sys/vfs/hammer/hammer_vnops.c

   1 /*
   2  * Copyright (c) 2007 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.20 2008/01/16 01:15:37 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
  81 static int hammer_vop_fifoclose (struct vop_close_args *);
  82 static int hammer_vop_fiforead (struct vop_read_args *);
  83 static int hammer_vop_fifowrite (struct vop_write_args *);
  84
  85 static int hammer_vop_specclose (struct vop_close_args *);
  86 static int hammer_vop_specread (struct vop_read_args *);
  87 static int hammer_vop_specwrite (struct vop_write_args *);
  88
  89 struct vop_ops hammer_vnode_vops = {
  90         .vop_default =          vop_defaultop,
  91         .vop_fsync =            hammer_vop_fsync,
  92         .vop_getpages =         vop_stdgetpages,
  93         .vop_putpages =         vop_stdputpages,
  94         .vop_read =             hammer_vop_read,
  95         .vop_write =            hammer_vop_write,
  96         .vop_access =           hammer_vop_access,
  97         .vop_advlock =          hammer_vop_advlock,
  98         .vop_close =            hammer_vop_close,
  99         .vop_ncreate =          hammer_vop_ncreate,
 100         .vop_getattr =          hammer_vop_getattr,
 101         .vop_inactive =         hammer_vop_inactive,
 102         .vop_reclaim =          hammer_vop_reclaim,
 103         .vop_nresolve =         hammer_vop_nresolve,
 104         .vop_nlookupdotdot =    hammer_vop_nlookupdotdot,
 105         .vop_nlink =            hammer_vop_nlink,
 106         .vop_nmkdir =           hammer_vop_nmkdir,
 107         .vop_nmknod =           hammer_vop_nmknod,
 108         .vop_open =             hammer_vop_open,
 109         .vop_pathconf =         hammer_vop_pathconf,
 110         .vop_print =            hammer_vop_print,
 111         .vop_readdir =          hammer_vop_readdir,
 112         .vop_readlink =         hammer_vop_readlink,
 113         .vop_nremove =          hammer_vop_nremove,
 114         .vop_nrename =          hammer_vop_nrename,
 115         .vop_nrmdir =           hammer_vop_nrmdir,
 116         .vop_setattr =          hammer_vop_setattr,
 117         .vop_strategy =         hammer_vop_strategy,
 118         .vop_nsymlink =         hammer_vop_nsymlink,
 119         .vop_nwhiteout =        hammer_vop_nwhiteout
 120 };
 121
 122 struct vop_ops hammer_spec_vops = {
 123         .vop_default =          spec_vnoperate,
 124         .vop_fsync =            hammer_vop_fsync,
 125         .vop_read =             hammer_vop_specread,
 126         .vop_write =            hammer_vop_specwrite,
 127         .vop_access =           hammer_vop_access,
 128         .vop_close =            hammer_vop_specclose,
 129         .vop_getattr =          hammer_vop_getattr,
 130         .vop_inactive =         hammer_vop_inactive,
 131         .vop_reclaim =          hammer_vop_reclaim,
 132         .vop_setattr =          hammer_vop_setattr
 133 };
 134
 135 struct vop_ops hammer_fifo_vops = {
 136         .vop_default =          fifo_vnoperate,
 137         .vop_fsync =            hammer_vop_fsync,
 138         .vop_read =             hammer_vop_fiforead,
 139         .vop_write =            hammer_vop_fifowrite,
 140         .vop_access =           hammer_vop_access,
 141         .vop_close =            hammer_vop_fifoclose,
 142         .vop_getattr =          hammer_vop_getattr,
 143         .vop_inactive =         hammer_vop_inactive,
 144         .vop_reclaim =          hammer_vop_reclaim,
 145         .vop_setattr =          hammer_vop_setattr
 146 };
 147
 148 static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp,
 149                            struct ucred *cred, int flags);
 150 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
 151 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
 152
 153 #if 0
 154 static
 155 int
 156 hammer_vop_vnoperate(struct vop_generic_args *)
 157 {
 158         return (VOCALL(&hammer_vnode_vops, ap));
 159 }
 160 #endif
 161
 162 /*
 163  * hammer_vop_fsync { vp, waitfor }
 164  */
 165 static
 166 int
 167 hammer_vop_fsync(struct vop_fsync_args *ap)
 168 {
 169         hammer_inode_t ip;
 170         int error;
 171
 172         ip = VTOI(ap->a_vp);
 173         error = hammer_sync_inode(ip, ap->a_waitfor, 0);
 174         return (error);
 175 }
 176
 177 /*
 178  * hammer_vop_read { vp, uio, ioflag, cred }
 179  */
 180 static
 181 int
 182 hammer_vop_read(struct vop_read_args *ap)
 183 {
 184         struct hammer_transaction trans;
 185         hammer_inode_t ip;
 186         off_t offset;
 187         struct buf *bp;
 188         struct uio *uio;
 189         int error;
 190         int n;
 191         int seqcount;
 192
 193         if (ap->a_vp->v_type != VREG)
 194                 return (EINVAL);
 195         ip = VTOI(ap->a_vp);
 196         error = 0;
 197         seqcount = ap->a_ioflag >> 16;
 198
 199         hammer_start_transaction(&trans, ip->hmp);
 200
 201         /*
 202          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 203          */
 204         uio = ap->a_uio;
 205         while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
 206                 offset = uio->uio_offset & HAMMER_BUFMASK;
 207 #if 0
 208                 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
 209                                      uio->uio_offset - offset, HAMMER_BUFSIZE,
 210                                      MAXBSIZE, seqcount, &bp);
 211 #endif
 212                 error = bread(ap->a_vp, uio->uio_offset - offset,
 213                               HAMMER_BUFSIZE, &bp);
 214                 if (error) {
 215                         brelse(bp);
 216                         break;
 217                 }
 218                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 219                 n = HAMMER_BUFSIZE - offset;
 220                 if (n > uio->uio_resid)
 221                         n = uio->uio_resid;
 222                 if (n > ip->ino_rec.ino_size - uio->uio_offset)
 223                         n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
 224                 error = uiomove((char *)bp->b_data + offset, n, uio);
 225                 if (error) {
 226                         bqrelse(bp);
 227                         break;
 228                 }
 229                 if ((ip->flags & HAMMER_INODE_RO) == 0) {
 230                         ip->ino_rec.ino_atime = trans.tid;
 231                         hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
 232                 }
 233                 bqrelse(bp);
 234         }
 235         hammer_commit_transaction(&trans);
 236         return (error);
 237 }
 238
 239 /*
 240  * hammer_vop_write { vp, uio, ioflag, cred }
 241  */
 242 static
 243 int
 244 hammer_vop_write(struct vop_write_args *ap)
 245 {
 246         struct hammer_transaction trans;
 247         struct hammer_inode *ip;
 248         struct uio *uio;
 249         off_t offset;
 250         struct buf *bp;
 251         int error;
 252         int n;
 253         int flags;
 254
 255         if (ap->a_vp->v_type != VREG)
 256                 return (EINVAL);
 257         ip = VTOI(ap->a_vp);
 258         error = 0;
 259
 260         if (ip->flags & HAMMER_INODE_RO)
 261                 return (EROFS);
 262
 263         /*
 264          * Create a transaction to cover the operations we perform.
