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 1.42 2008/04/27 21:07:15 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                 hammer_lock_sh(&ip->lock);
 376                 if (ip->ino_rec.ino_size < uio->uio_offset) {
 377                         ip->ino_rec.ino_size = uio->uio_offset;
 378                         flags = HAMMER_INODE_RDIRTY;
 379                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
 380                 } else {
 381                         flags = 0;
 382                 }
 383                 ip->ino_rec.ino_mtime = trans.time;
 384                 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
 385                 hammer_modify_inode(&trans, ip, flags);
 386                 hammer_unlock(&ip->lock);
 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_rec.ino_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                 kprintf("hammer_create_inode error %d\n", error);
 494                 hammer_done_transaction(&trans);
 495                 *ap->a_vpp = NULL;
 496                 return (error);
 497         }
 498         hammer_lock_sh(&nip->lock);
 499         hammer_lock_sh(&dip->lock);
 500
 501         /*
 502          * Add the new filesystem object to the directory.  This will also
 503          * bump the inode's link count.
 504          */
 505         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 506         hammer_finalize_inode(&trans, nip, error);
 507         if (error)
 508                 kprintf("hammer_ip_add_directory error %d\n", error);
 509         hammer_unlock(&dip->lock);
 510         hammer_unlock(&nip->lock);
 511
 512         /*
 513          * Finish up.
 514          */
 515         if (error) {
 516                 hammer_rel_inode(nip, 0);
 517                 hammer_done_transaction(&trans);
 518                 *ap->a_vpp = NULL;
 519         } else {
 520                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 521                 hammer_done_transaction(&trans);
 522                 hammer_rel_inode(nip, 0);
 523                 if (error == 0) {
 524                         cache_setunresolved(ap->a_nch);
 525                         cache_setvp(ap->a_nch, *ap->a_vpp);
 526                 }
 527         }
 528         return (error);
 529 }
 530
 531 /*
 532  * hammer_vop_getattr { vp, vap }
 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_rec.base.base.obj_id;
 553         vap->va_mode = ip->ino_data.mode;
 554         vap->va_nlink = ip->ino_rec.ino_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_rec.ino_size;
 560         hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
 561         hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
 562         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
 563         vap->va_flags = ip->ino_data.uflags;
 564         vap->va_gen = 1;        /* hammer inums are unique for all time */
 565         vap->va_blocksize = HAMMER_BUFSIZE;
 566         vap->va_bytes = (ip->ino_rec.ino_size + 63) & ~63;
 567         vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
 568         vap->va_filerev = 0;    /* XXX */
 569         /* mtime uniquely identifies any adjustments made to the file */
 570         vap->va_fsmid = ip->ino_rec.ino_mtime;
 571         vap->va_uid_uuid = ip->ino_data.uid;
 572         vap->va_gid_uuid = ip->ino_data.gid;
 573         vap->va_fsid_uuid = ip->hmp->fsid;
 574         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 575                           VA_FSID_UUID_VALID;
 576
 577         switch (ip->ino_rec.base.base.obj_type) {
 578         case HAMMER_OBJTYPE_CDEV:
 579         case HAMMER_OBJTYPE_BDEV:
 580                 vap->va_rmajor = ip->ino_data.rmajor;
 581                 vap->va_rminor = ip->ino_data.rminor;
 582                 break;
 583         default:
 584                 break;
 585         }
 586
 587         return(0);
 588 }
 589
 590 /*
 591  * hammer_vop_nresolve { nch, dvp, cred }
 592  *
 593  * Locate the requested directory entry.
 594  */
 595 static
 596 int
 597 hammer_vop_nresolve(struct vop_nresolve_args *ap)
 598 {
 599         struct hammer_transaction trans;
 600         struct namecache *ncp;
 601         hammer_inode_t dip;
 602         hammer_inode_t ip;
 603         hammer_tid_t asof;
 604         struct hammer_cursor cursor;
 605         union hammer_record_ondisk *rec;
 606         struct vnode *vp;
 607         int64_t namekey;
 608         int error;
 609         int i;
 610         int nlen;
 611         int flags;
 612         u_int64_t obj_id;
 613
 614         /*
 615          * Misc initialization, plus handle as-of name extensions.  Look for
 616          * the '@@' extension.  Note that as-of files and directories cannot
 617          * be modified.
 618          */
 619         dip = VTOI(ap->a_dvp);
 620         ncp = ap->a_nch->ncp;
 621         asof = dip->obj_asof;
 622         nlen = ncp->nc_nlen;
 623         flags = dip->flags;
 624
 625         hammer_simple_transaction(&trans, dip->hmp);
 626
 627         for (i = 0; i < nlen; ++i) {
 628                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
 629                         asof = hammer_str_to_tid(ncp->nc_name + i + 2);
 630                         flags |= HAMMER_INODE_RO;
 631                         break;
 632                 }
 633         }
 634         nlen = i;
 635
 636         /*
 637          * If there is no path component the time extension is relative to
 638          * dip.
 639          */
 640         if (nlen == 0) {
 641                 ip = hammer_get_inode(&trans, &dip->cache[1], dip->obj_id,
 642                                       asof, flags, &error);
 643                 if (error == 0) {
 644                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 645                         hammer_rel_inode(ip, 0);
 646                 } else {
 647                         vp = NULL;
 648                 }
 649                 if (error == 0) {
 650                         vn_unlock(vp);
 651                         cache_setvp(ap->a_nch, vp);
 652                         vrele(vp);
 653                 }
 654                 goto done;
 655         }
 656
 657         /*
 658          * Calculate the namekey and setup the key range for the scan.  This
 659          * works kinda like a chained hash table where the lower 32 bits
 660          * of the namekey synthesize the chain.
 661          *
 662          * The key range is inclusive of both key_beg and key_end.
 663          */
 664         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
 665
 666         error = hammer_init_cursor(&trans, &cursor, &dip->cache[0]);
 667         cursor.key_beg.obj_id = dip->obj_id;
 668         cursor.key_beg.key = namekey;
 669         cursor.key_beg.create_tid = 0;
 670         cursor.key_beg.delete_tid = 0;
 671         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 672         cursor.key_beg.obj_type = 0;
 673
 674         cursor.key_end = cursor.key_beg;
 675         cursor.key_end.key |= 0xFFFFFFFFULL;
 676         cursor.asof = asof;
 677         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
 678
 679         /*
 680          * Scan all matching records (the chain), locate the one matching
 681          * the requested path component.
