchain.core (vchain structure is not flagged
1582                  * ALLOCATED so it is cleaned out and then left to
1583                  * rot).
1584                  */
1585                 dumpcnt = 50;
1586                 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt);
1587                 dumpcnt = 50;
1588                 hammer2_dump_chain(&hmp->fchain, 0, &dumpcnt);
1589                 hammer2_mount_unlock(hmp);
1590                 hammer2_chain_drop(&hmp->vchain);
1591
1592                 hammer2_io_cleanup(hmp, &hmp->iotree);
1593                 if (hmp->iofree_count) {
1594                         kprintf("io_cleanup: %d I/O's left hanging\n",
1595                                 hmp->iofree_count);
1596                 }
1597
1598                 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
1599                 kmalloc_destroy(&hmp->mchain);
1600                 kfree(hmp, M_HAMMER2);
1601         } else {
1602                 hammer2_mount_unlock(hmp);
1603         }
1604 }
1605
1606 static
1607 int
1608 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
1609              ino_t ino, struct vnode **vpp)
1610 {
1611         kprintf("hammer2_vget\n");
1612         return (EOPNOTSUPP);
1613 }
1614
1615 static
1616 int
1617 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
1618 {
1619         hammer2_pfsmount_t *pmp;
1620         hammer2_chain_t *parent;
1621         int error;
1622         struct vnode *vp;
1623
1624         pmp = MPTOPMP(mp);
1625         if (pmp->iroot == NULL) {
1626                 *vpp = NULL;
1627                 error = EINVAL;
1628         } else {
1629                 parent = hammer2_inode_lock_sh(pmp->iroot);
1630                 vp = hammer2_igetv(pmp->iroot, &error);
1631                 hammer2_inode_unlock_sh(pmp->iroot, parent);
1632                 *vpp = vp;
1633                 if (vp == NULL)
1634                         kprintf("vnodefail\n");
1635         }
1636
1637         return (error);
1638 }
1639
1640 /*
1641  * Filesystem status
1642  *
1643  * XXX incorporate ipdata->inode_quota and data_quota
1644  */
1645 static
1646 int
1647 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1648 {
1649         hammer2_pfsmount_t *pmp;
1650         hammer2_mount_t *hmp;
1651
1652         pmp = MPTOPMP(mp);
1653         KKASSERT(pmp->cluster.nchains >= 1);
1654         hmp = pmp->cluster.chains[0]->hmp;      /* XXX */
1655
1656         mp->mnt_stat.f_files = pmp->inode_count;
1657         mp->mnt_stat.f_ffree = 0;
1658         mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
1659         mp->mnt_stat.f_bfree =  hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
1660         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1661
1662         *sbp = mp->mnt_stat;
1663         return (0);
1664 }
1665
1666 static
1667 int
1668 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1669 {
1670         hammer2_pfsmount_t *pmp;
1671         hammer2_mount_t *hmp;
1672
1673         pmp = MPTOPMP(mp);
1674         KKASSERT(pmp->cluster.nchains >= 1);
1675         hmp = pmp->cluster.chains[0]->hmp;      /* XXX */
1676
1677         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1678         mp->mnt_vstat.f_files = pmp->inode_count;
1679         mp->mnt_vstat.f_ffree = 0;
1680         mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
1681         mp->mnt_vstat.f_bfree =  hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
1682         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1683
1684         *sbp = mp->mnt_vstat;
1685         return (0);
1686 }
1687
1688 /*
1689  * Mount-time recovery (RW mounts)
1690  *
1691  * Updates to the free block table are allowed to lag flushes by one
1692  * transaction.  In case of a crash, then on a fresh mount we must do an
1693  * incremental scan of transaction id voldata.mirror_tid and make sure the
1694  * related blocks have been marked allocated.
