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 1.45 2008/05/03 05:28:55 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         if (ap->a_waitfor == MNT_WAIT)
 177                 hammer_wait_inode(ip);
 178         return (ip->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                 bqrelse(bp);
 234         }
 235         if ((ip->flags & HAMMER_INODE_RO) == 0 &&
 236             (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
 237                 ip->ino_rec.ino_atime = trans.time;
 238                 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
 239         }
 240         hammer_done_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         int count;
 260
 261         if (ap->a_vp->v_type != VREG)
 262                 return (EINVAL);
 263         ip = VTOI(ap->a_vp);
 264         error = 0;
 265
 266         if (ip->flags & HAMMER_INODE_RO)
 267                 return (EROFS);
 268
 269         /*
 270          * Create a transaction to cover the operations we perform.
 271          */
 272         hammer_start_transaction(&trans, ip->hmp);
 273         uio = ap->a_uio;
 274
 275         /*
 276          * Check append mode
 277          */
 278         if (ap->a_ioflag & IO_APPEND)
 279                 uio->uio_offset = ip->ino_rec.ino_size;
 280
 281         /*
 282          * Check for illegal write offsets.  Valid range is 0...2^63-1
 283          */
 284         if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
 285                 hammer_done_transaction(&trans);
 286                 return (EFBIG);
 287         }
 288
 289         /*
 290          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 291          */
 292         count = 0;
 293         while (uio->uio_resid > 0) {
 294                 int fixsize = 0;
 295
 296                 /*
 297                  * Do not allow huge writes to deadlock the buffer cache
 298                  */
 299                 if ((++count & 15) == 0) {
 300                         vn_unlock(ap->a_vp);
 301                         if ((ap->a_ioflag & IO_NOBWILL) == 0)
 302                                 bwillwrite();
 303                         vn_lock(ap->a_vp, LK_EXCLUSIVE|LK_RETRY);
 304                 }
 305
 306                 offset = uio->uio_offset & HAMMER_BUFMASK;
 307                 n = HAMMER_BUFSIZE - offset;
 308                 if (n > uio->uio_resid)
 309                         n = uio->uio_resid;
 310                 if (uio->uio_offset + n > ip->ino_rec.ino_size) {
 311                         vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
 312                         fixsize = 1;
 313                 }
 314
 315                 if (uio->uio_segflg == UIO_NOCOPY) {
 316                         /*
 317                          * Issuing a write with the same data backing the
 318                          * buffer.  Instantiate the buffer to collect the
 319                          * backing vm pages, then read-in any missing bits.
 320                          *
 321                          * This case is used by vop_stdputpages().
 322                          */
 323                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 324                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 325                         if ((bp->b_flags & B_CACHE) == 0) {
 326                                 bqrelse(bp);
 327                                 error = bread(ap->a_vp,
 328                                               uio->uio_offset - offset,
 329                                               HAMMER_BUFSIZE, &bp);
 330                         }
 331                 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
 332                         /*
 333                          * Even though we are entirely overwriting the buffer
 334                          * we may still have to zero it out to avoid a
 335                          * mmap/write visibility issue.
 336                          */
 337                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 338                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 339                         if ((bp->b_flags & B_CACHE) == 0)
 340                                 vfs_bio_clrbuf(bp);
 341                 } else if (uio->uio_offset - offset >= ip->ino_rec.ino_size) {
 342                         /*
 343                          * If the base offset of the buffer is beyond the
 344                          * file EOF, we don't have to issue a read.
 345                          */
 346                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 347                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 348                         vfs_bio_clrbuf(bp);
 349                 } else {
 350                         /*
 351                          * Partial overwrite, read in any missing bits then
 352                          * replace the portion being written.
 353                          */
 354                         error = bread(ap->a_vp, uio->uio_offset - offset,
 355                                       HAMMER_BUFSIZE, &bp);
 356                         if (error == 0)
 357                                 bheavy(bp);
 358                 }
 359                 if (error == 0)
 360                         error = uiomove((char *)bp->b_data + offset, n, uio);
 361
 362                 /*
 363                  * If we screwed up we have to undo any VM size changes we
 364                  * made.
 365                  */
 366                 if (error) {
 367                         brelse(bp);
 368                         if (fixsize) {
 369                                 vtruncbuf(ap->a_vp, ip->ino_rec.ino_size,
 370                                           HAMMER_BUFSIZE);
 371                         }
 372                         break;
 373                 }
 374                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 375                 if (ip->ino_rec.ino_size < uio->uio_offset) {
 376                         ip->ino_rec.ino_size = uio->uio_offset;
 377                         flags = HAMMER_INODE_RDIRTY;
 378                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
 379                 } else {
 380                         flags = 0;
 381                 }
 382                 ip->ino_rec.ino_mtime = trans.time;
 383                 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
 384                 hammer_modify_inode(&trans, ip, flags);
 385
 386                 if (ap->a_ioflag & IO_SYNC) {
 387                         bwrite(bp);
 388                 } else if (ap->a_ioflag & IO_DIRECT) {
 389                         bawrite(bp);
 390 #if 0
 391                 } else if ((ap->a_ioflag >> 16) == IO_SEQMAX &&
 392                            (uio->uio_offset & HAMMER_BUFMASK) == 0) {
 393                         /*
 394                          * XXX HAMMER can only fsync the whole inode,
 395                          * doing it on every buffer would be a bad idea.
 396                          */
 397                         /*
 398                          * If seqcount indicates sequential operation and
 399                          * we just finished filling a buffer, push it out
 400                          * now to prevent the buffer cache from becoming
 401                          * too full, which would trigger non-optimal
 402                          * flushes.
 403                          */
 404                         bdwrite(bp);
 405 #endif
 406                 } else {
 407                         bdwrite(bp);
 408                 }
 409         }
 410         hammer_done_transaction(&trans);
 411         return (error);
 412 }
 413
 414 /*
 415  * hammer_vop_access { vp, mode, cred }
 416  */
 417 static
 418 int
 419 hammer_vop_access(struct vop_access_args *ap)
 420 {
 421         struct hammer_inode *ip = VTOI(ap->a_vp);
 422         uid_t uid;
 423         gid_t gid;
 424         int error;
 425
 426         uid = hammer_to_unix_xid(&ip->ino_data.uid);
 427         gid = hammer_to_unix_xid(&ip->ino_data.gid);
 428
 429         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
 430                                   ip->ino_data.uflags);
 431         return (error);
 432 }
 433
 434 /*
 435  * hammer_vop_advlock { vp, id, op, fl, flags }
 436  */
 437 static
 438 int
 439 hammer_vop_advlock(struct vop_advlock_args *ap)
 440 {
 441         struct hammer_inode *ip = VTOI(ap->a_vp);
 442
 443         return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
 444 }
 445
 446 /*
 447  * hammer_vop_close { vp, fflag }
 448  */
 449 static
 450 int
 451 hammer_vop_close(struct vop_close_args *ap)
 452 {
 453         return (vop_stdclose(ap));
 454 }
 455
 456 /*
 457  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
 458  *
 459  * The operating system has already ensured that the directory entry
 460  * does not exist and done all appropriate namespace locking.
 461  */
 462 static
 463 int
 464 hammer_vop_ncreate(struct vop_ncreate_args *ap)
 465 {
 466         struct hammer_transaction trans;
 467         struct hammer_inode *dip;
 468         struct hammer_inode *nip;
 469         struct nchandle *nch;
 470         int error;
 471
 472         nch = ap->a_nch;
 473         dip = VTOI(ap->a_dvp);
 474
 475         if (dip->flags & HAMMER_INODE_RO)
 476                 return (EROFS);
 477
 478         /*
 479          * Create a transaction to cover the operations we perform.
