sys/vfs/hammer/hammer_vnops.c

   1 /*
   2  * Copyright (c) 2007-2008 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.53 2008/05/12 23:15:46 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(hammer_transaction_t trans, struct nchandle *nch,
 153                            struct vnode *dvp, 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 = VTOI(ap->a_vp);
 174
 175         hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
 176         vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL);
 177         if (ap->a_waitfor == MNT_WAIT)
 178                 hammer_wait_inode(ip);
 179         return (ip->error);
 180 }
 181
 182 /*
 183  * hammer_vop_read { vp, uio, ioflag, cred }
 184  */
 185 static
 186 int
 187 hammer_vop_read(struct vop_read_args *ap)
 188 {
 189         struct hammer_transaction trans;
 190         hammer_inode_t ip;
 191         off_t offset;
 192         struct buf *bp;
 193         struct uio *uio;
 194         int error;
 195         int n;
 196         int seqcount;
 197
 198         if (ap->a_vp->v_type != VREG)
 199                 return (EINVAL);
 200         ip = VTOI(ap->a_vp);
 201         error = 0;
 202         seqcount = ap->a_ioflag >> 16;
 203
 204         hammer_start_transaction(&trans, ip->hmp);
 205
 206         /*
 207          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 208          */
 209         uio = ap->a_uio;
 210         while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_data.size) {
 211                 offset = uio->uio_offset & HAMMER_BUFMASK;
 212 #if 0
 213                 error = cluster_read(ap->a_vp, ip->ino_data.size,
 214                                      uio->uio_offset - offset, HAMMER_BUFSIZE,
 215                                      MAXBSIZE, seqcount, &bp);
 216 #endif
 217                 error = bread(ap->a_vp, uio->uio_offset - offset,
 218                               HAMMER_BUFSIZE, &bp);
 219                 if (error) {
 220                         brelse(bp);
 221                         break;
 222                 }
 223                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 224                 n = HAMMER_BUFSIZE - offset;
 225                 if (n > uio->uio_resid)
 226                         n = uio->uio_resid;
 227                 if (n > ip->ino_data.size - uio->uio_offset)
 228                         n = (int)(ip->ino_data.size - uio->uio_offset);
 229                 error = uiomove((char *)bp->b_data + offset, n, uio);
 230                 if (error) {
 231                         bqrelse(bp);
 232                         break;
 233                 }
 234                 bqrelse(bp);
 235         }
 236         if ((ip->flags & HAMMER_INODE_RO) == 0 &&
 237             (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
 238                 ip->ino_leaf.atime = trans.time;
 239                 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
 240         }
 241         hammer_done_transaction(&trans);
 242         return (error);
 243 }
 244
 245 /*
 246  * hammer_vop_write { vp, uio, ioflag, cred }
 247  */
 248 static
 249 int
 250 hammer_vop_write(struct vop_write_args *ap)
 251 {
 252         struct hammer_transaction trans;
 253         struct hammer_inode *ip;
 254         struct uio *uio;
 255         off_t offset;
 256         struct buf *bp;
 257         int error;
 258         int n;
 259         int flags;
 260         int count;
 261
 262         if (ap->a_vp->v_type != VREG)
 263                 return (EINVAL);
 264         ip = VTOI(ap->a_vp);
 265         error = 0;
 266
 267         if (ip->flags & HAMMER_INODE_RO)
 268                 return (EROFS);
 269
 270         /*
 271          * Create a transaction to cover the operations we perform.
 272          */
 273         hammer_start_transaction(&trans, ip->hmp);
 274         uio = ap->a_uio;
 275
 276         /*
 277          * Check append mode
 278          */
 279         if (ap->a_ioflag & IO_APPEND)
 280                 uio->uio_offset = ip->ino_data.size;
 281
 282         /*
 283          * Check for illegal write offsets.  Valid range is 0...2^63-1
 284          */
 285         if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
 286                 hammer_done_transaction(&trans);
 287                 return (EFBIG);
 288         }
 289
 290         /*
 291          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 292          */
 293         count = 0;
 294         while (uio->uio_resid > 0) {
 295                 int fixsize = 0;
 296
 297                 /*
 298                  * Do not allow huge writes to deadlock the buffer cache
 299                  */
 300                 if ((++count & 15) == 0) {
 301                         vn_unlock(ap->a_vp);
 302                         if ((ap->a_ioflag & IO_NOBWILL) == 0)
 303                                 bwillwrite();
 304                         vn_lock(ap->a_vp, LK_EXCLUSIVE|LK_RETRY);
 305                 }
 306
 307                 offset = uio->uio_offset & HAMMER_BUFMASK;
 308                 n = HAMMER_BUFSIZE - offset;
 309                 if (n > uio->uio_resid)
 310                         n = uio->uio_resid;
 311                 if (uio->uio_offset + n > ip->ino_data.size) {
 312                         vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
 313                         fixsize = 1;
 314                 }
 315
 316                 if (uio->uio_segflg == UIO_NOCOPY) {
 317                         /*
 318                          * Issuing a write with the same data backing the
 319                          * buffer.  Instantiate the buffer to collect the
 320                          * backing vm pages, then read-in any missing bits.
 321                          *
 322                          * This case is used by vop_stdputpages().
 323                          */
 324                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 325                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 326                         if ((bp->b_flags & B_CACHE) == 0) {
 327                                 bqrelse(bp);
 328                                 error = bread(ap->a_vp,
 329                                               uio->uio_offset - offset,
 330                                               HAMMER_BUFSIZE, &bp);
 331                         }
 332                 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
 333                         /*
 334                          * Even though we are entirely overwriting the buffer
 335                          * we may still have to zero it out to avoid a
 336                          * mmap/write visibility issue.
 337                          */
 338                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 339                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 340                         if ((bp->b_flags & B_CACHE) == 0)
 341                                 vfs_bio_clrbuf(bp);
 342                 } else if (uio->uio_offset - offset >= ip->ino_data.size) {
 343                         /*
 344                          * If the base offset of the buffer is beyond the
 345                          * file EOF, we don't have to issue a read.
 346                          */
 347                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 348                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 349                         vfs_bio_clrbuf(bp);
 350                 } else {
 351                         /*
 352                          * Partial overwrite, read in any missing bits then
 353                          * replace the portion being written.
 354                          */
 355                         error = bread(ap->a_vp, uio->uio_offset - offset,
 356                                       HAMMER_BUFSIZE, &bp);
 357                         if (error == 0)
 358                                 bheavy(bp);
 359                 }
 360                 if (error == 0)
 361                         error = uiomove((char *)bp->b_data + offset, n, uio);
 362
 363                 /*
 364                  * If we screwed up we have to undo any VM size changes we
 365                  * made.
 366                  */
 367                 if (error) {
 368                         brelse(bp);
 369                         if (fixsize) {
 370                                 vtruncbuf(ap->a_vp, ip->ino_data.size,
 371                                           HAMMER_BUFSIZE);
 372                         }
 373                         break;
 374                 }
 375                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 376                 if (ip->ino_data.size < uio->uio_offset) {
 377                         ip->ino_data.size = uio->uio_offset;
 378                         flags = HAMMER_INODE_DDIRTY;
 379                         vnode_pager_setsize(ap->a_vp, ip->ino_data.size);
 380                 } else {
 381                         flags = 0;
 382                 }
 383                 ip->ino_data.mtime = trans.time;
 384                 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
 385                 flags |= HAMMER_INODE_DDIRTY;   /* XXX mtime */
 386                 hammer_modify_inode(&trans, ip, flags);
 387
 388                 if (ap->a_ioflag & IO_SYNC) {
 389                         bwrite(bp);
 390                 } else if (ap->a_ioflag & IO_DIRECT) {
 391                         bawrite(bp);
 392 #if 0
 393                 } else if ((ap->a_ioflag >> 16) == IO_SEQMAX &&
 394                            (uio->uio_offset & HAMMER_BUFMASK) == 0) {
 395                         /*
 396                          * XXX HAMMER can only fsync the whole inode,
 397                          * doing it on every buffer would be a bad idea.