1695  *
1696  */
1697 struct hammer2_recovery_elm {
1698         TAILQ_ENTRY(hammer2_recovery_elm) entry;
1699         hammer2_chain_t *chain;
1700 };
1701
1702 TAILQ_HEAD(hammer2_recovery_list, hammer2_recovery_elm);
1703
1704 static int hammer2_recovery_scan(hammer2_trans_t *trans, hammer2_mount_t *hmp,
1705                         hammer2_chain_t *parent,
1706                         struct hammer2_recovery_list *list, int depth);
1707
1708 #define HAMMER2_RECOVERY_MAXDEPTH       10
1709
1710 static
1711 int
1712 hammer2_recovery(hammer2_mount_t *hmp)
1713 {
1714         hammer2_trans_t trans;
1715         struct hammer2_recovery_list list;
1716         struct hammer2_recovery_elm *elm;
1717         hammer2_chain_t *parent;
1718         int error;
1719         int cumulative_error = 0;
1720
1721         hammer2_trans_init(&trans, NULL, hmp, 0);
1722
1723         TAILQ_INIT(&list);
1724         parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
1725         cumulative_error = hammer2_recovery_scan(&trans, hmp, parent, &list, 0);
1726         hammer2_chain_lookup_done(parent);
1727
1728         while ((elm = TAILQ_FIRST(&list)) != NULL) {
1729                 TAILQ_REMOVE(&list, elm, entry);
1730                 parent = elm->chain;
1731                 kfree(elm, M_HAMMER2);
1732
1733                 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS |
1734                                            HAMMER2_RESOLVE_NOREF);
1735                 error = hammer2_recovery_scan(&trans, hmp, parent, &list, 0);
1736                 hammer2_chain_unlock(parent);
1737                 if (error)
1738                         cumulative_error = error;
1739         }
1740         hammer2_trans_done(&trans);
1741
1742         return cumulative_error;
1743 }
1744
1745 static
1746 int
1747 hammer2_recovery_scan(hammer2_trans_t *trans, hammer2_mount_t *hmp,
1748                       hammer2_chain_t *parent,
1749                       struct hammer2_recovery_list *list, int depth)
1750 {
1751         hammer2_chain_t *chain;
1752         int cache_index;
1753         int cumulative_error = 0;
1754         int error;
1755
1756         /*
1757          * Defer operation if depth limit reached.
1758          */
1759         if (depth >= HAMMER2_RECOVERY_MAXDEPTH) {
1760                 struct hammer2_recovery_elm *elm;
1761
1762                 elm = kmalloc(sizeof(*elm), M_HAMMER2, M_ZERO | M_WAITOK);
1763                 elm->chain = parent;
1764                 hammer2_chain_ref(parent);
1765                 TAILQ_INSERT_TAIL(list, elm, entry);
1766                 /* unlocked by caller */
1767
1768                 return(0);
1769         }
1770
1771         /*
1772          * Adjust freemap to ensure that the block(s) are marked allocated.
1773          */
1774         if (parent->bref.type != HAMMER2_BREF_TYPE_VOLUME) {
1775                 hammer2_freemap_adjust(trans, hmp, &parent->bref,
1776                                        HAMMER2_FREEMAP_DORECOVER);
1777         }
1778
1779         /*
1780          * Check type for recursive scan
1781          */
1782         switch(parent->bref.type) {
1783         case HAMMER2_BREF_TYPE_VOLUME:
1784                 /* data already instantiated */
1785                 break;
1786         case HAMMER2_BREF_TYPE_INODE:
1787                 /*
1788                  * Must instantiate data for DIRECTDATA test and also
1789                  * for recursion.
1790                  */
1791                 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
1792                 hammer2_chain_unlock(parent);
1793                 if (parent->data->ipdata.op_flags & HAMMER2_OPFLAG_DIRECTDATA) {
1794                         /* not applicable to recovery scan */
1795                         return 0;
1796                 }
1797                 break;
1798         case HAMMER2_BREF_TYPE_INDIRECT:
1799                 /*
1800                  * Must instantiate data for recursion
1801                  */
1802                 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
1803                 hammer2_chain_unlock(parent);
1804                 break;
1805         case HAMMER2_BREF_TYPE_DATA:
1806         case HAMMER2_BREF_TYPE_FREEMAP:
1807         case HAMMER2_BREF_TYPE_FREEMAP_NODE:
1808         case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
1809                 /* not applicable to recovery scan */
1810                 return 0;
1811                 break;
1812         default:
1813                 return EDOM;
1814         }
1815
1816         /*
1817          * Recursive scan of the last flushed transaction only.  We are
1818          * doing this without pmp assignments so don't leave the chains
1819          * hanging around after we are done with them.
1820          */
1821         cache_index = 0;
1822         chain = hammer2_chain_scan(parent, NULL, &cache_index,
1823                                    HAMMER2_LOOKUP_NODATA);
1824         while (chain) {
1825                 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
1826                 if (chain->bref.mirror_tid >= hmp->voldata.mirror_tid) {
1827                         error = hammer2_recovery_scan(trans, hmp, chain,
1828                                                       list, depth + 1);
1829                         if (error)
1830                                 cumulative_error = error;
1831                 }
1832                 chain = hammer2_chain_scan(parent, chain, &cache_index,
1833                                            HAMMER2_LOOKUP_NODATA);
1834         }
1835
1836         return cumulative_error;
1837 }
1838
1839 /*
1840  * Sync the entire filesystem; this is called from the filesystem syncer
1841  * process periodically and whenever a user calls sync(1) on the hammer
1842  * mountpoint.