 682          *
 683          * The hammer_ip_*() functions merge in-memory records with on-disk
 684          * records for the purposes of the search.
 685          */
 686         if (error == 0)
 687                 error = hammer_ip_first(&cursor, dip);
 688
 689         rec = NULL;
 690         obj_id = 0;
 691
 692         while (error == 0) {
 693                 error = hammer_ip_resolve_data(&cursor);
 694                 if (error)
 695                         break;
 696                 rec = cursor.record;
 697                 if (nlen == rec->entry.base.data_len &&
 698                     bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
 699                         obj_id = rec->entry.obj_id;
 700                         break;
 701                 }
 702                 error = hammer_ip_next(&cursor);
 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, LK_EXCLUSIVE, &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, LK_EXCLUSIVE, 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         hammer_lock_sh(&ip->lock);
 819         hammer_lock_sh(&dip->lock);
 820         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 821         hammer_unlock(&dip->lock);
 822         hammer_unlock(&ip->lock);
 823
 824         /*
 825          * Finish up.
 826          */
 827         if (error == 0) {
 828                 cache_setunresolved(nch);
 829                 cache_setvp(nch, ap->a_vp);
 830         }
 831         hammer_done_transaction(&trans);
 832         return (error);
 833 }
 834
 835 /*
 836  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 837  *
 838  * The operating system has already ensured that the directory entry
 839  * does not exist and done all appropriate namespace locking.
 840  */
 841 static
 842 int
 843 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 844 {
 845         struct hammer_transaction trans;
 846         struct hammer_inode *dip;
 847         struct hammer_inode *nip;
 848         struct nchandle *nch;
 849         int error;
 850
 851         nch = ap->a_nch;
 852         dip = VTOI(ap->a_dvp);
 853
 854         if (dip->flags & HAMMER_INODE_RO)
 855                 return (EROFS);
 856
 857         /*
 858          * Create a transaction to cover the operations we perform.
 859          */
 860         hammer_start_transaction(&trans, dip->hmp);
 861
 862         /*
 863          * Create a new filesystem object of the requested type.  The
 864          * returned inode will be referenced but not locked.
 865          */
 866         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 867         if (error) {
 868                 kprintf("hammer_mkdir error %d\n", error);
 869                 hammer_done_transaction(&trans);
 870                 *ap->a_vpp = NULL;
 871                 return (error);
 872         }
 873         /*
 874          * Add the new filesystem object to the directory.  This will also
 875          * bump the inode's link count.
 876          */
 877         hammer_lock_sh(&nip->lock);
 878         hammer_lock_sh(&dip->lock);
 879         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 880         hammer_finalize_inode(&trans, nip, error);
 881         hammer_unlock(&dip->lock);
 882         hammer_unlock(&nip->lock);
 883         if (error)
 884                 kprintf("hammer_mkdir (add) error %d\n", error);
 885
 886         /*
 887          * Finish up.
 888          */
 889         if (error) {
 890                 hammer_rel_inode(nip, 0);
 891                 *ap->a_vpp = NULL;
 892         } else {
 893                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 894                 hammer_rel_inode(nip, 0);
 895                 if (error == 0) {
 896                         cache_setunresolved(ap->a_nch);
 897                         cache_setvp(ap->a_nch, *ap->a_vpp);
 898                 }
 899         }
 900         hammer_done_transaction(&trans);
 901         return (error);
 902 }
 903
 904 /*
 905  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 906  *
 907  * The operating system has already ensured that the directory entry
 908  * does not exist and done all appropriate namespace locking.
 909  */
 910 static
 911 int
 912 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 913 {
 914         struct hammer_transaction trans;
 915         struct hammer_inode *dip;
 916         struct hammer_inode *nip;
 917         struct nchandle *nch;
 918         int error;
 919
 920         nch = ap->a_nch;
 921         dip = VTOI(ap->a_dvp);
 922
 923         if (dip->flags & HAMMER_INODE_RO)
 924                 return (EROFS);
 925
 926         /*
 927          * Create a transaction to cover the operations we perform.
 928          */
 929         hammer_start_transaction(&trans, dip->hmp);
 930
 931         /*
 932          * Create a new filesystem object of the requested type.  The
 933          * returned inode will be referenced but not locked.
 934          */
 935         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 936         if (error) {
 937                 hammer_done_transaction(&trans);
 938                 *ap->a_vpp = NULL;
 939                 return (error);
 940         }
 941
 942         /*
 943          * Add the new filesystem object to the directory.  This will also
 944          * bump the inode's link count.
 945          */
 946         hammer_lock_sh(&nip->lock);
 947         hammer_lock_sh(&dip->lock);
 948         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 949         hammer_finalize_inode(&trans, nip, error);
 950         hammer_unlock(&dip->lock);
 951         hammer_unlock(&nip->lock);
 952
 953         /*
 954          * Finish up.