 480          */
 481         hammer_start_transaction(&trans, dip->hmp);
 482
 483         /*
 484          * Create a new filesystem object of the requested type.  The
 485          * returned inode will be referenced and shared-locked to prevent
 486          * it from being moved to the flusher.
 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                 hammer_done_transaction(&trans);
 493                 *ap->a_vpp = NULL;
 494                 return (error);
 495         }
 496
 497         /*
 498          * Add the new filesystem object to the directory.  This will also
 499          * bump the inode's link count.
 500          */
 501         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 502         if (error)
 503                 kprintf("hammer_ip_add_directory error %d\n", error);
 504
 505         /*
 506          * Finish up.
 507          */
 508         if (error) {
 509                 hammer_rel_inode(nip, 0);
 510                 hammer_done_transaction(&trans);
 511                 *ap->a_vpp = NULL;
 512         } else {
 513                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 514                 hammer_done_transaction(&trans);
 515                 hammer_rel_inode(nip, 0);
 516                 if (error == 0) {
 517                         cache_setunresolved(ap->a_nch);
 518                         cache_setvp(ap->a_nch, *ap->a_vpp);
 519                 }
 520         }
 521         return (error);
 522 }
 523
 524 /*
 525  * hammer_vop_getattr { vp, vap }
 526  */
 527 static
 528 int
 529 hammer_vop_getattr(struct vop_getattr_args *ap)
 530 {
 531         struct hammer_inode *ip = VTOI(ap->a_vp);
 532         struct vattr *vap = ap->a_vap;
 533
 534 #if 0
 535         if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
 536             (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
 537             ip->obj_asof == XXX
 538         ) {
 539                 /* LAZYMOD XXX */
 540         }
 541         hammer_itimes(ap->a_vp);
 542 #endif
 543
 544         vap->va_fsid = ip->hmp->fsid_udev;
 545         vap->va_fileid = ip->ino_rec.base.base.obj_id;
 546         vap->va_mode = ip->ino_data.mode;
 547         vap->va_nlink = ip->ino_rec.ino_nlinks;
 548         vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
 549         vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
 550         vap->va_rmajor = 0;
 551         vap->va_rminor = 0;
 552         vap->va_size = ip->ino_rec.ino_size;
 553         hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
 554         hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
 555         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
 556         vap->va_flags = ip->ino_data.uflags;
 557         vap->va_gen = 1;        /* hammer inums are unique for all time */
 558         vap->va_blocksize = HAMMER_BUFSIZE;
 559         vap->va_bytes = (ip->ino_rec.ino_size + 63) & ~63;
 560         vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
 561         vap->va_filerev = 0;    /* XXX */
 562         /* mtime uniquely identifies any adjustments made to the file */
 563         vap->va_fsmid = ip->ino_rec.ino_mtime;
 564         vap->va_uid_uuid = ip->ino_data.uid;
 565         vap->va_gid_uuid = ip->ino_data.gid;
 566         vap->va_fsid_uuid = ip->hmp->fsid;
 567         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 568                           VA_FSID_UUID_VALID;
 569
 570         switch (ip->ino_rec.base.base.obj_type) {
 571         case HAMMER_OBJTYPE_CDEV:
 572         case HAMMER_OBJTYPE_BDEV:
 573                 vap->va_rmajor = ip->ino_data.rmajor;
 574                 vap->va_rminor = ip->ino_data.rminor;
 575                 break;
 576         default:
 577                 break;
 578         }
 579
 580         return(0);
 581 }
 582
 583 /*
 584  * hammer_vop_nresolve { nch, dvp, cred }
 585  *
 586  * Locate the requested directory entry.
 587  */
 588 static
 589 int
 590 hammer_vop_nresolve(struct vop_nresolve_args *ap)
 591 {
 592         struct hammer_transaction trans;
 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         hammer_simple_transaction(&trans, dip->hmp);
 619
 620         for (i = 0; i < nlen; ++i) {
 621                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
 622                         asof = hammer_str_to_tid(ncp->nc_name + i + 2);
 623                         flags |= HAMMER_INODE_RO;
 624                         break;
 625                 }
 626         }
 627         nlen = i;
 628
 629         /*
 630          * If there is no path component the time extension is relative to
 631          * dip.
 632          */
 633         if (nlen == 0) {
 634                 ip = hammer_get_inode(&trans, &dip->cache[1], dip->obj_id,
 635                                       asof, flags, &error);
 636                 if (error == 0) {
 637                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 638                         hammer_rel_inode(ip, 0);
 639                 } else {
 640                         vp = NULL;
 641                 }
 642                 if (error == 0) {
 643                         vn_unlock(vp);
 644                         cache_setvp(ap->a_nch, vp);
 645                         vrele(vp);
 646                 }
 647                 goto done;
 648         }
 649
 650         /*
 651          * Calculate the namekey and setup the key range for the scan.  This
 652          * works kinda like a chained hash table where the lower 32 bits
 653          * of the namekey synthesize the chain.
 654          *
 655          * The key range is inclusive of both key_beg and key_end.
 656          */
 657         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
 658
 659         error = hammer_init_cursor(&trans, &cursor, &dip->cache[0], dip);
 660         cursor.key_beg.obj_id = dip->obj_id;
 661         cursor.key_beg.key = namekey;
 662         cursor.key_beg.create_tid = 0;
 663         cursor.key_beg.delete_tid = 0;
 664         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 665         cursor.key_beg.obj_type = 0;
 666
 667         cursor.key_end = cursor.key_beg;
 668         cursor.key_end.key |= 0xFFFFFFFFULL;
 669         cursor.asof = asof;
 670         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
 671
 672         /*
 673          * Scan all matching records (the chain), locate the one matching
 674          * the requested path component.
 675          *
 676          * The hammer_ip_*() functions merge in-memory records with on-disk
 677          * records for the purposes of the search.
 678          */
 679         obj_id = 0;
 680
 681         if (error == 0) {
 682                 rec = NULL;
 683                 error = hammer_ip_first(&cursor);
 684                 while (error == 0) {
 685                         error = hammer_ip_resolve_data(&cursor);
 686                         if (error)
 687                                 break;
 688                         rec = cursor.record;
 689                         if (nlen == rec->entry.base.data_len &&
 690                             bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
 691                                 obj_id = rec->entry.obj_id;
 692                                 break;
 693                         }
 694                         error = hammer_ip_next(&cursor);
 695                 }
 696         }
 697         hammer_done_cursor(&cursor);
 698         if (error == 0) {
 699                 ip = hammer_get_inode(&trans, &dip->cache[1],
 700                                       obj_id, asof, flags, &error);
 701                 if (error == 0) {
 702                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 703                         hammer_rel_inode(ip, 0);
 704                 } else {
 705                         kprintf("nresolve: lookup %s failed dip %p (%016llx) on"
 706                                 " inode %016llx error %d\n",
 707                                 ncp->nc_name,
 708                                 dip, dip->obj_id, obj_id, error);
 709                         Debugger("x");
 710                         vp = NULL;
 711                 }
 712                 if (error == 0) {
 713                         vn_unlock(vp);
 714                         cache_setvp(ap->a_nch, vp);
 715                         vrele(vp);
 716                 }
 717         } else if (error == ENOENT) {
 718                 cache_setvp(ap->a_nch, NULL);
 719         }
 720 done:
 721         hammer_done_transaction(&trans);
 722         return (error);
 723 }
 724
 725 /*
 726  * hammer_vop_nlookupdotdot { dvp, vpp, cred }
 727  *
 728  * Locate the parent directory of a directory vnode.
