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 1.15 2008/01/01 01:00:03 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                                     0, 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                                     0, 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                                     0, 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                 }
 333                 n = HAMMER_BUFSIZE - offset;
 334                 if (n > uio->uio_resid)
 335                         n = uio->uio_resid;
 336                 error = uiomove((char *)bp->b_data + offset, n, uio);
 337                 if (error) {
 338                         brelse(bp);
 339                         break;
 340                 }
 341                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 342                 if (ip->ino_rec.ino_size < uio->uio_offset) {
 343                         ip->ino_rec.ino_size = uio->uio_offset;
 344                         flags = HAMMER_INODE_RDIRTY;
 345                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
 346                 } else {
 347                         flags = 0;
 348                 }
 349                 ip->ino_rec.ino_mtime = trans.tid;
 350                 flags |= HAMMER_INODE_ITIMES;
 351                 hammer_modify_inode(&trans, ip, flags);
 352                 if (ap->a_ioflag & IO_SYNC) {
 353                         bwrite(bp);
 354                 } else if (ap->a_ioflag & IO_DIRECT) {
 355                         bawrite(bp);
 356                 } else {
 357                         bdwrite(bp);
 358                 }
 359         }
 360         if (error)
 361                 hammer_abort_transaction(&trans);
 362         else
 363                 hammer_commit_transaction(&trans);
 364         return (error);
 365 }
 366
 367 /*
 368  * hammer_vop_access { vp, mode, cred }
 369  */
 370 static
 371 int
 372 hammer_vop_access(struct vop_access_args *ap)
 373 {
 374         struct hammer_inode *ip = VTOI(ap->a_vp);
 375         uid_t uid;
 376         gid_t gid;
 377         int error;
 378
 379         uid = hammer_to_unix_xid(&ip->ino_data.uid);
 380         gid = hammer_to_unix_xid(&ip->ino_data.gid);
 381
 382         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
 383                                   ip->ino_data.uflags);
 384         return (error);
 385 }
 386
 387 /*
 388  * hammer_vop_advlock { vp, id, op, fl, flags }
 389  */
 390 static
 391 int
 392 hammer_vop_advlock(struct vop_advlock_args *ap)
 393 {
 394         struct hammer_inode *ip = VTOI(ap->a_vp);
 395
 396         return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
 397 }
 398
 399 /*
 400  * hammer_vop_close { vp, fflag }
 401  */
 402 static
 403 int
 404 hammer_vop_close(struct vop_close_args *ap)
 405 {
 406         return (vop_stdclose(ap));
 407 }
 408
 409 /*
 410  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
 411  *
 412  * The operating system has already ensured that the directory entry
 413  * does not exist and done all appropriate namespace locking.
 414  */
 415 static
 416 int
 417 hammer_vop_ncreate(struct vop_ncreate_args *ap)
 418 {
 419         struct hammer_transaction trans;
 420         struct hammer_inode *dip;
 421         struct hammer_inode *nip;
 422         struct nchandle *nch;
 423         int error;
 424
 425         nch = ap->a_nch;
 426         dip = VTOI(ap->a_dvp);
 427
 428         if (dip->flags & HAMMER_INODE_RO)
 429                 return (EROFS);
 430
 431         /*
 432          * Create a transaction to cover the operations we perform.
 433          */
 434         hammer_start_transaction(&trans, dip->hmp);
 435
 436         /*
 437          * Create a new filesystem object of the requested type.  The
 438          * returned inode will be referenced but not locked.
 439          */
 440
 441         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 442         if (error)
 443                 kprintf("hammer_create_inode error %d\n", error);
 444         if (error) {
 445                 hammer_abort_transaction(&trans);
 446                 *ap->a_vpp = NULL;
 447                 return (error);
 448         }
 449
 450         /*
 451          * Add the new filesystem object to the directory.  This will also
 452          * bump the inode's link count.
 453          */
 454         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 455         if (error)
 456                 kprintf("hammer_ip_add_directory error %d\n", error);
 457
 458         /*
 459          * Finish up.
 460          */
 461         if (error) {
 462                 hammer_rel_inode(nip, 0);
 463                 hammer_abort_transaction(&trans);
 464                 *ap->a_vpp = NULL;
 465         } else {
 466                 hammer_commit_transaction(&trans);
 467                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 468                 hammer_rel_inode(nip, 0);
 469                 if (error == 0) {
 470                         cache_setunresolved(ap->a_nch);
 471                         cache_setvp(ap->a_nch, *ap->a_vpp);
 472                 }
 473         }
 474         return (error);
 475 }
 476
 477 /*
 478  * hammer_vop_getattr { vp, vap }
 479  */
 480 static
 481 int
 482 hammer_vop_getattr(struct vop_getattr_args *ap)
 483 {
 484         struct hammer_inode *ip = VTOI(ap->a_vp);
 485         struct vattr *vap = ap->a_vap;
 486
 487 #if 0
 488         if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
 489             (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
 490             ip->obj_asof == XXX
 491         ) {
 492                 /* LAZYMOD XXX */
 493         }
 494         hammer_itimes(ap->a_vp);
 495 #endif
 496
 497         vap->va_fsid = ip->hmp->fsid_udev;
 498         vap->va_fileid = ip->ino_rec.base.base.obj_id;
 499         vap->va_mode = ip->ino_data.mode;
 500         vap->va_nlink = ip->ino_rec.ino_nlinks;
 501         vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
 502         vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
 503         vap->va_rmajor = 0;
 504         vap->va_rminor = 0;
 505         vap->va_size = ip->ino_rec.ino_size;
 506         hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
 507         hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
 508         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
 509         vap->va_flags = ip->ino_data.uflags;
 510         vap->va_gen = 1;        /* hammer inums are unique for all time */
 511         vap->va_blocksize = 32768; /* XXX - extract from root volume */
 512         vap->va_bytes = ip->ino_rec.ino_size;
 513         vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
 514         vap->va_filerev = 0;    /* XXX */
 515         /* mtime uniquely identifies any adjustments made to the file */
 516         vap->va_fsmid = ip->ino_rec.ino_mtime;
 517         vap->va_uid_uuid = ip->ino_data.uid;
 518         vap->va_gid_uuid = ip->ino_data.gid;
 519         vap->va_fsid_uuid = ip->hmp->fsid;
 520         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 521                           VA_FSID_UUID_VALID;
 522
 523         switch (ip->ino_rec.base.base.obj_type) {
 524         case HAMMER_OBJTYPE_CDEV:
 525         case HAMMER_OBJTYPE_BDEV:
 526                 vap->va_rmajor = ip->ino_data.rmajor;
 527                 vap->va_rminor = ip->ino_data.rminor;
 528                 break;
 529         default:
 530                 break;
 531         }
 532
 533         return(0);
 534 }
 535
 536 /*
 537  * hammer_vop_nresolve { nch, dvp, cred }
 538  *
 539  * Locate the requested directory entry.
