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 1.24 2008/01/25 10:36:04 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                 int fixsize = 0;
 288
 289                 offset = uio->uio_offset & HAMMER_BUFMASK;
 290                 n = HAMMER_BUFSIZE - offset;
 291                 if (n > uio->uio_resid)
 292                         n = uio->uio_resid;
 293                 if (uio->uio_offset + n > ip->ino_rec.ino_size) {
 294                         vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
 295                         fixsize = 1;
 296                 }
 297
 298                 if (uio->uio_segflg == UIO_NOCOPY) {
 299                         /*
 300                          * Issuing a write with the same data backing the
 301                          * buffer.  Instantiate the buffer to collect the
 302                          * backing vm pages, then read-in any missing bits.
 303                          *
 304                          * This case is used by vop_stdputpages().
 305                          */
 306                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 307                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 308                         if ((bp->b_flags & B_CACHE) == 0) {
 309                                 bqrelse(bp);
 310                                 error = bread(ap->a_vp,
 311                                               uio->uio_offset - offset,
 312                                               HAMMER_BUFSIZE, &bp);
 313                         }
 314                 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
 315                         /*
 316                          * entirely overwrite the buffer
 317                          */
 318                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 319                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 320                 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
 321                         /*
 322                          * XXX
 323                          */
 324                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 325                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 326                         vfs_bio_clrbuf(bp);
 327                 } else {
 328                         /*
 329                          * Partial overwrite, read in any missing bits then
 330                          * replace the portion being written.
 331                          */
 332                         error = bread(ap->a_vp, uio->uio_offset - offset,
 333                                       HAMMER_BUFSIZE, &bp);
 334                         if (error == 0)
 335                                 bheavy(bp);
 336                 }
 337                 if (error == 0)
 338                         error = uiomove((char *)bp->b_data + offset, n, uio);
 339
 340                 /*
 341                  * If we screwed up we have to undo any VM size changes we
 342                  * made.
 343                  */
 344                 if (error) {
 345                         brelse(bp);
 346                         if (fixsize) {
 347                                 vtruncbuf(ap->a_vp, ip->ino_rec.ino_size,
 348                                           HAMMER_BUFSIZE);
 349                         }
 350                         break;
 351                 }
 352                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 353                 if (ip->ino_rec.ino_size < uio->uio_offset) {
 354                         ip->ino_rec.ino_size = uio->uio_offset;
 355                         flags = HAMMER_INODE_RDIRTY;
 356                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
 357                 } else {
 358                         flags = 0;
 359                 }
 360                 ip->ino_rec.ino_mtime = trans.tid;
 361                 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
 362                 hammer_modify_inode(&trans, ip, flags);
 363                 if (ap->a_ioflag & IO_SYNC) {
 364                         bwrite(bp);
 365                 } else if (ap->a_ioflag & IO_DIRECT) {
 366                         bawrite(bp);
 367                 } else {
 368                         bdwrite(bp);
 369                 }
 370         }
 371         if (error)
 372                 hammer_abort_transaction(&trans);
 373         else
 374                 hammer_commit_transaction(&trans);
 375         return (error);
 376 }
 377
 378 /*
 379  * hammer_vop_access { vp, mode, cred }
 380  */
 381 static
 382 int
 383 hammer_vop_access(struct vop_access_args *ap)
 384 {
 385         struct hammer_inode *ip = VTOI(ap->a_vp);
 386         uid_t uid;
 387         gid_t gid;
 388         int error;
 389
 390         uid = hammer_to_unix_xid(&ip->ino_data.uid);
 391         gid = hammer_to_unix_xid(&ip->ino_data.gid);
 392
 393         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
 394                                   ip->ino_data.uflags);
 395         return (error);
 396 }
 397
 398 /*
 399  * hammer_vop_advlock { vp, id, op, fl, flags }
 400  */
 401 static
 402 int
 403 hammer_vop_advlock(struct vop_advlock_args *ap)
 404 {
 405         struct hammer_inode *ip = VTOI(ap->a_vp);
 406
 407         return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
 408 }
 409
 410 /*
 411  * hammer_vop_close { vp, fflag }
 412  */
 413 static
 414 int
 415 hammer_vop_close(struct vop_close_args *ap)
 416 {
 417         return (vop_stdclose(ap));
 418 }
 419
 420 /*
 421  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
 422  *
 423  * The operating system has already ensured that the directory entry
 424  * does not exist and done all appropriate namespace locking.
 425  */
 426 static
 427 int
 428 hammer_vop_ncreate(struct vop_ncreate_args *ap)
 429 {
 430         struct hammer_transaction trans;
 431         struct hammer_inode *dip;
 432         struct hammer_inode *nip;
 433         struct nchandle *nch;
 434         int error;
 435
 436         nch = ap->a_nch;
 437         dip = VTOI(ap->a_dvp);
 438
 439         if (dip->flags & HAMMER_INODE_RO)
 440                 return (EROFS);
 441
 442         /*
 443          * Create a transaction to cover the operations we perform.