 955          */
 956         if (error) {
 957                 hammer_rel_inode(nip, 0);
 958                 *ap->a_vpp = NULL;
 959         } else {
 960                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 961                 hammer_rel_inode(nip, 0);
 962                 if (error == 0) {
 963                         cache_setunresolved(ap->a_nch);
 964                         cache_setvp(ap->a_nch, *ap->a_vpp);
 965                 }
 966         }
 967         hammer_done_transaction(&trans);
 968         return (error);
 969 }
 970
 971 /*
 972  * hammer_vop_open { vp, mode, cred, fp }
 973  */
 974 static
 975 int
 976 hammer_vop_open(struct vop_open_args *ap)
 977 {
 978         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
 979                 return (EROFS);
 980
 981         return(vop_stdopen(ap));
 982 }
 983
 984 /*
 985  * hammer_vop_pathconf { vp, name, retval }
 986  */
 987 static
 988 int
 989 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 990 {
 991         return EOPNOTSUPP;
 992 }
 993
 994 /*
 995  * hammer_vop_print { vp }
 996  */
 997 static
 998 int
 999 hammer_vop_print(struct vop_print_args *ap)
1000 {
1001         return EOPNOTSUPP;
1002 }
1003
1004 /*
1005  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
1006  */
1007 static
1008 int
1009 hammer_vop_readdir(struct vop_readdir_args *ap)
1010 {
1011         struct hammer_transaction trans;
1012         struct hammer_cursor cursor;
1013         struct hammer_inode *ip;
1014         struct uio *uio;
1015         hammer_record_ondisk_t rec;
1016         hammer_base_elm_t base;
1017         int error;
1018         int cookie_index;
1019         int ncookies;
1020         off_t *cookies;
1021         off_t saveoff;
1022         int r;
1023
1024         ip = VTOI(ap->a_vp);
1025         uio = ap->a_uio;
1026         saveoff = uio->uio_offset;
1027
1028         if (ap->a_ncookies) {
1029                 ncookies = uio->uio_resid / 16 + 1;
1030                 if (ncookies > 1024)
1031                         ncookies = 1024;
1032                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1033                 cookie_index = 0;
1034         } else {
1035                 ncookies = -1;
1036                 cookies = NULL;
1037                 cookie_index = 0;
1038         }
1039
1040         hammer_simple_transaction(&trans, ip->hmp);
1041
1042         /*
1043          * Handle artificial entries
1044          */
1045         error = 0;
1046         if (saveoff == 0) {
1047                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1048                 if (r)
1049                         goto done;
1050                 if (cookies)
1051                         cookies[cookie_index] = saveoff;
1052                 ++saveoff;
1053                 ++cookie_index;
1054                 if (cookie_index == ncookies)
1055                         goto done;
1056         }
1057         if (saveoff == 1) {
1058                 if (ip->ino_data.parent_obj_id) {
1059                         r = vop_write_dirent(&error, uio,
1060                                              ip->ino_data.parent_obj_id,
1061                                              DT_DIR, 2, "..");
1062                 } else {
1063                         r = vop_write_dirent(&error, uio,
1064                                              ip->obj_id, DT_DIR, 2, "..");
1065                 }
1066                 if (r)
1067                         goto done;
1068                 if (cookies)
1069                         cookies[cookie_index] = saveoff;
1070                 ++saveoff;
1071                 ++cookie_index;
1072                 if (cookie_index == ncookies)
1073                         goto done;
1074         }
1075
1076         /*
1077          * Key range (begin and end inclusive) to scan.  Directory keys
1078          * directly translate to a 64 bit 'seek' position.
1079          */
1080         hammer_init_cursor(&trans, &cursor, &ip->cache[0]);
1081         cursor.key_beg.obj_id = ip->obj_id;
1082         cursor.key_beg.create_tid = 0;
1083         cursor.key_beg.delete_tid = 0;
1084         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1085         cursor.key_beg.obj_type = 0;
1086         cursor.key_beg.key = saveoff;
1087
1088         cursor.key_end = cursor.key_beg;
1089         cursor.key_end.key = HAMMER_MAX_KEY;
1090         cursor.asof = ip->obj_asof;
1091         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1092
1093         error = hammer_ip_first(&cursor, ip);
1094
1095         while (error == 0) {
1096                 error = hammer_ip_resolve_record_and_data(&cursor);
1097                 if (error)
1098                         break;
1099                 rec = cursor.record;
1100                 base = &rec->base.base;
1101                 saveoff = base->key;
1102
1103                 if (base->obj_id != ip->obj_id)
1104                         panic("readdir: bad record at %p", cursor.node);
1105
1106                 r = vop_write_dirent(
1107                              &error, uio, rec->entry.obj_id,
1108                              hammer_get_dtype(rec->entry.base.base.obj_type),
1109                              rec->entry.base.data_len,
1110                              (void *)cursor.data);
1111                 if (r)
1112                         break;
1113                 ++saveoff;
1114                 if (cookies)
1115                         cookies[cookie_index] = base->key;
1116                 ++cookie_index;
1117                 if (cookie_index == ncookies)
1118                         break;
1119                 error = hammer_ip_next(&cursor);
1120         }
1121         hammer_done_cursor(&cursor);
1122
1123 done:
1124         hammer_done_transaction(&trans);
1125
1126         if (ap->a_eofflag)
1127                 *ap->a_eofflag = (error == ENOENT);
1128         uio->uio_offset = saveoff;
1129         if (error && cookie_index == 0) {
1130                 if (error == ENOENT)
1131                         error = 0;
1132                 if (cookies) {
1133                         kfree(cookies, M_TEMP);
1134                         *ap->a_ncookies = 0;
1135                         *ap->a_cookies = NULL;
1136                 }
1137         } else {
1138                 if (error == ENOENT)
1139                         error = 0;
1140                 if (cookies) {
1141                         *ap->a_ncookies = cookie_index;
1142                         *ap->a_cookies = cookies;
1143                 }
1144         }
1145         return(error);
1146 }
1147
1148 /*
1149  * hammer_vop_readlink { vp, uio, cred }
1150  */
1151 static
1152 int
1153 hammer_vop_readlink(struct vop_readlink_args *ap)
1154 {
1155         struct hammer_transaction trans;
1156         struct hammer_cursor cursor;
1157         struct hammer_inode *ip;
1158         int error;
1159
1160         ip = VTOI(ap->a_vp);
1161
1162         hammer_simple_transaction(&trans, ip->hmp);
1163
1164         hammer_init_cursor(&trans, &cursor, &ip->cache[0]);
1165
1166         /*
1167          * Key range (begin and end inclusive) to scan.  Directory keys
1168          * directly translate to a 64 bit 'seek' position.