 540  */
 541 static
 542 int
 543 hammer_vop_nresolve(struct vop_nresolve_args *ap)
 544 {
 545         struct namecache *ncp;
 546         hammer_inode_t dip;
 547         hammer_inode_t ip;
 548         hammer_tid_t asof;
 549         struct hammer_cursor cursor;
 550         union hammer_record_ondisk *rec;
 551         struct vnode *vp;
 552         int64_t namekey;
 553         int error;
 554         int i;
 555         int nlen;
 556         int flags;
 557
 558         /*
 559          * Misc initialization, plus handle as-of name extensions.  Look for
 560          * the '@@' extension.  Note that as-of files and directories cannot
 561          * be modified.
 562          */
 563         dip = VTOI(ap->a_dvp);
 564         ncp = ap->a_nch->ncp;
 565         asof = dip->obj_asof;
 566         nlen = ncp->nc_nlen;
 567         flags = dip->flags;
 568
 569         for (i = 0; i < nlen; ++i) {
 570                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
 571                         asof = hammer_str_to_tid(ncp->nc_name + i + 2);
 572                         kprintf("ASOF %016llx\n", asof);
 573                         flags |= HAMMER_INODE_RO;
 574                         break;
 575                 }
 576         }
 577         nlen = i;
 578
 579         /*
 580          * If there is no path component the time extension is relative to
 581          * dip.
 582          */
 583         if (nlen == 0) {
 584                 ip = hammer_get_inode(dip->hmp, dip->obj_id, asof,
 585                                       flags, &error);
 586                 if (error == 0) {
 587                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 588                         hammer_rel_inode(ip, 0);
 589                 } else {
 590                         vp = NULL;
 591                 }
 592                 if (error == 0) {
 593                         vn_unlock(vp);
 594                         cache_setvp(ap->a_nch, vp);
 595                         vrele(vp);
 596                 }
 597                 return(error);
 598         }
 599
 600         /*
 601          * Calculate the namekey and setup the key range for the scan.  This
 602          * works kinda like a chained hash table where the lower 32 bits
 603          * of the namekey synthesize the chain.
 604          *
 605          * The key range is inclusive of both key_beg and key_end.
 606          */
 607         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
 608
 609         hammer_init_cursor_ip(&cursor, dip);
 610         cursor.key_beg.obj_id = dip->obj_id;
 611         cursor.key_beg.key = namekey;
 612         cursor.key_beg.create_tid = asof;
 613         cursor.key_beg.delete_tid = 0;
 614         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 615         cursor.key_beg.obj_type = 0;
 616
 617         cursor.key_end = cursor.key_beg;
 618         cursor.key_end.key |= 0xFFFFFFFFULL;
 619         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
 620
 621         /*
 622          * Scan all matching records (the chain), locate the one matching
 623          * the requested path component.
 624          *
 625          * The hammer_ip_*() functions merge in-memory records with on-disk
 626          * records for the purposes of the search.
 627          */
 628         error = hammer_ip_first(&cursor, dip);
 629         rec = NULL;
 630         while (error == 0) {
 631                 error = hammer_ip_resolve_data(&cursor);
 632                 if (error)
 633                         break;
 634                 rec = cursor.record;
 635                 if (nlen == rec->entry.base.data_len &&
 636                     bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
 637                         break;
 638                 }
 639                 error = hammer_ip_next(&cursor);
 640         }
 641         if (error == 0) {
 642                 ip = hammer_get_inode(dip->hmp, rec->entry.obj_id, asof,
 643                                       flags, &error);
 644                 if (error == 0) {
 645                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 646                         hammer_rel_inode(ip, 0);
 647                 } else {
 648                         vp = NULL;
 649                 }
 650                 if (error == 0) {
 651                         vn_unlock(vp);
 652                         cache_setvp(ap->a_nch, vp);
 653                         vrele(vp);
 654                 }
 655         } else if (error == ENOENT) {
 656                 cache_setvp(ap->a_nch, NULL);
 657         }
 658         hammer_done_cursor(&cursor);
 659         return (error);
 660 }
 661
 662 /*
 663  * hammer_vop_nlookupdotdot { dvp, vpp, cred }
 664  *
 665  * Locate the parent directory of a directory vnode.
 666  *
 667  * dvp is referenced but not locked.  *vpp must be returned referenced and
 668  * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
 669  * at the root, instead it could indicate that the directory we were in was
 670  * removed.