 444          */
 445         hammer_start_transaction(&trans, dip->hmp);
 446
 447         /*
 448          * Create a new filesystem object of the requested type.  The
 449          * returned inode will be referenced but not locked.
 450          */
 451
 452         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 453         if (error)
 454                 kprintf("hammer_create_inode error %d\n", error);
 455         if (error) {
 456                 hammer_abort_transaction(&trans);
 457                 *ap->a_vpp = NULL;
 458                 return (error);
 459         }
 460
 461         /*
 462          * Add the new filesystem object to the directory.  This will also
 463          * bump the inode's link count.
 464          */
 465         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 466         if (error)
 467                 kprintf("hammer_ip_add_directory error %d\n", error);
 468
 469         /*
 470          * Finish up.
 471          */
 472         if (error) {
 473                 hammer_rel_inode(nip, 0);
 474                 hammer_abort_transaction(&trans);
 475                 *ap->a_vpp = NULL;
 476         } else {
 477                 hammer_commit_transaction(&trans);
 478                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 479                 hammer_rel_inode(nip, 0);
 480                 if (error == 0) {
 481                         cache_setunresolved(ap->a_nch);
 482                         cache_setvp(ap->a_nch, *ap->a_vpp);
 483                 }
 484         }
 485         return (error);
 486 }
 487
 488 /*
 489  * hammer_vop_getattr { vp, vap }
 490  */
 491 static
 492 int
 493 hammer_vop_getattr(struct vop_getattr_args *ap)
 494 {
 495         struct hammer_inode *ip = VTOI(ap->a_vp);
 496         struct vattr *vap = ap->a_vap;
 497
 498 #if 0
 499         if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
 500             (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
 501             ip->obj_asof == XXX
 502         ) {
 503                 /* LAZYMOD XXX */
 504         }
 505         hammer_itimes(ap->a_vp);
 506 #endif
 507
 508         vap->va_fsid = ip->hmp->fsid_udev;
 509         vap->va_fileid = ip->ino_rec.base.base.obj_id;
 510         vap->va_mode = ip->ino_data.mode;
 511         vap->va_nlink = ip->ino_rec.ino_nlinks;
 512         vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
 513         vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
 514         vap->va_rmajor = 0;
 515         vap->va_rminor = 0;
 516         vap->va_size = ip->ino_rec.ino_size;
 517         hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
 518         hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
 519         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
 520         vap->va_flags = ip->ino_data.uflags;
 521         vap->va_gen = 1;        /* hammer inums are unique for all time */
 522         vap->va_blocksize = 32768; /* XXX - extract from root volume */
 523         vap->va_bytes = ip->ino_rec.ino_size;
 524         vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
 525         vap->va_filerev = 0;    /* XXX */
 526         /* mtime uniquely identifies any adjustments made to the file */
 527         vap->va_fsmid = ip->ino_rec.ino_mtime;
 528         vap->va_uid_uuid = ip->ino_data.uid;
 529         vap->va_gid_uuid = ip->ino_data.gid;
 530         vap->va_fsid_uuid = ip->hmp->fsid;
 531         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 532                           VA_FSID_UUID_VALID;
 533
 534         switch (ip->ino_rec.base.base.obj_type) {
 535         case HAMMER_OBJTYPE_CDEV:
 536         case HAMMER_OBJTYPE_BDEV:
 537                 vap->va_rmajor = ip->ino_data.rmajor;
 538                 vap->va_rminor = ip->ino_data.rminor;
 539                 break;
 540         default:
 541                 break;
 542         }
 543
 544         return(0);
 545 }
 546
 547 /*
 548  * hammer_vop_nresolve { nch, dvp, cred }
 549  *
 550  * Locate the requested directory entry.
 551  */
 552 static
 553 int
 554 hammer_vop_nresolve(struct vop_nresolve_args *ap)
 555 {
 556         struct namecache *ncp;
 557         hammer_inode_t dip;
 558         hammer_inode_t ip;
 559         hammer_tid_t asof;
 560         struct hammer_cursor cursor;
 561         union hammer_record_ondisk *rec;
 562         struct vnode *vp;
 563         int64_t namekey;
 564         int error;
 565         int i;
 566         int nlen;
 567         int flags;
 568         u_int64_t obj_id;
 569
 570         /*
 571          * Misc initialization, plus handle as-of name extensions.  Look for
 572          * the '@@' extension.  Note that as-of files and directories cannot
 573          * be modified.
