sys/vfs/hammer/hammer_vnops.c

   1 /*
   2  * Copyright (c) 2007 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.35 2008/03/18 05:19:16 dillon Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/fcntl.h>
  41 #include <sys/namecache.h>
  42 #include <sys/vnode.h>
  43 #include <sys/lockf.h>
  44 #include <sys/event.h>
  45 #include <sys/stat.h>
  46 #include <sys/dirent.h>
  47 #include <vm/vm_extern.h>
  48 #include <vfs/fifofs/fifo.h>
  49 #include "hammer.h"
  50
  51 /*
  52  * USERFS VNOPS
  53  */
  54 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
  55 static int hammer_vop_fsync(struct vop_fsync_args *);
  56 static int hammer_vop_read(struct vop_read_args *);
  57 static int hammer_vop_write(struct vop_write_args *);
  58 static int hammer_vop_access(struct vop_access_args *);
  59 static int hammer_vop_advlock(struct vop_advlock_args *);
  60 static int hammer_vop_close(struct vop_close_args *);
  61 static int hammer_vop_ncreate(struct vop_ncreate_args *);
  62 static int hammer_vop_getattr(struct vop_getattr_args *);
  63 static int hammer_vop_nresolve(struct vop_nresolve_args *);
  64 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
  65 static int hammer_vop_nlink(struct vop_nlink_args *);
  66 static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
  67 static int hammer_vop_nmknod(struct vop_nmknod_args *);
  68 static int hammer_vop_open(struct vop_open_args *);
  69 static int hammer_vop_pathconf(struct vop_pathconf_args *);
  70 static int hammer_vop_print(struct vop_print_args *);
  71 static int hammer_vop_readdir(struct vop_readdir_args *);
  72 static int hammer_vop_readlink(struct vop_readlink_args *);
  73 static int hammer_vop_nremove(struct vop_nremove_args *);
  74 static int hammer_vop_nrename(struct vop_nrename_args *);
  75 static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
  76 static int hammer_vop_setattr(struct vop_setattr_args *);
  77 static int hammer_vop_strategy(struct vop_strategy_args *);
  78 static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
  79 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
  80 static int hammer_vop_ioctl(struct vop_ioctl_args *);
  81 static int hammer_vop_mountctl(struct vop_mountctl_args *);
  82
  83 static int hammer_vop_fifoclose (struct vop_close_args *);
  84 static int hammer_vop_fiforead (struct vop_read_args *);
  85 static int hammer_vop_fifowrite (struct vop_write_args *);
  86
  87 static int hammer_vop_specclose (struct vop_close_args *);
  88 static int hammer_vop_specread (struct vop_read_args *);
  89 static int hammer_vop_specwrite (struct vop_write_args *);
  90
  91 struct vop_ops hammer_vnode_vops = {
  92         .vop_default =          vop_defaultop,
  93         .vop_fsync =            hammer_vop_fsync,
  94         .vop_getpages =         vop_stdgetpages,
  95         .vop_putpages =         vop_stdputpages,
  96         .vop_read =             hammer_vop_read,
  97         .vop_write =            hammer_vop_write,
  98         .vop_access =           hammer_vop_access,
  99         .vop_advlock =          hammer_vop_advlock,
 100         .vop_close =            hammer_vop_close,
 101         .vop_ncreate =          hammer_vop_ncreate,
 102         .vop_getattr =          hammer_vop_getattr,
 103         .vop_inactive =         hammer_vop_inactive,
 104         .vop_reclaim =          hammer_vop_reclaim,
 105         .vop_nresolve =         hammer_vop_nresolve,
 106         .vop_nlookupdotdot =    hammer_vop_nlookupdotdot,
 107         .vop_nlink =            hammer_vop_nlink,
 108         .vop_nmkdir =           hammer_vop_nmkdir,
 109         .vop_nmknod =           hammer_vop_nmknod,
 110         .vop_open =             hammer_vop_open,
 111         .vop_pathconf =         hammer_vop_pathconf,
 112         .vop_print =            hammer_vop_print,
 113         .vop_readdir =          hammer_vop_readdir,
 114         .vop_readlink =         hammer_vop_readlink,
 115         .vop_nremove =          hammer_vop_nremove,
 116         .vop_nrename =          hammer_vop_nrename,
 117         .vop_nrmdir =           hammer_vop_nrmdir,
 118         .vop_setattr =          hammer_vop_setattr,
 119         .vop_strategy =         hammer_vop_strategy,
 120         .vop_nsymlink =         hammer_vop_nsymlink,
 121         .vop_nwhiteout =        hammer_vop_nwhiteout,
 122         .vop_ioctl =            hammer_vop_ioctl,
 123         .vop_mountctl =         hammer_vop_mountctl
 124 };
 125
 126 struct vop_ops hammer_spec_vops = {
 127         .vop_default =          spec_vnoperate,
 128         .vop_fsync =            hammer_vop_fsync,
 129         .vop_read =             hammer_vop_specread,
 130         .vop_write =            hammer_vop_specwrite,
 131         .vop_access =           hammer_vop_access,
 132         .vop_close =            hammer_vop_specclose,
 133         .vop_getattr =          hammer_vop_getattr,
 134         .vop_inactive =         hammer_vop_inactive,
 135         .vop_reclaim =          hammer_vop_reclaim,
 136         .vop_setattr =          hammer_vop_setattr
 137 };
 138
 139 struct vop_ops hammer_fifo_vops = {
 140         .vop_default =          fifo_vnoperate,
 141         .vop_fsync =            hammer_vop_fsync,
 142         .vop_read =             hammer_vop_fiforead,
 143         .vop_write =            hammer_vop_fifowrite,
 144         .vop_access =           hammer_vop_access,
 145         .vop_close =            hammer_vop_fifoclose,
 146         .vop_getattr =          hammer_vop_getattr,
 147         .vop_inactive =         hammer_vop_inactive,
 148         .vop_reclaim =          hammer_vop_reclaim,
 149         .vop_setattr =          hammer_vop_setattr
 150 };
 151
 152 static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp,
 153                            struct ucred *cred, int flags);
 154 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
 155 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
 156
 157 #if 0
 158 static
 159 int
 160 hammer_vop_vnoperate(struct vop_generic_args *)
 161 {
 162         return (VOCALL(&hammer_vnode_vops, ap));
 163 }
 164 #endif
 165
 166 /*
 167  * hammer_vop_fsync { vp, waitfor }
 168  */
 169 static
 170 int
 171 hammer_vop_fsync(struct vop_fsync_args *ap)
 172 {
 173         hammer_inode_t ip;
 174         int error;
 175
 176         ip = VTOI(ap->a_vp);
 177         error = hammer_sync_inode(ip, ap->a_waitfor, 0);
 178         return (error);
 179 }
 180
 181 /*
 182  * hammer_vop_read { vp, uio, ioflag, cred }
 183  */
 184 static
 185 int
 186 hammer_vop_read(struct vop_read_args *ap)
 187 {
 188         struct hammer_transaction trans;
 189         hammer_inode_t ip;
 190         off_t offset;
 191         struct buf *bp;
 192         struct uio *uio;
 193         int error;
 194         int n;
 195         int seqcount;
 196
 197         if (ap->a_vp->v_type != VREG)
 198                 return (EINVAL);
 199         ip = VTOI(ap->a_vp);
 200         error = 0;
 201         seqcount = ap->a_ioflag >> 16;
 202
 203         hammer_start_transaction(&trans, ip->hmp);
 204
 205         /*
 206          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 207          */
 208         uio = ap->a_uio;
 209         while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
 210                 offset = uio->uio_offset & HAMMER_BUFMASK;
 211 #if 0
 212                 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
 213                                      uio->uio_offset - offset, HAMMER_BUFSIZE,
 214                                      MAXBSIZE, seqcount, &bp);
 215 #endif
 216                 error = bread(ap->a_vp, uio->uio_offset - offset,
 217                               HAMMER_BUFSIZE, &bp);
 218                 if (error) {
 219                         brelse(bp);
 220                         break;
 221                 }
 222                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 223                 n = HAMMER_BUFSIZE - offset;
 224                 if (n > uio->uio_resid)
 225                         n = uio->uio_resid;
 226                 if (n > ip->ino_rec.ino_size - uio->uio_offset)
 227                         n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
 228                 error = uiomove((char *)bp->b_data + offset, n, uio);
 229                 if (error) {
 230                         bqrelse(bp);
 231                         break;
 232                 }
 233                 if ((ip->flags & HAMMER_INODE_RO) == 0 &&
 234                     (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
 235                         ip->ino_rec.ino_atime = trans.tid;
 236                         hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
 237                 }
 238                 bqrelse(bp);
 239         }
 240         hammer_commit_transaction(&trans);
 241         return (error);
 242 }
 243
 244 /*
 245  * hammer_vop_write { vp, uio, ioflag, cred }
 246  */
 247 static
 248 int
 249 hammer_vop_write(struct vop_write_args *ap)
 250 {
 251         struct hammer_transaction trans;
 252         struct hammer_inode *ip;
 253         struct uio *uio;
 254         off_t offset;
 255         struct buf *bp;
 256         int error;
 257         int n;
 258         int flags;
 259
 260         if (ap->a_vp->v_type != VREG)
 261                 return (EINVAL);
 262         ip = VTOI(ap->a_vp);
 263         error = 0;
 264
 265         if (ip->flags & HAMMER_INODE_RO)
 266                 return (EROFS);
 267
 268         /*
 269          * Create a transaction to cover the operations we perform.