 398                          */
 399                         /*
 400                          * If seqcount indicates sequential operation and
 401                          * we just finished filling a buffer, push it out
 402                          * now to prevent the buffer cache from becoming
 403                          * too full, which would trigger non-optimal
 404                          * flushes.
 405                          */
 406                         bdwrite(bp);
 407 #endif
 408                 } else {
 409                         bdwrite(bp);
 410                 }
 411         }
 412         hammer_done_transaction(&trans);
 413         return (error);
 414 }
 415
 416 /*
 417  * hammer_vop_access { vp, mode, cred }
 418  */
 419 static
 420 int
 421 hammer_vop_access(struct vop_access_args *ap)
 422 {
 423         struct hammer_inode *ip = VTOI(ap->a_vp);
 424         uid_t uid;
 425         gid_t gid;
 426         int error;
 427
 428         uid = hammer_to_unix_xid(&ip->ino_data.uid);
 429         gid = hammer_to_unix_xid(&ip->ino_data.gid);
 430
 431         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
 432                                   ip->ino_data.uflags);
 433         return (error);
 434 }
 435
 436 /*
 437  * hammer_vop_advlock { vp, id, op, fl, flags }
 438  */
 439 static
 440 int
 441 hammer_vop_advlock(struct vop_advlock_args *ap)
 442 {
 443         struct hammer_inode *ip = VTOI(ap->a_vp);
 444
 445         return (lf_advlock(ap, &ip->advlock, ip->ino_data.size));
 446 }
 447
 448 /*
 449  * hammer_vop_close { vp, fflag }
 450  */
 451 static
 452 int
 453 hammer_vop_close(struct vop_close_args *ap)
 454 {
 455         return (vop_stdclose(ap));
 456 }
 457
 458 /*
 459  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
 460  *
 461  * The operating system has already ensured that the directory entry
 462  * does not exist and done all appropriate namespace locking.
 463  */
 464 static
 465 int
 466 hammer_vop_ncreate(struct vop_ncreate_args *ap)
 467 {
 468         struct hammer_transaction trans;
 469         struct hammer_inode *dip;
 470         struct hammer_inode *nip;
 471         struct nchandle *nch;
 472         int error;
 473
 474         nch = ap->a_nch;
 475         dip = VTOI(ap->a_dvp);
 476
 477         if (dip->flags & HAMMER_INODE_RO)
 478                 return (EROFS);
 479
 480         /*
 481          * Create a transaction to cover the operations we perform.
 482          */
 483         hammer_start_transaction(&trans, dip->hmp);
 484
 485         /*
 486          * Create a new filesystem object of the requested type.  The
 487          * returned inode will be referenced and shared-locked to prevent
 488          * it from being moved to the flusher.
 489          */
 490
 491         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 492         if (error) {
 493                 hkprintf("hammer_create_inode error %d\n", error);
 494                 hammer_done_transaction(&trans);
 495                 *ap->a_vpp = NULL;
 496                 return (error);
 497         }
 498
 499         /*
 500          * Add the new filesystem object to the directory.  This will also
 501          * bump the inode's link count.
 502          */
 503         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 504         if (error)
 505                 hkprintf("hammer_ip_add_directory error %d\n", error);
 506
 507         /*
 508          * Finish up.
 509          */
 510         if (error) {
 511                 hammer_rel_inode(nip, 0);
 512                 hammer_done_transaction(&trans);
 513                 *ap->a_vpp = NULL;
 514         } else {
 515                 error = hammer_get_vnode(nip, ap->a_vpp);
 516                 hammer_done_transaction(&trans);
 517                 hammer_rel_inode(nip, 0);
 518                 if (error == 0) {
 519                         cache_setunresolved(ap->a_nch);
 520                         cache_setvp(ap->a_nch, *ap->a_vpp);
 521                 }
 522         }
 523         return (error);
 524 }
 525
 526 /*
 527  * hammer_vop_getattr { vp, vap }
 528  *
 529  * Retrieve an inode's attribute information.  When accessing inodes
 530  * historically we fake the atime field to ensure consistent results.
 531  * The atime field is stored in the B-Tree element and allowed to be
 532  * updated without cycling the element.
 533  */
 534 static
 535 int
 536 hammer_vop_getattr(struct vop_getattr_args *ap)
 537 {
 538         struct hammer_inode *ip = VTOI(ap->a_vp);
 539         struct vattr *vap = ap->a_vap;
 540
 541 #if 0
 542         if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
 543             (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
 544             ip->obj_asof == XXX
 545         ) {
 546                 /* LAZYMOD XXX */
 547         }
 548         hammer_itimes(ap->a_vp);
 549 #endif
 550
 551         vap->va_fsid = ip->hmp->fsid_udev;
 552         vap->va_fileid = ip->ino_leaf.base.obj_id;
 553         vap->va_mode = ip->ino_data.mode;
 554         vap->va_nlink = ip->ino_data.nlinks;
 555         vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
 556         vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
 557         vap->va_rmajor = 0;
 558         vap->va_rminor = 0;
 559         vap->va_size = ip->ino_data.size;
 560         if (ip->flags & HAMMER_INODE_RO)
 561                 hammer_to_timespec(ip->ino_data.mtime, &vap->va_atime);
 562         else
 563                 hammer_to_timespec(ip->ino_leaf.atime, &vap->va_atime);
 564         hammer_to_timespec(ip->ino_data.mtime, &vap->va_mtime);
 565         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
 566         vap->va_flags = ip->ino_data.uflags;
 567         vap->va_gen = 1;        /* hammer inums are unique for all time */
 568         vap->va_blocksize = HAMMER_BUFSIZE;
 569         vap->va_bytes = (ip->ino_data.size + 63) & ~63;
 570         vap->va_type = hammer_get_vnode_type(ip->ino_data.obj_type);
 571         vap->va_filerev = 0;    /* XXX */
 572         /* mtime uniquely identifies any adjustments made to the file */
 573         vap->va_fsmid = ip->ino_data.mtime;
 574         vap->va_uid_uuid = ip->ino_data.uid;
 575         vap->va_gid_uuid = ip->ino_data.gid;
 576         vap->va_fsid_uuid = ip->hmp->fsid;
 577         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 578                           VA_FSID_UUID_VALID;
 579
 580         switch (ip->ino_data.obj_type) {
 581         case HAMMER_OBJTYPE_CDEV:
 582         case HAMMER_OBJTYPE_BDEV:
 583                 vap->va_rmajor = ip->ino_data.rmajor;
 584                 vap->va_rminor = ip->ino_data.rminor;
 585                 break;
 586         default:
 587                 break;
 588         }
 589
 590         return(0);
 591 }
 592
 593 /*
 594  * hammer_vop_nresolve { nch, dvp, cred }
 595  *
 596  * Locate the requested directory entry.
 597  */
 598 static
 599 int
 600 hammer_vop_nresolve(struct vop_nresolve_args *ap)
 601 {
 602         struct hammer_transaction trans;
 603         struct namecache *ncp;
 604         hammer_inode_t dip;
 605         hammer_inode_t ip;
 606         hammer_tid_t asof;
 607         struct hammer_cursor cursor;
 608         struct vnode *vp;
 609         int64_t namekey;
 610         int error;
 611         int i;
 612         int nlen;
 613         int flags;
 614         u_int64_t obj_id;
 615
 616         /*
 617          * Misc initialization, plus handle as-of name extensions.  Look for
 618          * the '@@' extension.  Note that as-of files and directories cannot
 619          * be modified.
 620          */
 621         dip = VTOI(ap->a_dvp);
 622         ncp = ap->a_nch->ncp;
 623         asof = dip->obj_asof;
 624         nlen = ncp->nc_nlen;
 625         flags = dip->flags;
 626
 627         hammer_simple_transaction(&trans, dip->hmp);
 628
 629         for (i = 0; i < nlen; ++i) {
 630                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
 631                         asof = hammer_str_to_tid(ncp->nc_name + i + 2);
 632                         flags |= HAMMER_INODE_RO;
 633                         break;
 634                 }
 635         }
 636         nlen = i;
 637
 638         /*
 639          * If there is no path component the time extension is relative to
 640          * dip.