 729  *
 730  * dvp is referenced but not locked.  *vpp must be returned referenced and
 731  * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
 732  * at the root, instead it could indicate that the directory we were in was
 733  * removed.
 734  *
 735  * NOTE: as-of sequences are not linked into the directory structure.  If
 736  * we are at the root with a different asof then the mount point, reload
 737  * the same directory with the mount point's asof.   I'm not sure what this
 738  * will do to NFS.  We encode ASOF stamps in NFS file handles so it might not
 739  * get confused, but it hasn't been tested.
 740  */
 741 static
 742 int
 743 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
 744 {
 745         struct hammer_transaction trans;
 746         struct hammer_inode *dip;
 747         struct hammer_inode *ip;
 748         int64_t parent_obj_id;
 749         hammer_tid_t asof;
 750         int error;
 751
 752         dip = VTOI(ap->a_dvp);
 753         asof = dip->obj_asof;
 754         parent_obj_id = dip->ino_data.parent_obj_id;
 755
 756         if (parent_obj_id == 0) {
 757                 if (dip->obj_id == HAMMER_OBJID_ROOT &&
 758                    asof != dip->hmp->asof) {
 759                         parent_obj_id = dip->obj_id;
 760                         asof = dip->hmp->asof;
 761                         *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
 762                         ksnprintf(*ap->a_fakename, 19, "0x%016llx",
 763                                    dip->obj_asof);
 764                 } else {
 765                         *ap->a_vpp = NULL;
 766                         return ENOENT;
 767                 }
 768         }
 769
 770         hammer_simple_transaction(&trans, dip->hmp);
 771
 772         ip = hammer_get_inode(&trans, &dip->cache[1], parent_obj_id,
 773                               asof, dip->flags, &error);
 774         if (ip) {
 775                 error = hammer_get_vnode(ip, LK_EXCLUSIVE, ap->a_vpp);
 776                 hammer_rel_inode(ip, 0);
 777         } else {
 778                 *ap->a_vpp = NULL;
 779         }
 780         hammer_done_transaction(&trans);
 781         return (error);
 782 }
 783
 784 /*
 785  * hammer_vop_nlink { nch, dvp, vp, cred }
 786  */
 787 static
 788 int
 789 hammer_vop_nlink(struct vop_nlink_args *ap)
 790 {
 791         struct hammer_transaction trans;
 792         struct hammer_inode *dip;
 793         struct hammer_inode *ip;
 794         struct nchandle *nch;
 795         int error;
 796
 797         nch = ap->a_nch;
 798         dip = VTOI(ap->a_dvp);
 799         ip = VTOI(ap->a_vp);
 800
 801         if (dip->flags & HAMMER_INODE_RO)
 802                 return (EROFS);
 803         if (ip->flags & HAMMER_INODE_RO)
 804                 return (EROFS);
 805
 806         /*
 807          * Create a transaction to cover the operations we perform.
 808          */
 809         hammer_start_transaction(&trans, dip->hmp);
 810
 811         /*
 812          * Add the filesystem object to the directory.  Note that neither
 813          * dip nor ip are referenced or locked, but their vnodes are
 814          * referenced.  This function will bump the inode's link count.
 815          */
 816         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 817
 818         /*
 819          * Finish up.
 820          */
 821         if (error == 0) {
 822                 cache_setunresolved(nch);
 823                 cache_setvp(nch, ap->a_vp);
 824         }
 825         hammer_done_transaction(&trans);
 826         return (error);
 827 }
 828
 829 /*
 830  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 831  *
 832  * The operating system has already ensured that the directory entry
 833  * does not exist and done all appropriate namespace locking.
 834  */
 835 static
 836 int
 837 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 838 {
 839         struct hammer_transaction trans;
 840         struct hammer_inode *dip;
 841         struct hammer_inode *nip;
 842         struct nchandle *nch;
 843         int error;
 844
 845         nch = ap->a_nch;
 846         dip = VTOI(ap->a_dvp);
 847
 848         if (dip->flags & HAMMER_INODE_RO)
 849                 return (EROFS);
 850
 851         /*
 852          * Create a transaction to cover the operations we perform.
 853          */
 854         hammer_start_transaction(&trans, dip->hmp);
 855
 856         /*
 857          * Create a new filesystem object of the requested type.  The
 858          * returned inode will be referenced but not locked.
 859          */
 860         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 861         if (error) {
 862                 kprintf("hammer_mkdir error %d\n", error);
 863                 hammer_done_transaction(&trans);
 864                 *ap->a_vpp = NULL;
 865                 return (error);
 866         }
 867         /*
 868          * Add the new filesystem object to the directory.  This will also
 869          * bump the inode's link count.
 870          */
 871         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 872         if (error)
 873                 kprintf("hammer_mkdir (add) error %d\n", error);
 874
 875         /*
 876          * Finish up.
 877          */
 878         if (error) {
 879                 hammer_rel_inode(nip, 0);
 880                 *ap->a_vpp = NULL;
 881         } else {
 882                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 883                 hammer_rel_inode(nip, 0);
 884                 if (error == 0) {
 885                         cache_setunresolved(ap->a_nch);
 886                         cache_setvp(ap->a_nch, *ap->a_vpp);
 887                 }
 888         }
 889         hammer_done_transaction(&trans);
 890         return (error);
 891 }
 892
 893 /*
 894  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 895  *
 896  * The operating system has already ensured that the directory entry
 897  * does not exist and done all appropriate namespace locking.
 898  */
 899 static
 900 int
 901 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 902 {
 903         struct hammer_transaction trans;
 904         struct hammer_inode *dip;
 905         struct hammer_inode *nip;
 906         struct nchandle *nch;
 907         int error;
 908
 909         nch = ap->a_nch;
 910         dip = VTOI(ap->a_dvp);
 911
 912         if (dip->flags & HAMMER_INODE_RO)
 913                 return (EROFS);
 914
 915         /*
 916          * Create a transaction to cover the operations we perform.
 917          */
 918         hammer_start_transaction(&trans, dip->hmp);
 919
 920         /*
 921          * Create a new filesystem object of the requested type.  The
 922          * returned inode will be referenced but not locked.
 923          */
 924         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 925         if (error) {
 926                 hammer_done_transaction(&trans);
 927                 *ap->a_vpp = NULL;
 928                 return (error);
 929         }
 930
 931         /*
 932          * Add the new filesystem object to the directory.  This will also
 933          * bump the inode's link count.
 934          */
 935         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 936
 937         /*
 938          * Finish up.