1843  *
1844  * Currently is actually called from the syncer! \o/
1845  *
1846  * This task will have to snapshot the state of the dirty inode chain.
1847  * From that, it will have to make sure all of the inodes on the dirty
1848  * chain have IO initiated. We make sure that io is initiated for the root
1849  * block.
1850  *
1851  * If waitfor is set, we wait for media to acknowledge the new rootblock.
1852  *
1853  * THINKS: side A vs side B, to have sync not stall all I/O?
1854  */
1855 int
1856 hammer2_vfs_sync(struct mount *mp, int waitfor)
1857 {
1858         struct hammer2_sync_info info;
1859         hammer2_chain_t *chain;
1860         hammer2_pfsmount_t *pmp;
1861         hammer2_mount_t *hmp;
1862         int flags;
1863         int error;
1864         int total_error;
1865         int force_fchain;
1866         int i;
1867
1868         pmp = MPTOPMP(mp);
1869
1870         /*
1871          * We can't acquire locks on existing vnodes while in a transaction
1872          * without risking a deadlock.  This assumes that vfsync() can be
1873          * called without the vnode locked (which it can in DragonFly).
1874          * Otherwise we'd have to implement a multi-pass or flag the lock
1875          * failures and retry.
1876          *
1877          * The reclamation code interlocks with the sync list's token
1878          * (by removing the vnode from the scan list) before unlocking
1879          * the inode, giving us time to ref the inode.
1880          */
1881         /*flags = VMSC_GETVP;*/
1882         flags = 0;
1883         if (waitfor & MNT_LAZY)
1884                 flags |= VMSC_ONEPASS;
1885
1886         /*
1887          * Initialize a normal transaction and sync everything out, then
1888          * wait for pending I/O to finish (so it gets a transaction id
1889          * that the meta-data flush will catch).
1890          */
1891         hammer2_trans_init(&info.trans, pmp, NULL, 0);
1892         info.error = 0;
1893         info.waitfor = MNT_NOWAIT;
1894         vsyncscan(mp, flags | VMSC_NOWAIT, hammer2_sync_scan2, &info);
1895
1896         if (info.error == 0 && (waitfor & MNT_WAIT)) {
1897                 info.waitfor = waitfor;
1898                     vsyncscan(mp, flags, hammer2_sync_scan2, &info);
1899
1900         }
1901         hammer2_trans_done(&info.trans);
1902         hammer2_bioq_sync(info.trans.pmp);
1903
1904         /*
1905          * Start the flush transaction and flush all meta-data.
1906          */
1907         hammer2_trans_init(&info.trans, pmp, NULL, HAMMER2_TRANS_ISFLUSH);
1908
1909         total_error = 0;
1910         for (i = 0; i < pmp->cluster.nchains; ++i) {
1911                 hmp = pmp->cluster.chains[i]->hmp;
1912
1913                 /*
1914                  * Media mounts have two 'roots', vchain for the topology
1915                  * and fchain for the free block table.  Flush both.
1916                  *
1917                  * Note that the topology and free block table are handled
1918                  * independently, so the free block table can wind up being
1919                  * ahead of the topology.  We depend on the bulk free scan
1920                  * code to deal with any loose ends.