 270          */
 271         hammer_start_transaction(&trans, ip->hmp);
 272         uio = ap->a_uio;
 273
 274         /*
 275          * Check append mode
 276          */
 277         if (ap->a_ioflag & IO_APPEND)
 278                 uio->uio_offset = ip->ino_rec.ino_size;
 279
 280         /*
 281          * Check for illegal write offsets.  Valid range is 0...2^63-1
 282          */
 283         if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
 284                 hammer_commit_transaction(&trans);
 285                 return (EFBIG);
 286         }
 287
 288         /*
 289          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 290          */
 291         while (uio->uio_resid > 0) {
 292                 int fixsize = 0;
 293
 294                 offset = uio->uio_offset & HAMMER_BUFMASK;
 295                 n = HAMMER_BUFSIZE - offset;
 296                 if (n > uio->uio_resid)
 297                         n = uio->uio_resid;
 298                 if (uio->uio_offset + n > ip->ino_rec.ino_size) {
 299                         vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
 300                         fixsize = 1;
 301                 }
 302
 303                 if (uio->uio_segflg == UIO_NOCOPY) {
 304                         /*
 305                          * Issuing a write with the same data backing the
 306                          * buffer.  Instantiate the buffer to collect the
 307                          * backing vm pages, then read-in any missing bits.
 308                          *
 309                          * This case is used by vop_stdputpages().
 310                          */
 311                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 312                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 313                         if ((bp->b_flags & B_CACHE) == 0) {
 314                                 bqrelse(bp);
 315                                 error = bread(ap->a_vp,
 316                                               uio->uio_offset - offset,
 317                                               HAMMER_BUFSIZE, &bp);
 318                         }
 319                 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
 320                         /*
 321                          * entirely overwrite the buffer
 322                          */
 323                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 324                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 325                 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
 326                         /*
 327                          * XXX
 328                          */
 329                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 330                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 331                         vfs_bio_clrbuf(bp);
 332                 } else {
 333                         /*
 334                          * Partial overwrite, read in any missing bits then
 335                          * replace the portion being written.
 336                          */
 337                         error = bread(ap->a_vp, uio->uio_offset - offset,
 338                                       HAMMER_BUFSIZE, &bp);
 339                         if (error == 0)
 340                                 bheavy(bp);
 341                 }
 342                 if (error == 0)
 343                         error = uiomove((char *)bp->b_data + offset, n, uio);
 344
 345                 /*
 346                  * If we screwed up we have to undo any VM size changes we
 347                  * made.
 348                  */
 349                 if (error) {
 350                         brelse(bp);
 351                         if (fixsize) {
 352                                 vtruncbuf(ap->a_vp, ip->ino_rec.ino_size,
 353                                           HAMMER_BUFSIZE);
 354                         }
 355                         break;
 356                 }
 357                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 358                 if (ip->ino_rec.ino_size < uio->uio_offset) {
 359                         ip->ino_rec.ino_size = uio->uio_offset;
 360                         flags = HAMMER_INODE_RDIRTY;
 361                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
 362                 } else {
 363                         flags = 0;
 364                 }
 365                 ip->ino_rec.ino_mtime = trans.tid;
 366                 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
 367                 hammer_modify_inode(&trans, ip, flags);
 368
 369 #if 0
 370                 /*
 371                  * The file write must be tagged with the same TID as the
 372                  * inode, for consistency in case the inode changed size.
 373                  * This guarantees the on-disk data records will have a
 374                  * TID <= the inode TID representing the size change.
 375                  *
 376                  * If a prior write has not yet flushed, retain its TID.
 377                  */
 378                 if (bp->b_tid == 0)
 379                         bp->b_tid = ip->last_tid;
 380 #endif
 381                 /*
 382                  * For now we can't use ip->last_tid because we may wind
 383                  * up trying to flush the same buffer with the same TID
 384                  * (but different data) multiple times, which will cause
 385                  * a panic.
 386                  */
 387                 if (bp->b_tid == 0)
 388                         bp->b_tid = trans.tid;
 389
 390                 if (ap->a_ioflag & IO_SYNC) {
 391                         bwrite(bp);
 392                 } else if (ap->a_ioflag & IO_DIRECT) {
 393                         bawrite(bp);
 394                 } else if ((ap->a_ioflag >> 16) > 1 &&
 395                            (uio->uio_offset & HAMMER_BUFMASK) == 0) {
 396                         /*
 397                          * If seqcount indicates sequential operation and
 398                          * we just finished filling a buffer, push it out
 399                          * now to prevent the buffer cache from becoming
 400                          * too full, which would trigger non-optimal
 401                          * flushes.
 402                          */
 403                         bawrite(bp);
 404                 } else {
 405                         bdwrite(bp);
 406                 }
 407         }
 408         if (error)
 409                 hammer_abort_transaction(&trans);
 410         else
 411                 hammer_commit_transaction(&trans);
 412         return (error);
 413 }
 414
 415 /*
 416  * hammer_vop_access { vp, mode, cred }
 417  */
 418 static
 419 int
 420 hammer_vop_access(struct vop_access_args *ap)
 421 {
 422         struct hammer_inode *ip = VTOI(ap->a_vp);
 423         uid_t uid;
 424         gid_t gid;
 425         int error;
 426
 427         uid = hammer_to_unix_xid(&ip->ino_data.uid);
 428         gid = hammer_to_unix_xid(&ip->ino_data.gid);
 429
 430         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
 431                                   ip->ino_data.uflags);
 432         return (error);
 433 }
 434
 435 /*
 436  * hammer_vop_advlock { vp, id, op, fl, flags }
 437  */
 438 static
 439 int
 440 hammer_vop_advlock(struct vop_advlock_args *ap)
 441 {
 442         struct hammer_inode *ip = VTOI(ap->a_vp);
 443
 444         return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
 445 }
 446
 447 /*
 448  * hammer_vop_close { vp, fflag }
 449  */
 450 static
 451 int
 452 hammer_vop_close(struct vop_close_args *ap)
 453 {
 454         return (vop_stdclose(ap));
 455 }
 456
 457 /*
 458  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
 459  *
 460  * The operating system has already ensured that the directory entry
 461  * does not exist and done all appropriate namespace locking.
 462  */
 463 static
 464 int
 465 hammer_vop_ncreate(struct vop_ncreate_args *ap)
 466 {
 467         struct hammer_transaction trans;
 468         struct hammer_inode *dip;
 469         struct hammer_inode *nip;
 470         struct nchandle *nch;
 471         int error;
 472
 473         nch = ap->a_nch;
 474         dip = VTOI(ap->a_dvp);
 475
 476         if (dip->flags & HAMMER_INODE_RO)
 477                 return (EROFS);
 478
 479         /*
 480          * Create a transaction to cover the operations we perform.