 939          */
 940         if (error) {
 941                 hammer_rel_inode(nip, 0);
 942                 *ap->a_vpp = NULL;
 943         } else {
 944                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 945                 hammer_rel_inode(nip, 0);
 946                 if (error == 0) {
 947                         cache_setunresolved(ap->a_nch);
 948                         cache_setvp(ap->a_nch, *ap->a_vpp);
 949                 }
 950         }
 951         hammer_done_transaction(&trans);
 952         return (error);
 953 }
 954
 955 /*
 956  * hammer_vop_open { vp, mode, cred, fp }
 957  */
 958 static
 959 int
 960 hammer_vop_open(struct vop_open_args *ap)
 961 {
 962         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
 963                 return (EROFS);
 964
 965         return(vop_stdopen(ap));
 966 }
 967
 968 /*
 969  * hammer_vop_pathconf { vp, name, retval }
 970  */
 971 static
 972 int
 973 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 974 {
 975         return EOPNOTSUPP;
 976 }
 977
 978 /*
 979  * hammer_vop_print { vp }
 980  */
 981 static
 982 int
 983 hammer_vop_print(struct vop_print_args *ap)
 984 {
 985         return EOPNOTSUPP;
 986 }
 987
 988 /*
 989  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 990  */
 991 static
 992 int
 993 hammer_vop_readdir(struct vop_readdir_args *ap)
 994 {
 995         struct hammer_transaction trans;
 996         struct hammer_cursor cursor;
 997         struct hammer_inode *ip;
 998         struct uio *uio;
 999         hammer_record_ondisk_t rec;
1000         hammer_base_elm_t base;
1001         int error;
1002         int cookie_index;
1003         int ncookies;
1004         off_t *cookies;
1005         off_t saveoff;
1006         int r;
1007
1008         ip = VTOI(ap->a_vp);
1009         uio = ap->a_uio;
1010         saveoff = uio->uio_offset;
1011
1012         if (ap->a_ncookies) {
1013                 ncookies = uio->uio_resid / 16 + 1;
1014                 if (ncookies > 1024)
1015                         ncookies = 1024;
1016                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1017                 cookie_index = 0;
1018         } else {
1019                 ncookies = -1;
1020                 cookies = NULL;
1021                 cookie_index = 0;
1022         }
1023
1024         hammer_simple_transaction(&trans, ip->hmp);
1025
1026         /*
1027          * Handle artificial entries
1028          */
1029         error = 0;
1030         if (saveoff == 0) {
1031                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1032                 if (r)
1033                         goto done;
1034                 if (cookies)
1035                         cookies[cookie_index] = saveoff;
1036                 ++saveoff;
1037                 ++cookie_index;
1038                 if (cookie_index == ncookies)
1039                         goto done;
1040         }
1041         if (saveoff == 1) {
1042                 if (ip->ino_data.parent_obj_id) {
1043                         r = vop_write_dirent(&error, uio,
1044                                              ip->ino_data.parent_obj_id,
1045                                              DT_DIR, 2, "..");
1046                 } else {
1047                         r = vop_write_dirent(&error, uio,
1048                                              ip->obj_id, DT_DIR, 2, "..");
1049                 }
1050                 if (r)
1051                         goto done;
1052                 if (cookies)
1053                         cookies[cookie_index] = saveoff;
1054                 ++saveoff;
1055                 ++cookie_index;
1056                 if (cookie_index == ncookies)
1057                         goto done;
1058         }
1059
1060         /*
1061          * Key range (begin and end inclusive) to scan.  Directory keys
1062          * directly translate to a 64 bit 'seek' position.
1063          */
1064         hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1065         cursor.key_beg.obj_id = ip->obj_id;
1066         cursor.key_beg.create_tid = 0;
1067         cursor.key_beg.delete_tid = 0;
1068         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1069         cursor.key_beg.obj_type = 0;
1070         cursor.key_beg.key = saveoff;
1071
1072         cursor.key_end = cursor.key_beg;
1073         cursor.key_end.key = HAMMER_MAX_KEY;
1074         cursor.asof = ip->obj_asof;
1075         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1076
1077         error = hammer_ip_first(&cursor);
1078
1079         while (error == 0) {
1080                 error = hammer_ip_resolve_record_and_data(&cursor);
1081                 if (error)
1082                         break;
1083                 rec = cursor.record;
1084                 base = &rec->base.base;
1085                 saveoff = base->key;
1086
1087                 if (base->obj_id != ip->obj_id)
1088                         panic("readdir: bad record at %p", cursor.node);
1089
1090                 r = vop_write_dirent(
1091                              &error, uio, rec->entry.obj_id,
1092                              hammer_get_dtype(rec->entry.base.base.obj_type),
1093                              rec->entry.base.data_len,
1094                              (void *)cursor.data);
1095                 if (r)
1096                         break;
1097                 ++saveoff;
1098                 if (cookies)
1099                         cookies[cookie_index] = base->key;
1100                 ++cookie_index;
1101                 if (cookie_index == ncookies)
1102                         break;
1103                 error = hammer_ip_next(&cursor);
1104         }
1105         hammer_done_cursor(&cursor);
1106
1107 done:
1108         hammer_done_transaction(&trans);
1109
1110         if (ap->a_eofflag)
1111                 *ap->a_eofflag = (error == ENOENT);
1112         uio->uio_offset = saveoff;
1113         if (error && cookie_index == 0) {
1114                 if (error == ENOENT)
1115                         error = 0;
1116                 if (cookies) {
1117                         kfree(cookies, M_TEMP);
1118                         *ap->a_ncookies = 0;
1119                         *ap->a_cookies = NULL;
1120                 }
1121         } else {
1122                 if (error == ENOENT)
1123                         error = 0;
1124                 if (cookies) {
1125                         *ap->a_ncookies = cookie_index;
1126                         *ap->a_cookies = cookies;
1127                 }
1128         }
1129         return(error);
1130 }
1131
1132 /*
1133  * hammer_vop_readlink { vp, uio, cred }
1134  */
1135 static
1136 int
1137 hammer_vop_readlink(struct vop_readlink_args *ap)
1138 {
1139         struct hammer_transaction trans;
1140         struct hammer_cursor cursor;
1141         struct hammer_inode *ip;
1142         int error;
1143
1144         ip = VTOI(ap->a_vp);
1145
1146         hammer_simple_transaction(&trans, ip->hmp);
1147
1148         hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1149
1150         /*
1151          * Key range (begin and end inclusive) to scan.  Directory keys
1152          * directly translate to a 64 bit 'seek' position.
1153          */
1154         cursor.key_beg.obj_id = ip->obj_id;
1155         cursor.key_beg.create_tid = 0;
1156         cursor.key_beg.delete_tid = 0;
1157         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1158         cursor.key_beg.obj_type = 0;
1159         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1160         cursor.asof = ip->obj_asof;
1161         cursor.flags |= HAMMER_CURSOR_ASOF;
1162
1163         error = hammer_ip_lookup(&cursor, ip);
1164         if (error == 0) {
1165                 error = hammer_ip_resolve_data(&cursor);
1166                 if (error == 0) {
1167                         error = uiomove((char *)cursor.data,
1168                                         cursor.record->base.data_len,
1169                                         ap->a_uio);
1170                 }
1171         }
1172         hammer_done_cursor(&cursor);
1173         hammer_done_transaction(&trans);
1174         return(error);
1175 }
1176
1177 /*
1178  * hammer_vop_nremove { nch, dvp, cred }
1179  */
1180 static
1181 int
1182 hammer_vop_nremove(struct vop_nremove_args *ap)
1183 {
1184         struct hammer_transaction trans;
1185         int error;
1186
1187         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1188         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1189         hammer_done_transaction(&trans);
1190
1191         return (error);
1192 }
1193
1194 /*
1195  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1196  */
1197 static
1198 int
1199 hammer_vop_nrename(struct vop_nrename_args *ap)
1200 {
1201         struct hammer_transaction trans;
1202         struct namecache *fncp;
1203         struct namecache *tncp;
1204         struct hammer_inode *fdip;
1205         struct hammer_inode *tdip;
1206         struct hammer_inode *ip;
1207         struct hammer_cursor cursor;
1208         union hammer_record_ondisk *rec;
1209         int64_t namekey;
1210         int error;
1211
1212         fdip = VTOI(ap->a_fdvp);
1213         tdip = VTOI(ap->a_tdvp);
1214         fncp = ap->a_fnch->ncp;
1215         tncp = ap->a_tnch->ncp;
1216         ip = VTOI(fncp->nc_vp);
1217         KKASSERT(ip != NULL);
1218
1219         if (fdip->flags & HAMMER_INODE_RO)
1220                 return (EROFS);
1221         if (tdip->flags & HAMMER_INODE_RO)
1222                 return (EROFS);
1223         if (ip->flags & HAMMER_INODE_RO)
1224                 return (EROFS);
1225
1226         hammer_start_transaction(&trans, fdip->hmp);
1227
1228         /*
1229          * Remove tncp from the target directory and then link ip as
1230          * tncp. XXX pass trans to dounlink
1231          *
1232          * Force the inode sync-time to match the transaction so it is
1233          * in-sync with the creation of the target directory entry.