1921                  */
1922 #if 1
1923                 hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
1924                 if ((hmp->fchain.flags & HAMMER2_CHAIN_MODIFIED) ||
1925                     hmp->fchain.core->update_hi > hmp->voldata.freemap_tid) {
1926                         /* this will also modify vchain as a side effect */
1927                         chain = &hmp->fchain;
1928                         hammer2_chain_flush(&info.trans, &chain);
1929                         KKASSERT(chain == &hmp->fchain);
1930                 }
1931                 hammer2_chain_unlock(&hmp->fchain);
1932 #endif
1933
1934                 hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
1935                 if ((hmp->vchain.flags & HAMMER2_CHAIN_MODIFIED) ||
1936                     hmp->vchain.core->update_hi > hmp->voldata.mirror_tid) {
1937                         chain = &hmp->vchain;
1938                         hammer2_chain_flush(&info.trans, &chain);
1939                         KKASSERT(chain == &hmp->vchain);
1940                         force_fchain = 1;
1941                 } else {
1942                         force_fchain = 0;
1943                 }
1944                 hammer2_chain_unlock(&hmp->vchain);
1945
1946 #if 0
1947                 hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
1948                 if ((hmp->fchain.flags & HAMMER2_CHAIN_MODIFIED) ||
1949                     hmp->fchain.core->update_hi > hmp->voldata.freemap_tid ||
1950                     force_fchain) {
1951                         /* this will also modify vchain as a side effect */
1952                         chain = &hmp->fchain;
1953                         hammer2_chain_flush(&info.trans, &chain);
1954                         KKASSERT(chain == &hmp->fchain);
1955                 }
1956                 hammer2_chain_unlock(&hmp->fchain);
1957 #endif
1958
1959                 error = 0;
1960
1961                 /*
1962                  * We can't safely flush the volume header until we have
1963                  * flushed any device buffers which have built up.
1964                  *
1965                  * XXX this isn't being incremental
1966                  */
1967                 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
1968                 error = VOP_FSYNC(hmp->devvp, MNT_WAIT, 0);
1969                 vn_unlock(hmp->devvp);
1970
1971                 /*
1972                  * The flush code sets CHAIN_VOLUMESYNC to indicate that the
1973                  * volume header needs synchronization via hmp->volsync.
1974                  *
1975                  * XXX synchronize the flag & data with only this flush XXX
1976                  */
1977                 if (error == 0 &&
1978                     (hmp->vchain.flags & HAMMER2_CHAIN_VOLUMESYNC)) {
1979                         struct buf *bp;
1980
1981                         /*
1982                          * Synchronize the disk before flushing the volume
1983                          * header.
1984                          */
1985                         bp = getpbuf(NULL);
1986                         bp->b_bio1.bio_offset = 0;
1987                         bp->b_bufsize = 0;
1988                         bp->b_bcount = 0;
1989                         bp->b_cmd = BUF_CMD_FLUSH;
1990                         bp->b_bio1.bio_done = biodone_sync;
1991                         bp->b_bio1.bio_flags |= BIO_SYNC;
1992                         vn_strategy(hmp->devvp, &bp->b_bio1);
1993                         biowait(&bp->b_bio1, "h2vol");
1994                         relpbuf(bp, NULL);
1995
1996                         /*
1997                          * Then we can safely flush the version of the
1998                          * volume header synchronized by the flush code.
1999                          */
2000                         i = hmp->volhdrno + 1;
2001                         if (i >= HAMMER2_NUM_VOLHDRS)
2002                                 i = 0;
2003                         if (i * HAMMER2_ZONE_BYTES64 + HAMMER2_SEGSIZE >
2004                             hmp->volsync.volu_size) {
2005                                 i = 0;
2006                         }
2007                         kprintf("sync volhdr %d %jd\n",
2008                                 i, (intmax_t)hmp->volsync.volu_size);
2009                         bp = getblk(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
2010                                     HAMMER2_PBUFSIZE, 0, 0);
2011                         atomic_clear_int(&hmp->vchain.flags,
2012                                          HAMMER2_CHAIN_VOLUMESYNC);
2013                         bcopy(&hmp->volsync, bp->b_data, HAMMER2_PBUFSIZE);
2014                         bawrite(bp);
2015                         hmp->volhdrno = i;
2016                 }
2017                 if (error)
2018                         total_error = error;
2019         }
2020         hammer2_trans_done(&info.trans);
2021
2022         return (total_error);
2023 }
2024
2025 /*
2026  * Sync passes.
2027  *
2028  * NOTE: We don't test update_lo/update_hi or MOVED here because the fsync
2029  *       code won't flush on those flags.  The syncer code above will do a
2030  *       general meta-data flush globally that will catch these flags.
2031  */
2032
2033 static int
2034 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
2035 {
2036         struct hammer2_sync_info *info = data;
2037         hammer2_inode_t *ip;
2038         int error;
2039
2040         /*
2041          *
2042          */
2043         ip = VTOI(vp);
2044         if (ip == NULL)
2045                 return(0);
2046         if (vp->v_type == VNON || vp->v_type == VBAD) {
2047                 vclrisdirty(vp);
2048                 return(0);
2049         }
2050         if ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
2051             RB_EMPTY(&vp->v_rbdirty_tree)) {
2052                 vclrisdirty(vp);
2053                 return(0);
2054         }
2055
2056         /*
2057          * VOP_FSYNC will start a new transaction so replicate some code
2058          * here to do it inline (see hammer2_vop_fsync()).