 481          */
 482         hammer_start_transaction(&trans, dip->hmp);
 483
 484         /*
 485          * Create a new filesystem object of the requested type.  The
 486          * returned inode will be referenced but not locked.
 487          */
 488
 489         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 490         if (error)
 491                 kprintf("hammer_create_inode error %d\n", error);
 492         if (error) {
 493                 hammer_abort_transaction(&trans);
 494                 *ap->a_vpp = NULL;
 495                 return (error);
 496         }
 497
 498         /*
 499          * Add the new filesystem object to the directory.  This will also
 500          * bump the inode's link count.
 501          */
 502         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 503         if (error)
 504                 kprintf("hammer_ip_add_directory error %d\n", error);
 505
 506         /*
 507          * Finish up.
 508          */
 509         if (error) {
 510                 hammer_rel_inode(nip, 0);
 511                 hammer_abort_transaction(&trans);
 512                 *ap->a_vpp = NULL;
 513         } else {
 514                 hammer_commit_transaction(&trans);
 515                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 516                 hammer_rel_inode(nip, 0);
 517                 if (error == 0) {
 518                         cache_setunresolved(ap->a_nch);
 519                         cache_setvp(ap->a_nch, *ap->a_vpp);
 520                 }
 521         }
 522         return (error);
 523 }
 524
 525 /*
 526  * hammer_vop_getattr { vp, vap }
 527  */
 528 static
 529 int
 530 hammer_vop_getattr(struct vop_getattr_args *ap)
 531 {
 532         struct hammer_inode *ip = VTOI(ap->a_vp);
 533         struct vattr *vap = ap->a_vap;
 534
 535 #if 0
 536         if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
 537             (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
 538             ip->obj_asof == XXX
 539         ) {
 540                 /* LAZYMOD XXX */
 541         }
 542         hammer_itimes(ap->a_vp);
 543 #endif
 544
 545         vap->va_fsid = ip->hmp->fsid_udev;
 546         vap->va_fileid = ip->ino_rec.base.base.obj_id;
 547         vap->va_mode = ip->ino_data.mode;
 548         vap->va_nlink = ip->ino_rec.ino_nlinks;
 549         vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
 550         vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
 551         vap->va_rmajor = 0;
 552         vap->va_rminor = 0;
 553         vap->va_size = ip->ino_rec.ino_size;
 554         hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
 555         hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
 556         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
 557         vap->va_flags = ip->ino_data.uflags;
 558         vap->va_gen = 1;        /* hammer inums are unique for all time */
 559         vap->va_blocksize = HAMMER_BUFSIZE;
 560         vap->va_bytes = (ip->ino_rec.ino_size + 63) & ~63;
 561         vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
 562         vap->va_filerev = 0;    /* XXX */
 563         /* mtime uniquely identifies any adjustments made to the file */
 564         vap->va_fsmid = ip->ino_rec.ino_mtime;
 565         vap->va_uid_uuid = ip->ino_data.uid;
 566         vap->va_gid_uuid = ip->ino_data.gid;
 567         vap->va_fsid_uuid = ip->hmp->fsid;
 568         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 569                           VA_FSID_UUID_VALID;
 570
 571         switch (ip->ino_rec.base.base.obj_type) {
 572         case HAMMER_OBJTYPE_CDEV:
 573         case HAMMER_OBJTYPE_BDEV:
 574                 vap->va_rmajor = ip->ino_data.rmajor;
 575                 vap->va_rminor = ip->ino_data.rminor;
 576                 break;
 577         default:
 578                 break;
 579         }
 580
 581         return(0);
 582 }
 583
 584 /*
 585  * hammer_vop_nresolve { nch, dvp, cred }
 586  *
 587  * Locate the requested directory entry.
 588  */
 589 static
 590 int
 591 hammer_vop_nresolve(struct vop_nresolve_args *ap)
 592 {
 593         struct namecache *ncp;
 594         hammer_inode_t dip;
 595         hammer_inode_t ip;
 596         hammer_tid_t asof;
 597         struct hammer_cursor cursor;
 598         union hammer_record_ondisk *rec;
 599         struct vnode *vp;
 600         int64_t namekey;
 601         int error;
 602         int i;
 603         int nlen;
 604         int flags;
 605         u_int64_t obj_id;
 606
 607         /*
 608          * Misc initialization, plus handle as-of name extensions.  Look for
 609          * the '@@' extension.  Note that as-of files and directories cannot
 610          * be modified.
 611          */
 612         dip = VTOI(ap->a_dvp);
 613         ncp = ap->a_nch->ncp;
 614         asof = dip->obj_asof;
 615         nlen = ncp->nc_nlen;
 616         flags = dip->flags;
 617
 618         for (i = 0; i < nlen; ++i) {
 619                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
 620                         asof = hammer_str_to_tid(ncp->nc_name + i + 2);
 621                         flags |= HAMMER_INODE_RO;
 622                         break;
 623                 }
 624         }
 625         nlen = i;
 626
 627         /*
 628          * If there is no path component the time extension is relative to
 629          * dip.
 630          */
 631         if (nlen == 0) {
 632                 ip = hammer_get_inode(dip->hmp, &dip->cache[1], dip->obj_id,
 633                                       asof, flags, &error);
 634                 if (error == 0) {
 635                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 636                         hammer_rel_inode(ip, 0);
 637                 } else {
 638                         vp = NULL;
 639                 }
 640                 if (error == 0) {
 641                         vn_unlock(vp);
 642                         cache_setvp(ap->a_nch, vp);
 643                         vrele(vp);
 644                 }
 645                 return(error);
 646         }
 647
 648         /*
 649          * Calculate the namekey and setup the key range for the scan.  This
 650          * works kinda like a chained hash table where the lower 32 bits
 651          * of the namekey synthesize the chain.
 652          *
 653          * The key range is inclusive of both key_beg and key_end.
 654          */
 655         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
 656
 657         error = hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
 658         cursor.key_beg.obj_id = dip->obj_id;
 659         cursor.key_beg.key = namekey;
 660         cursor.key_beg.create_tid = 0;
 661         cursor.key_beg.delete_tid = 0;
 662         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 663         cursor.key_beg.obj_type = 0;
 664
 665         cursor.key_end = cursor.key_beg;
 666         cursor.key_end.key |= 0xFFFFFFFFULL;
 667         cursor.asof = asof;
 668         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
 669
 670         /*
 671          * Scan all matching records (the chain), locate the one matching
 672          * the requested path component.
 673          *
 674          * The hammer_ip_*() functions merge in-memory records with on-disk
 675          * records for the purposes of the search.
 676          */
 677         if (error == 0)
 678                 error = hammer_ip_first(&cursor, dip);
 679
 680         rec = NULL;
 681         obj_id = 0;
 682
 683         while (error == 0) {
 684                 error = hammer_ip_resolve_data(&cursor);
 685                 if (error)
 686                         break;
 687                 rec = cursor.record;
 688                 if (nlen == rec->entry.base.data_len &&
 689                     bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
 690                         obj_id = rec->entry.obj_id;
 691                         break;
 692                 }
 693                 error = hammer_ip_next(&cursor);
 694         }
 695         hammer_done_cursor(&cursor);
 696         if (error == 0) {
 697                 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
 698                                       obj_id, asof, flags, &error);
 699                 if (error == 0) {
 700                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 701                         hammer_rel_inode(ip, 0);
 702                 } else {
 703                         vp = NULL;
 704                 }
 705                 if (error == 0) {
 706                         vn_unlock(vp);
 707                         cache_setvp(ap->a_nch, vp);
 708                         vrele(vp);
 709                 }
 710         } else if (error == ENOENT) {
 711                 cache_setvp(ap->a_nch, NULL);
 712         }
 713         return (error);
 714 }
 715
 716 /*
 717  * hammer_vop_nlookupdotdot { dvp, vpp, cred }
 718  *
 719  * Locate the parent directory of a directory vnode.