 671  */
 672 static
 673 int
 674 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
 675 {
 676         struct hammer_inode *dip;
 677         struct hammer_inode *ip;
 678         u_int64_t parent_obj_id;
 679         int error;
 680
 681         dip = VTOI(ap->a_dvp);
 682         if ((parent_obj_id = dip->ino_data.parent_obj_id) == 0) {
 683                 *ap->a_vpp = NULL;
 684                 return ENOENT;
 685         }
 686
 687         ip = hammer_get_inode(dip->hmp, parent_obj_id, dip->obj_asof,
 688                               dip->flags, &error);
 689         if (ip == NULL) {
 690                 *ap->a_vpp = NULL;
 691                 return(error);
 692         }
 693         error = hammer_get_vnode(ip, LK_EXCLUSIVE, ap->a_vpp);
 694         hammer_rel_inode(ip, 0);
 695         return (error);
 696 }
 697
 698 /*
 699  * hammer_vop_nlink { nch, dvp, vp, cred }
 700  */
 701 static
 702 int
 703 hammer_vop_nlink(struct vop_nlink_args *ap)
 704 {
 705         struct hammer_transaction trans;
 706         struct hammer_inode *dip;
 707         struct hammer_inode *ip;
 708         struct nchandle *nch;
 709         int error;
 710
 711         nch = ap->a_nch;
 712         dip = VTOI(ap->a_dvp);
 713         ip = VTOI(ap->a_vp);
 714
 715         if (dip->flags & HAMMER_INODE_RO)
 716                 return (EROFS);
 717         if (ip->flags & HAMMER_INODE_RO)
 718                 return (EROFS);
 719
 720         /*
 721          * Create a transaction to cover the operations we perform.
 722          */
 723         hammer_start_transaction(&trans, dip->hmp);
 724
 725         /*
 726          * Add the filesystem object to the directory.  Note that neither
 727          * dip nor ip are referenced or locked, but their vnodes are
 728          * referenced.  This function will bump the inode's link count.
 729          */
 730         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 731
 732         /*
 733          * Finish up.
 734          */
 735         if (error) {
 736                 hammer_abort_transaction(&trans);
 737         } else {
 738                 cache_setunresolved(nch);
 739                 cache_setvp(nch, ap->a_vp);
 740                 hammer_commit_transaction(&trans);
 741         }
 742         return (error);
 743 }
 744
 745 /*
 746  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 747  *
 748  * The operating system has already ensured that the directory entry
 749  * does not exist and done all appropriate namespace locking.
 750  */
 751 static
 752 int
 753 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 754 {
 755         struct hammer_transaction trans;
 756         struct hammer_inode *dip;
 757         struct hammer_inode *nip;
 758         struct nchandle *nch;
 759         int error;
 760
 761         nch = ap->a_nch;
 762         dip = VTOI(ap->a_dvp);
 763
 764         if (dip->flags & HAMMER_INODE_RO)
 765                 return (EROFS);
 766
 767         /*
 768          * Create a transaction to cover the operations we perform.
 769          */
 770         hammer_start_transaction(&trans, dip->hmp);
 771
 772         /*
 773          * Create a new filesystem object of the requested type.  The
 774          * returned inode will be referenced but not locked.
 775          */
 776         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 777         if (error)
 778                 kprintf("hammer_mkdir error %d\n", error);
 779         if (error) {
 780                 hammer_abort_transaction(&trans);
 781                 *ap->a_vpp = NULL;
 782                 return (error);
 783         }
 784
 785         /*
 786          * Add the new filesystem object to the directory.  This will also
 787          * bump the inode's link count.
 788          */
 789         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 790         if (error)
 791                 kprintf("hammer_mkdir (add) error %d\n", error);
 792
 793         /*
 794          * Finish up.
 795          */
 796         if (error) {
 797                 hammer_rel_inode(nip, 0);
 798                 hammer_abort_transaction(&trans);
 799                 *ap->a_vpp = NULL;
 800         } else {
 801                 hammer_commit_transaction(&trans);
 802                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 803                 hammer_rel_inode(nip, 0);
 804                 if (error == 0) {
 805                         cache_setunresolved(ap->a_nch);
 806                         cache_setvp(ap->a_nch, *ap->a_vpp);
 807                 }
 808         }
 809         return (error);
 810 }
 811
 812 /*
 813  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 814  *
 815  * The operating system has already ensured that the directory entry
 816  * does not exist and done all appropriate namespace locking.
 817  */
 818 static
 819 int
 820 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 821 {
 822         struct hammer_transaction trans;
 823         struct hammer_inode *dip;
 824         struct hammer_inode *nip;
 825         struct nchandle *nch;
 826         int error;
 827
 828         nch = ap->a_nch;
 829         dip = VTOI(ap->a_dvp);
 830
 831         if (dip->flags & HAMMER_INODE_RO)
 832                 return (EROFS);
 833
 834         /*
 835          * Create a transaction to cover the operations we perform.
 836          */
 837         hammer_start_transaction(&trans, dip->hmp);
 838
 839         /*
 840          * Create a new filesystem object of the requested type.  The
 841          * returned inode will be referenced but not locked.
 842          */
 843         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 844         if (error) {
 845                 hammer_abort_transaction(&trans);
 846                 *ap->a_vpp = NULL;
 847                 return (error);
 848         }
 849
 850         /*
 851          * Add the new filesystem object to the directory.  This will also
 852          * bump the inode's link count.
 853          */
 854         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 855
 856         /*
 857          * Finish up.