2059          *
2060          * WARNING: The vfsync interacts with the buffer cache and might
2061          *          block, we can't hold the inode lock at that time.
2062          *          However, we MUST ref ip before blocking to ensure that
2063          *          it isn't ripped out from under us (since we do not
2064          *          hold a lock on the vnode).
2065          */
2066         hammer2_inode_ref(ip);
2067         atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
2068         if (vp)
2069                 vfsync(vp, MNT_NOWAIT, 1, NULL, NULL);
2070
2071 #if 0
2072         /*
2073          * XXX this interferes with flush operations mainly because the
2074          *     same transaction id is being used by asynchronous buffer
2075          *     operations above and can be reordered after the flush
2076          *     below.
2077          */
2078         parent = hammer2_inode_lock_ex(ip);
2079         hammer2_chain_flush(&info->trans, &parent);
2080         hammer2_inode_unlock_ex(ip, parent);
2081 #endif
2082         hammer2_inode_drop(ip);
2083         error = 0;
2084 #if 0
2085         error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
2086 #endif
2087         if (error)
2088                 info->error = error;
2089         return(0);
2090 }
2091
2092 static
2093 int
2094 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
2095 {
2096         return (0);
2097 }
2098
2099 static
2100 int
2101 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
2102                struct fid *fhp, struct vnode **vpp)
2103 {
2104         return (0);
2105 }
2106
2107 static
2108 int
2109 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
2110                  int *exflagsp, struct ucred **credanonp)
2111 {
2112         return (0);
2113 }
2114
2115 /*
2116  * Support code for hammer2_mount().  Read, verify, and install the volume
2117  * header into the HMP
2118  *
2119  * XXX read four volhdrs and use the one with the highest TID whos CRC
2120  *     matches.
2121  *
2122  * XXX check iCRCs.
2123  *
2124  * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
2125  *     nonexistant locations.
2126  *
2127  * XXX Record selected volhdr and ring updates to each of 4 volhdrs
2128  */
2129 static
2130 int
2131 hammer2_install_volume_header(hammer2_mount_t *hmp)
2132 {
2133         hammer2_volume_data_t *vd;
2134         struct buf *bp;
2135         hammer2_crc32_t crc0, crc, bcrc0, bcrc;
2136         int error_reported;
2137         int error;
2138         int valid;
2139         int i;
2140
2141         error_reported = 0;
2142         error = 0;
2143         valid = 0;
2144         bp = NULL;
2145
2146         /*
2147          * There are up to 4 copies of the volume header (syncs iterate
2148          * between them so there is no single master).  We don't trust the
2149          * volu_size field so we don't know precisely how large the filesystem
2150          * is, so depend on the OS to return an error if we go beyond the
2151          * block device's EOF.
2152          */
2153         for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
2154                 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
2155                               HAMMER2_VOLUME_BYTES, &bp);
2156                 if (error) {
2157                         brelse(bp);
2158                         bp = NULL;
2159                         continue;
2160                 }
2161
2162                 vd = (struct hammer2_volume_data *) bp->b_data;
2163                 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
2164                     (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
2165                         brelse(bp);
2166                         bp = NULL;
2167                         continue;
2168                 }
2169
2170                 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
2171                         /* XXX: Reversed-endianness filesystem */
2172                         kprintf("hammer2: reverse-endian filesystem detected");
2173                         brelse(bp);
2174                         bp = NULL;
2175                         continue;
2176                 }
2177
2178                 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
2179                 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
2180                                       HAMMER2_VOLUME_ICRC0_SIZE);
2181                 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
2182                 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
2183                                        HAMMER2_VOLUME_ICRC1_SIZE);
2184                 if ((crc0 != crc) || (bcrc0 != bcrc)) {
2185                         kprintf("hammer2 volume header crc "
2186                                 "mismatch copy #%d %08x/%08x\n",
2187                                 i, crc0, crc);
2188                         error_reported = 1;
2189                         brelse(bp);
2190                         bp = NULL;
2191                         continue;
2192                 }
2193                 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
2194                         valid = 1;
2195                         hmp->voldata = *vd;
2196                         hmp->volhdrno = i;
2197                 }
2198                 brelse(bp);
2199                 bp = NULL;
2200         }
2201         if (valid) {
2202                 hmp->volsync = hmp->voldata;
2203                 error = 0;
2204                 if (error_reported || bootverbose || 1) { /* 1/DEBUG */
2205                         kprintf("hammer2: using volume header #%d\n",
2206                                 hmp->volhdrno);
2207                 }
2208         } else {
2209                 error = EINVAL;
2210                 kprintf("hammer2: no valid volume headers found!\n");
2211         }
2212         return (error);
2213 }
2214
2215 /*
2216  * Reconnect using the passed file pointer.  The caller must ref the
2217  * fp for us.