1169          */
1170         cursor.key_beg.obj_id = ip->obj_id;
1171         cursor.key_beg.create_tid = 0;
1172         cursor.key_beg.delete_tid = 0;
1173         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1174         cursor.key_beg.obj_type = 0;
1175         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1176         cursor.asof = ip->obj_asof;
1177         cursor.flags |= HAMMER_CURSOR_ASOF;
1178
1179         error = hammer_ip_lookup(&cursor, ip);
1180         if (error == 0) {
1181                 error = hammer_ip_resolve_data(&cursor);
1182                 if (error == 0) {
1183                         error = uiomove((char *)cursor.data,
1184                                         cursor.record->base.data_len,
1185                                         ap->a_uio);
1186                 }
1187         }
1188         hammer_done_cursor(&cursor);
1189         hammer_done_transaction(&trans);
1190         return(error);
1191 }
1192
1193 /*
1194  * hammer_vop_nremove { nch, dvp, cred }
1195  */
1196 static
1197 int
1198 hammer_vop_nremove(struct vop_nremove_args *ap)
1199 {
1200         struct hammer_transaction trans;
1201         int error;
1202
1203         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1204         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1205         hammer_done_transaction(&trans);
1206
1207         return (error);
1208 }
1209
1210 /*
1211  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1212  */
1213 static
1214 int
1215 hammer_vop_nrename(struct vop_nrename_args *ap)
1216 {
1217         struct hammer_transaction trans;
1218         struct namecache *fncp;
1219         struct namecache *tncp;
1220         struct hammer_inode *fdip;
1221         struct hammer_inode *tdip;
1222         struct hammer_inode *ip;
1223         struct hammer_cursor cursor;
1224         union hammer_record_ondisk *rec;
1225         int64_t namekey;
1226         int error;
1227
1228         fdip = VTOI(ap->a_fdvp);
1229         tdip = VTOI(ap->a_tdvp);
1230         fncp = ap->a_fnch->ncp;
1231         tncp = ap->a_tnch->ncp;
1232         ip = VTOI(fncp->nc_vp);
1233         KKASSERT(ip != NULL);
1234
1235         if (fdip->flags & HAMMER_INODE_RO)
1236                 return (EROFS);
1237         if (tdip->flags & HAMMER_INODE_RO)
1238                 return (EROFS);
1239         if (ip->flags & HAMMER_INODE_RO)
1240                 return (EROFS);
1241
1242         hammer_start_transaction(&trans, fdip->hmp);
1243
1244         hammer_lock_sh(&ip->lock);
1245         if (fdip->obj_id < tdip->obj_id) {
1246                 hammer_lock_sh(&fdip->lock);
1247                 hammer_lock_sh(&tdip->lock);
1248         } else {
1249                 hammer_lock_sh(&tdip->lock);
1250                 hammer_lock_sh(&fdip->lock);
1251         }
1252
1253         /*
1254          * Remove tncp from the target directory and then link ip as
1255          * tncp. XXX pass trans to dounlink
1256          *
1257          * Force the inode sync-time to match the transaction so it is
1258          * in-sync with the creation of the target directory entry.
1259          */
1260         error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1261         if (error == 0 || error == ENOENT) {
1262                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1263                 if (error == 0) {
1264                         ip->ino_data.parent_obj_id = tdip->obj_id;
1265                         hammer_modify_inode(&trans, ip, HAMMER_INODE_DDIRTY);
1266                 }
1267         }
1268         if (error)
1269                 goto failed; /* XXX */
1270
1271         /*
1272          * Locate the record in the originating directory and remove it.
1273          *
1274          * Calculate the namekey and setup the key range for the scan.  This
1275          * works kinda like a chained hash table where the lower 32 bits
1276          * of the namekey synthesize the chain.
1277          *
1278          * The key range is inclusive of both key_beg and key_end.
1279          */
1280         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1281 retry:
1282         hammer_init_cursor(&trans, &cursor, &fdip->cache[0]);
1283         cursor.key_beg.obj_id = fdip->obj_id;
1284         cursor.key_beg.key = namekey;
1285         cursor.key_beg.create_tid = 0;
1286         cursor.key_beg.delete_tid = 0;
1287         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1288         cursor.key_beg.obj_type = 0;
1289
1290         cursor.key_end = cursor.key_beg;
1291         cursor.key_end.key |= 0xFFFFFFFFULL;
1292         cursor.asof = fdip->obj_asof;
1293         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1294
1295         /*
1296          * Scan all matching records (the chain), locate the one matching
1297          * the requested path component.
1298          *
1299          * The hammer_ip_*() functions merge in-memory records with on-disk
1300          * records for the purposes of the search.
1301          */
1302         error = hammer_ip_first(&cursor, fdip);
1303         while (error == 0) {
1304                 if (hammer_ip_resolve_data(&cursor) != 0)
1305                         break;
1306                 rec = cursor.record;
1307                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1308                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1309                         break;
1310                 }
1311                 error = hammer_ip_next(&cursor);
1312         }
1313
1314         /*
1315          * If all is ok we have to get the inode so we can adjust nlinks.
1316          *
1317          * WARNING: hammer_ip_del_directory() may have to terminate the
1318          * cursor to avoid a recursion.  It's ok to call hammer_done_cursor()
1319          * twice.
1320          */
1321         if (error == 0)
1322                 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1323
1324         /*
1325          * XXX A deadlock here will break rename's atomicy for the purposes
1326          * of crash recovery.
1327          */
1328         if (error == EDEADLK) {
1329                 hammer_unlock(&ip->lock);
1330                 hammer_unlock(&fdip->lock);
1331                 hammer_unlock(&tdip->lock);
1332                 hammer_done_cursor(&cursor);
1333                 hammer_lock_sh(&ip->lock);
1334                 if (fdip->obj_id < tdip->obj_id) {
1335                         hammer_lock_sh(&fdip->lock);
1336                         hammer_lock_sh(&tdip->lock);
1337                 } else {
1338                         hammer_lock_sh(&tdip->lock);
1339                         hammer_lock_sh(&fdip->lock);
1340                 }
1341                 goto retry;
1342         }
1343
1344         /*
1345          * Cleanup and tell the kernel that the rename succeeded.