 858          */
 859         if (error) {
 860                 hammer_rel_inode(nip, 0);
 861                 hammer_abort_transaction(&trans);
 862                 *ap->a_vpp = NULL;
 863         } else {
 864                 hammer_commit_transaction(&trans);
 865                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 866                 hammer_rel_inode(nip, 0);
 867                 if (error == 0) {
 868                         cache_setunresolved(ap->a_nch);
 869                         cache_setvp(ap->a_nch, *ap->a_vpp);
 870                 }
 871         }
 872         return (error);
 873 }
 874
 875 /*
 876  * hammer_vop_open { vp, mode, cred, fp }
 877  */
 878 static
 879 int
 880 hammer_vop_open(struct vop_open_args *ap)
 881 {
 882         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
 883                 return (EROFS);
 884
 885         return(vop_stdopen(ap));
 886 }
 887
 888 /*
 889  * hammer_vop_pathconf { vp, name, retval }
 890  */
 891 static
 892 int
 893 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 894 {
 895         return EOPNOTSUPP;
 896 }
 897
 898 /*
 899  * hammer_vop_print { vp }
 900  */
 901 static
 902 int
 903 hammer_vop_print(struct vop_print_args *ap)
 904 {
 905         return EOPNOTSUPP;
 906 }
 907
 908 /*
 909  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 910  */
 911 static
 912 int
 913 hammer_vop_readdir(struct vop_readdir_args *ap)
 914 {
 915         struct hammer_cursor cursor;
 916         struct hammer_inode *ip;
 917         struct uio *uio;
 918         hammer_record_ondisk_t rec;
 919         hammer_base_elm_t base;
 920         int error;
 921         int cookie_index;
 922         int ncookies;
 923         off_t *cookies;
 924         off_t saveoff;
 925         int r;
 926
 927         ip = VTOI(ap->a_vp);
 928         uio = ap->a_uio;
 929         saveoff = uio->uio_offset;
 930
 931         if (ap->a_ncookies) {
 932                 ncookies = uio->uio_resid / 16 + 1;
 933                 if (ncookies > 1024)
 934                         ncookies = 1024;
 935                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 936                 cookie_index = 0;
 937         } else {
 938                 ncookies = -1;
 939                 cookies = NULL;
 940                 cookie_index = 0;
 941         }
 942
 943         /*
 944          * Handle artificial entries
 945          */
 946         error = 0;
 947         if (saveoff == 0) {
 948                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
 949                 if (r)
 950                         goto done;
 951                 if (cookies)
 952                         cookies[cookie_index] = saveoff;
 953                 ++saveoff;
 954                 ++cookie_index;
 955                 if (cookie_index == ncookies)
 956                         goto done;
 957         }
 958         if (saveoff == 1) {
 959                 if (ip->ino_data.parent_obj_id) {
 960                         r = vop_write_dirent(&error, uio,
 961                                              ip->ino_data.parent_obj_id,
 962                                              DT_DIR, 2, "..");
 963                 } else {
 964                         r = vop_write_dirent(&error, uio,
 965                                              ip->obj_id, DT_DIR, 2, "..");
 966                 }
 967                 if (r)
 968                         goto done;
 969                 if (cookies)
 970                         cookies[cookie_index] = saveoff;
 971                 ++saveoff;
 972                 ++cookie_index;
 973                 if (cookie_index == ncookies)
 974                         goto done;
 975         }
 976
 977         /*
 978          * Key range (begin and end inclusive) to scan.  Directory keys
 979          * directly translate to a 64 bit 'seek' position.
 980          */
 981         hammer_init_cursor_ip(&cursor, ip);
 982         cursor.key_beg.obj_id = ip->obj_id;
 983         cursor.key_beg.create_tid = ip->obj_asof;
 984         cursor.key_beg.delete_tid = 0;
 985         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 986         cursor.key_beg.obj_type = 0;
 987         cursor.key_beg.key = saveoff;
 988
 989         cursor.key_end = cursor.key_beg;
 990         cursor.key_end.key = HAMMER_MAX_KEY;
 991         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
 992
 993         error = hammer_ip_first(&cursor, ip);
 994
 995         while (error == 0) {
 996                 error = hammer_ip_resolve_data(&cursor);
 997                 if (error)
 998                         break;
 999                 rec = cursor.record;
1000                 base = &rec->base.base;
1001                 saveoff = base->key;
1002
1003                 if (base->obj_id != ip->obj_id)
1004                         panic("readdir: bad record at %p", cursor.node);
1005
1006                 r = vop_write_dirent(
1007                              &error, uio, rec->entry.obj_id,
1008                              hammer_get_dtype(rec->entry.base.base.obj_type),
1009                              rec->entry.base.data_len,
1010                              (void *)cursor.data);
1011                 if (r)
1012                         break;
1013                 ++saveoff;
1014                 if (cookies)
1015                         cookies[cookie_index] = base->key;
1016                 ++cookie_index;
1017                 if (cookie_index == ncookies)
1018                         break;
1019                 error = hammer_ip_next(&cursor);
1020         }
1021         hammer_done_cursor(&cursor);
1022
1023 done:
1024         if (ap->a_eofflag)
1025                 *ap->a_eofflag = (error == ENOENT);
1026         uio->uio_offset = saveoff;
1027         if (error && cookie_index == 0) {
1028                 if (error == ENOENT)
1029                         error = 0;
1030                 if (cookies) {
1031                         kfree(cookies, M_TEMP);
1032                         *ap->a_ncookies = 0;
1033                         *ap->a_cookies = NULL;
1034                 }
1035         } else {
1036                 if (error == ENOENT)
1037                         error = 0;
1038                 if (cookies) {
1039                         *ap->a_ncookies = cookie_index;
1040                         *ap->a_cookies = cookies;
1041                 }
1042         }
1043         return(error);
1044 }
1045
1046 /*
1047  * hammer_vop_readlink { vp, uio, cred }
1048  */
1049 static
1050 int
1051 hammer_vop_readlink(struct vop_readlink_args *ap)
1052 {
1053         struct hammer_cursor cursor;
1054         struct hammer_inode *ip;
1055         int error;
1056
1057         ip = VTOI(ap->a_vp);
1058         hammer_init_cursor_ip(&cursor, ip);
1059
1060         /*
1061          * Key range (begin and end inclusive) to scan.  Directory keys
1062          * directly translate to a 64 bit 'seek' position.