 265          */
 266         hammer_start_transaction(&trans, ip->hmp);
 267         uio = ap->a_uio;
 268
 269         /*
 270          * Check append mode
 271          */
 272         if (ap->a_ioflag & IO_APPEND)
 273                 uio->uio_offset = ip->ino_rec.ino_size;
 274
 275         /*
 276          * Check for illegal write offsets.  Valid range is 0...2^63-1
 277          */
 278         if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
 279                 hammer_commit_transaction(&trans);
 280                 return (EFBIG);
 281         }
 282
 283         /*
 284          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 285          */
 286         while (uio->uio_resid > 0) {
 287                 offset = uio->uio_offset & HAMMER_BUFMASK;
 288                 if (uio->uio_segflg == UIO_NOCOPY) {
 289                         /*
 290                          * Issuing a write with the same data backing the
 291                          * buffer.  Instantiate the buffer to collect the
 292                          * backing vm pages, then read-in any missing bits.
 293                          *
 294                          * This case is used by vop_stdputpages().
 295                          */
 296                         bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
 297                                     GETBLK_BHEAVY, 0);
 298                         if ((bp->b_flags & B_CACHE) == 0) {
 299                                 bqrelse(bp);
 300                                 error = bread(ap->a_vp,
 301                                               uio->uio_offset - offset,
 302                                               HAMMER_BUFSIZE, &bp);
 303                                 if (error) {
 304                                         brelse(bp);
 305                                         break;
 306                                 }
 307                         }
 308                 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
 309                         /*
 310                          * entirely overwrite the buffer
 311                          */
 312                         bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
 313                                     GETBLK_BHEAVY, 0);
 314                 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
 315                         /*
 316                          * XXX
 317                          */
 318                         bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
 319                                     GETBLK_BHEAVY, 0);
 320                         vfs_bio_clrbuf(bp);
 321                 } else {
 322                         /*
 323                          * Partial overwrite, read in any missing bits then
 324                          * replace the portion being written.
 325                          */
 326                         error = bread(ap->a_vp, uio->uio_offset - offset,
 327                                       HAMMER_BUFSIZE, &bp);
 328                         if (error) {
 329                                 brelse(bp);
 330                                 break;
 331                         }
 332                         bheavy(bp);
 333                 }
 334                 n = HAMMER_BUFSIZE - offset;
 335                 if (n > uio->uio_resid)
 336                         n = uio->uio_resid;
 337                 error = uiomove((char *)bp->b_data + offset, n, uio);
 338                 if (error) {
 339                         brelse(bp);
 340                         break;
 341                 }
 342                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 343                 if (ip->ino_rec.ino_size < uio->uio_offset) {
 344                         ip->ino_rec.ino_size = uio->uio_offset;
 345                         flags = HAMMER_INODE_RDIRTY;
 346                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
 347                 } else {
 348                         flags = 0;
 349                 }
 350                 ip->ino_rec.ino_mtime = trans.tid;
 351                 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
 352                 hammer_modify_inode(&trans, ip, flags);
 353                 if (ap->a_ioflag & IO_SYNC) {
 354                         bwrite(bp);
 355                 } else if (ap->a_ioflag & IO_DIRECT) {
 356                         bawrite(bp);
 357                 } else {
 358                         bdwrite(bp);
 359                 }
 360         }
 361         if (error)
 362                 hammer_abort_transaction(&trans);
 363         else
 364                 hammer_commit_transaction(&trans);
 365         return (error);
 366 }
 367
 368 /*
 369  * hammer_vop_access { vp, mode, cred }
 370  */
 371 static
 372 int
 373 hammer_vop_access(struct vop_access_args *ap)
 374 {
 375         struct hammer_inode *ip = VTOI(ap->a_vp);
 376         uid_t uid;
 377         gid_t gid;
 378         int error;
 379
 380         uid = hammer_to_unix_xid(&ip->ino_data.uid);
 381         gid = hammer_to_unix_xid(&ip->ino_data.gid);
 382
 383         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
 384                                   ip->ino_data.uflags);
 385         return (error);
 386 }
 387
 388 /*
 389  * hammer_vop_advlock { vp, id, op, fl, flags }
 390  */
 391 static
 392 int
 393 hammer_vop_advlock(struct vop_advlock_args *ap)
 394 {
 395         struct hammer_inode *ip = VTOI(ap->a_vp);
 396
 397         return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
 398 }
 399
 400 /*
 401  * hammer_vop_close { vp, fflag }
 402  */
 403 static
 404 int
 405 hammer_vop_close(struct vop_close_args *ap)
 406 {
 407         return (vop_stdclose(ap));
 408 }
 409
 410 /*
 411  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
 412  *
 413  * The operating system has already ensured that the directory entry
 414  * does not exist and done all appropriate namespace locking.
 415  */
 416 static
 417 int
 418 hammer_vop_ncreate(struct vop_ncreate_args *ap)
 419 {
 420         struct hammer_transaction trans;
 421         struct hammer_inode *dip;
 422         struct hammer_inode *nip;
 423         struct nchandle *nch;
 424         int error;
 425
 426         nch = ap->a_nch;
 427         dip = VTOI(ap->a_dvp);
 428
 429         if (dip->flags & HAMMER_INODE_RO)
 430                 return (EROFS);
 431
 432         /*
 433          * Create a transaction to cover the operations we perform.
 434          */
 435         hammer_start_transaction(&trans, dip->hmp);
 436
 437         /*
 438          * Create a new filesystem object of the requested type.  The
 439          * returned inode will be referenced but not locked.
 440          */
 441
 442         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 443         if (error)
 444                 kprintf("hammer_create_inode error %d\n", error);
 445         if (error) {
 446                 hammer_abort_transaction(&trans);
 447                 *ap->a_vpp = NULL;
 448                 return (error);
 449         }
 450
 451         /*
 452          * Add the new filesystem object to the directory.  This will also
 453          * bump the inode's link count.
 454          */
 455         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 456         if (error)
 457                 kprintf("hammer_ip_add_directory error %d\n", error);
 458
 459         /*
 460          * Finish up.