1234          */
1235         error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1236         if (error == 0 || error == ENOENT) {
1237                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1238                 if (error == 0) {
1239                         ip->ino_data.parent_obj_id = tdip->obj_id;
1240                         hammer_modify_inode(&trans, ip, HAMMER_INODE_DDIRTY);
1241                 }
1242         }
1243         if (error)
1244                 goto failed; /* XXX */
1245
1246         /*
1247          * Locate the record in the originating directory and remove it.
1248          *
1249          * Calculate the namekey and setup the key range for the scan.  This
1250          * works kinda like a chained hash table where the lower 32 bits
1251          * of the namekey synthesize the chain.
1252          *
1253          * The key range is inclusive of both key_beg and key_end.
1254          */
1255         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1256 retry:
1257         hammer_init_cursor(&trans, &cursor, &fdip->cache[0], fdip);
1258         cursor.key_beg.obj_id = fdip->obj_id;
1259         cursor.key_beg.key = namekey;
1260         cursor.key_beg.create_tid = 0;
1261         cursor.key_beg.delete_tid = 0;
1262         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1263         cursor.key_beg.obj_type = 0;
1264
1265         cursor.key_end = cursor.key_beg;
1266         cursor.key_end.key |= 0xFFFFFFFFULL;
1267         cursor.asof = fdip->obj_asof;
1268         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1269
1270         /*
1271          * Scan all matching records (the chain), locate the one matching
1272          * the requested path component.
1273          *
1274          * The hammer_ip_*() functions merge in-memory records with on-disk
1275          * records for the purposes of the search.
1276          */
1277         error = hammer_ip_first(&cursor);
1278         while (error == 0) {
1279                 if (hammer_ip_resolve_data(&cursor) != 0)
1280                         break;
1281                 rec = cursor.record;
1282                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1283                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1284                         break;
1285                 }
1286                 error = hammer_ip_next(&cursor);
1287         }
1288
1289         /*
1290          * If all is ok we have to get the inode so we can adjust nlinks.
1291          *
1292          * WARNING: hammer_ip_del_directory() may have to terminate the
1293          * cursor to avoid a recursion.  It's ok to call hammer_done_cursor()
1294          * twice.
1295          */
1296         if (error == 0)
1297                 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1298
1299         /*
1300          * XXX A deadlock here will break rename's atomicy for the purposes
1301          * of crash recovery.
1302          */
1303         if (error == EDEADLK) {
1304                 hammer_done_cursor(&cursor);
1305                 goto retry;
1306         }
1307
1308         /*
1309          * Cleanup and tell the kernel that the rename succeeded.
1310          */
1311         hammer_done_cursor(&cursor);
1312         if (error == 0)
1313                 cache_rename(ap->a_fnch, ap->a_tnch);
1314
1315 failed:
1316         hammer_done_transaction(&trans);
1317         return (error);
1318 }
1319
1320 /*
1321  * hammer_vop_nrmdir { nch, dvp, cred }
1322  */
1323 static
1324 int
1325 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1326 {
1327         struct hammer_transaction trans;
1328         int error;
1329
1330         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1331         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1332         hammer_done_transaction(&trans);
1333
1334         return (error);
1335 }
1336
1337 /*
1338  * hammer_vop_setattr { vp, vap, cred }
1339  */
1340 static
1341 int
1342 hammer_vop_setattr(struct vop_setattr_args *ap)
1343 {
1344         struct hammer_transaction trans;
1345         struct vattr *vap;
1346         struct hammer_inode *ip;
1347         int modflags;
1348         int error;
1349         int truncating;
1350         off_t aligned_size;
1351         u_int32_t flags;
1352         uuid_t uuid;
1353
1354         vap = ap->a_vap;
1355         ip = ap->a_vp->v_data;
1356         modflags = 0;
1357
1358         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1359                 return(EROFS);
1360         if (ip->flags & HAMMER_INODE_RO)
1361                 return (EROFS);
1362
1363         hammer_start_transaction(&trans, ip->hmp);
1364         error = 0;
1365
1366         if (vap->va_flags != VNOVAL) {
1367                 flags = ip->ino_data.uflags;
1368                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1369                                          hammer_to_unix_xid(&ip->ino_data.uid),
1370                                          ap->a_cred);
1371                 if (error == 0) {
1372                         if (ip->ino_data.uflags != flags) {
1373                                 ip->ino_data.uflags = flags;
1374                                 modflags |= HAMMER_INODE_DDIRTY;
1375                         }
1376                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1377                                 error = 0;
1378                                 goto done;
1379                         }
1380                 }
1381                 goto done;
1382         }
1383         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1384                 error = EPERM;
1385                 goto done;
1386         }
1387         if (vap->va_uid != (uid_t)VNOVAL) {
1388                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1389                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1390                         ip->ino_data.uid = uuid;
1391                         modflags |= HAMMER_INODE_DDIRTY;
1392                 }
1393         }
1394         if (vap->va_gid != (uid_t)VNOVAL) {
1395                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1396                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1397                         ip->ino_data.gid = uuid;
1398                         modflags |= HAMMER_INODE_DDIRTY;
1399                 }
1400         }
1401         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1402                 switch(ap->a_vp->v_type) {
1403                 case VREG:
1404                         if (vap->va_size == ip->ino_rec.ino_size)
1405                                 break;
1406                         /*
1407                          * XXX break atomicy, we can deadlock the backend
1408                          * if we do not release the lock.  Probably not a
1409                          * big deal here.
1410                          */
1411                         if (vap->va_size < ip->ino_rec.ino_size) {
1412                                 vtruncbuf(ap->a_vp, vap->va_size,
1413                                           HAMMER_BUFSIZE);
1414                                 truncating = 1;
1415                         } else {
1416                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1417                                 truncating = 0;
1418                         }
1419                         ip->ino_rec.ino_size = vap->va_size;
1420                         modflags |= HAMMER_INODE_RDIRTY;
1421                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1422                                        ~HAMMER_BUFMASK64;
1423
1424                         /*
1425                          * on-media truncation is cached in the inode until
1426                          * the inode is synchronized.
1427                          */
1428                         if (truncating) {
1429                                 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1430                                         ip->flags |= HAMMER_INODE_TRUNCATED;
1431                                         ip->trunc_off = vap->va_size;
1432                                 } else if (ip->trunc_off > vap->va_size) {
1433                                         ip->trunc_off = vap->va_size;
1434                                 }
1435                         }
1436
1437                         /*
1438                          * If truncating we have to clean out a portion of
1439                          * the last block on-disk.  We do this in the
1440                          * front-end buffer cache.