 641          */
 642         if (nlen == 0) {
 643                 ip = hammer_get_inode(&trans, &dip->cache[1], dip->obj_id,
 644                                       asof, flags, &error);
 645                 if (error == 0) {
 646                         error = hammer_get_vnode(ip, &vp);
 647                         hammer_rel_inode(ip, 0);
 648                 } else {
 649                         vp = NULL;
 650                 }
 651                 if (error == 0) {
 652                         vn_unlock(vp);
 653                         cache_setvp(ap->a_nch, vp);
 654                         vrele(vp);
 655                 }
 656                 goto done;
 657         }
 658
 659         /*
 660          * Calculate the namekey and setup the key range for the scan.  This
 661          * works kinda like a chained hash table where the lower 32 bits
 662          * of the namekey synthesize the chain.
 663          *
 664          * The key range is inclusive of both key_beg and key_end.
 665          */
 666         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
 667
 668         error = hammer_init_cursor(&trans, &cursor, &dip->cache[0], dip);
 669         cursor.key_beg.obj_id = dip->obj_id;
 670         cursor.key_beg.key = namekey;
 671         cursor.key_beg.create_tid = 0;
 672         cursor.key_beg.delete_tid = 0;
 673         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 674         cursor.key_beg.obj_type = 0;
 675
 676         cursor.key_end = cursor.key_beg;
 677         cursor.key_end.key |= 0xFFFFFFFFULL;
 678         cursor.asof = asof;
 679         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
 680
 681         /*
 682          * Scan all matching records (the chain), locate the one matching
 683          * the requested path component.
 684          *
 685          * The hammer_ip_*() functions merge in-memory records with on-disk
 686          * records for the purposes of the search.
 687          */
 688         obj_id = 0;
 689
 690         if (error == 0) {
 691                 error = hammer_ip_first(&cursor);
 692                 while (error == 0) {
 693                         error = hammer_ip_resolve_data(&cursor);
 694                         if (error)
 695                                 break;
 696                         if (nlen == cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF &&
 697                             bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
 698                                 obj_id = cursor.data->entry.obj_id;
 699                                 break;
 700                         }
 701                         error = hammer_ip_next(&cursor);
 702                 }
 703         }
 704         hammer_done_cursor(&cursor);
 705         if (error == 0) {
 706                 ip = hammer_get_inode(&trans, &dip->cache[1],
 707                                       obj_id, asof, flags, &error);
 708                 if (error == 0) {
 709                         error = hammer_get_vnode(ip, &vp);
 710                         hammer_rel_inode(ip, 0);
 711                 } else {
 712                         vp = NULL;
 713                 }
 714                 if (error == 0) {
 715                         vn_unlock(vp);
 716                         cache_setvp(ap->a_nch, vp);
 717                         vrele(vp);
 718                 }
 719         } else if (error == ENOENT) {
 720                 cache_setvp(ap->a_nch, NULL);
 721         }
 722 done:
 723         hammer_done_transaction(&trans);
 724         return (error);
 725 }
 726
 727 /*
 728  * hammer_vop_nlookupdotdot { dvp, vpp, cred }
 729  *
 730  * Locate the parent directory of a directory vnode.
 731  *
 732  * dvp is referenced but not locked.  *vpp must be returned referenced and
 733  * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
 734  * at the root, instead it could indicate that the directory we were in was
 735  * removed.
 736  *
 737  * NOTE: as-of sequences are not linked into the directory structure.  If
 738  * we are at the root with a different asof then the mount point, reload
 739  * the same directory with the mount point's asof.   I'm not sure what this
 740  * will do to NFS.  We encode ASOF stamps in NFS file handles so it might not
 741  * get confused, but it hasn't been tested.
 742  */
 743 static
 744 int
 745 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
 746 {
 747         struct hammer_transaction trans;
 748         struct hammer_inode *dip;
 749         struct hammer_inode *ip;
 750         int64_t parent_obj_id;
 751         hammer_tid_t asof;
 752         int error;
 753
 754         dip = VTOI(ap->a_dvp);
 755         asof = dip->obj_asof;
 756         parent_obj_id = dip->ino_data.parent_obj_id;
 757
 758         if (parent_obj_id == 0) {
 759                 if (dip->obj_id == HAMMER_OBJID_ROOT &&
 760                    asof != dip->hmp->asof) {
 761                         parent_obj_id = dip->obj_id;
 762                         asof = dip->hmp->asof;
 763                         *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
 764                         ksnprintf(*ap->a_fakename, 19, "0x%016llx",
 765                                    dip->obj_asof);
 766                 } else {
 767                         *ap->a_vpp = NULL;
 768                         return ENOENT;
 769                 }
 770         }
 771
 772         hammer_simple_transaction(&trans, dip->hmp);
 773
 774         ip = hammer_get_inode(&trans, &dip->cache[1], parent_obj_id,
 775                               asof, dip->flags, &error);
 776         if (ip) {
 777                 error = hammer_get_vnode(ip, ap->a_vpp);
 778                 hammer_rel_inode(ip, 0);
 779         } else {
 780                 *ap->a_vpp = NULL;
 781         }
 782         hammer_done_transaction(&trans);
 783         return (error);
 784 }
 785
 786 /*
 787  * hammer_vop_nlink { nch, dvp, vp, cred }
 788  */
 789 static
 790 int
 791 hammer_vop_nlink(struct vop_nlink_args *ap)
 792 {
 793         struct hammer_transaction trans;
 794         struct hammer_inode *dip;
 795         struct hammer_inode *ip;
 796         struct nchandle *nch;
 797         int error;
 798
 799         nch = ap->a_nch;
 800         dip = VTOI(ap->a_dvp);
 801         ip = VTOI(ap->a_vp);
 802
 803         if (dip->flags & HAMMER_INODE_RO)
 804                 return (EROFS);
 805         if (ip->flags & HAMMER_INODE_RO)
 806                 return (EROFS);
 807
 808         /*
 809          * Create a transaction to cover the operations we perform.
 810          */
 811         hammer_start_transaction(&trans, dip->hmp);
 812
 813         /*
 814          * Add the filesystem object to the directory.  Note that neither
 815          * dip nor ip are referenced or locked, but their vnodes are
 816          * referenced.  This function will bump the inode's link count.
 817          */
 818         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 819
 820         /*
 821          * Finish up.
 822          */
 823         if (error == 0) {
 824                 cache_setunresolved(nch);
 825                 cache_setvp(nch, ap->a_vp);
 826         }
 827         hammer_done_transaction(&trans);
 828         return (error);
 829 }
 830
 831 /*
 832  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 833  *
 834  * The operating system has already ensured that the directory entry
 835  * does not exist and done all appropriate namespace locking.
 836  */
 837 static
 838 int
 839 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 840 {
 841         struct hammer_transaction trans;
 842         struct hammer_inode *dip;
 843         struct hammer_inode *nip;
 844         struct nchandle *nch;
 845         int error;
 846
 847         nch = ap->a_nch;
 848         dip = VTOI(ap->a_dvp);
 849
 850         if (dip->flags & HAMMER_INODE_RO)
 851                 return (EROFS);
 852
 853         /*
 854          * Create a transaction to cover the operations we perform.
 855          */
 856         hammer_start_transaction(&trans, dip->hmp);
 857
 858         /*
 859          * Create a new filesystem object of the requested type.  The
 860          * returned inode will be referenced but not locked.
 861          */
 862         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 863         if (error) {
 864                 hkprintf("hammer_mkdir error %d\n", error);
 865                 hammer_done_transaction(&trans);
 866                 *ap->a_vpp = NULL;
 867                 return (error);
 868         }
 869         /*
 870          * Add the new filesystem object to the directory.  This will also
 871          * bump the inode's link count.
 872          */
 873         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 874         if (error)
 875                 hkprintf("hammer_mkdir (add) error %d\n", error);
 876
 877         /*
 878          * Finish up.