 720  *
 721  * dvp is referenced but not locked.  *vpp must be returned referenced and
 722  * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
 723  * at the root, instead it could indicate that the directory we were in was
 724  * removed.
 725  *
 726  * NOTE: as-of sequences are not linked into the directory structure.  If
 727  * we are at the root with a different asof then the mount point, reload
 728  * the same directory with the mount point's asof.   I'm not sure what this
 729  * will do to NFS.  We encode ASOF stamps in NFS file handles so it might not
 730  * get confused, but it hasn't been tested.
 731  */
 732 static
 733 int
 734 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
 735 {
 736         struct hammer_inode *dip;
 737         struct hammer_inode *ip;
 738         int64_t parent_obj_id;
 739         hammer_tid_t asof;
 740         int error;
 741
 742         dip = VTOI(ap->a_dvp);
 743         asof = dip->obj_asof;
 744         parent_obj_id = dip->ino_data.parent_obj_id;
 745
 746         if (parent_obj_id == 0) {
 747                 if (dip->obj_id == HAMMER_OBJID_ROOT &&
 748                    asof != dip->hmp->asof) {
 749                         parent_obj_id = dip->obj_id;
 750                         asof = dip->hmp->asof;
 751                         *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
 752                         ksnprintf(*ap->a_fakename, 19, "0x%016llx",
 753                                    dip->obj_asof);
 754                 } else {
 755                         *ap->a_vpp = NULL;
 756                         return ENOENT;
 757                 }
 758         }
 759
 760         ip = hammer_get_inode(dip->hmp, &dip->cache[1], parent_obj_id,
 761                               asof, dip->flags, &error);
 762         if (ip == NULL) {
 763                 *ap->a_vpp = NULL;
 764                 return(error);
 765         }
 766         error = hammer_get_vnode(ip, LK_EXCLUSIVE, ap->a_vpp);
 767         hammer_rel_inode(ip, 0);
 768         return (error);
 769 }
 770
 771 /*
 772  * hammer_vop_nlink { nch, dvp, vp, cred }
 773  */
 774 static
 775 int
 776 hammer_vop_nlink(struct vop_nlink_args *ap)
 777 {
 778         struct hammer_transaction trans;
 779         struct hammer_inode *dip;
 780         struct hammer_inode *ip;
 781         struct nchandle *nch;
 782         int error;
 783
 784         nch = ap->a_nch;
 785         dip = VTOI(ap->a_dvp);
 786         ip = VTOI(ap->a_vp);
 787
 788         if (dip->flags & HAMMER_INODE_RO)
 789                 return (EROFS);
 790         if (ip->flags & HAMMER_INODE_RO)
 791                 return (EROFS);
 792
 793         /*
 794          * Create a transaction to cover the operations we perform.
 795          */
 796         hammer_start_transaction(&trans, dip->hmp);
 797
 798         /*
 799          * Add the filesystem object to the directory.  Note that neither
 800          * dip nor ip are referenced or locked, but their vnodes are
 801          * referenced.  This function will bump the inode's link count.
 802          */
 803         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 804
 805         /*
 806          * Finish up.
 807          */
 808         if (error) {
 809                 hammer_abort_transaction(&trans);
 810         } else {
 811                 cache_setunresolved(nch);
 812                 cache_setvp(nch, ap->a_vp);
 813                 hammer_commit_transaction(&trans);
 814         }
 815         return (error);
 816 }
 817
 818 /*
 819  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 820  *
 821  * The operating system has already ensured that the directory entry
 822  * does not exist and done all appropriate namespace locking.
 823  */
 824 static
 825 int
 826 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 827 {
 828         struct hammer_transaction trans;
 829         struct hammer_inode *dip;
 830         struct hammer_inode *nip;
 831         struct nchandle *nch;
 832         int error;
 833
 834         nch = ap->a_nch;
 835         dip = VTOI(ap->a_dvp);
 836
 837         if (dip->flags & HAMMER_INODE_RO)
 838                 return (EROFS);
 839
 840         /*
 841          * Create a transaction to cover the operations we perform.
 842          */
 843         hammer_start_transaction(&trans, dip->hmp);
 844
 845         /*
 846          * Create a new filesystem object of the requested type.  The
 847          * returned inode will be referenced but not locked.
 848          */
 849         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 850         if (error)
 851                 kprintf("hammer_mkdir error %d\n", error);
 852         if (error) {
 853                 hammer_abort_transaction(&trans);
 854                 *ap->a_vpp = NULL;
 855                 return (error);
 856         }
 857
 858         /*
 859          * Add the new filesystem object to the directory.  This will also
 860          * bump the inode's link count.
 861          */
 862         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 863         if (error)
 864                 kprintf("hammer_mkdir (add) error %d\n", error);
 865
 866         /*
 867          * Finish up.
 868          */
 869         if (error) {
 870                 hammer_rel_inode(nip, 0);
 871                 hammer_abort_transaction(&trans);
 872                 *ap->a_vpp = NULL;
 873         } else {
 874                 hammer_commit_transaction(&trans);
 875                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 876                 hammer_rel_inode(nip, 0);
 877                 if (error == 0) {
 878                         cache_setunresolved(ap->a_nch);
 879                         cache_setvp(ap->a_nch, *ap->a_vpp);
 880                 }
 881         }
 882         return (error);
 883 }
 884
 885 /*
 886  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 887  *
 888  * The operating system has already ensured that the directory entry
 889  * does not exist and done all appropriate namespace locking.
 890  */
 891 static
 892 int
 893 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 894 {
 895         struct hammer_transaction trans;
 896         struct hammer_inode *dip;
 897         struct hammer_inode *nip;
 898         struct nchandle *nch;
 899         int error;
 900
 901         nch = ap->a_nch;
 902         dip = VTOI(ap->a_dvp);
 903
 904         if (dip->flags & HAMMER_INODE_RO)
 905                 return (EROFS);
 906
 907         /*
 908          * Create a transaction to cover the operations we perform.
 909          */
 910         hammer_start_transaction(&trans, dip->hmp);
 911
 912         /*
 913          * Create a new filesystem object of the requested type.  The
 914          * returned inode will be referenced but not locked.
 915          */
 916         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 917         if (error) {
 918                 hammer_abort_transaction(&trans);
 919                 *ap->a_vpp = NULL;
 920                 return (error);
 921         }
 922
 923         /*
 924          * Add the new filesystem object to the directory.  This will also
 925          * bump the inode's link count.
 926          */
 927         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 928
 929         /*
 930          * Finish up.