 461          */
 462         if (error) {
 463                 hammer_rel_inode(nip, 0);
 464                 hammer_abort_transaction(&trans);
 465                 *ap->a_vpp = NULL;
 466         } else {
 467                 hammer_commit_transaction(&trans);
 468                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 469                 hammer_rel_inode(nip, 0);
 470                 if (error == 0) {
 471                         cache_setunresolved(ap->a_nch);
 472                         cache_setvp(ap->a_nch, *ap->a_vpp);
 473                 }
 474         }
 475         return (error);
 476 }
 477
 478 /*
 479  * hammer_vop_getattr { vp, vap }
 480  */
 481 static
 482 int
 483 hammer_vop_getattr(struct vop_getattr_args *ap)
 484 {
 485         struct hammer_inode *ip = VTOI(ap->a_vp);
 486         struct vattr *vap = ap->a_vap;
 487
 488 #if 0
 489         if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
 490             (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
 491             ip->obj_asof == XXX
 492         ) {
 493                 /* LAZYMOD XXX */
 494         }
 495         hammer_itimes(ap->a_vp);
 496 #endif
 497
 498         vap->va_fsid = ip->hmp->fsid_udev;
 499         vap->va_fileid = ip->ino_rec.base.base.obj_id;
 500         vap->va_mode = ip->ino_data.mode;
 501         vap->va_nlink = ip->ino_rec.ino_nlinks;
 502         vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
 503         vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
 504         vap->va_rmajor = 0;
 505         vap->va_rminor = 0;
 506         vap->va_size = ip->ino_rec.ino_size;
 507         hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
 508         hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
 509         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
 510         vap->va_flags = ip->ino_data.uflags;
 511         vap->va_gen = 1;        /* hammer inums are unique for all time */
 512         vap->va_blocksize = 32768; /* XXX - extract from root volume */
 513         vap->va_bytes = ip->ino_rec.ino_size;
 514         vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
 515         vap->va_filerev = 0;    /* XXX */
 516         /* mtime uniquely identifies any adjustments made to the file */
 517         vap->va_fsmid = ip->ino_rec.ino_mtime;
 518         vap->va_uid_uuid = ip->ino_data.uid;
 519         vap->va_gid_uuid = ip->ino_data.gid;
 520         vap->va_fsid_uuid = ip->hmp->fsid;
 521         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 522                           VA_FSID_UUID_VALID;
 523
 524         switch (ip->ino_rec.base.base.obj_type) {
 525         case HAMMER_OBJTYPE_CDEV:
 526         case HAMMER_OBJTYPE_BDEV:
 527                 vap->va_rmajor = ip->ino_data.rmajor;
 528                 vap->va_rminor = ip->ino_data.rminor;
 529                 break;
 530         default:
 531                 break;
 532         }
 533
 534         return(0);
 535 }
 536
 537 /*
 538  * hammer_vop_nresolve { nch, dvp, cred }
 539  *
 540  * Locate the requested directory entry.
 541  */
 542 static
 543 int
 544 hammer_vop_nresolve(struct vop_nresolve_args *ap)
 545 {
 546         struct namecache *ncp;
 547         hammer_inode_t dip;
 548         hammer_inode_t ip;
 549         hammer_tid_t asof;
 550         struct hammer_cursor cursor;
 551         union hammer_record_ondisk *rec;
 552         struct vnode *vp;
 553         int64_t namekey;
 554         int error;
 555         int i;
 556         int nlen;
 557         int flags;
 558
 559         /*
 560          * Misc initialization, plus handle as-of name extensions.  Look for
 561          * the '@@' extension.  Note that as-of files and directories cannot
 562          * be modified.
 563          */
 564         dip = VTOI(ap->a_dvp);
 565         ncp = ap->a_nch->ncp;
 566         asof = dip->obj_asof;
 567         nlen = ncp->nc_nlen;
 568         flags = dip->flags;
 569
 570         for (i = 0; i < nlen; ++i) {
 571                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
 572                         asof = hammer_str_to_tid(ncp->nc_name + i + 2);
 573                         kprintf("ASOF %016llx\n", asof);
 574                         flags |= HAMMER_INODE_RO;
 575                         break;
 576                 }
 577         }
 578         nlen = i;
 579
 580         /*
 581          * If there is no path component the time extension is relative to
 582          * dip.
 583          */
 584         if (nlen == 0) {
 585                 ip = hammer_get_inode(dip->hmp, &dip->cache[1], dip->obj_id,
 586                                       asof, flags, &error);
 587                 if (error == 0) {
 588                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 589                         hammer_rel_inode(ip, 0);
 590                 } else {
 591                         vp = NULL;
 592                 }
 593                 if (error == 0) {
 594                         vn_unlock(vp);
 595                         cache_setvp(ap->a_nch, vp);
 596                         vrele(vp);
 597                 }
 598                 return(error);
 599         }
 600
 601         /*
 602          * Calculate the namekey and setup the key range for the scan.  This
 603          * works kinda like a chained hash table where the lower 32 bits
 604          * of the namekey synthesize the chain.
 605          *
 606          * The key range is inclusive of both key_beg and key_end.
 607          */
 608         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
 609
 610         hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
 611         cursor.key_beg.obj_id = dip->obj_id;
 612         cursor.key_beg.key = namekey;
 613         cursor.key_beg.create_tid = 0;
 614         cursor.key_beg.delete_tid = 0;
 615         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 616         cursor.key_beg.obj_type = 0;
 617
 618         cursor.key_end = cursor.key_beg;
 619         cursor.key_end.key |= 0xFFFFFFFFULL;
 620         cursor.asof = asof;
 621         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
 622
 623         /*
 624          * Scan all matching records (the chain), locate the one matching
 625          * the requested path component.
 626          *
 627          * The hammer_ip_*() functions merge in-memory records with on-disk
 628          * records for the purposes of the search.
 629          */
 630         error = hammer_ip_first(&cursor, dip);
 631         rec = NULL;
 632         while (error == 0) {
 633                 error = hammer_ip_resolve_data(&cursor);
 634                 if (error)
 635                         break;
 636                 rec = cursor.record;
 637                 if (nlen == rec->entry.base.data_len &&
 638                     bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
 639                         break;
 640                 }
 641                 error = hammer_ip_next(&cursor);
 642         }
 643         if (error == 0) {
 644                 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
 645                                       rec->entry.obj_id, asof,
 646                                       flags, &error);
 647                 if (error == 0) {
 648                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 649                         hammer_rel_inode(ip, 0);
 650                 } else {
 651                         vp = NULL;
 652                 }
 653                 if (error == 0) {
 654                         vn_unlock(vp);
 655                         cache_setvp(ap->a_nch, vp);
 656                         vrele(vp);
 657                 }
 658         } else if (error == ENOENT) {
 659                 cache_setvp(ap->a_nch, NULL);
 660         }
 661         hammer_done_cursor(&cursor);
 662         return (error);
 663 }
 664
 665 /*
 666  * hammer_vop_nlookupdotdot { dvp, vpp, cred }
 667  *
 668  * Locate the parent directory of a directory vnode.
 669  *
 670  * dvp is referenced but not locked.  *vpp must be returned referenced and
 671  * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
 672  * at the root, instead it could indicate that the directory we were in was
 673  * removed.