1346          */
1347         hammer_done_cursor(&cursor);
1348         if (error == 0)
1349                 cache_rename(ap->a_fnch, ap->a_tnch);
1350
1351 failed:
1352         hammer_unlock(&ip->lock);
1353         hammer_unlock(&fdip->lock);
1354         hammer_unlock(&tdip->lock);
1355         hammer_done_transaction(&trans);
1356         return (error);
1357 }
1358
1359 /*
1360  * hammer_vop_nrmdir { nch, dvp, cred }
1361  */
1362 static
1363 int
1364 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1365 {
1366         struct hammer_transaction trans;
1367         int error;
1368
1369         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1370         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1371         hammer_done_transaction(&trans);
1372
1373         return (error);
1374 }
1375
1376 /*
1377  * hammer_vop_setattr { vp, vap, cred }
1378  */
1379 static
1380 int
1381 hammer_vop_setattr(struct vop_setattr_args *ap)
1382 {
1383         struct hammer_transaction trans;
1384         struct vattr *vap;
1385         struct hammer_inode *ip;
1386         int modflags;
1387         int error;
1388         int truncating;
1389         off_t aligned_size;
1390         u_int32_t flags;
1391         uuid_t uuid;
1392
1393         vap = ap->a_vap;
1394         ip = ap->a_vp->v_data;
1395         modflags = 0;
1396
1397         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1398                 return(EROFS);
1399         if (ip->flags & HAMMER_INODE_RO)
1400                 return (EROFS);
1401
1402         hammer_start_transaction(&trans, ip->hmp);
1403         hammer_lock_sh(&ip->lock);
1404         error = 0;
1405
1406         if (vap->va_flags != VNOVAL) {
1407                 flags = ip->ino_data.uflags;
1408                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1409                                          hammer_to_unix_xid(&ip->ino_data.uid),
1410                                          ap->a_cred);
1411                 if (error == 0) {
1412                         if (ip->ino_data.uflags != flags) {
1413                                 ip->ino_data.uflags = flags;
1414                                 modflags |= HAMMER_INODE_DDIRTY;
1415                         }
1416                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1417                                 error = 0;
1418                                 goto done;
1419                         }
1420                 }
1421                 goto done;
1422         }
1423         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1424                 error = EPERM;
1425                 goto done;
1426         }
1427         if (vap->va_uid != (uid_t)VNOVAL) {
1428                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1429                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1430                         ip->ino_data.uid = uuid;
1431                         modflags |= HAMMER_INODE_DDIRTY;
1432                 }
1433         }
1434         if (vap->va_gid != (uid_t)VNOVAL) {
1435                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1436                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1437                         ip->ino_data.gid = uuid;
1438                         modflags |= HAMMER_INODE_DDIRTY;
1439                 }
1440         }
1441         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1442                 switch(ap->a_vp->v_type) {
1443                 case VREG:
1444                         if (vap->va_size == ip->ino_rec.ino_size)
1445                                 break;
1446                         /*
1447                          * XXX break atomicy, we can deadlock the backend
1448                          * if we do not release the lock.  Probably not a
1449                          * big deal here.
1450                          */
1451                         hammer_unlock(&ip->lock);
1452                         if (vap->va_size < ip->ino_rec.ino_size) {
1453                                 vtruncbuf(ap->a_vp, vap->va_size,
1454                                           HAMMER_BUFSIZE);
1455                                 truncating = 1;
1456                         } else {
1457                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1458                                 truncating = 0;
1459                         }
1460                         hammer_lock_sh(&ip->lock);
1461                         ip->ino_rec.ino_size = vap->va_size;
1462                         modflags |= HAMMER_INODE_RDIRTY;
1463                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1464                                        ~HAMMER_BUFMASK64;
1465
1466                         /*
1467                          * on-media truncation is cached in the inode until
1468                          * the inode is synchronized.
1469                          */
1470                         if (truncating) {
1471                                 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1472                                         ip->flags |= HAMMER_INODE_TRUNCATED;
1473                                         ip->trunc_off = vap->va_size;
1474                                 } else if (ip->trunc_off > vap->va_size) {
1475                                         ip->trunc_off = vap->va_size;
1476                                 }
1477                         }
1478
1479                         /*
1480                          * If truncating we have to clean out a portion of
1481                          * the last block on-disk.  We do this in the
1482                          * front-end buffer cache.
1483                          */
1484                         if (truncating && vap->va_size < aligned_size) {
1485                                 struct buf *bp;
1486                                 int offset;
1487
1488                                 offset = vap->va_size & HAMMER_BUFMASK;
1489                                 error = bread(ap->a_vp,
1490                                               aligned_size - HAMMER_BUFSIZE,
1491                                               HAMMER_BUFSIZE, &bp);
1492                                 if (error == 0) {
1493                                         bzero(bp->b_data + offset,
1494                                               HAMMER_BUFSIZE - offset);
1495                                         bdwrite(bp);
1496                                 } else {
1497                                         brelse(bp);
1498                                 }
1499                         }
1500                         break;
1501                 case VDATABASE:
1502                         if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1503                                 ip->flags |= HAMMER_INODE_TRUNCATED;
1504                                 ip->trunc_off = vap->va_size;
1505                         } else if (ip->trunc_off > vap->va_size) {
1506                                 ip->trunc_off = vap->va_size;
1507                         }
1508                         ip->ino_rec.ino_size = vap->va_size;
1509                         modflags |= HAMMER_INODE_RDIRTY;
1510                         break;
1511                 default:
1512                         error = EINVAL;
1513                         goto done;
1514                 }
1515                 break;
1516         }
1517         if (vap->va_atime.tv_sec != VNOVAL) {
1518                 ip->ino_rec.ino_atime =
1519                         hammer_timespec_to_transid(&vap->va_atime);
1520                 modflags |= HAMMER_INODE_ITIMES;
1521         }
1522         if (vap->va_mtime.tv_sec != VNOVAL) {
1523                 ip->ino_rec.ino_mtime =
1524                         hammer_timespec_to_transid(&vap->va_mtime);
1525                 modflags |= HAMMER_INODE_ITIMES;
1526         }
1527         if (vap->va_mode != (mode_t)VNOVAL) {
1528                 if (ip->ino_data.mode != vap->va_mode) {
1529                         ip->ino_data.mode = vap->va_mode;
1530                         modflags |= HAMMER_INODE_DDIRTY;
1531                 }
1532         }
1533 done:
1534         if (error == 0)
1535                 hammer_modify_inode(&trans, ip, modflags);
1536         hammer_unlock(&ip->lock);
1537         hammer_done_transaction(&trans);
1538         return (error);
1539 }
1540
1541 /*
1542  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1543  */
1544 static
1545 int
1546 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1547 {
1548         struct hammer_transaction trans;
1549         struct hammer_inode *dip;
1550         struct hammer_inode *nip;
1551         struct nchandle *nch;
1552         hammer_record_t record;
1553         int error;
1554         int bytes;
1555
1556         ap->a_vap->va_type = VLNK;
1557
1558         nch = ap->a_nch;
1559         dip = VTOI(ap->a_dvp);
1560
1561         if (dip->flags & HAMMER_INODE_RO)
1562                 return (EROFS);
1563
1564         /*
1565          * Create a transaction to cover the operations we perform.