 931          */
 932         if (error) {
 933                 hammer_rel_inode(nip, 0);
 934                 hammer_abort_transaction(&trans);
 935                 *ap->a_vpp = NULL;
 936         } else {
 937                 hammer_commit_transaction(&trans);
 938                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 939                 hammer_rel_inode(nip, 0);
 940                 if (error == 0) {
 941                         cache_setunresolved(ap->a_nch);
 942                         cache_setvp(ap->a_nch, *ap->a_vpp);
 943                 }
 944         }
 945         return (error);
 946 }
 947
 948 /*
 949  * hammer_vop_open { vp, mode, cred, fp }
 950  */
 951 static
 952 int
 953 hammer_vop_open(struct vop_open_args *ap)
 954 {
 955         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
 956                 return (EROFS);
 957
 958         return(vop_stdopen(ap));
 959 }
 960
 961 /*
 962  * hammer_vop_pathconf { vp, name, retval }
 963  */
 964 static
 965 int
 966 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 967 {
 968         return EOPNOTSUPP;
 969 }
 970
 971 /*
 972  * hammer_vop_print { vp }
 973  */
 974 static
 975 int
 976 hammer_vop_print(struct vop_print_args *ap)
 977 {
 978         return EOPNOTSUPP;
 979 }
 980
 981 /*
 982  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 983  */
 984 static
 985 int
 986 hammer_vop_readdir(struct vop_readdir_args *ap)
 987 {
 988         struct hammer_cursor cursor;
 989         struct hammer_inode *ip;
 990         struct uio *uio;
 991         hammer_record_ondisk_t rec;
 992         hammer_base_elm_t base;
 993         int error;
 994         int cookie_index;
 995         int ncookies;
 996         off_t *cookies;
 997         off_t saveoff;
 998         int r;
 999
1000         ip = VTOI(ap->a_vp);
1001         uio = ap->a_uio;
1002         saveoff = uio->uio_offset;
1003
1004         if (ap->a_ncookies) {
1005                 ncookies = uio->uio_resid / 16 + 1;
1006                 if (ncookies > 1024)
1007                         ncookies = 1024;
1008                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1009                 cookie_index = 0;
1010         } else {
1011                 ncookies = -1;
1012                 cookies = NULL;
1013                 cookie_index = 0;
1014         }
1015
1016         /*
1017          * Handle artificial entries
1018          */
1019         error = 0;
1020         if (saveoff == 0) {
1021                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1022                 if (r)
1023                         goto done;
1024                 if (cookies)
1025                         cookies[cookie_index] = saveoff;
1026                 ++saveoff;
1027                 ++cookie_index;
1028                 if (cookie_index == ncookies)
1029                         goto done;
1030         }
1031         if (saveoff == 1) {
1032                 if (ip->ino_data.parent_obj_id) {
1033                         r = vop_write_dirent(&error, uio,
1034                                              ip->ino_data.parent_obj_id,
1035                                              DT_DIR, 2, "..");
1036                 } else {
1037                         r = vop_write_dirent(&error, uio,
1038                                              ip->obj_id, DT_DIR, 2, "..");
1039                 }
1040                 if (r)
1041                         goto done;
1042                 if (cookies)
1043                         cookies[cookie_index] = saveoff;
1044                 ++saveoff;
1045                 ++cookie_index;
1046                 if (cookie_index == ncookies)
1047                         goto done;
1048         }
1049
1050         /*
1051          * Key range (begin and end inclusive) to scan.  Directory keys
1052          * directly translate to a 64 bit 'seek' position.
1053          */
1054         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1055         cursor.key_beg.obj_id = ip->obj_id;
1056         cursor.key_beg.create_tid = 0;
1057         cursor.key_beg.delete_tid = 0;
1058         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1059         cursor.key_beg.obj_type = 0;
1060         cursor.key_beg.key = saveoff;
1061
1062         cursor.key_end = cursor.key_beg;
1063         cursor.key_end.key = HAMMER_MAX_KEY;
1064         cursor.asof = ip->obj_asof;
1065         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1066
1067         error = hammer_ip_first(&cursor, ip);
1068
1069         while (error == 0) {
1070                 error = hammer_ip_resolve_record_and_data(&cursor);
1071                 if (error)
1072                         break;
1073                 rec = cursor.record;
1074                 base = &rec->base.base;
1075                 saveoff = base->key;
1076
1077                 if (base->obj_id != ip->obj_id)
1078                         panic("readdir: bad record at %p", cursor.node);
1079
1080                 r = vop_write_dirent(
1081                              &error, uio, rec->entry.obj_id,
1082                              hammer_get_dtype(rec->entry.base.base.obj_type),
1083                              rec->entry.base.data_len,
1084                              (void *)cursor.data);
1085                 if (r)
1086                         break;
1087                 ++saveoff;
1088                 if (cookies)
1089                         cookies[cookie_index] = base->key;
1090                 ++cookie_index;
1091                 if (cookie_index == ncookies)
1092                         break;
1093                 error = hammer_ip_next(&cursor);
1094         }
1095         hammer_done_cursor(&cursor);
1096
1097 done:
1098         if (ap->a_eofflag)
1099                 *ap->a_eofflag = (error == ENOENT);
1100         uio->uio_offset = saveoff;
1101         if (error && cookie_index == 0) {
1102                 if (error == ENOENT)
1103                         error = 0;
1104                 if (cookies) {
1105                         kfree(cookies, M_TEMP);
1106                         *ap->a_ncookies = 0;
1107                         *ap->a_cookies = NULL;
1108                 }
1109         } else {
1110                 if (error == ENOENT)
1111                         error = 0;
1112                 if (cookies) {
1113                         *ap->a_ncookies = cookie_index;
1114                         *ap->a_cookies = cookies;
1115                 }
1116         }
1117         return(error);
1118 }
1119
1120 /*
1121  * hammer_vop_readlink { vp, uio, cred }
1122  */
1123 static
1124 int
1125 hammer_vop_readlink(struct vop_readlink_args *ap)
1126 {
1127         struct hammer_cursor cursor;
1128         struct hammer_inode *ip;
1129         int error;
1130
1131         ip = VTOI(ap->a_vp);
1132         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1133
1134         /*
1135          * Key range (begin and end inclusive) to scan.  Directory keys
1136          * directly translate to a 64 bit 'seek' position.
1137          */
1138         cursor.key_beg.obj_id = ip->obj_id;
1139         cursor.key_beg.create_tid = 0;
1140         cursor.key_beg.delete_tid = 0;
1141         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1142         cursor.key_beg.obj_type = 0;
1143         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1144         cursor.asof = ip->obj_asof;
1145         cursor.flags |= HAMMER_CURSOR_ASOF;
1146
1147         error = hammer_ip_lookup(&cursor, ip);
1148         if (error == 0) {
1149                 error = hammer_ip_resolve_data(&cursor);
1150                 if (error == 0) {
1151                         error = uiomove((char *)cursor.data,
1152                                         cursor.record->base.data_len,
1153                                         ap->a_uio);
1154                 }
1155         }
1156         hammer_done_cursor(&cursor);
1157         return(error);
1158 }
1159
1160 /*
1161  * hammer_vop_nremove { nch, dvp, cred }
1162  */
1163 static
1164 int
1165 hammer_vop_nremove(struct vop_nremove_args *ap)
1166 {
1167         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1168 }
1169
1170 /*
1171  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1172  */
1173 static
1174 int
1175 hammer_vop_nrename(struct vop_nrename_args *ap)
1176 {
1177         struct hammer_transaction trans;
1178         struct namecache *fncp;
1179         struct namecache *tncp;
1180         struct hammer_inode *fdip;
1181         struct hammer_inode *tdip;
1182         struct hammer_inode *ip;
1183         struct hammer_cursor cursor;
1184         union hammer_record_ondisk *rec;
1185         int64_t namekey;
1186         int error;
1187
1188         fdip = VTOI(ap->a_fdvp);
1189         tdip = VTOI(ap->a_tdvp);
1190         fncp = ap->a_fnch->ncp;
1191         tncp = ap->a_tnch->ncp;
1192         ip = VTOI(fncp->nc_vp);
1193         KKASSERT(ip != NULL);
1194
1195         if (fdip->flags & HAMMER_INODE_RO)
1196                 return (EROFS);
1197         if (tdip->flags & HAMMER_INODE_RO)
1198                 return (EROFS);
1199         if (ip->flags & HAMMER_INODE_RO)
1200                 return (EROFS);
1201
1202         hammer_start_transaction(&trans, fdip->hmp);
1203
1204         /*
1205          * Remove tncp from the target directory and then link ip as
1206          * tncp. XXX pass trans to dounlink
1207          *
1208          * Force the inode sync-time to match the transaction so it is
1209          * in-sync with the creation of the target directory entry.
1210          */
1211         error = hammer_dounlink(ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1212         if (error == 0 || error == ENOENT) {
1213                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1214                 if (error == 0) {
1215                         ip->ino_data.parent_obj_id = tdip->obj_id;
1216                         hammer_modify_inode(&trans, ip,
1217                                 HAMMER_INODE_DDIRTY | HAMMER_INODE_TIDLOCKED);
1218                 }
1219         }
1220         if (error)
1221                 goto failed; /* XXX */
1222
1223         /*
1224          * Locate the record in the originating directory and remove it.
1225          *
1226          * Calculate the namekey and setup the key range for the scan.  This
1227          * works kinda like a chained hash table where the lower 32 bits
1228          * of the namekey synthesize the chain.
1229          *
1230          * The key range is inclusive of both key_beg and key_end.