 674  */
 675 static
 676 int
 677 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
 678 {
 679         struct hammer_inode *dip;
 680         struct hammer_inode *ip;
 681         u_int64_t parent_obj_id;
 682         int error;
 683
 684         dip = VTOI(ap->a_dvp);
 685         if ((parent_obj_id = dip->ino_data.parent_obj_id) == 0) {
 686                 *ap->a_vpp = NULL;
 687                 return ENOENT;
 688         }
 689
 690         ip = hammer_get_inode(dip->hmp, &dip->cache[1], parent_obj_id,
 691                               dip->obj_asof, dip->flags, &error);
 692         if (ip == NULL) {
 693                 *ap->a_vpp = NULL;
 694                 return(error);
 695         }
 696         error = hammer_get_vnode(ip, LK_EXCLUSIVE, ap->a_vpp);
 697         hammer_rel_inode(ip, 0);
 698         return (error);
 699 }
 700
 701 /*
 702  * hammer_vop_nlink { nch, dvp, vp, cred }
 703  */
 704 static
 705 int
 706 hammer_vop_nlink(struct vop_nlink_args *ap)
 707 {
 708         struct hammer_transaction trans;
 709         struct hammer_inode *dip;
 710         struct hammer_inode *ip;
 711         struct nchandle *nch;
 712         int error;
 713
 714         nch = ap->a_nch;
 715         dip = VTOI(ap->a_dvp);
 716         ip = VTOI(ap->a_vp);
 717
 718         if (dip->flags & HAMMER_INODE_RO)
 719                 return (EROFS);
 720         if (ip->flags & HAMMER_INODE_RO)
 721                 return (EROFS);
 722
 723         /*
 724          * Create a transaction to cover the operations we perform.
 725          */
 726         hammer_start_transaction(&trans, dip->hmp);
 727
 728         /*
 729          * Add the filesystem object to the directory.  Note that neither
 730          * dip nor ip are referenced or locked, but their vnodes are
 731          * referenced.  This function will bump the inode's link count.
 732          */
 733         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 734
 735         /*
 736          * Finish up.
 737          */
 738         if (error) {
 739                 hammer_abort_transaction(&trans);
 740         } else {
 741                 cache_setunresolved(nch);
 742                 cache_setvp(nch, ap->a_vp);
 743                 hammer_commit_transaction(&trans);
 744         }
 745         return (error);
 746 }
 747
 748 /*
 749  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 750  *
 751  * The operating system has already ensured that the directory entry
 752  * does not exist and done all appropriate namespace locking.
 753  */
 754 static
 755 int
 756 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 757 {
 758         struct hammer_transaction trans;
 759         struct hammer_inode *dip;
 760         struct hammer_inode *nip;
 761         struct nchandle *nch;
 762         int error;
 763
 764         nch = ap->a_nch;
 765         dip = VTOI(ap->a_dvp);
 766
 767         if (dip->flags & HAMMER_INODE_RO)
 768                 return (EROFS);
 769
 770         /*
 771          * Create a transaction to cover the operations we perform.
 772          */
 773         hammer_start_transaction(&trans, dip->hmp);
 774
 775         /*
 776          * Create a new filesystem object of the requested type.  The
 777          * returned inode will be referenced but not locked.
 778          */
 779         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 780         if (error)
 781                 kprintf("hammer_mkdir error %d\n", error);
 782         if (error) {
 783                 hammer_abort_transaction(&trans);
 784                 *ap->a_vpp = NULL;
 785                 return (error);
 786         }
 787
 788         /*
 789          * Add the new filesystem object to the directory.  This will also
 790          * bump the inode's link count.
 791          */
 792         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 793         if (error)
 794                 kprintf("hammer_mkdir (add) error %d\n", error);
 795
 796         /*
 797          * Finish up.
 798          */
 799         if (error) {
 800                 hammer_rel_inode(nip, 0);
 801                 hammer_abort_transaction(&trans);
 802                 *ap->a_vpp = NULL;
 803         } else {
 804                 hammer_commit_transaction(&trans);
 805                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 806                 hammer_rel_inode(nip, 0);
 807                 if (error == 0) {
 808                         cache_setunresolved(ap->a_nch);
 809                         cache_setvp(ap->a_nch, *ap->a_vpp);
 810                 }
 811         }
 812         return (error);
 813 }
 814
 815 /*
 816  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 817  *
 818  * The operating system has already ensured that the directory entry
 819  * does not exist and done all appropriate namespace locking.
 820  */
 821 static
 822 int
 823 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 824 {
 825         struct hammer_transaction trans;
 826         struct hammer_inode *dip;
 827         struct hammer_inode *nip;
 828         struct nchandle *nch;
 829         int error;
 830
 831         nch = ap->a_nch;
 832         dip = VTOI(ap->a_dvp);
 833
 834         if (dip->flags & HAMMER_INODE_RO)
 835                 return (EROFS);
 836
 837         /*
 838          * Create a transaction to cover the operations we perform.
 839          */
 840         hammer_start_transaction(&trans, dip->hmp);
 841
 842         /*
 843          * Create a new filesystem object of the requested type.  The
 844          * returned inode will be referenced but not locked.
 845          */
 846         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 847         if (error) {
 848                 hammer_abort_transaction(&trans);
 849                 *ap->a_vpp = NULL;
 850                 return (error);
 851         }
 852
 853         /*
 854          * Add the new filesystem object to the directory.  This will also
 855          * bump the inode's link count.
 856          */
 857         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 858
 859         /*
 860          * Finish up.
 861          */
 862         if (error) {
 863                 hammer_rel_inode(nip, 0);
 864                 hammer_abort_transaction(&trans);
 865                 *ap->a_vpp = NULL;
 866         } else {
 867                 hammer_commit_transaction(&trans);
 868                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 869                 hammer_rel_inode(nip, 0);
 870                 if (error == 0) {
 871                         cache_setunresolved(ap->a_nch);
 872                         cache_setvp(ap->a_nch, *ap->a_vpp);
 873                 }
 874         }
 875         return (error);
 876 }
 877
 878 /*
 879  * hammer_vop_open { vp, mode, cred, fp }
 880  */
 881 static
 882 int
 883 hammer_vop_open(struct vop_open_args *ap)
 884 {
 885         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
 886                 return (EROFS);
 887
 888         return(vop_stdopen(ap));
 889 }
 890
 891 /*
 892  * hammer_vop_pathconf { vp, name, retval }
 893  */
 894 static
 895 int
 896 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 897 {
 898         return EOPNOTSUPP;
 899 }
 900
 901 /*
 902  * hammer_vop_print { vp }
 903  */
 904 static
 905 int
 906 hammer_vop_print(struct vop_print_args *ap)
 907 {
 908         return EOPNOTSUPP;
 909 }
 910
 911 /*
 912  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 913  */
 914 static
 915 int
 916 hammer_vop_readdir(struct vop_readdir_args *ap)
 917 {
 918         struct hammer_cursor cursor;
 919         struct hammer_inode *ip;
 920         struct uio *uio;
 921         hammer_record_ondisk_t rec;
 922         hammer_base_elm_t base;
 923         int error;
 924         int cookie_index;
 925         int ncookies;
 926         off_t *cookies;
 927         off_t saveoff;
 928         int r;
 929
 930         ip = VTOI(ap->a_vp);
 931         uio = ap->a_uio;
 932         saveoff = uio->uio_offset;
 933
 934         if (ap->a_ncookies) {
 935                 ncookies = uio->uio_resid / 16 + 1;
 936                 if (ncookies > 1024)
 937                         ncookies = 1024;
 938                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 939                 cookie_index = 0;
 940         } else {
 941                 ncookies = -1;
 942                 cookies = NULL;
 943                 cookie_index = 0;
 944         }
 945
 946         /*
 947          * Handle artificial entries
 948          */
 949         error = 0;
 950         if (saveoff == 0) {
 951                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
 952                 if (r)
 953                         goto done;
 954                 if (cookies)
 955                         cookies[cookie_index] = saveoff;
 956                 ++saveoff;
 957                 ++cookie_index;
 958                 if (cookie_index == ncookies)
 959                         goto done;
 960         }
 961         if (saveoff == 1) {
 962                 if (ip->ino_data.parent_obj_id) {
 963                         r = vop_write_dirent(&error, uio,
 964                                              ip->ino_data.parent_obj_id,
 965                                              DT_DIR, 2, "..");
 966                 } else {
 967                         r = vop_write_dirent(&error, uio,
 968                                              ip->obj_id, DT_DIR, 2, "..");
 969                 }
 970                 if (r)
 971                         goto done;
 972                 if (cookies)
 973                         cookies[cookie_index] = saveoff;
 974                 ++saveoff;
 975                 ++cookie_index;
 976                 if (cookie_index == ncookies)
 977                         goto done;
 978         }
 979
 980         /*
 981          * Key range (begin and end inclusive) to scan.  Directory keys
 982          * directly translate to a 64 bit 'seek' position.