1441                          */
1442                         if (truncating && vap->va_size < aligned_size) {
1443                                 struct buf *bp;
1444                                 int offset;
1445
1446                                 offset = vap->va_size & HAMMER_BUFMASK;
1447                                 error = bread(ap->a_vp,
1448                                               aligned_size - HAMMER_BUFSIZE,
1449                                               HAMMER_BUFSIZE, &bp);
1450                                 if (error == 0) {
1451                                         bzero(bp->b_data + offset,
1452                                               HAMMER_BUFSIZE - offset);
1453                                         bdwrite(bp);
1454                                 } else {
1455                                         brelse(bp);
1456                                 }
1457                         }
1458                         break;
1459                 case VDATABASE:
1460                         if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1461                                 ip->flags |= HAMMER_INODE_TRUNCATED;
1462                                 ip->trunc_off = vap->va_size;
1463                         } else if (ip->trunc_off > vap->va_size) {
1464                                 ip->trunc_off = vap->va_size;
1465                         }
1466                         ip->ino_rec.ino_size = vap->va_size;
1467                         modflags |= HAMMER_INODE_RDIRTY;
1468                         break;
1469                 default:
1470                         error = EINVAL;
1471                         goto done;
1472                 }
1473                 break;
1474         }
1475         if (vap->va_atime.tv_sec != VNOVAL) {
1476                 ip->ino_rec.ino_atime =
1477                         hammer_timespec_to_transid(&vap->va_atime);
1478                 modflags |= HAMMER_INODE_ITIMES;
1479         }
1480         if (vap->va_mtime.tv_sec != VNOVAL) {
1481                 ip->ino_rec.ino_mtime =
1482                         hammer_timespec_to_transid(&vap->va_mtime);
1483                 modflags |= HAMMER_INODE_ITIMES;
1484         }
1485         if (vap->va_mode != (mode_t)VNOVAL) {
1486                 if (ip->ino_data.mode != vap->va_mode) {
1487                         ip->ino_data.mode = vap->va_mode;
1488                         modflags |= HAMMER_INODE_DDIRTY;
1489                 }
1490         }
1491 done:
1492         if (error == 0)
1493                 hammer_modify_inode(&trans, ip, modflags);
1494         hammer_done_transaction(&trans);
1495         return (error);
1496 }
1497
1498 /*
1499  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1500  */
1501 static
1502 int
1503 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1504 {
1505         struct hammer_transaction trans;
1506         struct hammer_inode *dip;
1507         struct hammer_inode *nip;
1508         struct nchandle *nch;
1509         hammer_record_t record;
1510         int error;
1511         int bytes;
1512
1513         ap->a_vap->va_type = VLNK;
1514
1515         nch = ap->a_nch;
1516         dip = VTOI(ap->a_dvp);
1517
1518         if (dip->flags & HAMMER_INODE_RO)
1519                 return (EROFS);
1520
1521         /*
1522          * Create a transaction to cover the operations we perform.
1523          */
1524         hammer_start_transaction(&trans, dip->hmp);
1525
1526         /*
1527          * Create a new filesystem object of the requested type.  The
1528          * returned inode will be referenced but not locked.
1529          */
1530
1531         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1532         if (error) {
1533                 hammer_done_transaction(&trans);
1534                 *ap->a_vpp = NULL;
1535                 return (error);
1536         }
1537
1538         /*
1539          * Add a record representing the symlink.  symlink stores the link
1540          * as pure data, not a string, and is no \0 terminated.
1541          */
1542         if (error == 0) {
1543                 record = hammer_alloc_mem_record(nip);
1544                 bytes = strlen(ap->a_target);
1545
1546                 record->rec.base.base.key = HAMMER_FIXKEY_SYMLINK;
1547                 record->rec.base.base.rec_type = HAMMER_RECTYPE_FIX;
1548                 record->rec.base.data_len = bytes;
1549                 record->data = (void *)ap->a_target;
1550                 /* will be reallocated by routine below */
1551                 error = hammer_ip_add_record(&trans, record);
1552
1553                 /*
1554                  * Set the file size to the length of the link.
1555                  */
1556                 if (error == 0) {
1557                         nip->ino_rec.ino_size = bytes;
1558                         hammer_modify_inode(&trans, nip, HAMMER_INODE_RDIRTY);
1559                 }
1560         }
1561         if (error == 0)
1562                 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1563
1564         /*
1565          * Finish up.
1566          */
1567         if (error) {
1568                 hammer_rel_inode(nip, 0);
1569                 *ap->a_vpp = NULL;
1570         } else {
1571                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1572                 hammer_rel_inode(nip, 0);
1573                 if (error == 0) {
1574                         cache_setunresolved(ap->a_nch);
1575                         cache_setvp(ap->a_nch, *ap->a_vpp);
1576                 }
1577         }
1578         hammer_done_transaction(&trans);
1579         return (error);
1580 }
1581
1582 /*
1583  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1584  */
1585 static
1586 int
1587 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1588 {
1589         struct hammer_transaction trans;
1590         int error;
1591
1592         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1593         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1594                                 ap->a_cred, ap->a_flags);
1595         hammer_done_transaction(&trans);
1596
1597         return (error);
1598 }
1599
1600 /*
1601  * hammer_vop_ioctl { vp, command, data, fflag, cred }
1602  */
1603 static
1604 int
1605 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1606 {
1607         struct hammer_inode *ip = ap->a_vp->v_data;
1608
1609         return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1610                             ap->a_fflag, ap->a_cred));
1611 }
1612
1613 static
1614 int
1615 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1616 {
1617         struct mount *mp;
1618         int error;
1619
1620         mp = ap->a_head.a_ops->head.vv_mount;
1621
1622         switch(ap->a_op) {
1623         case MOUNTCTL_SET_EXPORT:
1624                 if (ap->a_ctllen != sizeof(struct export_args))
1625                         error = EINVAL;
1626                 error = hammer_vfs_export(mp, ap->a_op,
1627                                       (const struct export_args *)ap->a_ctl);
1628                 break;
1629         default:
1630                 error = journal_mountctl(ap);
1631                 break;
1632         }
1633         return(error);
1634 }
1635
1636 /*
1637  * hammer_vop_strategy { vp, bio }
1638  *
1639  * Strategy call, used for regular file read & write only.  Note that the
1640  * bp may represent a cluster.
1641  *
1642  * To simplify operation and allow better optimizations in the future,
1643  * this code does not make any assumptions with regards to buffer alignment
1644  * or size.
1645  */
1646 static
1647 int
1648 hammer_vop_strategy(struct vop_strategy_args *ap)
1649 {
1650         struct buf *bp;
1651         int error;
1652
1653         bp = ap->a_bio->bio_buf;
1654
1655         switch(bp->b_cmd) {
1656         case BUF_CMD_READ:
1657                 error = hammer_vop_strategy_read(ap);
1658                 break;
1659         case BUF_CMD_WRITE:
1660                 error = hammer_vop_strategy_write(ap);
1661                 break;
1662         default:
1663                 bp->b_error = error = EINVAL;
1664                 bp->b_flags |= B_ERROR;
1665                 biodone(ap->a_bio);
1666                 break;
1667         }
1668         return (error);
1669 }
1670
1671 /*
1672  * Read from a regular file.  Iterate the related records and fill in the
1673  * BIO/BUF.  Gaps are zero-filled.