 879          */
 880         if (error) {
 881                 hammer_rel_inode(nip, 0);
 882                 *ap->a_vpp = NULL;
 883         } else {
 884                 error = hammer_get_vnode(nip, ap->a_vpp);
 885                 hammer_rel_inode(nip, 0);
 886                 if (error == 0) {
 887                         cache_setunresolved(ap->a_nch);
 888                         cache_setvp(ap->a_nch, *ap->a_vpp);
 889                 }
 890         }
 891         hammer_done_transaction(&trans);
 892         return (error);
 893 }
 894
 895 /*
 896  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 897  *
 898  * The operating system has already ensured that the directory entry
 899  * does not exist and done all appropriate namespace locking.
 900  */
 901 static
 902 int
 903 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 904 {
 905         struct hammer_transaction trans;
 906         struct hammer_inode *dip;
 907         struct hammer_inode *nip;
 908         struct nchandle *nch;
 909         int error;
 910
 911         nch = ap->a_nch;
 912         dip = VTOI(ap->a_dvp);
 913
 914         if (dip->flags & HAMMER_INODE_RO)
 915                 return (EROFS);
 916
 917         /*
 918          * Create a transaction to cover the operations we perform.
 919          */
 920         hammer_start_transaction(&trans, dip->hmp);
 921
 922         /*
 923          * Create a new filesystem object of the requested type.  The
 924          * returned inode will be referenced but not locked.
 925          */
 926         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 927         if (error) {
 928                 hammer_done_transaction(&trans);
 929                 *ap->a_vpp = NULL;
 930                 return (error);
 931         }
 932
 933         /*
 934          * Add the new filesystem object to the directory.  This will also
 935          * bump the inode's link count.
 936          */
 937         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 938
 939         /*
 940          * Finish up.
 941          */
 942         if (error) {
 943                 hammer_rel_inode(nip, 0);
 944                 *ap->a_vpp = NULL;
 945         } else {
 946                 error = hammer_get_vnode(nip, ap->a_vpp);
 947                 hammer_rel_inode(nip, 0);
 948                 if (error == 0) {
 949                         cache_setunresolved(ap->a_nch);
 950                         cache_setvp(ap->a_nch, *ap->a_vpp);
 951                 }
 952         }
 953         hammer_done_transaction(&trans);
 954         return (error);
 955 }
 956
 957 /*
 958  * hammer_vop_open { vp, mode, cred, fp }
 959  */
 960 static
 961 int
 962 hammer_vop_open(struct vop_open_args *ap)
 963 {
 964         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
 965                 return (EROFS);
 966
 967         return(vop_stdopen(ap));
 968 }
 969
 970 /*
 971  * hammer_vop_pathconf { vp, name, retval }
 972  */
 973 static
 974 int
 975 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 976 {
 977         return EOPNOTSUPP;
 978 }
 979
 980 /*
 981  * hammer_vop_print { vp }
 982  */
 983 static
 984 int
 985 hammer_vop_print(struct vop_print_args *ap)
 986 {
 987         return EOPNOTSUPP;
 988 }
 989
 990 /*
 991  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 992  */
 993 static
 994 int
 995 hammer_vop_readdir(struct vop_readdir_args *ap)
 996 {
 997         struct hammer_transaction trans;
 998         struct hammer_cursor cursor;
 999         struct hammer_inode *ip;
1000         struct uio *uio;
1001         hammer_base_elm_t base;
1002         int error;
1003         int cookie_index;
1004         int ncookies;
1005         off_t *cookies;
1006         off_t saveoff;
1007         int r;
1008
1009         ip = VTOI(ap->a_vp);
1010         uio = ap->a_uio;
1011         saveoff = uio->uio_offset;
1012
1013         if (ap->a_ncookies) {
1014                 ncookies = uio->uio_resid / 16 + 1;
1015                 if (ncookies > 1024)
1016                         ncookies = 1024;
1017                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1018                 cookie_index = 0;
1019         } else {
1020                 ncookies = -1;
1021                 cookies = NULL;
1022                 cookie_index = 0;
1023         }
1024
1025         hammer_simple_transaction(&trans, ip->hmp);
1026
1027         /*
1028          * Handle artificial entries
1029          */
1030         error = 0;
1031         if (saveoff == 0) {
1032                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1033                 if (r)
1034                         goto done;
1035                 if (cookies)
1036                         cookies[cookie_index] = saveoff;
1037                 ++saveoff;
1038                 ++cookie_index;
1039                 if (cookie_index == ncookies)
1040                         goto done;
1041         }
1042         if (saveoff == 1) {
1043                 if (ip->ino_data.parent_obj_id) {
1044                         r = vop_write_dirent(&error, uio,
1045                                              ip->ino_data.parent_obj_id,
1046                                              DT_DIR, 2, "..");
1047                 } else {
1048                         r = vop_write_dirent(&error, uio,
1049                                              ip->obj_id, DT_DIR, 2, "..");
1050                 }
1051                 if (r)
1052                         goto done;
1053                 if (cookies)
1054                         cookies[cookie_index] = saveoff;
1055                 ++saveoff;
1056                 ++cookie_index;
1057                 if (cookie_index == ncookies)
1058                         goto done;
1059         }
1060
1061         /*
1062          * Key range (begin and end inclusive) to scan.  Directory keys
1063          * directly translate to a 64 bit 'seek' position.
1064          */
1065         hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1066         cursor.key_beg.obj_id = ip->obj_id;
1067         cursor.key_beg.create_tid = 0;
1068         cursor.key_beg.delete_tid = 0;
1069         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1070         cursor.key_beg.obj_type = 0;
1071         cursor.key_beg.key = saveoff;
1072
1073         cursor.key_end = cursor.key_beg;
1074         cursor.key_end.key = HAMMER_MAX_KEY;
1075         cursor.asof = ip->obj_asof;
1076         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1077
1078         error = hammer_ip_first(&cursor);
1079
1080         while (error == 0) {
1081                 error = hammer_ip_resolve_data(&cursor);
1082                 if (error)
1083                         break;
1084                 base = &cursor.leaf->base;
1085                 saveoff = base->key;
1086                 KKASSERT(cursor.leaf->data_len > HAMMER_ENTRY_NAME_OFF);
1087
1088                 if (base->obj_id != ip->obj_id)
1089                         panic("readdir: bad record at %p", cursor.node);
1090
1091                 r = vop_write_dirent(
1092                              &error, uio, cursor.data->entry.obj_id,
1093                              hammer_get_dtype(cursor.leaf->base.obj_type),
1094                              cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF ,
1095                              (void *)cursor.data->entry.name);
1096                 if (r)
1097                         break;
1098                 ++saveoff;
1099                 if (cookies)
1100                         cookies[cookie_index] = base->key;
1101                 ++cookie_index;
1102                 if (cookie_index == ncookies)
1103                         break;
1104                 error = hammer_ip_next(&cursor);
1105         }
1106         hammer_done_cursor(&cursor);
1107
1108 done:
1109         hammer_done_transaction(&trans);
1110
1111         if (ap->a_eofflag)
1112                 *ap->a_eofflag = (error == ENOENT);
1113         uio->uio_offset = saveoff;
1114         if (error && cookie_index == 0) {
1115                 if (error == ENOENT)
1116                         error = 0;
1117                 if (cookies) {
1118                         kfree(cookies, M_TEMP);
1119                         *ap->a_ncookies = 0;
1120                         *ap->a_cookies = NULL;
1121                 }
1122         } else {
1123                 if (error == ENOENT)
1124                         error = 0;
1125                 if (cookies) {
1126                         *ap->a_ncookies = cookie_index;
1127                         *ap->a_cookies = cookies;
1128                 }
1129         }
1130         return(error);
1131 }
1132
1133 /*
1134  * hammer_vop_readlink { vp, uio, cred }
1135  */
1136 static
1137 int
1138 hammer_vop_readlink(struct vop_readlink_args *ap)
1139 {
1140         struct hammer_transaction trans;
1141         struct hammer_cursor cursor;
1142         struct hammer_inode *ip;
1143         int error;
1144
1145         ip = VTOI(ap->a_vp);
1146
1147         hammer_simple_transaction(&trans, ip->hmp);
1148
1149         hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1150
1151         /*
1152          * Key range (begin and end inclusive) to scan.  Directory keys
1153          * directly translate to a 64 bit 'seek' position.