1231          */
1232         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1233 retry:
1234         hammer_init_cursor_hmp(&cursor, &fdip->cache[0], fdip->hmp);
1235         cursor.key_beg.obj_id = fdip->obj_id;
1236         cursor.key_beg.key = namekey;
1237         cursor.key_beg.create_tid = 0;
1238         cursor.key_beg.delete_tid = 0;
1239         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1240         cursor.key_beg.obj_type = 0;
1241
1242         cursor.key_end = cursor.key_beg;
1243         cursor.key_end.key |= 0xFFFFFFFFULL;
1244         cursor.asof = fdip->obj_asof;
1245         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1246
1247         /*
1248          * Scan all matching records (the chain), locate the one matching
1249          * the requested path component.
1250          *
1251          * The hammer_ip_*() functions merge in-memory records with on-disk
1252          * records for the purposes of the search.
1253          */
1254         error = hammer_ip_first(&cursor, fdip);
1255         while (error == 0) {
1256                 if (hammer_ip_resolve_data(&cursor) != 0)
1257                         break;
1258                 rec = cursor.record;
1259                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1260                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1261                         break;
1262                 }
1263                 error = hammer_ip_next(&cursor);
1264         }
1265
1266         /*
1267          * If all is ok we have to get the inode so we can adjust nlinks.
1268          *
1269          * WARNING: hammer_ip_del_directory() may have to terminate the
1270          * cursor to avoid a recursion.  It's ok to call hammer_done_cursor()
1271          * twice.
1272          */
1273         if (error == 0)
1274                 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1275         hammer_done_cursor(&cursor);
1276         if (error == 0)
1277                 cache_rename(ap->a_fnch, ap->a_tnch);
1278         if (error == EDEADLK)
1279                 goto retry;
1280 failed:
1281         if (error == 0) {
1282                 hammer_commit_transaction(&trans);
1283         } else {
1284                 hammer_abort_transaction(&trans);
1285         }
1286         return (error);
1287 }
1288
1289 /*
1290  * hammer_vop_nrmdir { nch, dvp, cred }
1291  */
1292 static
1293 int
1294 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1295 {
1296         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1297 }
1298
1299 /*
1300  * hammer_vop_setattr { vp, vap, cred }
1301  */
1302 static
1303 int
1304 hammer_vop_setattr(struct vop_setattr_args *ap)
1305 {
1306         struct hammer_transaction trans;
1307         struct vattr *vap;
1308         struct hammer_inode *ip;
1309         int modflags;
1310         int error;
1311         int truncating;
1312         int64_t aligned_size;
1313         u_int32_t flags;
1314         uuid_t uuid;
1315
1316         vap = ap->a_vap;
1317         ip = ap->a_vp->v_data;
1318         modflags = 0;
1319
1320         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1321                 return(EROFS);
1322         if (ip->flags & HAMMER_INODE_RO)
1323                 return (EROFS);
1324
1325         hammer_start_transaction(&trans, ip->hmp);
1326         error = 0;
1327
1328         if (vap->va_flags != VNOVAL) {
1329                 flags = ip->ino_data.uflags;
1330                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1331                                          hammer_to_unix_xid(&ip->ino_data.uid),
1332                                          ap->a_cred);
1333                 if (error == 0) {
1334                         if (ip->ino_data.uflags != flags) {
1335                                 ip->ino_data.uflags = flags;
1336                                 modflags |= HAMMER_INODE_DDIRTY;
1337                         }
1338                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1339                                 error = 0;
1340                                 goto done;
1341                         }
1342                 }
1343                 goto done;
1344         }
1345         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1346                 error = EPERM;
1347                 goto done;
1348         }
1349         if (vap->va_uid != (uid_t)VNOVAL) {
1350                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1351                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1352                         ip->ino_data.uid = uuid;
1353                         modflags |= HAMMER_INODE_DDIRTY;
1354                 }
1355         }
1356         if (vap->va_gid != (uid_t)VNOVAL) {
1357                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1358                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1359                         ip->ino_data.gid = uuid;
1360                         modflags |= HAMMER_INODE_DDIRTY;
1361                 }
1362         }
1363         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1364                 switch(ap->a_vp->v_type) {
1365                 case VREG:
1366                         if (vap->va_size == ip->ino_rec.ino_size)
1367                                 break;
1368                         if (vap->va_size < ip->ino_rec.ino_size) {
1369                                 vtruncbuf(ap->a_vp, vap->va_size,
1370                                           HAMMER_BUFSIZE);
1371                                 truncating = 1;
1372                         } else {
1373                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1374                                 truncating = 0;
1375                         }
1376                         ip->ino_rec.ino_size = vap->va_size;
1377                         modflags |= HAMMER_INODE_RDIRTY;
1378                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1379                                         ~(int64_t)HAMMER_BUFMASK;
1380
1381                         if (truncating) {
1382                                 error = hammer_ip_delete_range(&trans, ip,
1383                                                     aligned_size,
1384                                                     0x7FFFFFFFFFFFFFFFLL);
1385                         }
1386                         /*
1387                          * If truncating we have to clean out a portion of
1388                          * the last block on-disk.
1389                          */
1390                         if (truncating && error == 0 &&
1391                             vap->va_size < aligned_size) {
1392                                 struct buf *bp;
1393                                 int offset;
1394
1395                                 offset = vap->va_size & HAMMER_BUFMASK;
1396                                 error = bread(ap->a_vp,
1397                                               aligned_size - HAMMER_BUFSIZE,
1398                                               HAMMER_BUFSIZE, &bp);
1399                                 if (error == 0) {
1400                                         bzero(bp->b_data + offset,
1401                                               HAMMER_BUFSIZE - offset);
1402                                         if (bp->b_tid == 0)
1403                                                 bp->b_tid = trans.tid;
1404                                         bdwrite(bp);
1405                                 } else {
1406                                         brelse(bp);
1407                                 }
1408                         }
1409                         break;
1410                 case VDATABASE:
1411                         error = hammer_ip_delete_range(&trans, ip,
1412                                                     vap->va_size,
1413                                                     0x7FFFFFFFFFFFFFFFLL);
1414                         ip->ino_rec.ino_size = vap->va_size;
1415                         modflags |= HAMMER_INODE_RDIRTY;
1416                         break;
1417                 default:
1418                         error = EINVAL;
1419                         goto done;
1420                 }
1421                 break;
1422         }
1423         if (vap->va_atime.tv_sec != VNOVAL) {
1424                 ip->ino_rec.ino_atime =
1425                         hammer_timespec_to_transid(&vap->va_atime);
1426                 modflags |= HAMMER_INODE_ITIMES;
1427         }
1428         if (vap->va_mtime.tv_sec != VNOVAL) {
1429                 ip->ino_rec.ino_mtime =
1430                         hammer_timespec_to_transid(&vap->va_mtime);
1431                 modflags |= HAMMER_INODE_ITIMES;
1432         }
1433         if (vap->va_mode != (mode_t)VNOVAL) {
1434                 if (ip->ino_data.mode != vap->va_mode) {
1435                         ip->ino_data.mode = vap->va_mode;
1436                         modflags |= HAMMER_INODE_DDIRTY;
1437                 }
1438         }
1439 done:
1440         if (error) {
1441                 hammer_abort_transaction(&trans);
1442         } else {
1443                 hammer_modify_inode(&trans, ip, modflags);
1444                 hammer_commit_transaction(&trans);
1445         }
1446         return (error);
1447 }
1448
1449 /*
1450  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1451  */
1452 static
1453 int
1454 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1455 {
1456         struct hammer_transaction trans;
1457         struct hammer_inode *dip;
1458         struct hammer_inode *nip;
1459         struct nchandle *nch;
1460         hammer_record_t record;
1461         int error;
1462         int bytes;
1463
1464         ap->a_vap->va_type = VLNK;
1465
1466         nch = ap->a_nch;
1467         dip = VTOI(ap->a_dvp);
1468
1469         if (dip->flags & HAMMER_INODE_RO)
1470                 return (EROFS);
1471
1472         /*
1473          * Create a transaction to cover the operations we perform.