 983          */
 984         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
 985         cursor.key_beg.obj_id = ip->obj_id;
 986         cursor.key_beg.create_tid = 0;
 987         cursor.key_beg.delete_tid = 0;
 988         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 989         cursor.key_beg.obj_type = 0;
 990         cursor.key_beg.key = saveoff;
 991
 992         cursor.key_end = cursor.key_beg;
 993         cursor.key_end.key = HAMMER_MAX_KEY;
 994         cursor.asof = ip->obj_asof;
 995         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
 996
 997         error = hammer_ip_first(&cursor, ip);
 998
 999         while (error == 0) {
1000                 error = hammer_ip_resolve_data(&cursor);
1001                 if (error)
1002                         break;
1003                 rec = cursor.record;
1004                 base = &rec->base.base;
1005                 saveoff = base->key;
1006
1007                 if (base->obj_id != ip->obj_id)
1008                         panic("readdir: bad record at %p", cursor.node);
1009
1010                 r = vop_write_dirent(
1011                              &error, uio, rec->entry.obj_id,
1012                              hammer_get_dtype(rec->entry.base.base.obj_type),
1013                              rec->entry.base.data_len,
1014                              (void *)cursor.data);
1015                 if (r)
1016                         break;
1017                 ++saveoff;
1018                 if (cookies)
1019                         cookies[cookie_index] = base->key;
1020                 ++cookie_index;
1021                 if (cookie_index == ncookies)
1022                         break;
1023                 error = hammer_ip_next(&cursor);
1024         }
1025         hammer_done_cursor(&cursor);
1026
1027 done:
1028         if (ap->a_eofflag)
1029                 *ap->a_eofflag = (error == ENOENT);
1030         uio->uio_offset = saveoff;
1031         if (error && cookie_index == 0) {
1032                 if (error == ENOENT)
1033                         error = 0;
1034                 if (cookies) {
1035                         kfree(cookies, M_TEMP);
1036                         *ap->a_ncookies = 0;
1037                         *ap->a_cookies = NULL;
1038                 }
1039         } else {
1040                 if (error == ENOENT)
1041                         error = 0;
1042                 if (cookies) {
1043                         *ap->a_ncookies = cookie_index;
1044                         *ap->a_cookies = cookies;
1045                 }
1046         }
1047         return(error);
1048 }
1049
1050 /*
1051  * hammer_vop_readlink { vp, uio, cred }
1052  */
1053 static
1054 int
1055 hammer_vop_readlink(struct vop_readlink_args *ap)
1056 {
1057         struct hammer_cursor cursor;
1058         struct hammer_inode *ip;
1059         int error;
1060
1061         ip = VTOI(ap->a_vp);
1062         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1063
1064         /*
1065          * Key range (begin and end inclusive) to scan.  Directory keys
1066          * directly translate to a 64 bit 'seek' position.
1067          */
1068         cursor.key_beg.obj_id = ip->obj_id;
1069         cursor.key_beg.create_tid = 0;
1070         cursor.key_beg.delete_tid = 0;
1071         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1072         cursor.key_beg.obj_type = 0;
1073         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1074         cursor.asof = ip->obj_asof;
1075         cursor.flags |= HAMMER_CURSOR_ASOF;
1076
1077         error = hammer_ip_lookup(&cursor, ip);
1078         if (error == 0) {
1079                 error = hammer_ip_resolve_data(&cursor);
1080                 if (error == 0) {
1081                         error = uiomove((char *)cursor.data,
1082                                         cursor.record->generic.base.data_len,
1083                                         ap->a_uio);
1084                 }
1085         }
1086         hammer_done_cursor(&cursor);
1087         return(error);
1088 }
1089
1090 /*
1091  * hammer_vop_nremove { nch, dvp, cred }
1092  */
1093 static
1094 int
1095 hammer_vop_nremove(struct vop_nremove_args *ap)
1096 {
1097         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1098 }
1099
1100 /*
1101  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1102  */
1103 static
1104 int
1105 hammer_vop_nrename(struct vop_nrename_args *ap)
1106 {
1107         struct hammer_transaction trans;
1108         struct namecache *fncp;
1109         struct namecache *tncp;
1110         struct hammer_inode *fdip;
1111         struct hammer_inode *tdip;
1112         struct hammer_inode *ip;
1113         struct hammer_cursor cursor;
1114         union hammer_record_ondisk *rec;
1115         int64_t namekey;
1116         int error;
1117
1118         fdip = VTOI(ap->a_fdvp);
1119         tdip = VTOI(ap->a_tdvp);
1120         fncp = ap->a_fnch->ncp;
1121         tncp = ap->a_tnch->ncp;
1122         ip = VTOI(fncp->nc_vp);
1123         KKASSERT(ip != NULL);
1124
1125         if (fdip->flags & HAMMER_INODE_RO)
1126                 return (EROFS);
1127         if (tdip->flags & HAMMER_INODE_RO)
1128                 return (EROFS);
1129         if (ip->flags & HAMMER_INODE_RO)
1130                 return (EROFS);
1131
1132         hammer_start_transaction(&trans, fdip->hmp);
1133
1134         /*
1135          * Remove tncp from the target directory and then link ip as
1136          * tncp. XXX pass trans to dounlink
1137          */
1138         error = hammer_dounlink(ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1139         if (error == 0 || error == ENOENT)
1140                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1141         if (error)
1142                 goto failed; /* XXX */
1143
1144         /*
1145          * Locate the record in the originating directory and remove it.
1146          *
1147          * Calculate the namekey and setup the key range for the scan.  This
1148          * works kinda like a chained hash table where the lower 32 bits
1149          * of the namekey synthesize the chain.
1150          *
1151          * The key range is inclusive of both key_beg and key_end.