1154          */
1155         cursor.key_beg.obj_id = ip->obj_id;
1156         cursor.key_beg.create_tid = 0;
1157         cursor.key_beg.delete_tid = 0;
1158         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1159         cursor.key_beg.obj_type = 0;
1160         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1161         cursor.asof = ip->obj_asof;
1162         cursor.flags |= HAMMER_CURSOR_ASOF;
1163
1164         error = hammer_ip_lookup(&cursor);
1165         if (error == 0) {
1166                 error = hammer_ip_resolve_data(&cursor);
1167                 if (error == 0) {
1168                         KKASSERT(cursor.leaf->data_len >=
1169                                  HAMMER_SYMLINK_NAME_OFF);
1170                         error = uiomove(cursor.data->symlink.name,
1171                                         cursor.leaf->data_len -
1172                                                 HAMMER_SYMLINK_NAME_OFF,
1173                                         ap->a_uio);
1174                 }
1175         }
1176         hammer_done_cursor(&cursor);
1177         hammer_done_transaction(&trans);
1178         return(error);
1179 }
1180
1181 /*
1182  * hammer_vop_nremove { nch, dvp, cred }
1183  */
1184 static
1185 int
1186 hammer_vop_nremove(struct vop_nremove_args *ap)
1187 {
1188         struct hammer_transaction trans;
1189         int error;
1190
1191         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1192         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1193         hammer_done_transaction(&trans);
1194
1195         return (error);
1196 }
1197
1198 /*
1199  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1200  */
1201 static
1202 int
1203 hammer_vop_nrename(struct vop_nrename_args *ap)
1204 {
1205         struct hammer_transaction trans;
1206         struct namecache *fncp;
1207         struct namecache *tncp;
1208         struct hammer_inode *fdip;
1209         struct hammer_inode *tdip;
1210         struct hammer_inode *ip;
1211         struct hammer_cursor cursor;
1212         int64_t namekey;
1213         int nlen, error;
1214
1215         fdip = VTOI(ap->a_fdvp);
1216         tdip = VTOI(ap->a_tdvp);
1217         fncp = ap->a_fnch->ncp;
1218         tncp = ap->a_tnch->ncp;
1219         ip = VTOI(fncp->nc_vp);
1220         KKASSERT(ip != NULL);
1221
1222         if (fdip->flags & HAMMER_INODE_RO)
1223                 return (EROFS);
1224         if (tdip->flags & HAMMER_INODE_RO)
1225                 return (EROFS);
1226         if (ip->flags & HAMMER_INODE_RO)
1227                 return (EROFS);
1228
1229         hammer_start_transaction(&trans, fdip->hmp);
1230
1231         /*
1232          * Remove tncp from the target directory and then link ip as
1233          * tncp. XXX pass trans to dounlink
1234          *
1235          * Force the inode sync-time to match the transaction so it is
1236          * in-sync with the creation of the target directory entry.
1237          */
1238         error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1239         if (error == 0 || error == ENOENT) {
1240                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1241                 if (error == 0) {
1242                         ip->ino_data.parent_obj_id = tdip->obj_id;
1243                         hammer_modify_inode(&trans, ip, HAMMER_INODE_DDIRTY);
1244                 }
1245         }
1246         if (error)
1247                 goto failed; /* XXX */
1248
1249         /*
1250          * Locate the record in the originating directory and remove it.
1251          *
1252          * Calculate the namekey and setup the key range for the scan.  This
1253          * works kinda like a chained hash table where the lower 32 bits
1254          * of the namekey synthesize the chain.
1255          *
1256          * The key range is inclusive of both key_beg and key_end.
1257          */
1258         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1259 retry:
1260         hammer_init_cursor(&trans, &cursor, &fdip->cache[0], fdip);
1261         cursor.key_beg.obj_id = fdip->obj_id;
1262         cursor.key_beg.key = namekey;
1263         cursor.key_beg.create_tid = 0;
1264         cursor.key_beg.delete_tid = 0;
1265         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1266         cursor.key_beg.obj_type = 0;
1267
1268         cursor.key_end = cursor.key_beg;
1269         cursor.key_end.key |= 0xFFFFFFFFULL;
1270         cursor.asof = fdip->obj_asof;
1271         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1272
1273         /*
1274          * Scan all matching records (the chain), locate the one matching
1275          * the requested path component.
1276          *
1277          * The hammer_ip_*() functions merge in-memory records with on-disk
1278          * records for the purposes of the search.
1279          */
1280         error = hammer_ip_first(&cursor);
1281         while (error == 0) {
1282                 if (hammer_ip_resolve_data(&cursor) != 0)
1283                         break;
1284                 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
1285                 KKASSERT(nlen > 0);
1286                 if (fncp->nc_nlen == nlen &&
1287                     bcmp(fncp->nc_name, cursor.data->entry.name, nlen) == 0) {
1288                         break;
1289                 }
1290                 error = hammer_ip_next(&cursor);
1291         }
1292
1293         /*
1294          * If all is ok we have to get the inode so we can adjust nlinks.
1295          *
1296          * WARNING: hammer_ip_del_directory() may have to terminate the
1297          * cursor to avoid a recursion.  It's ok to call hammer_done_cursor()
1298          * twice.
1299          */
1300         if (error == 0)
1301                 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1302
1303         /*
1304          * XXX A deadlock here will break rename's atomicy for the purposes
1305          * of crash recovery.
1306          */
1307         if (error == EDEADLK) {
1308                 hammer_done_cursor(&cursor);
1309                 goto retry;
1310         }
1311
1312         /*
1313          * Cleanup and tell the kernel that the rename succeeded.
1314          */
1315         hammer_done_cursor(&cursor);
1316         if (error == 0)
1317                 cache_rename(ap->a_fnch, ap->a_tnch);
1318
1319 failed:
1320         hammer_done_transaction(&trans);
1321         return (error);
1322 }
1323
1324 /*
1325  * hammer_vop_nrmdir { nch, dvp, cred }
1326  */
1327 static
1328 int
1329 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1330 {
1331         struct hammer_transaction trans;
1332         int error;
1333
1334         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1335         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1336         hammer_done_transaction(&trans);
1337
1338         return (error);
1339 }
1340
1341 /*
1342  * hammer_vop_setattr { vp, vap, cred }
1343  */
1344 static
1345 int
1346 hammer_vop_setattr(struct vop_setattr_args *ap)
1347 {
1348         struct hammer_transaction trans;
1349         struct vattr *vap;
1350         struct hammer_inode *ip;
1351         int modflags;
1352         int error;
1353         int truncating;
1354         off_t aligned_size;
1355         u_int32_t flags;
1356         uuid_t uuid;
1357
1358         vap = ap->a_vap;
1359         ip = ap->a_vp->v_data;
1360         modflags = 0;
1361
1362         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1363                 return(EROFS);
1364         if (ip->flags & HAMMER_INODE_RO)
1365                 return (EROFS);
1366
1367         hammer_start_transaction(&trans, ip->hmp);
1368         error = 0;
1369
1370         if (vap->va_flags != VNOVAL) {
1371                 flags = ip->ino_data.uflags;
1372                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1373                                          hammer_to_unix_xid(&ip->ino_data.uid),
1374                                          ap->a_cred);
1375                 if (error == 0) {
1376                         if (ip->ino_data.uflags != flags) {
1377                                 ip->ino_data.uflags = flags;
1378                                 modflags |= HAMMER_INODE_DDIRTY;
1379                         }
1380                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1381                                 error = 0;
1382                                 goto done;
1383                         }
1384                 }
1385                 goto done;
1386         }
1387         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1388                 error = EPERM;
1389                 goto done;
1390         }
1391         if (vap->va_uid != (uid_t)VNOVAL) {
1392                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1393                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1394                         ip->ino_data.uid = uuid;
1395                         modflags |= HAMMER_INODE_DDIRTY;
1396                 }
1397         }
1398         if (vap->va_gid != (uid_t)VNOVAL) {
1399                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1400                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1401                         ip->ino_data.gid = uuid;
1402                         modflags |= HAMMER_INODE_DDIRTY;
1403                 }
1404         }
1405         while (vap->va_size != VNOVAL && ip->ino_data.size != vap->va_size) {
1406                 switch(ap->a_vp->v_type) {
1407                 case VREG:
1408                         if (vap->va_size == ip->ino_data.size)
1409                                 break;
1410                         /*
1411                          * XXX break atomicy, we can deadlock the backend
1412                          * if we do not release the lock.  Probably not a
1413                          * big deal here.