2218  */
2219 void
2220 hammer2_cluster_reconnect(hammer2_pfsmount_t *pmp, struct file *fp)
2221 {
2222         hammer2_inode_data_t *ipdata;
2223         hammer2_chain_t *parent;
2224         hammer2_mount_t *hmp;
2225         size_t name_len;
2226
2227         hmp = pmp->cluster.chains[0]->hmp;      /* XXX */
2228
2229         /*
2230          * Closes old comm descriptor, kills threads, cleans up
2231          * states, then installs the new descriptor and creates
2232          * new threads.
2233          */
2234         kdmsg_iocom_reconnect(&pmp->iocom, fp, "hammer2");
2235
2236         /*
2237          * Setup LNK_CONN fields for autoinitiated state machine
2238          */
2239         parent = hammer2_inode_lock_ex(pmp->iroot);
2240         ipdata = &parent->data->ipdata;
2241         pmp->iocom.auto_lnk_conn.pfs_clid = ipdata->pfs_clid;
2242         pmp->iocom.auto_lnk_conn.pfs_fsid = ipdata->pfs_fsid;
2243         pmp->iocom.auto_lnk_conn.pfs_type = ipdata->pfs_type;
2244         pmp->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
2245         pmp->iocom.auto_lnk_conn.peer_type = hmp->voldata.peer_type;
2246
2247         /*
2248          * Filter adjustment.  Clients do not need visibility into other
2249          * clients (otherwise millions of clients would present a serious
2250          * problem).  The fs_label also serves to restrict the namespace.
2251          */
2252         pmp->iocom.auto_lnk_conn.peer_mask = 1LLU << HAMMER2_PEER_HAMMER2;
2253         pmp->iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
2254         switch (ipdata->pfs_type) {
2255         case DMSG_PFSTYPE_CLIENT:
2256                 pmp->iocom.auto_lnk_conn.peer_mask &=
2257                                 ~(1LLU << DMSG_PFSTYPE_CLIENT);
2258                 break;
2259         default:
2260                 break;
2261         }
2262
2263         name_len = ipdata->name_len;
2264         if (name_len >= sizeof(pmp->iocom.auto_lnk_conn.fs_label))
2265                 name_len = sizeof(pmp->iocom.auto_lnk_conn.fs_label) - 1;
2266         bcopy(ipdata->filename,
2267               pmp->iocom.auto_lnk_conn.fs_label,
2268               name_len);
2269         pmp->iocom.auto_lnk_conn.fs_label[name_len] = 0;
2270
2271         /*
2272          * Setup LNK_SPAN fields for autoinitiated state machine
2273          */
2274         pmp->iocom.auto_lnk_span.pfs_clid = ipdata->pfs_clid;
2275         pmp->iocom.auto_lnk_span.pfs_fsid = ipdata->pfs_fsid;
2276         pmp->iocom.auto_lnk_span.pfs_type = ipdata->pfs_type;
2277         pmp->iocom.auto_lnk_span.peer_type = hmp->voldata.peer_type;
2278         pmp->iocom.auto_lnk_span.proto_version = DMSG_SPAN_PROTO_1;
2279         name_len = ipdata->name_len;
2280         if (name_len >= sizeof(pmp->iocom.auto_lnk_span.fs_label))
2281                 name_len = sizeof(pmp->iocom.auto_lnk_span.fs_label) - 1;
2282         bcopy(ipdata->filename,
2283               pmp->iocom.auto_lnk_span.fs_label,
2284               name_len);
2285         pmp->iocom.auto_lnk_span.fs_label[name_len] = 0;
2286         hammer2_inode_unlock_ex(pmp->iroot, parent);
2287
2288         kdmsg_iocom_autoinitiate(&pmp->iocom, hammer2_autodmsg);
2289 }
2290
2291 static int
2292 hammer2_rcvdmsg(kdmsg_msg_t *msg)
2293 {
2294         switch(msg->any.head.cmd & DMSGF_TRANSMASK) {
2295         case DMSG_DBG_SHELL:
2296                 /*
2297                  * (non-transaction)
2298                  * Execute shell command (not supported atm)
2299                  */
2300                 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
2301                 break;
2302         case DMSG_DBG_SHELL | DMSGF_REPLY:
2303                 /*
2304                  * (non-transaction)
2305                  */
2306                 if (msg->aux_data) {
2307                         msg->aux_data[msg->aux_size - 1] = 0;
2308                         kprintf("HAMMER2 DBG: %s\n", msg->aux_data);
2309                 }
2310                 break;
2311         default:
2312                 /*
2313                  * Unsupported message received.  We only need to
2314                  * reply if it's a transaction in order to close our end.