1566          */
1567         hammer_start_transaction(&trans, dip->hmp);
1568
1569         /*
1570          * Create a new filesystem object of the requested type.  The
1571          * returned inode will be referenced but not locked.
1572          */
1573
1574         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1575         if (error) {
1576                 hammer_done_transaction(&trans);
1577                 *ap->a_vpp = NULL;
1578                 return (error);
1579         }
1580
1581         /*
1582          * Add the new filesystem object to the directory.  This will also
1583          * bump the inode's link count.
1584          */
1585         hammer_lock_sh(&nip->lock);
1586         hammer_lock_sh(&dip->lock);
1587         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1588
1589         /*
1590          * Add a record representing the symlink.  symlink stores the link
1591          * as pure data, not a string, and is no \0 terminated.
1592          */
1593         if (error == 0) {
1594                 record = hammer_alloc_mem_record(nip);
1595                 bytes = strlen(ap->a_target);
1596
1597                 record->rec.base.base.key = HAMMER_FIXKEY_SYMLINK;
1598                 record->rec.base.base.rec_type = HAMMER_RECTYPE_FIX;
1599                 record->rec.base.data_len = bytes;
1600                 record->data = (void *)ap->a_target;
1601                 /* will be reallocated by routine below */
1602                 error = hammer_ip_add_record(&trans, record);
1603
1604                 /*
1605                  * Set the file size to the length of the link.
1606                  */
1607                 if (error == 0) {
1608                         nip->ino_rec.ino_size = bytes;
1609                         hammer_modify_inode(&trans, nip, HAMMER_INODE_RDIRTY);
1610                 }
1611         }
1612         hammer_finalize_inode(&trans, nip, error);
1613         hammer_unlock(&dip->lock);
1614         hammer_unlock(&nip->lock);
1615
1616         /*
1617          * Finish up.
1618          */
1619         if (error) {
1620                 hammer_rel_inode(nip, 0);
1621                 *ap->a_vpp = NULL;
1622         } else {
1623                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1624                 hammer_rel_inode(nip, 0);
1625                 if (error == 0) {
1626                         cache_setunresolved(ap->a_nch);
1627                         cache_setvp(ap->a_nch, *ap->a_vpp);
1628                 }
1629         }
1630         hammer_done_transaction(&trans);
1631         return (error);
1632 }
1633
1634 /*
1635  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1636  */
1637 static
1638 int
1639 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1640 {
1641         struct hammer_transaction trans;
1642         int error;
1643
1644         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1645         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1646                                 ap->a_cred, ap->a_flags);
1647         hammer_done_transaction(&trans);
1648
1649         return (error);
1650 }
1651
1652 /*
1653  * hammer_vop_ioctl { vp, command, data, fflag, cred }
1654  */
1655 static
1656 int
1657 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1658 {
1659         struct hammer_inode *ip = ap->a_vp->v_data;
1660
1661         return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1662                             ap->a_fflag, ap->a_cred));
1663 }
1664
1665 static
1666 int
1667 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1668 {
1669         struct mount *mp;
1670         int error;
1671
1672         mp = ap->a_head.a_ops->head.vv_mount;
1673
1674         switch(ap->a_op) {
1675         case MOUNTCTL_SET_EXPORT:
1676                 if (ap->a_ctllen != sizeof(struct export_args))
1677                         error = EINVAL;
1678                 error = hammer_vfs_export(mp, ap->a_op,
1679                                       (const struct export_args *)ap->a_ctl);
1680                 break;
1681         default:
1682                 error = journal_mountctl(ap);
1683                 break;
1684         }
1685         return(error);
1686 }
1687
1688 /*
1689  * hammer_vop_strategy { vp, bio }
1690  *
1691  * Strategy call, used for regular file read & write only.  Note that the
1692  * bp may represent a cluster.
1693  *
1694  * To simplify operation and allow better optimizations in the future,
1695  * this code does not make any assumptions with regards to buffer alignment
1696  * or size.
1697  */
1698 static
1699 int
1700 hammer_vop_strategy(struct vop_strategy_args *ap)
1701 {
1702         struct buf *bp;
1703         int error;
1704
1705         bp = ap->a_bio->bio_buf;
1706
1707         switch(bp->b_cmd) {
1708         case BUF_CMD_READ:
1709                 error = hammer_vop_strategy_read(ap);
1710                 break;
1711         case BUF_CMD_WRITE:
1712                 error = hammer_vop_strategy_write(ap);
1713                 break;
1714         default:
1715                 bp->b_error = error = EINVAL;
1716                 bp->b_flags |= B_ERROR;
1717                 biodone(ap->a_bio);
1718                 break;
1719         }
1720         return (error);
1721 }
1722
1723 /*
1724  * Read from a regular file.  Iterate the related records and fill in the
1725  * BIO/BUF.  Gaps are zero-filled.
1726  *
1727  * The support code in hammer_object.c should be used to deal with mixed
1728  * in-memory and on-disk records.
1729  *
1730  * XXX atime update
1731  */
1732 static
1733 int
1734 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1735 {
1736         struct hammer_transaction trans;
1737         struct hammer_inode *ip;
1738         struct hammer_cursor cursor;
1739         hammer_record_ondisk_t rec;
1740         hammer_base_elm_t base;
1741         struct bio *bio;
1742         struct buf *bp;
1743         int64_t rec_offset;
1744         int64_t ran_end;
1745         int64_t tmp64;
1746         int error;
1747         int boff;
1748         int roff;
1749         int n;
1750
1751         bio = ap->a_bio;
1752         bp = bio->bio_buf;
1753         ip = ap->a_vp->v_data;
1754
1755         hammer_simple_transaction(&trans, ip->hmp);
1756         hammer_init_cursor(&trans, &cursor, &ip->cache[0]);
1757
1758         /*
1759          * Key range (begin and end inclusive) to scan.  Note that the key's
1760          * stored in the actual records represent BASE+LEN, not BASE.  The
1761          * first record containing bio_offset will have a key > bio_offset.