1474          */
1475         hammer_start_transaction(&trans, dip->hmp);
1476
1477         /*
1478          * Create a new filesystem object of the requested type.  The
1479          * returned inode will be referenced but not locked.
1480          */
1481
1482         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1483         if (error) {
1484                 hammer_abort_transaction(&trans);
1485                 *ap->a_vpp = NULL;
1486                 return (error);
1487         }
1488
1489         /*
1490          * Add the new filesystem object to the directory.  This will also
1491          * bump the inode's link count.
1492          */
1493         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1494
1495         /*
1496          * Add a record representing the symlink.  symlink stores the link
1497          * as pure data, not a string, and is no \0 terminated.
1498          */
1499         if (error == 0) {
1500                 record = hammer_alloc_mem_record(nip);
1501                 bytes = strlen(ap->a_target);
1502
1503                 record->rec.base.base.key = HAMMER_FIXKEY_SYMLINK;
1504                 record->rec.base.base.rec_type = HAMMER_RECTYPE_FIX;
1505                 record->rec.base.data_len = bytes;
1506                 record->data = (void *)ap->a_target;
1507                 /* will be reallocated by routine below */
1508                 error = hammer_ip_add_record(&trans, record);
1509
1510                 /*
1511                  * Set the file size to the length of the link.
1512                  */
1513                 if (error == 0) {
1514                         nip->ino_rec.ino_size = bytes;
1515                         hammer_modify_inode(&trans, nip, HAMMER_INODE_RDIRTY);
1516                 }
1517         }
1518
1519         /*
1520          * Finish up.
1521          */
1522         if (error) {
1523                 hammer_rel_inode(nip, 0);
1524                 hammer_abort_transaction(&trans);
1525                 *ap->a_vpp = NULL;
1526         } else {
1527                 hammer_commit_transaction(&trans);
1528                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1529                 hammer_rel_inode(nip, 0);
1530                 if (error == 0) {
1531                         cache_setunresolved(ap->a_nch);
1532                         cache_setvp(ap->a_nch, *ap->a_vpp);
1533                 }
1534         }
1535         return (error);
1536 }
1537
1538 /*
1539  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1540  */
1541 static
1542 int
1543 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1544 {
1545         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1546 }
1547
1548 /*
1549  * hammer_vop_ioctl { vp, command, data, fflag, cred }
1550  */
1551 static
1552 int
1553 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1554 {
1555         struct hammer_inode *ip = ap->a_vp->v_data;
1556
1557         return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1558                             ap->a_fflag, ap->a_cred));
1559 }
1560
1561 static
1562 int
1563 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1564 {
1565         struct mount *mp;
1566         int error;
1567
1568         mp = ap->a_head.a_ops->head.vv_mount;
1569
1570         switch(ap->a_op) {
1571         case MOUNTCTL_SET_EXPORT:
1572                 if (ap->a_ctllen != sizeof(struct export_args))
1573                         error = EINVAL;
1574                 error = hammer_vfs_export(mp, ap->a_op,
1575                                       (const struct export_args *)ap->a_ctl);
1576                 break;
1577         default:
1578                 error = journal_mountctl(ap);
1579                 break;
1580         }
1581         return(error);
1582 }
1583
1584 /*
1585  * hammer_vop_strategy { vp, bio }
1586  *
1587  * Strategy call, used for regular file read & write only.  Note that the
1588  * bp may represent a cluster.
1589  *
1590  * To simplify operation and allow better optimizations in the future,
1591  * this code does not make any assumptions with regards to buffer alignment
1592  * or size.
1593  */
1594 static
1595 int
1596 hammer_vop_strategy(struct vop_strategy_args *ap)
1597 {
1598         struct buf *bp;
1599         int error;
1600
1601         bp = ap->a_bio->bio_buf;
1602
1603         switch(bp->b_cmd) {
1604         case BUF_CMD_READ:
1605                 error = hammer_vop_strategy_read(ap);
1606                 break;
1607         case BUF_CMD_WRITE:
1608                 error = hammer_vop_strategy_write(ap);
1609                 break;
1610         default:
1611                 error = EINVAL;
1612                 break;
1613         }
1614         bp->b_error = error;
1615         if (error)
1616                 bp->b_flags |= B_ERROR;
1617         biodone(ap->a_bio);
1618         return (error);
1619 }
1620
1621 /*
1622  * Read from a regular file.  Iterate the related records and fill in the
1623  * BIO/BUF.  Gaps are zero-filled.
1624  *
1625  * The support code in hammer_object.c should be used to deal with mixed
1626  * in-memory and on-disk records.
1627  *
1628  * XXX atime update
1629  */
1630 static
1631 int
1632 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1633 {
1634         struct hammer_inode *ip = ap->a_vp->v_data;
1635         struct hammer_cursor cursor;
1636         hammer_record_ondisk_t rec;
1637         hammer_base_elm_t base;
1638         struct bio *bio;
1639         struct buf *bp;
1640         int64_t rec_offset;
1641         int64_t ran_end;
1642         int64_t tmp64;
1643         int error;
1644         int boff;
1645         int roff;
1646         int n;
1647
1648         bio = ap->a_bio;
1649         bp = bio->bio_buf;
1650
1651         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1652
1653         /*
1654          * Key range (begin and end inclusive) to scan.  Note that the key's
1655          * stored in the actual records represent BASE+LEN, not BASE.  The
1656          * first record containing bio_offset will have a key > bio_offset.
1657          */
1658         cursor.key_beg.obj_id = ip->obj_id;
1659         cursor.key_beg.create_tid = 0;
1660         cursor.key_beg.delete_tid = 0;
1661         cursor.key_beg.obj_type = 0;
1662         cursor.key_beg.key = bio->bio_offset + 1;
1663         cursor.asof = ip->obj_asof;
1664         cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1665
1666         cursor.key_end = cursor.key_beg;
1667         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1668                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1669                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1670                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1671         } else {
1672                 ran_end = bio->bio_offset + bp->b_bufsize;
1673                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1674                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1675                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1676                 if (tmp64 < ran_end)
1677                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1678                 else
1679                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1680         }
1681         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1682
1683         error = hammer_ip_first(&cursor, ip);
1684         boff = 0;
1685
1686         while (error == 0) {
1687                 error = hammer_ip_resolve_data(&cursor);
1688                 if (error)
1689                         break;
1690                 rec = cursor.record;
1691                 base = &rec->base.base;
1692
1693                 rec_offset = base->key - rec->data.base.data_len;
1694
1695                 /*
1696                  * Calculate the gap, if any, and zero-fill it.
1697                  */
1698                 n = (int)(rec_offset - (bio->bio_offset + boff));
1699                 if (n > 0) {
1700                         if (n > bp->b_bufsize - boff)
1701                                 n = bp->b_bufsize - boff;
1702                         bzero((char *)bp->b_data + boff, n);
1703                         boff += n;
1704                         n = 0;
1705                 }
1706
1707                 /*
1708                  * Calculate the data offset in the record and the number
1709                  * of bytes we can copy.
1710                  *
1711                  * Note there is a degenerate case here where boff may
1712                  * already be at bp->b_bufsize.
1713                  */
1714                 roff = -n;
1715                 n = rec->data.base.data_len - roff;
1716                 KKASSERT(n > 0);
1717                 if (n > bp->b_bufsize - boff)
1718                         n = bp->b_bufsize - boff;
1719                 bcopy((char *)cursor.data + roff,
1720                       (char *)bp->b_data + boff, n);
1721                 boff += n;
1722                 if (boff == bp->b_bufsize)
1723                         break;
1724                 error = hammer_ip_next(&cursor);
1725         }
1726         hammer_done_cursor(&cursor);
1727
1728         /*
1729          * There may have been a gap after the last record
1730          */
1731         if (error == ENOENT)
1732                 error = 0;
1733         if (error == 0 && boff != bp->b_bufsize) {
1734                 KKASSERT(boff < bp->b_bufsize);
1735                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1736                 /* boff = bp->b_bufsize; */
1737         }
1738         bp->b_resid = 0;
1739         return(error);
1740 }
1741
1742 /*
1743  * Write to a regular file.  Iterate the related records and mark for
1744  * deletion.  If existing edge records (left and right side) overlap our
1745  * write they have to be marked deleted and new records created, usually
1746  * referencing a portion of the original data.  Then add a record to
1747  * represent the buffer.