2315                  * Ignore any one-way messages are any further messages
2316                  * associated with the transaction.
2317                  *
2318                  * NOTE: This case also includes DMSG_LNK_ERROR messages
2319                  *       which might be one-way, replying to those would
2320                  *       cause an infinite ping-pong.
2321                  */
2322                 if (msg->any.head.cmd & DMSGF_CREATE)
2323                         kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
2324                 break;
2325         }
2326         return(0);
2327 }
2328
2329 /*
2330  * This function is called after KDMSG has automatically handled processing
2331  * of a LNK layer message (typically CONN, SPAN, or CIRC).
2332  *
2333  * We tag off the LNK_CONN to trigger our LNK_VOLCONF messages which
2334  * advertises all available hammer2 super-root volumes.
2335  */
2336 static void
2337 hammer2_autodmsg(kdmsg_msg_t *msg)
2338 {
2339         hammer2_pfsmount_t *pmp = msg->iocom->handle;
2340         hammer2_mount_t *hmp = pmp->cluster.chains[0]->hmp; /* XXX */
2341         int copyid;
2342
2343         /*
2344          * We only care about replies to our LNK_CONN auto-request.  kdmsg
2345          * has already processed the reply, we use this calback as a shim
2346          * to know when we can advertise available super-root volumes.
2347          */
2348         if ((msg->any.head.cmd & DMSGF_TRANSMASK) !=
2349             (DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_REPLY) ||
2350             msg->state == NULL) {
2351                 return;
2352         }
2353
2354         kprintf("LNK_CONN REPLY RECEIVED CMD %08x\n", msg->any.head.cmd);
2355
2356         if (msg->any.head.cmd & DMSGF_CREATE) {
2357                 kprintf("HAMMER2: VOLDATA DUMP\n");
2358
2359                 /*
2360                  * Dump the configuration stored in the volume header
2361                  */
2362                 hammer2_voldata_lock(hmp);
2363                 for (copyid = 0; copyid < HAMMER2_COPYID_COUNT; ++copyid) {
2364                         if (hmp->voldata.copyinfo[copyid].copyid == 0)
2365                                 continue;
2366                         hammer2_volconf_update(pmp, copyid);
2367                 }
2368                 hammer2_voldata_unlock(hmp, 0);
2369         }
2370         if ((msg->any.head.cmd & DMSGF_DELETE) &&
2371             msg->state && (msg->state->txcmd & DMSGF_DELETE) == 0) {
2372                 kprintf("HAMMER2: CONN WAS TERMINATED\n");
2373         }
2374 }
2375
2376 /*
2377  * Volume configuration updates are passed onto the userland service
2378  * daemon via the open LNK_CONN transaction.
2379  */
2380 void
2381 hammer2_volconf_update(hammer2_pfsmount_t *pmp, int index)
2382 {
2383         hammer2_mount_t *hmp = pmp->cluster.chains[0]->hmp;     /* XXX */
2384         kdmsg_msg_t *msg;
2385
2386         /* XXX interlock against connection state termination */
2387         kprintf("volconf update %p\n", pmp->iocom.conn_state);
2388         if (pmp->iocom.conn_state) {
2389                 kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
2390                 msg = kdmsg_msg_alloc_state(pmp->iocom.conn_state,
2391                                             DMSG_LNK_VOLCONF, NULL, NULL);
2392                 msg->any.lnk_volconf.copy = hmp->voldata.copyinfo[index];
2393                 msg->any.lnk_volconf.mediaid = hmp->voldata.fsid;
2394                 msg->any.lnk_volconf.index = index;
2395                 kdmsg_msg_write(msg);
2396         }
2397 }
2398
2399 /*
2400  * This handles hysteresis on regular file flushes.  Because the BIOs are
2401  * routed to a thread it is possible for an excessive number to build up
2402  * and cause long front-end stalls long before the runningbuffspace limit
2403  * is hit, so we implement hammer2_flush_pipe to control the
2404  * hysteresis.