1762          */
1763         cursor.key_beg.obj_id = ip->obj_id;
1764         cursor.key_beg.create_tid = 0;
1765         cursor.key_beg.delete_tid = 0;
1766         cursor.key_beg.obj_type = 0;
1767         cursor.key_beg.key = bio->bio_offset + 1;
1768         cursor.asof = ip->obj_asof;
1769         cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1770
1771         cursor.key_end = cursor.key_beg;
1772         KKASSERT(ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_REGFILE);
1773 #if 0
1774         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1775                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1776                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1777                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1778         } else
1779 #endif
1780         {
1781                 ran_end = bio->bio_offset + bp->b_bufsize;
1782                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1783                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1784                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1785                 if (tmp64 < ran_end)
1786                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1787                 else
1788                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1789         }
1790         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1791
1792         error = hammer_ip_first(&cursor, ip);
1793         boff = 0;
1794
1795         while (error == 0) {
1796                 error = hammer_ip_resolve_data(&cursor);
1797                 if (error)
1798                         break;
1799                 rec = cursor.record;
1800                 base = &rec->base.base;
1801
1802                 rec_offset = base->key - rec->data.base.data_len;
1803
1804                 /*
1805                  * Calculate the gap, if any, and zero-fill it.
1806                  */
1807                 n = (int)(rec_offset - (bio->bio_offset + boff));
1808                 if (n > 0) {
1809                         if (n > bp->b_bufsize - boff)
1810                                 n = bp->b_bufsize - boff;
1811                         bzero((char *)bp->b_data + boff, n);
1812                         boff += n;
1813                         n = 0;
1814                 }
1815
1816                 /*
1817                  * Calculate the data offset in the record and the number
1818                  * of bytes we can copy.
1819                  *
1820                  * Note there is a degenerate case here where boff may
1821                  * already be at bp->b_bufsize.
1822                  */
1823                 roff = -n;
1824                 rec_offset += roff;
1825                 n = rec->data.base.data_len - roff;
1826                 KKASSERT(n > 0);
1827                 if (n > bp->b_bufsize - boff)
1828                         n = bp->b_bufsize - boff;
1829
1830                 /*
1831                  * If we cached a truncation point on our front-end the
1832                  * on-disk version may still have physical records beyond
1833                  * that point.  Truncate visibility.
1834                  */
1835                 if (ip->trunc_off <= rec_offset)
1836                         n = 0;
1837                 else if (ip->trunc_off < rec_offset + n)
1838                         n = (int)(ip->trunc_off - rec_offset);
1839
1840                 /*
1841                  * Copy
1842                  */
1843                 if (n) {
1844                         bcopy((char *)cursor.data + roff,
1845                               (char *)bp->b_data + boff, n);
1846                         boff += n;
1847                 }
1848                 if (boff == bp->b_bufsize)
1849                         break;
1850                 error = hammer_ip_next(&cursor);
1851         }
1852         hammer_done_cursor(&cursor);
1853         hammer_done_transaction(&trans);
1854
1855         /*
1856          * There may have been a gap after the last record
1857          */
1858         if (error == ENOENT)
1859                 error = 0;
1860         if (error == 0 && boff != bp->b_bufsize) {
1861                 KKASSERT(boff < bp->b_bufsize);
1862                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1863                 /* boff = bp->b_bufsize; */
1864         }
1865         bp->b_resid = 0;
1866         bp->b_error = error;
1867         if (error)
1868                 bp->b_flags |= B_ERROR;
1869         biodone(ap->a_bio);
1870         return(error);
1871 }
1872
1873 /*
1874  * Write to a regular file.   Because this is a strategy call the OS is
1875  * trying to actually sync data to the media.   HAMMER can only flush
1876  * the entire inode (so the TID remains properly synchronized).
1877  *
1878  * Basically all we do here is place the bio on the inode's flush queue
1879  * and activate the flusher.
1880  */
1881 static
1882 int
1883 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1884 {
1885         hammer_inode_t ip;
1886         struct bio *bio;
1887         struct buf *bp;
1888
1889         bio = ap->a_bio;
1890         bp = bio->bio_buf;
1891         ip = ap->a_vp->v_data;
1892
1893         if (ip->flags & HAMMER_INODE_RO) {
1894                 bp->b_error = EROFS;
1895                 bp->b_flags |= B_ERROR;
1896                 biodone(ap->a_bio);
1897                 return(EROFS);
1898         }
1899
1900         /*
1901          * If the inode is being flushed we cannot re-queue buffers
1902          * it may have already flushed, or it could result in duplicate
1903          * records in the database.
1904          */
1905         BUF_KERNPROC(bp);
1906         if (ip->flush_state == HAMMER_FST_FLUSH)
1907                 TAILQ_INSERT_TAIL(&ip->bio_alt_list, bio, bio_act);
1908         else
1909                 TAILQ_INSERT_TAIL(&ip->bio_list, bio, bio_act);
1910         hammer_modify_inode(NULL, ip, HAMMER_INODE_XDIRTY);
1911         hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1912         kprintf("a");
1913         return(0);
1914 }
1915
1916 /*
1917  * Backend code which actually performs the write to the media.  This
1918  * routine is typically called from the flusher.  The bio will be disposed
1919  * of (biodone'd) by this routine.
1920  *
1921  * Iterate the related records and mark for deletion.  If existing edge
1922  * records (left and right side) overlap our write they have to be marked
1923  * deleted and new records created, usually referencing a portion of the
1924  * original data.  Then add a record to represent the buffer.
1925  */
1926 int
1927 hammer_dowrite(hammer_transaction_t trans, hammer_inode_t ip, struct bio *bio)
1928 {
1929         struct buf *bp = bio->bio_buf;
1930         int error;
1931
1932         KKASSERT(ip->flush_state == HAMMER_FST_FLUSH);
1933
1934         /*
1935          * Delete any records overlapping our range.  This function will
1936          * (eventually) properly truncate partial overlaps.
1937          */
1938         if (ip->sync_ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1939                 error = hammer_ip_delete_range(trans, ip, bio->bio_offset,
1940                                                bio->bio_offset);
1941         } else {
1942                 error = hammer_ip_delete_range(trans, ip, bio->bio_offset,
1943                                                bio->bio_offset +
1944                                                 bp->b_bufsize - 1);
1945         }
1946
1947         /*
1948          * Add a single record to cover the write.  We can write a record
1949          * with only the actual file data - for example, a small 200 byte
1950          * file does not have to write out a 16K record.
1951          *
1952          * While the data size does not have to be aligned, we still do it
1953          * to reduce fragmentation in a future allocation model.