1152          */
1153         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1154
1155         hammer_init_cursor_hmp(&cursor, &fdip->cache[0], fdip->hmp);
1156         cursor.key_beg.obj_id = fdip->obj_id;
1157         cursor.key_beg.key = namekey;
1158         cursor.key_beg.create_tid = 0;
1159         cursor.key_beg.delete_tid = 0;
1160         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1161         cursor.key_beg.obj_type = 0;
1162
1163         cursor.key_end = cursor.key_beg;
1164         cursor.key_end.key |= 0xFFFFFFFFULL;
1165         cursor.asof = fdip->obj_asof;
1166         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1167
1168         /*
1169          * Scan all matching records (the chain), locate the one matching
1170          * the requested path component.
1171          *
1172          * The hammer_ip_*() functions merge in-memory records with on-disk
1173          * records for the purposes of the search.
1174          */
1175         error = hammer_ip_first(&cursor, fdip);
1176         while (error == 0) {
1177                 if (hammer_ip_resolve_data(&cursor) != 0)
1178                         break;
1179                 rec = cursor.record;
1180                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1181                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1182                         break;
1183                 }
1184                 error = hammer_ip_next(&cursor);
1185         }
1186
1187         /*
1188          * If all is ok we have to get the inode so we can adjust nlinks.
1189          */
1190         if (error)
1191                 goto done;
1192         error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1193
1194         if (error == 0)
1195                 cache_rename(ap->a_fnch, ap->a_tnch);
1196 done:
1197         hammer_done_cursor(&cursor);
1198 failed:
1199         if (error == 0) {
1200                 hammer_commit_transaction(&trans);
1201         } else {
1202                 hammer_abort_transaction(&trans);
1203         }
1204         return (error);
1205 }
1206
1207 /*
1208  * hammer_vop_nrmdir { nch, dvp, cred }
1209  */
1210 static
1211 int
1212 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1213 {
1214         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1215 }
1216
1217 /*
1218  * hammer_vop_setattr { vp, vap, cred }
1219  */
1220 static
1221 int
1222 hammer_vop_setattr(struct vop_setattr_args *ap)
1223 {
1224         struct hammer_transaction trans;
1225         struct hammer_cursor *spike = NULL;
1226         struct vattr *vap;
1227         struct hammer_inode *ip;
1228         int modflags;
1229         int error;
1230         int64_t aligned_size;
1231         u_int32_t flags;
1232         uuid_t uuid;
1233
1234         vap = ap->a_vap;
1235         ip = ap->a_vp->v_data;
1236         modflags = 0;
1237
1238         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1239                 return(EROFS);
1240         if (ip->flags & HAMMER_INODE_RO)
1241                 return (EROFS);
1242
1243         hammer_start_transaction(&trans, ip->hmp);
1244         error = 0;
1245
1246         if (vap->va_flags != VNOVAL) {
1247                 flags = ip->ino_data.uflags;
1248                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1249                                          hammer_to_unix_xid(&ip->ino_data.uid),
1250                                          ap->a_cred);
1251                 if (error == 0) {
1252                         if (ip->ino_data.uflags != flags) {
1253                                 ip->ino_data.uflags = flags;
1254                                 modflags |= HAMMER_INODE_DDIRTY;
1255                         }
1256                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1257                                 error = 0;
1258                                 goto done;
1259                         }
1260                 }
1261                 goto done;
1262         }
1263         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1264                 error = EPERM;
1265                 goto done;
1266         }
1267         if (vap->va_uid != (uid_t)VNOVAL) {
1268                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1269                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1270                         ip->ino_data.uid = uuid;
1271                         modflags |= HAMMER_INODE_DDIRTY;
1272                 }
1273         }
1274         if (vap->va_gid != (uid_t)VNOVAL) {
1275                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1276                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1277                         ip->ino_data.gid = uuid;
1278                         modflags |= HAMMER_INODE_DDIRTY;
1279                 }
1280         }
1281         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1282                 switch(ap->a_vp->v_type) {
1283                 case VREG:
1284                         if (vap->va_size < ip->ino_rec.ino_size) {
1285                                 vtruncbuf(ap->a_vp, vap->va_size,
1286                                           HAMMER_BUFSIZE);
1287                         } else if (vap->va_size > ip->ino_rec.ino_size) {
1288                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1289                         }
1290                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1291                                         ~(int64_t)HAMMER_BUFMASK;
1292                         error = hammer_ip_delete_range(&trans, ip,
1293                                                     aligned_size,
1294                                                     0x7FFFFFFFFFFFFFFFLL,
1295                                                     &spike);
1296                         ip->ino_rec.ino_size = vap->va_size;
1297                         modflags |= HAMMER_INODE_RDIRTY;
1298                         break;
1299                 case VDATABASE:
1300                         error = hammer_ip_delete_range(&trans, ip,
1301                                                     vap->va_size,
1302                                                     0x7FFFFFFFFFFFFFFFLL,
1303                                                     &spike);
1304                         ip->ino_rec.ino_size = vap->va_size;
1305                         modflags |= HAMMER_INODE_RDIRTY;
1306                         break;
1307                 default:
1308                         error = EINVAL;
1309                         goto done;
1310                 }
1311                 if (error == ENOSPC) {
1312                         error = hammer_spike(&spike);
1313                         if (error == 0)
1314                                 continue;
1315                 }
1316                 KKASSERT(spike == NULL);
1317                 break;
1318         }
1319         if (vap->va_atime.tv_sec != VNOVAL) {
1320                 ip->ino_rec.ino_atime =
1321                         hammer_timespec_to_transid(&vap->va_atime);
1322                 modflags |= HAMMER_INODE_ITIMES;
1323         }
1324         if (vap->va_mtime.tv_sec != VNOVAL) {
1325                 ip->ino_rec.ino_mtime =
1326                         hammer_timespec_to_transid(&vap->va_mtime);
1327                 modflags |= HAMMER_INODE_ITIMES;
1328         }
1329         if (vap->va_mode != (mode_t)VNOVAL) {
1330                 if (ip->ino_data.mode != vap->va_mode) {
1331                         ip->ino_data.mode = vap->va_mode;
1332                         modflags |= HAMMER_INODE_DDIRTY;
1333                 }
1334         }
1335 done:
1336         if (error) {
1337                 hammer_abort_transaction(&trans);
1338         } else {
1339                 hammer_modify_inode(&trans, ip, modflags);
1340                 hammer_commit_transaction(&trans);
1341         }
1342         return (error);
1343 }
1344
1345 /*
1346  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1347  */
1348 static
1349 int
1350 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1351 {
1352         struct hammer_transaction trans;
1353         struct hammer_inode *dip;
1354         struct hammer_inode *nip;
1355         struct nchandle *nch;
1356         hammer_record_t record;
1357         int error;
1358         int bytes;
1359
1360         ap->a_vap->va_type = VLNK;
1361
1362         nch = ap->a_nch;
1363         dip = VTOI(ap->a_dvp);
1364
1365         if (dip->flags & HAMMER_INODE_RO)
1366                 return (EROFS);
1367
1368         /*
1369          * Create a transaction to cover the operations we perform.