1063          */
1064         cursor.key_beg.obj_id = ip->obj_id;
1065         cursor.key_beg.create_tid = ip->obj_asof;
1066         cursor.key_beg.delete_tid = 0;
1067         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1068         cursor.key_beg.obj_type = 0;
1069         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1070
1071         error = hammer_ip_lookup(&cursor, ip);
1072         if (error == 0) {
1073                 error = hammer_ip_resolve_data(&cursor);
1074                 if (error == 0) {
1075                         error = uiomove((char *)cursor.data,
1076                                         cursor.record->generic.base.data_len,
1077                                         ap->a_uio);
1078                 }
1079         }
1080         hammer_done_cursor(&cursor);
1081         return(error);
1082 }
1083
1084 /*
1085  * hammer_vop_nremove { nch, dvp, cred }
1086  */
1087 static
1088 int
1089 hammer_vop_nremove(struct vop_nremove_args *ap)
1090 {
1091         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1092 }
1093
1094 /*
1095  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1096  */
1097 static
1098 int
1099 hammer_vop_nrename(struct vop_nrename_args *ap)
1100 {
1101         struct hammer_transaction trans;
1102         struct namecache *fncp;
1103         struct namecache *tncp;
1104         struct hammer_inode *fdip;
1105         struct hammer_inode *tdip;
1106         struct hammer_inode *ip;
1107         struct hammer_cursor cursor;
1108         union hammer_record_ondisk *rec;
1109         int64_t namekey;
1110         int error;
1111
1112         fdip = VTOI(ap->a_fdvp);
1113         tdip = VTOI(ap->a_tdvp);
1114         fncp = ap->a_fnch->ncp;
1115         tncp = ap->a_tnch->ncp;
1116         ip = VTOI(fncp->nc_vp);
1117         KKASSERT(ip != NULL);
1118
1119         if (fdip->flags & HAMMER_INODE_RO)
1120                 return (EROFS);
1121         if (tdip->flags & HAMMER_INODE_RO)
1122                 return (EROFS);
1123         if (ip->flags & HAMMER_INODE_RO)
1124                 return (EROFS);
1125
1126         hammer_start_transaction(&trans, fdip->hmp);
1127
1128         /*
1129          * Remove tncp from the target directory and then link ip as
1130          * tncp. XXX pass trans to dounlink
1131          */
1132         error = hammer_dounlink(ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1133         if (error == 0 || error == ENOENT)
1134                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1135         if (error)
1136                 goto failed; /* XXX */
1137
1138         /*
1139          * Locate the record in the originating directory and remove it.
1140          *
1141          * Calculate the namekey and setup the key range for the scan.  This
1142          * works kinda like a chained hash table where the lower 32 bits
1143          * of the namekey synthesize the chain.
1144          *
1145          * The key range is inclusive of both key_beg and key_end.
1146          */
1147         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1148
1149         hammer_init_cursor_ip(&cursor, fdip);
1150         cursor.key_beg.obj_id = fdip->obj_id;
1151         cursor.key_beg.key = namekey;
1152         cursor.key_beg.create_tid = fdip->obj_asof;
1153         cursor.key_beg.delete_tid = 0;
1154         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1155         cursor.key_beg.obj_type = 0;
1156
1157         cursor.key_end = cursor.key_beg;
1158         cursor.key_end.key |= 0xFFFFFFFFULL;
1159         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1160
1161         /*
1162          * Scan all matching records (the chain), locate the one matching
1163          * the requested path component.
1164          *
1165          * The hammer_ip_*() functions merge in-memory records with on-disk
1166          * records for the purposes of the search.
1167          */
1168         error = hammer_ip_first(&cursor, fdip);
1169         while (error == 0) {
1170                 if (hammer_ip_resolve_data(&cursor) != 0)
1171                         break;
1172                 rec = cursor.record;
1173                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1174                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1175                         break;
1176                 }
1177                 error = hammer_ip_next(&cursor);
1178         }
1179
1180         /*
1181          * If all is ok we have to get the inode so we can adjust nlinks.
1182          */
1183         if (error)
1184                 goto done;
1185         error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1186
1187         if (error == 0) {
1188                 cache_rename(ap->a_fnch, ap->a_tnch);
1189                 cache_setvp(ap->a_tnch, ip->vp);
1190         }
1191 done:
1192         hammer_done_cursor(&cursor);
1193 failed:
1194         if (error == 0) {
1195                 hammer_commit_transaction(&trans);
1196         } else {
1197                 hammer_abort_transaction(&trans);
1198         }
1199         return (error);
1200 }
1201
1202 /*
1203  * hammer_vop_nrmdir { nch, dvp, cred }
1204  */
1205 static
1206 int
1207 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1208 {
1209         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1210 }
1211
1212 /*
1213  * hammer_vop_setattr { vp, vap, cred }
1214  */
1215 static
1216 int
1217 hammer_vop_setattr(struct vop_setattr_args *ap)
1218 {
1219         struct hammer_transaction trans;
1220         struct hammer_cursor *spike = NULL;
1221         struct vattr *vap;
1222         struct hammer_inode *ip;
1223         int modflags;
1224         int error;
1225         int64_t aligned_size;
1226         u_int32_t flags;
1227         uuid_t uuid;
1228
1229         vap = ap->a_vap;
1230         ip = ap->a_vp->v_data;
1231         modflags = 0;
1232
1233         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1234                 return(EROFS);
1235         if (ip->flags & HAMMER_INODE_RO)
1236                 return (EROFS);
1237
1238         hammer_start_transaction(&trans, ip->hmp);
1239         error = 0;
1240
1241         if (vap->va_flags != VNOVAL) {
1242                 flags = ip->ino_data.uflags;
1243                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1244                                          hammer_to_unix_xid(&ip->ino_data.uid),
1245                                          ap->a_cred);
1246                 if (error == 0) {
1247                         if (ip->ino_data.uflags != flags) {
1248                                 ip->ino_data.uflags = flags;
1249                                 modflags |= HAMMER_INODE_DDIRTY;
1250                         }
1251                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1252                                 error = 0;
1253                                 goto done;
1254                         }
1255                 }
1256                 goto done;
1257         }
1258         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1259                 error = EPERM;
1260                 goto done;
1261         }
1262         if (vap->va_uid != (uid_t)VNOVAL) {