1414                          */
1415                         if (vap->va_size < ip->ino_data.size) {
1416                                 vtruncbuf(ap->a_vp, vap->va_size,
1417                                           HAMMER_BUFSIZE);
1418                                 truncating = 1;
1419                         } else {
1420                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1421                                 truncating = 0;
1422                         }
1423                         ip->ino_data.size = vap->va_size;
1424                         modflags |= HAMMER_INODE_DDIRTY;
1425                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1426                                        ~HAMMER_BUFMASK64;
1427
1428                         /*
1429                          * on-media truncation is cached in the inode until
1430                          * the inode is synchronized.
1431                          */
1432                         if (truncating) {
1433                                 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1434                                         ip->flags |= HAMMER_INODE_TRUNCATED;
1435                                         ip->trunc_off = vap->va_size;
1436                                 } else if (ip->trunc_off > vap->va_size) {
1437                                         ip->trunc_off = vap->va_size;
1438                                 }
1439                         }
1440
1441                         /*
1442                          * If truncating we have to clean out a portion of
1443                          * the last block on-disk.  We do this in the
1444                          * front-end buffer cache.
1445                          */
1446                         if (truncating && vap->va_size < aligned_size) {
1447                                 struct buf *bp;
1448                                 int offset;
1449
1450                                 offset = vap->va_size & HAMMER_BUFMASK;
1451                                 error = bread(ap->a_vp,
1452                                               aligned_size - HAMMER_BUFSIZE,
1453                                               HAMMER_BUFSIZE, &bp);
1454                                 if (error == 0) {
1455                                         bzero(bp->b_data + offset,
1456                                               HAMMER_BUFSIZE - offset);
1457                                         bdwrite(bp);
1458                                 } else {
1459                                         brelse(bp);
1460                                 }
1461                         }
1462                         break;
1463                 case VDATABASE:
1464                         if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1465                                 ip->flags |= HAMMER_INODE_TRUNCATED;
1466                                 ip->trunc_off = vap->va_size;
1467                         } else if (ip->trunc_off > vap->va_size) {
1468                                 ip->trunc_off = vap->va_size;
1469                         }
1470                         ip->ino_data.size = vap->va_size;
1471                         modflags |= HAMMER_INODE_DDIRTY;
1472                         break;
1473                 default:
1474                         error = EINVAL;
1475                         goto done;
1476                 }
1477                 break;
1478         }
1479         if (vap->va_atime.tv_sec != VNOVAL) {
1480                 ip->ino_leaf.atime =
1481                         hammer_timespec_to_transid(&vap->va_atime);
1482                 modflags |= HAMMER_INODE_ITIMES;
1483         }
1484         if (vap->va_mtime.tv_sec != VNOVAL) {
1485                 ip->ino_data.mtime =
1486                         hammer_timespec_to_transid(&vap->va_mtime);
1487                 modflags |= HAMMER_INODE_ITIMES;
1488                 modflags |= HAMMER_INODE_DDIRTY;        /* XXX mtime */
1489         }
1490         if (vap->va_mode != (mode_t)VNOVAL) {
1491                 if (ip->ino_data.mode != vap->va_mode) {
1492                         ip->ino_data.mode = vap->va_mode;
1493                         modflags |= HAMMER_INODE_DDIRTY;
1494                 }
1495         }
1496 done:
1497         if (error == 0)
1498                 hammer_modify_inode(&trans, ip, modflags);
1499         hammer_done_transaction(&trans);
1500         return (error);
1501 }
1502
1503 /*
1504  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1505  */
1506 static
1507 int
1508 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1509 {
1510         struct hammer_transaction trans;
1511         struct hammer_inode *dip;
1512         struct hammer_inode *nip;
1513         struct nchandle *nch;
1514         hammer_record_t record;
1515         int error;
1516         int bytes;
1517
1518         ap->a_vap->va_type = VLNK;
1519
1520         nch = ap->a_nch;
1521         dip = VTOI(ap->a_dvp);
1522
1523         if (dip->flags & HAMMER_INODE_RO)
1524                 return (EROFS);
1525
1526         /*
1527          * Create a transaction to cover the operations we perform.
1528          */
1529         hammer_start_transaction(&trans, dip->hmp);
1530
1531         /*
1532          * Create a new filesystem object of the requested type.  The
1533          * returned inode will be referenced but not locked.
1534          */
1535
1536         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1537         if (error) {
1538                 hammer_done_transaction(&trans);
1539                 *ap->a_vpp = NULL;
1540                 return (error);
1541         }
1542
1543         /*
1544          * Add a record representing the symlink.  symlink stores the link
1545          * as pure data, not a string, and is no \0 terminated.
1546          */
1547         if (error == 0) {
1548                 bytes = strlen(ap->a_target);
1549                 record = hammer_alloc_mem_record(nip, bytes);
1550                 record->type = HAMMER_MEM_RECORD_GENERAL;
1551
1552                 record->leaf.base.key = HAMMER_FIXKEY_SYMLINK;
1553                 record->leaf.base.rec_type = HAMMER_RECTYPE_FIX;
1554                 record->leaf.data_len = bytes;
1555                 KKASSERT(HAMMER_SYMLINK_NAME_OFF == 0);
1556                 bcopy(ap->a_target, record->data->symlink.name, bytes);
1557                 error = hammer_ip_add_record(&trans, record);
1558
1559                 /*
1560                  * Set the file size to the length of the link.
1561                  */
1562                 if (error == 0) {
1563                         nip->ino_data.size = bytes;
1564                         hammer_modify_inode(&trans, nip, HAMMER_INODE_DDIRTY);
1565                 }
1566         }
1567         if (error == 0)
1568                 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1569
1570         /*
1571          * Finish up.
1572          */
1573         if (error) {
1574                 hammer_rel_inode(nip, 0);
1575                 *ap->a_vpp = NULL;
1576         } else {
1577                 error = hammer_get_vnode(nip, ap->a_vpp);
1578                 hammer_rel_inode(nip, 0);
1579                 if (error == 0) {
1580                         cache_setunresolved(ap->a_nch);
1581                         cache_setvp(ap->a_nch, *ap->a_vpp);
1582                 }
1583         }
1584         hammer_done_transaction(&trans);
1585         return (error);
1586 }
1587
1588 /*
1589  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1590  */
1591 static
1592 int
1593 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1594 {
1595         struct hammer_transaction trans;
1596         int error;
1597
1598         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1599         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1600                                 ap->a_cred, ap->a_flags);
1601         hammer_done_transaction(&trans);
1602
1603         return (error);
1604 }
1605
1606 /*
1607  * hammer_vop_ioctl { vp, command, data, fflag, cred }
1608  */
1609 static
1610 int
1611 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1612 {
1613         struct hammer_inode *ip = ap->a_vp->v_data;
1614
1615         return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1616                             ap->a_fflag, ap->a_cred));
1617 }
1618
1619 static
1620 int
1621 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1622 {
1623         struct mount *mp;
1624         int error;
1625
1626         mp = ap->a_head.a_ops->head.vv_mount;
1627
1628         switch(ap->a_op) {
1629         case MOUNTCTL_SET_EXPORT:
1630                 if (ap->a_ctllen != sizeof(struct export_args))
1631                         error = EINVAL;
1632                 error = hammer_vfs_export(mp, ap->a_op,
1633                                       (const struct export_args *)ap->a_ctl);
1634                 break;
1635         default:
1636                 error = journal_mountctl(ap);
1637                 break;
1638         }
1639         return(error);
1640 }
1641
1642 /*
1643  * hammer_vop_strategy { vp, bio }
1644  *
1645  * Strategy call, used for regular file read & write only.  Note that the
1646  * bp may represent a cluster.