2405  *
2406  * This is a particular problem when compression is used.
2407  */
2408 void
2409 hammer2_lwinprog_ref(hammer2_pfsmount_t *pmp)
2410 {
2411         atomic_add_int(&pmp->count_lwinprog, 1);
2412 }
2413
2414 void
2415 hammer2_lwinprog_drop(hammer2_pfsmount_t *pmp)
2416 {
2417         int lwinprog;
2418
2419         lwinprog = atomic_fetchadd_int(&pmp->count_lwinprog, -1);
2420         if ((lwinprog & HAMMER2_LWINPROG_WAITING) &&
2421             (lwinprog & HAMMER2_LWINPROG_MASK) <= hammer2_flush_pipe * 2 / 3) {
2422                 atomic_clear_int(&pmp->count_lwinprog,
2423                                  HAMMER2_LWINPROG_WAITING);
2424                 wakeup(&pmp->count_lwinprog);
2425         }
2426 }
2427
2428 void
2429 hammer2_lwinprog_wait(hammer2_pfsmount_t *pmp)
2430 {
2431         int lwinprog;
2432
2433         for (;;) {
2434                 lwinprog = pmp->count_lwinprog;
2435                 cpu_ccfence();
2436                 if ((lwinprog & HAMMER2_LWINPROG_MASK) < hammer2_flush_pipe)
2437                         break;
2438                 tsleep_interlock(&pmp->count_lwinprog, 0);
2439                 atomic_set_int(&pmp->count_lwinprog, HAMMER2_LWINPROG_WAITING);
2440                 lwinprog = pmp->count_lwinprog;
2441                 if ((lwinprog & HAMMER2_LWINPROG_MASK) < hammer2_flush_pipe)
2442                         break;
2443                 tsleep(&pmp->count_lwinprog, PINTERLOCKED, "h2wpipe", hz);
2444         }
2445 }
2446
2447 void
2448 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp)
2449 {
2450         hammer2_chain_layer_t *layer;
2451         hammer2_chain_t *scan;
2452         hammer2_chain_t *first_parent;
2453
2454         --*countp;
2455         if (*countp == 0) {
2456                 kprintf("%*.*s...\n", tab, tab, "");
2457                 return;
2458         }
2459         if (*countp < 0)
2460                 return;
2461         first_parent = chain->core ? TAILQ_FIRST(&chain->core->ownerq) : NULL;
2462         kprintf("%*.*schain %p.%d %016jx/%d mir=%016jx\n",
2463                 tab, tab, "",
2464                 chain, chain->bref.type,
2465                 chain->bref.key, chain->bref.keybits,
2466                 chain->bref.mirror_tid);
2467
2468         kprintf("%*.*s      [%08x] (%s) dt=%016jx refs=%d\n",
2469                 tab, tab, "",
2470                 chain->flags,
2471                 ((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
2472                 chain->data) ?  (char *)chain->data->ipdata.filename : "?"),
2473                 chain->delete_tid,
2474                 chain->refs);
2475
2476         kprintf("%*.*s      core %p [%08x] lo=%08jx hi=%08jx fp=%p np=%p",
2477                 tab, tab, "",
2478                 chain->core, (chain->core ? chain->core->flags : 0),
2479                 (chain->core ? chain->core->update_lo : -1),
2480                 (chain->core ? chain->core->update_hi : -1),
2481                 first_parent,
2482                 (first_parent ? TAILQ_NEXT(chain, core_entry) : NULL));
2483
2484         if (first_parent)
2485                 kprintf(" [fpflags %08x fprefs %d\n",
2486                         first_parent->flags,
2487                         first_parent->refs);
2488         if (chain->core == NULL || TAILQ_EMPTY(&chain->core->layerq))
2489                 kprintf("\n");
2490         else
2491                 kprintf(" {\n");
2492         if (chain->core) {
2493                 TAILQ_FOREACH(layer, &chain->core->layerq, entry) {
2494                         RB_FOREACH(scan, hammer2_chain_tree, &layer->rbtree) {
2495                                 hammer2_dump_chain(scan, tab + 4, countp);
2496                         }
2497                 }
2498         }
2499         if (chain->core && !TAILQ_EMPTY(&chain->core->layerq)) {
2500                 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
2501                         kprintf("%*.*s}(%s)\n", tab, tab, "",
2502                                 chain->data->ipdata.filename);
2503                 else
2504                         kprintf("%*.*s}\n", tab, tab, "");
2505         }
2506 }