1674  *
1675  * The support code in hammer_object.c should be used to deal with mixed
1676  * in-memory and on-disk records.
1677  *
1678  * XXX atime update
1679  */
1680 static
1681 int
1682 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1683 {
1684         struct hammer_transaction trans;
1685         struct hammer_inode *ip;
1686         struct hammer_cursor cursor;
1687         hammer_record_ondisk_t rec;
1688         hammer_base_elm_t base;
1689         struct bio *bio;
1690         struct buf *bp;
1691         int64_t rec_offset;
1692         int64_t ran_end;
1693         int64_t tmp64;
1694         int error;
1695         int boff;
1696         int roff;
1697         int n;
1698
1699         bio = ap->a_bio;
1700         bp = bio->bio_buf;
1701         ip = ap->a_vp->v_data;
1702
1703         hammer_simple_transaction(&trans, ip->hmp);
1704         hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1705
1706         /*
1707          * Key range (begin and end inclusive) to scan.  Note that the key's
1708          * stored in the actual records represent BASE+LEN, not BASE.  The
1709          * first record containing bio_offset will have a key > bio_offset.
1710          */
1711         cursor.key_beg.obj_id = ip->obj_id;
1712         cursor.key_beg.create_tid = 0;
1713         cursor.key_beg.delete_tid = 0;
1714         cursor.key_beg.obj_type = 0;
1715         cursor.key_beg.key = bio->bio_offset + 1;
1716         cursor.asof = ip->obj_asof;
1717         cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1718
1719         cursor.key_end = cursor.key_beg;
1720         KKASSERT(ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_REGFILE);
1721 #if 0
1722         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1723                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1724                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1725                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1726         } else
1727 #endif
1728         {
1729                 ran_end = bio->bio_offset + bp->b_bufsize;
1730                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1731                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1732                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1733                 if (tmp64 < ran_end)
1734                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1735                 else
1736                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1737         }
1738         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1739
1740         error = hammer_ip_first(&cursor);
1741         boff = 0;
1742
1743         while (error == 0) {
1744                 error = hammer_ip_resolve_data(&cursor);
1745                 if (error)
1746                         break;
1747                 rec = cursor.record;
1748                 base = &rec->base.base;
1749
1750                 rec_offset = base->key - rec->data.base.data_len;
1751
1752                 /*
1753                  * Calculate the gap, if any, and zero-fill it.
1754                  */
1755                 n = (int)(rec_offset - (bio->bio_offset + boff));
1756                 if (n > 0) {
1757                         if (n > bp->b_bufsize - boff)
1758                                 n = bp->b_bufsize - boff;
1759                         bzero((char *)bp->b_data + boff, n);
1760                         boff += n;
1761                         n = 0;
1762                 }
1763
1764                 /*
1765                  * Calculate the data offset in the record and the number
1766                  * of bytes we can copy.
1767                  *
1768                  * Note there is a degenerate case here where boff may
1769                  * already be at bp->b_bufsize.
1770                  */
1771                 roff = -n;
1772                 rec_offset += roff;
1773                 n = rec->data.base.data_len - roff;
1774                 KKASSERT(n > 0);
1775                 if (n > bp->b_bufsize - boff)
1776                         n = bp->b_bufsize - boff;
1777
1778                 /*
1779                  * If we cached a truncation point on our front-end the
1780                  * on-disk version may still have physical records beyond
1781                  * that point.  Truncate visibility.
1782                  */
1783                 if (ip->trunc_off <= rec_offset)
1784                         n = 0;
1785                 else if (ip->trunc_off < rec_offset + n)
1786                         n = (int)(ip->trunc_off - rec_offset);
1787
1788                 /*
1789                  * Copy
1790                  */
1791                 if (n) {
1792                         bcopy((char *)cursor.data + roff,
1793                               (char *)bp->b_data + boff, n);
1794                         boff += n;
1795                 }
1796                 if (boff == bp->b_bufsize)
1797                         break;
1798                 error = hammer_ip_next(&cursor);
1799         }
1800         hammer_done_cursor(&cursor);
1801         hammer_done_transaction(&trans);
1802
1803         /*
1804          * There may have been a gap after the last record
1805          */
1806         if (error == ENOENT)
1807                 error = 0;
1808         if (error == 0 && boff != bp->b_bufsize) {
1809                 KKASSERT(boff < bp->b_bufsize);
1810                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1811                 /* boff = bp->b_bufsize; */
1812         }
1813         bp->b_resid = 0;
1814         bp->b_error = error;
1815         if (error)
1816                 bp->b_flags |= B_ERROR;
1817         biodone(ap->a_bio);
1818         return(error);
1819 }
1820
1821 /*
1822  * Write to a regular file.   Because this is a strategy call the OS is
1823  * trying to actually sync data to the media.   HAMMER can only flush
1824  * the entire inode (so the TID remains properly synchronized).
1825  *
1826  * Basically all we do here is place the bio on the inode's flush queue
1827  * and activate the flusher.
1828  */
1829 static
1830 int
1831 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1832 {
1833         hammer_inode_t ip;
1834         struct bio *bio;
1835         struct buf *bp;
1836
1837         bio = ap->a_bio;
1838         bp = bio->bio_buf;
1839         ip = ap->a_vp->v_data;
1840
1841         if (ip->flags & HAMMER_INODE_RO) {
1842                 bp->b_error = EROFS;
1843                 bp->b_flags |= B_ERROR;
1844                 biodone(ap->a_bio);
1845                 return(EROFS);
1846         }
1847
1848         /*
1849          * If the inode is being flushed we cannot re-queue buffers
1850          * it may have already flushed, or it could result in duplicate
1851          * records in the database.
1852          */
1853         BUF_KERNPROC(bp);
1854         if (ip->flags & HAMMER_INODE_WRITE_ALT)
1855                 TAILQ_INSERT_TAIL(&ip->bio_alt_list, bio, bio_act);
1856         else
1857                 TAILQ_INSERT_TAIL(&ip->bio_list, bio, bio_act);
1858         ++hammer_bio_count;
1859         hammer_modify_inode(NULL, ip, HAMMER_INODE_BUFS);
1860
1861         hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1862 #if 0
1863         /*
1864          * XXX
1865          *
1866          * If the write was not part of an integrated flush operation then
1867          * signal a flush.
1868          */
1869         if (ip->flush_state != HAMMER_FST_FLUSH ||
1870             (ip->flags & HAMMER_INODE_WRITE_ALT)) {
1871                 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1872         }
1873 #endif
1874         return(0);
1875 }
1876
1877 /*
1878  * Backend code which actually performs the write to the media.  This
1879  * routine is typically called from the flusher.  The bio will be disposed
1880  * of (biodone'd) by this routine.
1881  *
1882  * Iterate the related records and mark for deletion.  If existing edge
1883  * records (left and right side) overlap our write they have to be marked
1884  * deleted and new records created, usually referencing a portion of the
1885  * original data.  Then add a record to represent the buffer.
1886  */
1887 int
1888 hammer_dowrite(hammer_cursor_t cursor, hammer_inode_t ip, struct bio *bio)
1889 {
1890         struct buf *bp = bio->bio_buf;
1891         int error;
1892
1893         KKASSERT(ip->flush_state == HAMMER_FST_FLUSH);
1894
1895         /*
1896          * If the inode is going or gone, just throw away any frontend
1897          * buffers.