1647  *
1648  * To simplify operation and allow better optimizations in the future,
1649  * this code does not make any assumptions with regards to buffer alignment
1650  * or size.
1651  */
1652 static
1653 int
1654 hammer_vop_strategy(struct vop_strategy_args *ap)
1655 {
1656         struct buf *bp;
1657         int error;
1658
1659         bp = ap->a_bio->bio_buf;
1660
1661         switch(bp->b_cmd) {
1662         case BUF_CMD_READ:
1663                 error = hammer_vop_strategy_read(ap);
1664                 break;
1665         case BUF_CMD_WRITE:
1666                 error = hammer_vop_strategy_write(ap);
1667                 break;
1668         default:
1669                 bp->b_error = error = EINVAL;
1670                 bp->b_flags |= B_ERROR;
1671                 biodone(ap->a_bio);
1672                 break;
1673         }
1674         return (error);
1675 }
1676
1677 /*
1678  * Read from a regular file.  Iterate the related records and fill in the
1679  * BIO/BUF.  Gaps are zero-filled.
1680  *
1681  * The support code in hammer_object.c should be used to deal with mixed
1682  * in-memory and on-disk records.
1683  *
1684  * XXX atime update
1685  */
1686 static
1687 int
1688 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1689 {
1690         struct hammer_transaction trans;
1691         struct hammer_inode *ip;
1692         struct hammer_cursor cursor;
1693         hammer_base_elm_t base;
1694         struct bio *bio;
1695         struct buf *bp;
1696         int64_t rec_offset;
1697         int64_t ran_end;
1698         int64_t tmp64;
1699         int error;
1700         int boff;
1701         int roff;
1702         int n;
1703
1704         bio = ap->a_bio;
1705         bp = bio->bio_buf;
1706         ip = ap->a_vp->v_data;
1707
1708         hammer_simple_transaction(&trans, ip->hmp);
1709         hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1710
1711         /*
1712          * Key range (begin and end inclusive) to scan.  Note that the key's
1713          * stored in the actual records represent BASE+LEN, not BASE.  The
1714          * first record containing bio_offset will have a key > bio_offset.
1715          */
1716         cursor.key_beg.obj_id = ip->obj_id;
1717         cursor.key_beg.create_tid = 0;
1718         cursor.key_beg.delete_tid = 0;
1719         cursor.key_beg.obj_type = 0;
1720         cursor.key_beg.key = bio->bio_offset + 1;
1721         cursor.asof = ip->obj_asof;
1722         cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1723
1724         cursor.key_end = cursor.key_beg;
1725         KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);
1726 #if 0
1727         if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
1728                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1729                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1730                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1731         } else
1732 #endif
1733         {
1734                 ran_end = bio->bio_offset + bp->b_bufsize;
1735                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1736                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1737                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1738                 if (tmp64 < ran_end)
1739                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1740                 else
1741                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1742         }
1743         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1744
1745         error = hammer_ip_first(&cursor);
1746         boff = 0;
1747
1748         while (error == 0) {
1749                 error = hammer_ip_resolve_data(&cursor);
1750                 if (error)
1751                         break;
1752                 base = &cursor.leaf->base;
1753
1754                 rec_offset = base->key - cursor.leaf->data_len;
1755
1756                 /*
1757                  * Calculate the gap, if any, and zero-fill it.
1758                  */
1759                 n = (int)(rec_offset - (bio->bio_offset + boff));
1760                 if (n > 0) {
1761                         if (n > bp->b_bufsize - boff)
1762                                 n = bp->b_bufsize - boff;
1763                         bzero((char *)bp->b_data + boff, n);
1764                         boff += n;
1765                         n = 0;
1766                 }
1767
1768                 /*
1769                  * Calculate the data offset in the record and the number
1770                  * of bytes we can copy.
1771                  *
1772                  * Note there is a degenerate case here where boff may
1773                  * already be at bp->b_bufsize.
1774                  */
1775                 roff = -n;
1776                 rec_offset += roff;
1777                 n = cursor.leaf->data_len - roff;
1778                 KKASSERT(n > 0);
1779                 if (n > bp->b_bufsize - boff)
1780                         n = bp->b_bufsize - boff;
1781
1782                 /*
1783                  * If we cached a truncation point on our front-end the
1784                  * on-disk version may still have physical records beyond
1785                  * that point.  Truncate visibility.
1786                  */
1787                 if (ip->trunc_off <= rec_offset)
1788                         n = 0;
1789                 else if (ip->trunc_off < rec_offset + n)
1790                         n = (int)(ip->trunc_off - rec_offset);
1791
1792                 /*
1793                  * Copy
1794                  */
1795                 if (n) {
1796                         bcopy((char *)cursor.data + roff,
1797                               (char *)bp->b_data + boff, n);
1798                         boff += n;
1799                 }
1800                 if (boff == bp->b_bufsize)
1801                         break;
1802                 error = hammer_ip_next(&cursor);
1803         }
1804         hammer_done_cursor(&cursor);
1805         hammer_done_transaction(&trans);
1806
1807         /*
1808          * There may have been a gap after the last record
1809          */
1810         if (error == ENOENT)
1811                 error = 0;
1812         if (error == 0 && boff != bp->b_bufsize) {
1813                 KKASSERT(boff < bp->b_bufsize);
1814                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1815                 /* boff = bp->b_bufsize; */
1816         }
1817         bp->b_resid = 0;
1818         bp->b_error = error;
1819         if (error)
1820                 bp->b_flags |= B_ERROR;
1821         biodone(ap->a_bio);
1822         return(error);
1823 }
1824
1825 /*
1826  * Write to a regular file.   Because this is a strategy call the OS is
1827  * trying to actually sync data to the media.   HAMMER can only flush
1828  * the entire inode (so the TID remains properly synchronized).
1829  *
1830  * Basically all we do here is place the bio on the inode's flush queue
1831  * and activate the flusher.
1832  */
1833 static
1834 int
1835 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1836 {
1837         hammer_inode_t ip;
1838         struct bio *bio;
1839         struct buf *bp;
1840
1841         bio = ap->a_bio;
1842         bp = bio->bio_buf;
1843         ip = ap->a_vp->v_data;
1844
1845         if (ip->flags & HAMMER_INODE_RO) {
1846                 bp->b_error = EROFS;
1847                 bp->b_flags |= B_ERROR;
1848                 biodone(ap->a_bio);
1849                 return(EROFS);
1850         }
1851
1852         /*
1853          * If the inode is being flushed we cannot re-queue buffers
1854          * it may have already flushed, or it could result in duplicate
1855          * records in the database.
1856          */
1857         BUF_KERNPROC(bp);
1858         if (ip->flags & HAMMER_INODE_WRITE_ALT)
1859                 TAILQ_INSERT_TAIL(&ip->bio_alt_list, bio, bio_act);
1860         else
1861                 TAILQ_INSERT_TAIL(&ip->bio_list, bio, bio_act);
1862         ++hammer_bio_count;
1863         hammer_modify_inode(NULL, ip, HAMMER_INODE_BUFS);
1864
1865         hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1866 #if 0
1867         /*
1868          * XXX
1869          *
1870          * If the write was not part of an integrated flush operation then
1871          * signal a flush.
1872          */
1873         if (ip->flush_state != HAMMER_FST_FLUSH ||
1874             (ip->flags & HAMMER_INODE_WRITE_ALT)) {
1875                 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1876         }
1877 #endif
1878         return(0);
1879 }
1880
1881 /*
1882  * Backend code which actually performs the write to the media.  This
1883  * routine is typically called from the flusher.  The bio will be disposed
1884  * of (biodone'd) by this routine.