 574          */
 575         dip = VTOI(ap->a_dvp);
 576         ncp = ap->a_nch->ncp;
 577         asof = dip->obj_asof;
 578         nlen = ncp->nc_nlen;
 579         flags = dip->flags;
 580
 581         for (i = 0; i < nlen; ++i) {
 582                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
 583                         asof = hammer_str_to_tid(ncp->nc_name + i + 2);
 584                         kprintf("ASOF %016llx\n", asof);
 585                         flags |= HAMMER_INODE_RO;
 586                         break;
 587                 }
 588         }
 589         nlen = i;
 590
 591         /*
 592          * If there is no path component the time extension is relative to
 593          * dip.
 594          */
 595         if (nlen == 0) {
 596                 ip = hammer_get_inode(dip->hmp, &dip->cache[1], dip->obj_id,
 597                                       asof, flags, &error);
 598                 if (error == 0) {
 599                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 600                         hammer_rel_inode(ip, 0);
 601                 } else {
 602                         vp = NULL;
 603                 }
 604                 if (error == 0) {
 605                         vn_unlock(vp);
 606                         cache_setvp(ap->a_nch, vp);
 607                         vrele(vp);
 608                 }
 609                 return(error);
 610         }
 611
 612         /*
 613          * Calculate the namekey and setup the key range for the scan.  This
 614          * works kinda like a chained hash table where the lower 32 bits
 615          * of the namekey synthesize the chain.
 616          *
 617          * The key range is inclusive of both key_beg and key_end.
 618          */
 619         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
 620
 621         hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
 622         cursor.key_beg.obj_id = dip->obj_id;
 623         cursor.key_beg.key = namekey;
 624         cursor.key_beg.create_tid = 0;
 625         cursor.key_beg.delete_tid = 0;
 626         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 627         cursor.key_beg.obj_type = 0;
 628
 629         cursor.key_end = cursor.key_beg;
 630         cursor.key_end.key |= 0xFFFFFFFFULL;
 631         cursor.asof = asof;
 632         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
 633
 634         /*
 635          * Scan all matching records (the chain), locate the one matching
 636          * the requested path component.
 637          *
 638          * The hammer_ip_*() functions merge in-memory records with on-disk
 639          * records for the purposes of the search.
 640          */
 641         error = hammer_ip_first(&cursor, dip);
 642         rec = NULL;
 643         obj_id = 0;
 644
 645         while (error == 0) {
 646                 error = hammer_ip_resolve_data(&cursor);
 647                 if (error)
 648                         break;
 649                 rec = cursor.record;
 650                 if (nlen == rec->entry.base.data_len &&
 651                     bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
 652                         obj_id = rec->entry.obj_id;
 653                         break;
 654                 }
 655                 error = hammer_ip_next(&cursor);
 656         }
 657         hammer_done_cursor(&cursor);
 658         if (error == 0) {
 659                 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
 660                                       obj_id, asof, flags, &error);
 661                 if (error == 0) {
 662                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 663                         hammer_rel_inode(ip, 0);
 664                 } else {
 665                         vp = NULL;
 666                 }
 667                 if (error == 0) {
 668                         vn_unlock(vp);
 669                         cache_setvp(ap->a_nch, vp);
 670                         vrele(vp);
 671                 }
 672         } else if (error == ENOENT) {
 673                 cache_setvp(ap->a_nch, NULL);
 674         }
 675         return (error);
 676 }
 677
 678 /*
 679  * hammer_vop_nlookupdotdot { dvp, vpp, cred }
 680  *
 681  * Locate the parent directory of a directory vnode.
 682  *
 683  * dvp is referenced but not locked.  *vpp must be returned referenced and
 684  * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
 685  * at the root, instead it could indicate that the directory we were in was
 686  * removed.
 687  */
 688 static
 689 int
 690 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
 691 {
 692         struct hammer_inode *dip;
 693         struct hammer_inode *ip;
 694         u_int64_t parent_obj_id;
 695         int error;
 696
 697         dip = VTOI(ap->a_dvp);
 698         if ((parent_obj_id = dip->ino_data.parent_obj_id) == 0) {
 699                 *ap->a_vpp = NULL;
 700                 return ENOENT;
 701         }
 702
 703         ip = hammer_get_inode(dip->hmp, &dip->cache[1], parent_obj_id,
 704                               dip->obj_asof, dip->flags, &error);
 705         if (ip == NULL) {
 706                 *ap->a_vpp = NULL;
 707                 return(error);
 708         }
 709         error = hammer_get_vnode(ip, LK_EXCLUSIVE, ap->a_vpp);
 710         hammer_rel_inode(ip, 0);
 711         return (error);
 712 }
 713
 714 /*
 715  * hammer_vop_nlink { nch, dvp, vp, cred }
 716  */
 717 static
 718 int
 719 hammer_vop_nlink(struct vop_nlink_args *ap)
 720 {
 721         struct hammer_transaction trans;
 722         struct hammer_inode *dip;
 723         struct hammer_inode *ip;
 724         struct nchandle *nch;
 725         int error;
 726
 727         nch = ap->a_nch;
 728         dip = VTOI(ap->a_dvp);
 729         ip = VTOI(ap->a_vp);
 730
 731         if (dip->flags & HAMMER_INODE_RO)
 732                 return (EROFS);
 733         if (ip->flags & HAMMER_INODE_RO)
 734                 return (EROFS);
 735
 736         /*
 737          * Create a transaction to cover the operations we perform.