1263                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1264                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1265                         ip->ino_data.uid = uuid;
1266                         modflags |= HAMMER_INODE_DDIRTY;
1267                 }
1268         }
1269         if (vap->va_gid != (uid_t)VNOVAL) {
1270                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1271                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1272                         ip->ino_data.gid = uuid;
1273                         modflags |= HAMMER_INODE_DDIRTY;
1274                 }
1275         }
1276         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1277                 switch(ap->a_vp->v_type) {
1278                 case VREG:
1279                         if (vap->va_size < ip->ino_rec.ino_size) {
1280                                 vtruncbuf(ap->a_vp, vap->va_size,
1281                                           HAMMER_BUFSIZE);
1282                         } else if (vap->va_size > ip->ino_rec.ino_size) {
1283                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1284                         }
1285                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1286                                         ~(int64_t)HAMMER_BUFMASK;
1287                         error = hammer_ip_delete_range(&trans, ip,
1288                                                     aligned_size,
1289                                                     0x7FFFFFFFFFFFFFFFLL,
1290                                                     &spike);
1291                         ip->ino_rec.ino_size = vap->va_size;
1292                         modflags |= HAMMER_INODE_RDIRTY;
1293                         break;
1294                 case VDATABASE:
1295                         error = hammer_ip_delete_range(&trans, ip,
1296                                                     vap->va_size,
1297                                                     0x7FFFFFFFFFFFFFFFLL,
1298                                                     &spike);
1299                         ip->ino_rec.ino_size = vap->va_size;
1300                         modflags |= HAMMER_INODE_RDIRTY;
1301                         break;
1302                 default:
1303                         error = EINVAL;
1304                         goto done;
1305                 }
1306                 if (error == ENOSPC) {
1307                         error = hammer_spike(&spike);
1308                         if (error == 0)
1309                                 continue;
1310                 }
1311                 KKASSERT(spike == NULL);
1312                 break;
1313         }
1314         if (vap->va_atime.tv_sec != VNOVAL) {
1315                 ip->ino_rec.ino_atime =
1316                         hammer_timespec_to_transid(&vap->va_atime);
1317                 modflags |= HAMMER_INODE_ITIMES;
1318         }
1319         if (vap->va_mtime.tv_sec != VNOVAL) {
1320                 ip->ino_rec.ino_mtime =
1321                         hammer_timespec_to_transid(&vap->va_mtime);
1322                 modflags |= HAMMER_INODE_ITIMES;
1323         }
1324         if (vap->va_mode != (mode_t)VNOVAL) {
1325                 if (ip->ino_data.mode != vap->va_mode) {
1326                         ip->ino_data.mode = vap->va_mode;
1327                         modflags |= HAMMER_INODE_DDIRTY;
1328                 }
1329         }
1330 done:
1331         if (error) {
1332                 hammer_abort_transaction(&trans);
1333         } else {
1334                 hammer_modify_inode(&trans, ip, modflags);
1335                 hammer_commit_transaction(&trans);
1336         }
1337         return (error);
1338 }
1339
1340 /*
1341  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1342  */
1343 static
1344 int
1345 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1346 {
1347         struct hammer_transaction trans;
1348         struct hammer_inode *dip;
1349         struct hammer_inode *nip;
1350         struct nchandle *nch;
1351         hammer_record_t record;
1352         int error;
1353         int bytes;
1354
1355         ap->a_vap->va_type = VLNK;
1356
1357         nch = ap->a_nch;
1358         dip = VTOI(ap->a_dvp);
1359
1360         if (dip->flags & HAMMER_INODE_RO)
1361                 return (EROFS);
1362
1363         /*
1364          * Create a transaction to cover the operations we perform.
1365          */
1366         hammer_start_transaction(&trans, dip->hmp);
1367
1368         /*
1369          * Create a new filesystem object of the requested type.  The
1370          * returned inode will be referenced but not locked.
1371          */
1372
1373         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1374         if (error) {
1375                 hammer_abort_transaction(&trans);
1376                 *ap->a_vpp = NULL;
1377                 return (error);
1378         }
1379
1380         /*
1381          * Add the new filesystem object to the directory.  This will also
1382          * bump the inode's link count.
1383          */
1384         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1385
1386         /*
1387          * Add a record representing the symlink.  symlink stores the link
1388          * as pure data, not a string, and is no \0 terminated.
1389          */
1390         if (error == 0) {
1391                 record = hammer_alloc_mem_record(nip);
1392                 bytes = strlen(ap->a_target);
1393
1394                 record->rec.generic.base.base.key = HAMMER_FIXKEY_SYMLINK;
1395                 record->rec.generic.base.base.rec_type = HAMMER_RECTYPE_FIX;
1396                 record->rec.generic.base.data_len = bytes;
1397                 if (bytes <= sizeof(record->rec.generic.filler)) {
1398                         record->data = (void *)record->rec.generic.filler;
1399                         bcopy(ap->a_target, record->data, bytes);
1400                 } else {
1401                         record->data = (void *)ap->a_target;
1402                         /* will be reallocated by routine below */
1403                 }
1404                 error = hammer_ip_add_record(&trans, record);
1405         }
1406
1407         /*
1408          * Finish up.
1409          */
1410         if (error) {
1411                 hammer_rel_inode(nip, 0);
1412                 hammer_abort_transaction(&trans);
1413                 *ap->a_vpp = NULL;
1414         } else {
1415                 hammer_commit_transaction(&trans);
1416                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1417                 hammer_rel_inode(nip, 0);
1418                 if (error == 0) {
1419                         cache_setunresolved(ap->a_nch);
1420                         cache_setvp(ap->a_nch, *ap->a_vpp);
1421                 }
1422         }
1423         return (error);
1424 }
1425
1426 /*
1427  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1428  */
1429 static
1430 int
1431 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1432 {
1433         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1434 }
1435
1436 /*
1437  * hammer_vop_strategy { vp, bio }
1438  *
1439  * Strategy call, used for regular file read & write only.  Note that the
1440  * bp may represent a cluster.