1885  *
1886  * Iterate the related records and mark for deletion.  If existing edge
1887  * records (left and right side) overlap our write they have to be marked
1888  * deleted and new records created, usually referencing a portion of the
1889  * original data.  Then add a record to represent the buffer.
1890  */
1891 int
1892 hammer_dowrite(hammer_cursor_t cursor, hammer_inode_t ip, struct bio *bio)
1893 {
1894         struct buf *bp = bio->bio_buf;
1895         int error;
1896
1897         KKASSERT(ip->flush_state == HAMMER_FST_FLUSH);
1898
1899         /*
1900          * If the inode is going or gone, just throw away any frontend
1901          * buffers.
1902          */
1903         if (ip->flags & HAMMER_INODE_DELETED) {
1904                 bp->b_resid = 0;
1905                 biodone(bio);
1906                 --hammer_bio_count;
1907                 return(0);
1908         }
1909
1910         /*
1911          * Delete any records overlapping our range.  This function will
1912          * (eventually) properly truncate partial overlaps.
1913          */
1914         if (ip->sync_ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
1915                 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
1916                                                bio->bio_offset);
1917         } else {
1918                 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
1919                                                bio->bio_offset +
1920                                                 bp->b_bufsize - 1);
1921         }
1922
1923         /*
1924          * Add a single record to cover the write.  We can write a record
1925          * with only the actual file data - for example, a small 200 byte
1926          * file does not have to write out a 16K record.
1927          *
1928          * While the data size does not have to be aligned, we still do it
1929          * to reduce fragmentation in a future allocation model.
1930          */
1931         if (error == 0) {
1932                 int limit_size;
1933
1934                 if (ip->sync_ino_data.size - bio->bio_offset >
1935                     bp->b_bufsize) {
1936                             limit_size = bp->b_bufsize;
1937                 } else {
1938                         limit_size = (int)(ip->sync_ino_data.size -
1939                                            bio->bio_offset);
1940                         KKASSERT(limit_size >= 0);
1941                         limit_size = (limit_size + 63) & ~63;
1942                 }
1943                 if (limit_size) {
1944                         error = hammer_ip_sync_data(cursor, ip, bio->bio_offset,
1945                                                     bp->b_data, limit_size);
1946                 }
1947         }
1948         if (error)
1949                 Debugger("hammer_dowrite: error");
1950
1951         if (error) {
1952                 bp->b_resid = bp->b_bufsize;
1953                 bp->b_error = error;
1954                 bp->b_flags |= B_ERROR;
1955         } else {
1956                 bp->b_resid = 0;
1957         }
1958         biodone(bio);
1959         --hammer_bio_count;
1960         return(error);
1961 }
1962
1963 /*
1964  * dounlink - disconnect a directory entry
1965  *
1966  * XXX whiteout support not really in yet
1967  */
1968 static int
1969 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
1970                 struct vnode *dvp, struct ucred *cred, int flags)
1971 {
1972         struct namecache *ncp;
1973         hammer_inode_t dip;
1974         hammer_inode_t ip;
1975         struct hammer_cursor cursor;
1976         int64_t namekey;
1977         int nlen, error;
1978
1979         /*
1980          * Calculate the namekey and setup the key range for the scan.  This
1981          * works kinda like a chained hash table where the lower 32 bits
1982          * of the namekey synthesize the chain.
1983          *
1984          * The key range is inclusive of both key_beg and key_end.
1985          */
1986         dip = VTOI(dvp);
1987         ncp = nch->ncp;
1988
1989         if (dip->flags & HAMMER_INODE_RO)
1990                 return (EROFS);
1991
1992         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1993 retry:
1994         hammer_init_cursor(trans, &cursor, &dip->cache[0], dip);
1995         cursor.key_beg.obj_id = dip->obj_id;
1996         cursor.key_beg.key = namekey;
1997         cursor.key_beg.create_tid = 0;
1998         cursor.key_beg.delete_tid = 0;
1999         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
2000         cursor.key_beg.obj_type = 0;
2001
2002         cursor.key_end = cursor.key_beg;
2003         cursor.key_end.key |= 0xFFFFFFFFULL;
2004         cursor.asof = dip->obj_asof;
2005         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
2006
2007         /*
2008          * Scan all matching records (the chain), locate the one matching
2009          * the requested path component.  info->last_error contains the
2010          * error code on search termination and could be 0, ENOENT, or
2011          * something else.
2012          *
2013          * The hammer_ip_*() functions merge in-memory records with on-disk
2014          * records for the purposes of the search.
2015          */
2016         error = hammer_ip_first(&cursor);
2017
2018         while (error == 0) {
2019                 error = hammer_ip_resolve_data(&cursor);
2020                 if (error)
2021                         break;
2022                 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
2023                 KKASSERT(nlen > 0);
2024                 if (ncp->nc_nlen == nlen &&
2025                     bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
2026                         break;
2027                 }
2028                 error = hammer_ip_next(&cursor);
2029         }
2030
2031         /*
2032          * If all is ok we have to get the inode so we can adjust nlinks.
2033          *
2034          * If the target is a directory, it must be empty.
2035          */
2036         if (error == 0) {
2037                 ip = hammer_get_inode(trans, &dip->cache[1],
2038                                       cursor.data->entry.obj_id,
2039                                       dip->hmp->asof, 0, &error);
2040                 if (error == ENOENT) {
2041                         kprintf("obj_id %016llx\n", cursor.data->entry.obj_id);
2042                         Debugger("ENOENT unlinking object that should exist");
2043                 }
2044
2045                 /*
2046                  * If we are trying to remove a directory the directory must
2047                  * be empty.
2048                  *
2049                  * WARNING: hammer_ip_check_directory_empty() may have to
2050                  * terminate the cursor to avoid a deadlock.  It is ok to
2051                  * call hammer_done_cursor() twice.
2052                  */
2053                 if (error == 0 && ip->ino_data.obj_type ==
2054                                   HAMMER_OBJTYPE_DIRECTORY) {
2055                         error = hammer_ip_check_directory_empty(trans, ip);
2056                 }
2057
2058                 /*
2059                  * Delete the directory entry.
2060                  *
2061                  * WARNING: hammer_ip_del_directory() may have to terminate
2062                  * the cursor to avoid a deadlock.  It is ok to call
2063                  * hammer_done_cursor() twice.
2064                  */
2065                 if (error == 0) {
2066                         error = hammer_ip_del_directory(trans, &cursor,
2067                                                         dip, ip);
2068                 }
2069                 if (error == 0) {
2070                         cache_setunresolved(nch);
2071                         cache_setvp(nch, NULL);
2072                         /* XXX locking */
2073                         if (ip->vp)
2074                                 cache_inval_vp(ip->vp, CINV_DESTROY);
2075                 }
2076                 hammer_rel_inode(ip, 0);
2077         }
2078         hammer_done_cursor(&cursor);
2079         if (error == EDEADLK)
2080                 goto retry;
2081
2082         return (error);
2083 }
2084
2085 /************************************************************************
2086  *                          FIFO AND SPECFS OPS                         *
2087  ************************************************************************
2088  *
2089  */
2090
2091 static int
2092 hammer_vop_fifoclose (struct vop_close_args *ap)
2093 {
2094         /* XXX update itimes */
2095         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2096 }
2097
2098 static int
2099 hammer_vop_fiforead (struct vop_read_args *ap)
2100 {
2101         int error;
2102
2103         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2104         /* XXX update access time */
2105         return (error);
2106 }
2107
2108 static int
2109 hammer_vop_fifowrite (struct vop_write_args *ap)
2110 {
2111         int error;
2112
2113         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2114         /* XXX update access time */
2115         return (error);
2116 }
2117
2118 static int
2119 hammer_vop_specclose (struct vop_close_args *ap)
2120 {
2121         /* XXX update itimes */
2122         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2123 }
2124
2125 static int
2126 hammer_vop_specread (struct vop_read_args *ap)
2127 {
2128         /* XXX update access time */
2129         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2130 }
2131
2132 static int
2133 hammer_vop_specwrite (struct vop_write_args *ap)
2134 {
2135         /* XXX update last change time */
2136         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2137 }
2138