1954          */
1955         if (error == 0) {
1956                 int limit_size;
1957
1958                 if (ip->sync_ino_rec.ino_size - bio->bio_offset >
1959                     bp->b_bufsize) {
1960                             limit_size = bp->b_bufsize;
1961                 } else {
1962                         limit_size = (int)(ip->sync_ino_rec.ino_size -
1963                                            bio->bio_offset);
1964                         KKASSERT(limit_size >= 0);
1965                         limit_size = (limit_size + 63) & ~63;
1966                 }
1967
1968                 error = hammer_ip_sync_data(trans, ip, bio->bio_offset,
1969                                             bp->b_data, limit_size);
1970
1971         }
1972         if (error)
1973                 Debugger("hammer_dowrite: error");
1974
1975         if (error) {
1976                 bp->b_resid = bp->b_bufsize;
1977                 bp->b_error = error;
1978                 bp->b_flags |= B_ERROR;
1979         } else {
1980                 bp->b_resid = 0;
1981         }
1982         biodone(bio);
1983         return(error);
1984 }
1985
1986 /*
1987  * dounlink - disconnect a directory entry
1988  *
1989  * XXX whiteout support not really in yet
1990  */
1991 static int
1992 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
1993                 struct vnode *dvp, struct ucred *cred, int flags)
1994 {
1995         struct namecache *ncp;
1996         hammer_inode_t dip;
1997         hammer_inode_t ip;
1998         hammer_record_ondisk_t rec;
1999         struct hammer_cursor cursor;
2000         int64_t namekey;
2001         int error;
2002
2003         /*
2004          * Calculate the namekey and setup the key range for the scan.  This
2005          * works kinda like a chained hash table where the lower 32 bits
2006          * of the namekey synthesize the chain.
2007          *
2008          * The key range is inclusive of both key_beg and key_end.
2009          */
2010         dip = VTOI(dvp);
2011         ncp = nch->ncp;
2012
2013         if (dip->flags & HAMMER_INODE_RO)
2014                 return (EROFS);
2015
2016         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
2017 retry:
2018         hammer_init_cursor(trans, &cursor, &dip->cache[0]);
2019         cursor.key_beg.obj_id = dip->obj_id;
2020         cursor.key_beg.key = namekey;
2021         cursor.key_beg.create_tid = 0;
2022         cursor.key_beg.delete_tid = 0;
2023         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
2024         cursor.key_beg.obj_type = 0;
2025
2026         cursor.key_end = cursor.key_beg;
2027         cursor.key_end.key |= 0xFFFFFFFFULL;
2028         cursor.asof = dip->obj_asof;
2029         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
2030
2031         /*
2032          * Scan all matching records (the chain), locate the one matching
2033          * the requested path component.  info->last_error contains the
2034          * error code on search termination and could be 0, ENOENT, or
2035          * something else.
2036          *
2037          * The hammer_ip_*() functions merge in-memory records with on-disk
2038          * records for the purposes of the search.
2039          */
2040         error = hammer_ip_first(&cursor, dip);
2041         while (error == 0) {
2042                 error = hammer_ip_resolve_data(&cursor);
2043                 if (error)
2044                         break;
2045                 rec = cursor.record;
2046                 if (ncp->nc_nlen == rec->entry.base.data_len &&
2047                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
2048                         break;
2049                 }
2050                 error = hammer_ip_next(&cursor);
2051         }
2052
2053         /*
2054          * If all is ok we have to get the inode so we can adjust nlinks.
2055          *
2056          * If the target is a directory, it must be empty.
2057          */
2058         if (error == 0) {
2059                 ip = hammer_get_inode(trans, &dip->cache[1],
2060                                       rec->entry.obj_id,
2061                                       dip->hmp->asof, 0, &error);
2062                 if (error == ENOENT) {
2063                         kprintf("obj_id %016llx\n", rec->entry.obj_id);
2064                         Debugger("ENOENT unlinking object that should exist");
2065                 }
2066                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
2067                                   HAMMER_OBJTYPE_DIRECTORY) {
2068                         error = hammer_ip_check_directory_empty(trans, ip);
2069                 }
2070                 /*
2071                  * WARNING: hammer_ip_del_directory() may have to terminate
2072                  * the cursor to avoid a lock recursion.  It's ok to call
2073                  * hammer_done_cursor() twice.
2074                  */
2075                 if (error == 0) {
2076                         hammer_lock_sh(&ip->lock);
2077                         hammer_lock_sh(&dip->lock);
2078                         error = hammer_ip_del_directory(trans, &cursor,
2079                                                         dip, ip);
2080                         hammer_unlock(&dip->lock);
2081                         hammer_unlock(&ip->lock);
2082                 }
2083                 if (error == 0) {
2084                         cache_setunresolved(nch);
2085                         cache_setvp(nch, NULL);
2086                         /* XXX locking */
2087                         if (ip->vp)
2088                                 cache_inval_vp(ip->vp, CINV_DESTROY);
2089                 }
2090                 hammer_rel_inode(ip, 0);
2091         }
2092         hammer_done_cursor(&cursor);
2093         if (error == EDEADLK)
2094                 goto retry;
2095
2096         return (error);
2097 }
2098
2099 /************************************************************************
2100  *                          FIFO AND SPECFS OPS                         *
2101  ************************************************************************
2102  *
2103  */
2104
2105 static int
2106 hammer_vop_fifoclose (struct vop_close_args *ap)
2107 {
2108         /* XXX update itimes */
2109         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2110 }
2111
2112 static int
2113 hammer_vop_fiforead (struct vop_read_args *ap)
2114 {
2115         int error;
2116
2117         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2118         /* XXX update access time */
2119         return (error);
2120 }
2121
2122 static int
2123 hammer_vop_fifowrite (struct vop_write_args *ap)
2124 {
2125         int error;
2126
2127         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2128         /* XXX update access time */
2129         return (error);
2130 }
2131
2132 static int
2133 hammer_vop_specclose (struct vop_close_args *ap)
2134 {
2135         /* XXX update itimes */
2136         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2137 }
2138
2139 static int
2140 hammer_vop_specread (struct vop_read_args *ap)
2141 {
2142         /* XXX update access time */
2143         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2144 }
2145
2146 static int
2147 hammer_vop_specwrite (struct vop_write_args *ap)
2148 {
2149         /* XXX update last change time */
2150         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2151 }
2152