1441  *
1442  * To simplify operation and allow better optimizations in the future,
1443  * this code does not make any assumptions with regards to buffer alignment
1444  * or size.
1445  */
1446 static
1447 int
1448 hammer_vop_strategy(struct vop_strategy_args *ap)
1449 {
1450         struct buf *bp;
1451         int error;
1452
1453         bp = ap->a_bio->bio_buf;
1454
1455         switch(bp->b_cmd) {
1456         case BUF_CMD_READ:
1457                 error = hammer_vop_strategy_read(ap);
1458                 break;
1459         case BUF_CMD_WRITE:
1460                 error = hammer_vop_strategy_write(ap);
1461                 break;
1462         default:
1463                 error = EINVAL;
1464                 break;
1465         }
1466         bp->b_error = error;
1467         if (error)
1468                 bp->b_flags |= B_ERROR;
1469         biodone(ap->a_bio);
1470         return (error);
1471 }
1472
1473 /*
1474  * Read from a regular file.  Iterate the related records and fill in the
1475  * BIO/BUF.  Gaps are zero-filled.
1476  *
1477  * The support code in hammer_object.c should be used to deal with mixed
1478  * in-memory and on-disk records.
1479  *
1480  * XXX atime update
1481  */
1482 static
1483 int
1484 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1485 {
1486         struct hammer_inode *ip = ap->a_vp->v_data;
1487         struct hammer_cursor cursor;
1488         hammer_record_ondisk_t rec;
1489         hammer_base_elm_t base;
1490         struct bio *bio;
1491         struct buf *bp;
1492         int64_t rec_offset;
1493         int64_t ran_end;
1494         int64_t tmp64;
1495         int error;
1496         int boff;
1497         int roff;
1498         int n;
1499
1500         bio = ap->a_bio;
1501         bp = bio->bio_buf;
1502
1503         hammer_init_cursor_ip(&cursor, ip);
1504
1505         /*
1506          * Key range (begin and end inclusive) to scan.  Note that the key's
1507          * stored in the actual records represent BASE+LEN, not BASE.  The
1508          * first record containing bio_offset will have a key > bio_offset.
1509          */
1510         cursor.key_beg.obj_id = ip->obj_id;
1511         cursor.key_beg.create_tid = ip->obj_asof;
1512         cursor.key_beg.delete_tid = 0;
1513         cursor.key_beg.obj_type = 0;
1514         cursor.key_beg.key = bio->bio_offset + 1;
1515
1516         cursor.key_end = cursor.key_beg;
1517         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1518                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1519                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1520                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1521         } else {
1522                 ran_end = bio->bio_offset + bp->b_bufsize;
1523                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1524                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1525                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1526                 if (tmp64 < ran_end)
1527                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1528                 else
1529                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1530         }
1531         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1532
1533         error = hammer_ip_first(&cursor, ip);
1534         boff = 0;
1535
1536         while (error == 0) {
1537                 error = hammer_ip_resolve_data(&cursor);
1538                 if (error)
1539                         break;
1540                 rec = cursor.record;
1541                 base = &rec->base.base;
1542
1543                 rec_offset = base->key - rec->data.base.data_len;
1544
1545                 /*
1546                  * Calculate the gap, if any, and zero-fill it.
1547                  */
1548                 n = (int)(rec_offset - (bio->bio_offset + boff));
1549                 if (n > 0) {
1550                         if (n > bp->b_bufsize - boff)
1551                                 n = bp->b_bufsize - boff;
1552                         kprintf("zfill %d bytes\n", n);
1553                         bzero((char *)bp->b_data + boff, n);
1554                         boff += n;
1555                         n = 0;
1556                 }
1557
1558                 /*
1559                  * Calculate the data offset in the record and the number
1560                  * of bytes we can copy.
1561                  *
1562                  * Note there is a degenerate case here where boff may
1563                  * already be at bp->b_bufsize.
1564                  */
1565                 roff = -n;
1566                 n = rec->data.base.data_len - roff;
1567                 KKASSERT(n > 0);
1568                 if (n > bp->b_bufsize - boff)
1569                         n = bp->b_bufsize - boff;
1570                 bcopy((char *)cursor.data + roff, (char *)bp->b_data + boff, n);
1571                 boff += n;
1572                 if (boff == bp->b_bufsize)
1573                         break;
1574                 error = hammer_ip_next(&cursor);
1575         }
1576         hammer_done_cursor(&cursor);
1577
1578         /*
1579          * There may have been a gap after the last record
1580          */
1581         if (error == ENOENT)
1582                 error = 0;
1583         if (error == 0 && boff != bp->b_bufsize) {
1584                 KKASSERT(boff < bp->b_bufsize);
1585                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1586                 /* boff = bp->b_bufsize; */
1587         }
1588         bp->b_resid = 0;
1589         return(error);
1590 }
1591
1592 /*
1593  * Write to a regular file.  Iterate the related records and mark for
1594  * deletion.  If existing edge records (left and right side) overlap our
1595  * write they have to be marked deleted and new records created, usually
1596  * referencing a portion of the original data.  Then add a record to
1597  * represent the buffer.
1598  *
1599  * The support code in hammer_object.c should be used to deal with mixed
1600  * in-memory and on-disk records.
1601  */
1602 static
1603 int
1604 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1605 {
1606         struct hammer_transaction trans;
1607         struct hammer_cursor *spike = NULL;
1608         hammer_inode_t ip;
1609         struct bio *bio;
1610         struct buf *bp;
1611         int error;
1612
1613         bio = ap->a_bio;
1614         bp = bio->bio_buf;
1615         ip = ap->a_vp->v_data;
1616
1617         if (ip->flags & HAMMER_INODE_RO)
1618                 return (EROFS);
1619
1620         hammer_start_transaction(&trans, ip->hmp);
1621
1622 retry:
1623         /*
1624          * Delete any records overlapping our range.  This function will
1625          * (eventually) properly truncate partial overlaps.