1748  *
1749  * The support code in hammer_object.c should be used to deal with mixed
1750  * in-memory and on-disk records.
1751  */
1752 static
1753 int
1754 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1755 {
1756         struct hammer_transaction trans;
1757         hammer_inode_t ip;
1758         struct bio *bio;
1759         struct buf *bp;
1760         int error;
1761
1762         bio = ap->a_bio;
1763         bp = bio->bio_buf;
1764         ip = ap->a_vp->v_data;
1765
1766         if (ip->flags & HAMMER_INODE_RO)
1767                 return (EROFS);
1768
1769         /*
1770          * Start a transaction using the TID stored with the bp.
1771          */
1772         KKASSERT(bp->b_tid != 0);
1773         hammer_start_transaction_tid(&trans, ip->hmp, bp->b_tid);
1774
1775         /*
1776          * Delete any records overlapping our range.  This function will
1777          * (eventually) properly truncate partial overlaps.
1778          */
1779         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1780                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1781                                                bio->bio_offset);
1782         } else {
1783                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1784                                                bio->bio_offset +
1785                                                 bp->b_bufsize - 1);
1786         }
1787
1788         /*
1789          * Add a single record to cover the write.  We can write a record
1790          * with only the actual file data - for example, a small 200 byte
1791          * file does not have to write out a 16K record.
1792          *
1793          * While the data size does not have to be aligned, we still do it
1794          * to reduce fragmentation in a future allocation model.
1795          */
1796         if (error == 0) {
1797                 int limit_size;
1798
1799                 if (ip->ino_rec.ino_size - bio->bio_offset > bp->b_bufsize) {
1800                         limit_size = bp->b_bufsize;
1801                 } else {
1802                         limit_size = (int)(ip->ino_rec.ino_size -
1803                                            bio->bio_offset);
1804                         KKASSERT(limit_size >= 0);
1805                         limit_size = (limit_size + 63) & ~63;
1806                 }
1807                 error = hammer_ip_sync_data(&trans, ip, bio->bio_offset,
1808                                             bp->b_data, limit_size);
1809         }
1810
1811         /*
1812          * If an error occured abort the transaction
1813          */
1814         if (error) {
1815                 /* XXX undo deletion */
1816                 hammer_abort_transaction(&trans);
1817                 bp->b_resid = bp->b_bufsize;
1818         } else {
1819                 hammer_commit_transaction(&trans);
1820                 bp->b_resid = 0;
1821                 bp->b_tid = 0;
1822         }
1823         return(error);
1824 }
1825
1826 /*
1827  * dounlink - disconnect a directory entry
1828  *
1829  * XXX whiteout support not really in yet
1830  */
1831 static int
1832 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1833                 int flags)
1834 {
1835         struct hammer_transaction trans;
1836         struct namecache *ncp;
1837         hammer_inode_t dip;
1838         hammer_inode_t ip;
1839         hammer_record_ondisk_t rec;
1840         struct hammer_cursor cursor;
1841         int64_t namekey;
1842         int error;
1843
1844         /*
1845          * Calculate the namekey and setup the key range for the scan.  This
1846          * works kinda like a chained hash table where the lower 32 bits
1847          * of the namekey synthesize the chain.
1848          *
1849          * The key range is inclusive of both key_beg and key_end.
1850          */
1851         dip = VTOI(dvp);
1852         ncp = nch->ncp;
1853
1854         if (dip->flags & HAMMER_INODE_RO)
1855                 return (EROFS);
1856
1857         hammer_start_transaction(&trans, dip->hmp);
1858
1859         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1860 retry:
1861         hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
1862         cursor.key_beg.obj_id = dip->obj_id;
1863         cursor.key_beg.key = namekey;
1864         cursor.key_beg.create_tid = 0;
1865         cursor.key_beg.delete_tid = 0;
1866         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1867         cursor.key_beg.obj_type = 0;
1868
1869         cursor.key_end = cursor.key_beg;
1870         cursor.key_end.key |= 0xFFFFFFFFULL;
1871         cursor.asof = dip->obj_asof;
1872         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1873
1874         /*
1875          * Scan all matching records (the chain), locate the one matching
1876          * the requested path component.  info->last_error contains the
1877          * error code on search termination and could be 0, ENOENT, or
1878          * something else.
1879          *
1880          * The hammer_ip_*() functions merge in-memory records with on-disk
1881          * records for the purposes of the search.
1882          */
1883         error = hammer_ip_first(&cursor, dip);
1884         while (error == 0) {
1885                 error = hammer_ip_resolve_data(&cursor);
1886                 if (error)
1887                         break;
1888                 rec = cursor.record;
1889                 if (ncp->nc_nlen == rec->entry.base.data_len &&
1890                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1891                         break;
1892                 }
1893                 error = hammer_ip_next(&cursor);
1894         }
1895
1896         /*
1897          * If all is ok we have to get the inode so we can adjust nlinks.
1898          *
1899          * If the target is a directory, it must be empty.
1900          */
1901         if (error == 0) {
1902                 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
1903                                       rec->entry.obj_id,
1904                                       dip->hmp->asof, 0, &error);
1905                 if (error == ENOENT) {
1906                         kprintf("obj_id %016llx\n", rec->entry.obj_id);
1907                         Debugger("ENOENT unlinking object that should exist, cont to sync");
1908                         hammer_sync_hmp(dip->hmp, MNT_NOWAIT);
1909                         Debugger("ENOENT - sync done");
1910                 }
1911                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
1912                                   HAMMER_OBJTYPE_DIRECTORY) {
1913                         error = hammer_ip_check_directory_empty(&trans, ip);
1914                 }
1915                 /*
1916                  * WARNING: hammer_ip_del_directory() may have to terminate
1917                  * the cursor to avoid a lock recursion.  It's ok to call
1918                  * hammer_done_cursor() twice.
1919                  */
1920                 if (error == 0)
1921                         error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1922                 if (error == 0) {
1923                         cache_setunresolved(nch);
1924                         cache_setvp(nch, NULL);
1925                         /* XXX locking */
1926                         if (ip->vp)
1927                                 cache_inval_vp(ip->vp, CINV_DESTROY);
1928                 }
1929                 hammer_rel_inode(ip, 0);
1930         }
1931         hammer_done_cursor(&cursor);
1932         if (error == EDEADLK)
1933                 goto retry;
1934
1935         if (error == 0)
1936                 hammer_commit_transaction(&trans);
1937         else
1938                 hammer_abort_transaction(&trans);
1939         return (error);
1940 }
1941
1942 /************************************************************************
1943  *                          FIFO AND SPECFS OPS                         *
1944  ************************************************************************
1945  *
1946  */
1947
1948 static int
1949 hammer_vop_fifoclose (struct vop_close_args *ap)
1950 {
1951         /* XXX update itimes */
1952         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1953 }
1954
1955 static int
1956 hammer_vop_fiforead (struct vop_read_args *ap)
1957 {
1958         int error;
1959
1960         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1961         /* XXX update access time */
1962         return (error);
1963 }
1964
1965 static int
1966 hammer_vop_fifowrite (struct vop_write_args *ap)
1967 {
1968         int error;
1969
1970         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1971         /* XXX update access time */
1972         return (error);
1973 }
1974
1975 static int
1976 hammer_vop_specclose (struct vop_close_args *ap)
1977 {
1978         /* XXX update itimes */
1979         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1980 }
1981
1982 static int
1983 hammer_vop_specread (struct vop_read_args *ap)
1984 {
1985         /* XXX update access time */
1986         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1987 }
1988
1989 static int
1990 hammer_vop_specwrite (struct vop_write_args *ap)
1991 {
1992         /* XXX update last change time */
1993         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1994 }
1995