1898          */
1899         if (ip->flags & HAMMER_INODE_DELETED) {
1900                 bp->b_resid = 0;
1901                 biodone(bio);
1902         }
1903
1904         /*
1905          * Delete any records overlapping our range.  This function will
1906          * (eventually) properly truncate partial overlaps.
1907          */
1908         if (ip->sync_ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1909                 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
1910                                                bio->bio_offset);
1911         } else {
1912                 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
1913                                                bio->bio_offset +
1914                                                 bp->b_bufsize - 1);
1915         }
1916
1917         /*
1918          * Add a single record to cover the write.  We can write a record
1919          * with only the actual file data - for example, a small 200 byte
1920          * file does not have to write out a 16K record.
1921          *
1922          * While the data size does not have to be aligned, we still do it
1923          * to reduce fragmentation in a future allocation model.
1924          */
1925         if (error == 0) {
1926                 int limit_size;
1927
1928                 if (ip->sync_ino_rec.ino_size - bio->bio_offset >
1929                     bp->b_bufsize) {
1930                             limit_size = bp->b_bufsize;
1931                 } else {
1932                         limit_size = (int)(ip->sync_ino_rec.ino_size -
1933                                            bio->bio_offset);
1934                         KKASSERT(limit_size >= 0);
1935                         limit_size = (limit_size + 63) & ~63;
1936                 }
1937                 if (limit_size) {
1938                         error = hammer_ip_sync_data(cursor, ip, bio->bio_offset,
1939                                                     bp->b_data, limit_size);
1940                 }
1941         }
1942         if (error)
1943                 Debugger("hammer_dowrite: error");
1944
1945         if (error) {
1946                 bp->b_resid = bp->b_bufsize;
1947                 bp->b_error = error;
1948                 bp->b_flags |= B_ERROR;
1949         } else {
1950                 bp->b_resid = 0;
1951         }
1952         biodone(bio);
1953         --hammer_bio_count;
1954         return(error);
1955 }
1956
1957 /*
1958  * dounlink - disconnect a directory entry
1959  *
1960  * XXX whiteout support not really in yet
1961  */
1962 static int
1963 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
1964                 struct vnode *dvp, struct ucred *cred, int flags)
1965 {
1966         struct namecache *ncp;
1967         hammer_inode_t dip;
1968         hammer_inode_t ip;
1969         hammer_record_ondisk_t rec;
1970         struct hammer_cursor cursor;
1971         int64_t namekey;
1972         int error;
1973
1974         /*
1975          * Calculate the namekey and setup the key range for the scan.  This
1976          * works kinda like a chained hash table where the lower 32 bits
1977          * of the namekey synthesize the chain.
1978          *
1979          * The key range is inclusive of both key_beg and key_end.
1980          */
1981         dip = VTOI(dvp);
1982         ncp = nch->ncp;
1983
1984         if (dip->flags & HAMMER_INODE_RO)
1985                 return (EROFS);
1986
1987         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1988 retry:
1989         hammer_init_cursor(trans, &cursor, &dip->cache[0], dip);
1990         cursor.key_beg.obj_id = dip->obj_id;
1991         cursor.key_beg.key = namekey;
1992         cursor.key_beg.create_tid = 0;
1993         cursor.key_beg.delete_tid = 0;
1994         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1995         cursor.key_beg.obj_type = 0;
1996
1997         cursor.key_end = cursor.key_beg;
1998         cursor.key_end.key |= 0xFFFFFFFFULL;
1999         cursor.asof = dip->obj_asof;
2000         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
2001
2002         /*
2003          * Scan all matching records (the chain), locate the one matching
2004          * the requested path component.  info->last_error contains the
2005          * error code on search termination and could be 0, ENOENT, or
2006          * something else.
2007          *
2008          * The hammer_ip_*() functions merge in-memory records with on-disk
2009          * records for the purposes of the search.
2010          */
2011         rec = NULL;
2012         error = hammer_ip_first(&cursor);
2013
2014         while (error == 0) {
2015                 error = hammer_ip_resolve_data(&cursor);
2016                 if (error)
2017                         break;
2018                 rec = cursor.record;
2019                 if (ncp->nc_nlen == rec->entry.base.data_len &&
2020                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
2021                         break;
2022                 }
2023                 error = hammer_ip_next(&cursor);
2024         }
2025
2026         /*
2027          * If all is ok we have to get the inode so we can adjust nlinks.
2028          *
2029          * If the target is a directory, it must be empty.
2030          */
2031         if (error == 0) {
2032                 ip = hammer_get_inode(trans, &dip->cache[1],
2033                                       rec->entry.obj_id,
2034                                       dip->hmp->asof, 0, &error);
2035                 if (error == ENOENT) {
2036                         kprintf("obj_id %016llx\n", rec->entry.obj_id);
2037                         Debugger("ENOENT unlinking object that should exist");
2038                 }
2039
2040                 /*
2041                  * If we are trying to remove a directory the directory must
2042                  * be empty.
2043                  *
2044                  * WARNING: hammer_ip_check_directory_empty() may have to
2045                  * terminate the cursor to avoid a deadlock.  It is ok to
2046                  * call hammer_done_cursor() twice.
2047                  */
2048                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
2049                                   HAMMER_OBJTYPE_DIRECTORY) {
2050                         error = hammer_ip_check_directory_empty(trans, &cursor,
2051                                                                 ip);
2052                 }
2053
2054                 /*
2055                  * Delete the directory entry.
2056                  *
2057                  * WARNING: hammer_ip_del_directory() may have to terminate
2058                  * the cursor to avoid a deadlock.  It is ok to call
2059                  * hammer_done_cursor() twice.
2060                  */
2061                 if (error == 0) {
2062                         error = hammer_ip_del_directory(trans, &cursor,
2063                                                         dip, ip);
2064                 }
2065                 if (error == 0) {
2066                         cache_setunresolved(nch);
2067                         cache_setvp(nch, NULL);
2068                         /* XXX locking */
2069                         if (ip->vp)
2070                                 cache_inval_vp(ip->vp, CINV_DESTROY);
2071                 }
2072                 hammer_rel_inode(ip, 0);
2073         }
2074         hammer_done_cursor(&cursor);
2075         if (error == EDEADLK)
2076                 goto retry;
2077
2078         return (error);
2079 }
2080
2081 /************************************************************************
2082  *                          FIFO AND SPECFS OPS                         *
2083  ************************************************************************
2084  *
2085  */
2086
2087 static int
2088 hammer_vop_fifoclose (struct vop_close_args *ap)
2089 {
2090         /* XXX update itimes */
2091         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2092 }
2093
2094 static int
2095 hammer_vop_fiforead (struct vop_read_args *ap)
2096 {
2097         int error;
2098
2099         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2100         /* XXX update access time */
2101         return (error);
2102 }
2103
2104 static int
2105 hammer_vop_fifowrite (struct vop_write_args *ap)
2106 {
2107         int error;
2108
2109         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2110         /* XXX update access time */
2111         return (error);
2112 }
2113
2114 static int
2115 hammer_vop_specclose (struct vop_close_args *ap)
2116 {
2117         /* XXX update itimes */
2118         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2119 }
2120
2121 static int
2122 hammer_vop_specread (struct vop_read_args *ap)
2123 {
2124         /* XXX update access time */
2125         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2126 }
2127
2128 static int
2129 hammer_vop_specwrite (struct vop_write_args *ap)
2130 {
2131         /* XXX update last change time */
2132         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2133 }
2134