1626          */
1627         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1628                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1629                                                bio->bio_offset, &spike);
1630         } else {
1631                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1632                                                bio->bio_offset +
1633                                                 bp->b_bufsize - 1,
1634                                                &spike);
1635         }
1636
1637         /*
1638          * Add a single record to cover the write
1639          */
1640         if (error == 0) {
1641                 error = hammer_ip_sync_data(&trans, ip, bio->bio_offset,
1642                                             bp->b_data, bp->b_bufsize,
1643                                             &spike);
1644         }
1645
1646         /*
1647          * If we ran out of space the spike structure will be filled in
1648          * and we must call hammer_spike with it, then retry.
1649          */
1650         if (error == ENOSPC) {
1651                 error = hammer_spike(&spike);
1652                 if (error == 0)
1653                         goto retry;
1654         }
1655         KKASSERT(spike == NULL);
1656
1657         /*
1658          * If an error occured abort the transaction
1659          */
1660         if (error) {
1661                 /* XXX undo deletion */
1662                 hammer_abort_transaction(&trans);
1663                 bp->b_resid = bp->b_bufsize;
1664         } else {
1665                 hammer_commit_transaction(&trans);
1666                 bp->b_resid = 0;
1667         }
1668         return(error);
1669 }
1670
1671 /*
1672  * dounlink - disconnect a directory entry
1673  *
1674  * XXX whiteout support not really in yet
1675  */
1676 static int
1677 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1678                 int flags)
1679 {
1680         struct hammer_transaction trans;
1681         struct namecache *ncp;
1682         hammer_inode_t dip;
1683         hammer_inode_t ip;
1684         hammer_record_ondisk_t rec;
1685         struct hammer_cursor cursor;
1686         int64_t namekey;
1687         int error;
1688
1689         /*
1690          * Calculate the namekey and setup the key range for the scan.  This
1691          * works kinda like a chained hash table where the lower 32 bits
1692          * of the namekey synthesize the chain.
1693          *
1694          * The key range is inclusive of both key_beg and key_end.
1695          */
1696         dip = VTOI(dvp);
1697         ncp = nch->ncp;
1698
1699         if (dip->flags & HAMMER_INODE_RO)
1700                 return (EROFS);
1701
1702         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1703
1704         hammer_init_cursor_ip(&cursor, dip);
1705         cursor.key_beg.obj_id = dip->obj_id;
1706         cursor.key_beg.key = namekey;
1707         cursor.key_beg.create_tid = dip->obj_asof;
1708         cursor.key_beg.delete_tid = 0;
1709         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1710         cursor.key_beg.obj_type = 0;
1711
1712         cursor.key_end = cursor.key_beg;
1713         cursor.key_end.key |= 0xFFFFFFFFULL;
1714         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1715
1716         hammer_start_transaction(&trans, dip->hmp);
1717
1718         /*
1719          * Scan all matching records (the chain), locate the one matching
1720          * the requested path component.  info->last_error contains the
1721          * error code on search termination and could be 0, ENOENT, or
1722          * something else.
1723          *
1724          * The hammer_ip_*() functions merge in-memory records with on-disk
1725          * records for the purposes of the search.
1726          */
1727         error = hammer_ip_first(&cursor, dip);
1728         while (error == 0) {
1729                 error = hammer_ip_resolve_data(&cursor);
1730                 if (error)
1731                         break;
1732                 rec = cursor.record;
1733                 if (ncp->nc_nlen == rec->entry.base.data_len &&
1734                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1735                         break;
1736                 }
1737                 error = hammer_ip_next(&cursor);
1738         }
1739
1740         /*
1741          * If all is ok we have to get the inode so we can adjust nlinks.
1742          *
1743          * If the target is a directory, it must be empty.
1744          */
1745         if (error == 0) {
1746                 ip = hammer_get_inode(dip->hmp, rec->entry.obj_id,
1747                                       dip->hmp->asof, 0, &error);
1748                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
1749                                   HAMMER_OBJTYPE_DIRECTORY) {
1750                         error = hammer_ip_check_directory_empty(&trans, ip);
1751                 }
1752                 if (error == 0)
1753                         error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1754                 if (error == 0) {
1755                         cache_setunresolved(nch);
1756                         cache_setvp(nch, NULL);
1757                         /* XXX locking */
1758                         if (ip->vp)
1759                                 cache_inval_vp(ip->vp, CINV_DESTROY);
1760                 }
1761                 hammer_rel_inode(ip, 0);
1762         }
1763
1764         if (error == 0)
1765                 hammer_commit_transaction(&trans);
1766         else
1767                 hammer_abort_transaction(&trans);
1768         hammer_done_cursor(&cursor);
1769         return (error);
1770 }
1771
1772 /************************************************************************
1773  *                          FIFO AND SPECFS OPS                         *
1774  ************************************************************************
1775  *
1776  */
1777
1778 static int
1779 hammer_vop_fifoclose (struct vop_close_args *ap)
1780 {
1781         /* XXX update itimes */
1782         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1783 }
1784
1785 static int
1786 hammer_vop_fiforead (struct vop_read_args *ap)
1787 {
1788         int error;
1789
1790         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1791         /* XXX update access time */
1792         return (error);
1793 }
1794
1795 static int
1796 hammer_vop_fifowrite (struct vop_write_args *ap)
1797 {
1798         int error;
1799
1800         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1801         /* XXX update access time */
1802         return (error);
1803 }
1804
1805 static int
1806 hammer_vop_specclose (struct vop_close_args *ap)
1807 {
1808         /* XXX update itimes */
1809         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1810 }
1811
1812 static int
1813 hammer_vop_specread (struct vop_read_args *ap)
1814 {
1815         /* XXX update access time */
1816         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1817 }
1818
1819 static int
1820 hammer_vop_specwrite (struct vop_write_args *ap)
1821 {
1822         /* XXX update last change time */
1823         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1824 }
1825