1370          */
1371         hammer_start_transaction(&trans, dip->hmp);
1372
1373         /*
1374          * Create a new filesystem object of the requested type.  The
1375          * returned inode will be referenced but not locked.
1376          */
1377
1378         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1379         if (error) {
1380                 hammer_abort_transaction(&trans);
1381                 *ap->a_vpp = NULL;
1382                 return (error);
1383         }
1384
1385         /*
1386          * Add the new filesystem object to the directory.  This will also
1387          * bump the inode's link count.
1388          */
1389         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1390
1391         /*
1392          * Add a record representing the symlink.  symlink stores the link
1393          * as pure data, not a string, and is no \0 terminated.
1394          */
1395         if (error == 0) {
1396                 record = hammer_alloc_mem_record(nip);
1397                 bytes = strlen(ap->a_target);
1398
1399                 record->rec.generic.base.base.key = HAMMER_FIXKEY_SYMLINK;
1400                 record->rec.generic.base.base.rec_type = HAMMER_RECTYPE_FIX;
1401                 record->rec.generic.base.data_len = bytes;
1402                 if (bytes <= sizeof(record->rec.generic.filler)) {
1403                         record->data = (void *)record->rec.generic.filler;
1404                         bcopy(ap->a_target, record->data, bytes);
1405                 } else {
1406                         record->data = (void *)ap->a_target;
1407                         /* will be reallocated by routine below */
1408                 }
1409                 error = hammer_ip_add_record(&trans, record);
1410         }
1411
1412         /*
1413          * Finish up.
1414          */
1415         if (error) {
1416                 hammer_rel_inode(nip, 0);
1417                 hammer_abort_transaction(&trans);
1418                 *ap->a_vpp = NULL;
1419         } else {
1420                 hammer_commit_transaction(&trans);
1421                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1422                 hammer_rel_inode(nip, 0);
1423                 if (error == 0) {
1424                         cache_setunresolved(ap->a_nch);
1425                         cache_setvp(ap->a_nch, *ap->a_vpp);
1426                 }
1427         }
1428         return (error);
1429 }
1430
1431 /*
1432  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1433  */
1434 static
1435 int
1436 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1437 {
1438         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1439 }
1440
1441 /*
1442  * hammer_vop_strategy { vp, bio }
1443  *
1444  * Strategy call, used for regular file read & write only.  Note that the
1445  * bp may represent a cluster.
1446  *
1447  * To simplify operation and allow better optimizations in the future,
1448  * this code does not make any assumptions with regards to buffer alignment
1449  * or size.
1450  */
1451 static
1452 int
1453 hammer_vop_strategy(struct vop_strategy_args *ap)
1454 {
1455         struct buf *bp;
1456         int error;
1457
1458         bp = ap->a_bio->bio_buf;
1459
1460         switch(bp->b_cmd) {
1461         case BUF_CMD_READ:
1462                 error = hammer_vop_strategy_read(ap);
1463                 break;
1464         case BUF_CMD_WRITE:
1465                 error = hammer_vop_strategy_write(ap);
1466                 break;
1467         default:
1468                 error = EINVAL;
1469                 break;
1470         }
1471         bp->b_error = error;
1472         if (error)
1473                 bp->b_flags |= B_ERROR;
1474         biodone(ap->a_bio);
1475         return (error);
1476 }
1477
1478 /*
1479  * Read from a regular file.  Iterate the related records and fill in the
1480  * BIO/BUF.  Gaps are zero-filled.
1481  *
1482  * The support code in hammer_object.c should be used to deal with mixed
1483  * in-memory and on-disk records.
1484  *
1485  * XXX atime update
1486  */
1487 static
1488 int
1489 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1490 {
1491         struct hammer_inode *ip = ap->a_vp->v_data;
1492         struct hammer_cursor cursor;
1493         hammer_record_ondisk_t rec;
1494         hammer_base_elm_t base;
1495         struct bio *bio;
1496         struct buf *bp;
1497         int64_t rec_offset;
1498         int64_t ran_end;
1499         int64_t tmp64;
1500         int error;
1501         int boff;
1502         int roff;
1503         int n;
1504
1505         bio = ap->a_bio;
1506         bp = bio->bio_buf;
1507
1508         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1509
1510         /*
1511          * Key range (begin and end inclusive) to scan.  Note that the key's
1512          * stored in the actual records represent BASE+LEN, not BASE.  The
1513          * first record containing bio_offset will have a key > bio_offset.
1514          */
1515         cursor.key_beg.obj_id = ip->obj_id;
1516         cursor.key_beg.create_tid = 0;
1517         cursor.key_beg.delete_tid = 0;
1518         cursor.key_beg.obj_type = 0;
1519         cursor.key_beg.key = bio->bio_offset + 1;
1520         cursor.asof = ip->obj_asof;
1521         cursor.flags |= HAMMER_CURSOR_ASOF;
1522
1523         cursor.key_end = cursor.key_beg;
1524         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1525                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1526                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1527                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1528         } else {
1529                 ran_end = bio->bio_offset + bp->b_bufsize;
1530                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1531                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1532                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1533                 if (tmp64 < ran_end)
1534                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1535                 else
1536                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1537         }
1538         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1539
1540         error = hammer_ip_first(&cursor, ip);
1541         boff = 0;
1542
1543         while (error == 0) {
1544                 error = hammer_ip_resolve_data(&cursor);
1545                 if (error)
1546                         break;
1547                 rec = cursor.record;
1548                 base = &rec->base.base;
1549
1550                 rec_offset = base->key - rec->data.base.data_len;
1551
1552                 /*
1553                  * Calculate the gap, if any, and zero-fill it.
1554                  */
1555                 n = (int)(rec_offset - (bio->bio_offset + boff));
1556                 if (n > 0) {
1557                         if (n > bp->b_bufsize - boff)
1558                                 n = bp->b_bufsize - boff;
1559                         kprintf("zfill %d bytes\n", n);
1560                         bzero((char *)bp->b_data + boff, n);
1561                         boff += n;
1562                         n = 0;
1563                 }
1564
1565                 /*
1566                  * Calculate the data offset in the record and the number
1567                  * of bytes we can copy.
1568                  *
1569                  * Note there is a degenerate case here where boff may
1570                  * already be at bp->b_bufsize.
1571                  */
1572                 roff = -n;
1573                 n = rec->data.base.data_len - roff;
1574                 KKASSERT(n > 0);
1575                 if (n > bp->b_bufsize - boff)
1576                         n = bp->b_bufsize - boff;
1577                 bcopy((char *)cursor.data + roff, (char *)bp->b_data + boff, n);
1578                 boff += n;
1579                 if (boff == bp->b_bufsize)
1580                         break;
1581                 error = hammer_ip_next(&cursor);
1582         }
1583         hammer_done_cursor(&cursor);
1584
1585         /*
1586          * There may have been a gap after the last record
1587          */
1588         if (error == ENOENT)
1589                 error = 0;
1590         if (error == 0 && boff != bp->b_bufsize) {
1591                 KKASSERT(boff < bp->b_bufsize);
1592                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1593                 /* boff = bp->b_bufsize; */
1594         }
1595         bp->b_resid = 0;
1596         return(error);
1597 }
1598
1599 /*
1600  * Write to a regular file.  Iterate the related records and mark for
1601  * deletion.  If existing edge records (left and right side) overlap our
1602  * write they have to be marked deleted and new records created, usually
1603  * referencing a portion of the original data.  Then add a record to
1604  * represent the buffer.
1605  *
1606  * The support code in hammer_object.c should be used to deal with mixed
1607  * in-memory and on-disk records.
1608  */
1609 static
1610 int
1611 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1612 {
1613         struct hammer_transaction trans;
1614         struct hammer_cursor *spike = NULL;
1615         hammer_inode_t ip;
1616         struct bio *bio;
1617         struct buf *bp;
1618         int error;
1619
1620         bio = ap->a_bio;
1621         bp = bio->bio_buf;
1622         ip = ap->a_vp->v_data;
1623
1624         if (ip->flags & HAMMER_INODE_RO)
1625                 return (EROFS);
1626
1627         hammer_start_transaction(&trans, ip->hmp);
1628
1629 retry:
1630         /*
1631          * Delete any records overlapping our range.  This function will
1632          * (eventually) properly truncate partial overlaps.
1633          */
1634         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1635                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1636                                                bio->bio_offset, &spike);
1637         } else {
1638                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1639                                                bio->bio_offset +
1640                                                 bp->b_bufsize - 1,
1641                                                &spike);
1642         }
1643
1644         /*
1645          * Add a single record to cover the write
1646          */
1647         if (error == 0) {
1648                 error = hammer_ip_sync_data(&trans, ip, bio->bio_offset,
1649                                             bp->b_data, bp->b_bufsize,
1650                                             &spike);
1651         }
1652
1653         /*
1654          * If we ran out of space the spike structure will be filled in
1655          * and we must call hammer_spike with it, then retry.
1656          */
1657         if (error == ENOSPC) {
1658                 error = hammer_spike(&spike);
1659                 if (error == 0)
1660                         goto retry;
1661         }
1662         KKASSERT(spike == NULL);
1663
1664         /*
1665          * If an error occured abort the transaction
1666          */
1667         if (error) {
1668                 /* XXX undo deletion */
1669                 hammer_abort_transaction(&trans);
1670                 bp->b_resid = bp->b_bufsize;
1671         } else {
1672                 hammer_commit_transaction(&trans);
1673                 bp->b_resid = 0;
1674         }
1675         return(error);
1676 }
1677
1678 /*
1679  * dounlink - disconnect a directory entry
1680  *
1681  * XXX whiteout support not really in yet
1682  */
1683 static int
1684 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1685                 int flags)
1686 {
1687         struct hammer_transaction trans;
1688         struct namecache *ncp;
1689         hammer_inode_t dip;
1690         hammer_inode_t ip;
1691         hammer_record_ondisk_t rec;
1692         struct hammer_cursor cursor;
1693         int64_t namekey;
1694         int error;
1695
1696         /*
1697          * Calculate the namekey and setup the key range for the scan.  This
1698          * works kinda like a chained hash table where the lower 32 bits
1699          * of the namekey synthesize the chain.
1700          *
1701          * The key range is inclusive of both key_beg and key_end.
1702          */
1703         dip = VTOI(dvp);
1704         ncp = nch->ncp;
1705
1706         if (dip->flags & HAMMER_INODE_RO)
1707                 return (EROFS);
1708
1709         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1710
1711         hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
1712         cursor.key_beg.obj_id = dip->obj_id;
1713         cursor.key_beg.key = namekey;
1714         cursor.key_beg.create_tid = 0;
1715         cursor.key_beg.delete_tid = 0;
1716         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1717         cursor.key_beg.obj_type = 0;
1718
1719         cursor.key_end = cursor.key_beg;
1720         cursor.key_end.key |= 0xFFFFFFFFULL;
1721         cursor.asof = dip->obj_asof;
1722         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1723
1724         hammer_start_transaction(&trans, dip->hmp);
1725
1726         /*
1727          * Scan all matching records (the chain), locate the one matching
1728          * the requested path component.  info->last_error contains the
1729          * error code on search termination and could be 0, ENOENT, or
1730          * something else.
1731          *
1732          * The hammer_ip_*() functions merge in-memory records with on-disk
1733          * records for the purposes of the search.
1734          */
1735         error = hammer_ip_first(&cursor, dip);
1736         while (error == 0) {
1737                 error = hammer_ip_resolve_data(&cursor);
1738                 if (error)
1739                         break;
1740                 rec = cursor.record;
1741                 if (ncp->nc_nlen == rec->entry.base.data_len &&
1742                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1743                         break;
1744                 }
1745                 error = hammer_ip_next(&cursor);
1746         }
1747
1748         /*
1749          * If all is ok we have to get the inode so we can adjust nlinks.
1750          *
1751          * If the target is a directory, it must be empty.
1752          */
1753         if (error == 0) {
1754                 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
1755                                       rec->entry.obj_id,
1756                                       dip->hmp->asof, 0, &error);
1757                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
1758                                   HAMMER_OBJTYPE_DIRECTORY) {
1759                         error = hammer_ip_check_directory_empty(&trans, ip);
1760                 }
1761                 if (error == 0)
1762                         error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1763                 if (error == 0) {
1764                         cache_setunresolved(nch);
1765                         cache_setvp(nch, NULL);
1766                         /* XXX locking */
1767                         if (ip->vp)
1768                                 cache_inval_vp(ip->vp, CINV_DESTROY);
1769                 }
1770                 hammer_rel_inode(ip, 0);
1771         }
1772
1773         if (error == 0)
1774                 hammer_commit_transaction(&trans);
1775         else
1776                 hammer_abort_transaction(&trans);
1777         hammer_done_cursor(&cursor);
1778         return (error);
1779 }
1780
1781 /************************************************************************
1782  *                          FIFO AND SPECFS OPS                         *
1783  ************************************************************************
1784  *
1785  */
1786
1787 static int
1788 hammer_vop_fifoclose (struct vop_close_args *ap)
1789 {
1790         /* XXX update itimes */
1791         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1792 }
1793
1794 static int
1795 hammer_vop_fiforead (struct vop_read_args *ap)
1796 {
1797         int error;
1798
1799         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1800         /* XXX update access time */
1801         return (error);
1802 }
1803
1804 static int
1805 hammer_vop_fifowrite (struct vop_write_args *ap)
1806 {
1807         int error;
1808
1809         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1810         /* XXX update access time */
1811         return (error);
1812 }
1813
1814 static int
1815 hammer_vop_specclose (struct vop_close_args *ap)
1816 {
1817         /* XXX update itimes */
1818         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1819 }
1820
1821 static int
1822 hammer_vop_specread (struct vop_read_args *ap)
1823 {
1824         /* XXX update access time */
1825         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1826 }
1827
1828 static int
1829 hammer_vop_specwrite (struct vop_write_args *ap)
1830 {
1831         /* XXX update last change time */
1832         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1833 }
1834