 738          */
 739         hammer_start_transaction(&trans, dip->hmp);
 740
 741         /*
 742          * Add the filesystem object to the directory.  Note that neither
 743          * dip nor ip are referenced or locked, but their vnodes are
 744          * referenced.  This function will bump the inode's link count.
 745          */
 746         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 747
 748         /*
 749          * Finish up.
 750          */
 751         if (error) {
 752                 hammer_abort_transaction(&trans);
 753         } else {
 754                 cache_setunresolved(nch);
 755                 cache_setvp(nch, ap->a_vp);
 756                 hammer_commit_transaction(&trans);
 757         }
 758         return (error);
 759 }
 760
 761 /*
 762  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 763  *
 764  * The operating system has already ensured that the directory entry
 765  * does not exist and done all appropriate namespace locking.
 766  */
 767 static
 768 int
 769 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 770 {
 771         struct hammer_transaction trans;
 772         struct hammer_inode *dip;
 773         struct hammer_inode *nip;
 774         struct nchandle *nch;
 775         int error;
 776
 777         nch = ap->a_nch;
 778         dip = VTOI(ap->a_dvp);
 779
 780         if (dip->flags & HAMMER_INODE_RO)
 781                 return (EROFS);
 782
 783         /*
 784          * Create a transaction to cover the operations we perform.
 785          */
 786         hammer_start_transaction(&trans, dip->hmp);
 787
 788         /*
 789          * Create a new filesystem object of the requested type.  The
 790          * returned inode will be referenced but not locked.
 791          */
 792         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 793         if (error)
 794                 kprintf("hammer_mkdir error %d\n", error);
 795         if (error) {
 796                 hammer_abort_transaction(&trans);
 797                 *ap->a_vpp = NULL;
 798                 return (error);
 799         }
 800
 801         /*
 802          * Add the new filesystem object to the directory.  This will also
 803          * bump the inode's link count.
 804          */
 805         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 806         if (error)
 807                 kprintf("hammer_mkdir (add) error %d\n", error);
 808
 809         /*
 810          * Finish up.
 811          */
 812         if (error) {
 813                 hammer_rel_inode(nip, 0);
 814                 hammer_abort_transaction(&trans);
 815                 *ap->a_vpp = NULL;
 816         } else {
 817                 hammer_commit_transaction(&trans);
 818                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 819                 hammer_rel_inode(nip, 0);
 820                 if (error == 0) {
 821                         cache_setunresolved(ap->a_nch);
 822                         cache_setvp(ap->a_nch, *ap->a_vpp);
 823                 }
 824         }
 825         return (error);
 826 }
 827
 828 /*
 829  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 830  *
 831  * The operating system has already ensured that the directory entry
 832  * does not exist and done all appropriate namespace locking.
 833  */
 834 static
 835 int
 836 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 837 {
 838         struct hammer_transaction trans;
 839         struct hammer_inode *dip;
 840         struct hammer_inode *nip;
 841         struct nchandle *nch;
 842         int error;
 843
 844         nch = ap->a_nch;
 845         dip = VTOI(ap->a_dvp);
 846
 847         if (dip->flags & HAMMER_INODE_RO)
 848                 return (EROFS);
 849
 850         /*
 851          * Create a transaction to cover the operations we perform.
 852          */
 853         hammer_start_transaction(&trans, dip->hmp);
 854
 855         /*
 856          * Create a new filesystem object of the requested type.  The
 857          * returned inode will be referenced but not locked.
 858          */
 859         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 860         if (error) {
 861                 hammer_abort_transaction(&trans);
 862                 *ap->a_vpp = NULL;
 863                 return (error);
 864         }
 865
 866         /*
 867          * Add the new filesystem object to the directory.  This will also
 868          * bump the inode's link count.
 869          */
 870         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 871
 872         /*
 873          * Finish up.
 874          */
 875         if (error) {
 876                 hammer_rel_inode(nip, 0);
 877                 hammer_abort_transaction(&trans);
 878                 *ap->a_vpp = NULL;
 879         } else {
 880                 hammer_commit_transaction(&trans);
 881                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 882                 hammer_rel_inode(nip, 0);
 883                 if (error == 0) {
 884                         cache_setunresolved(ap->a_nch);
 885                         cache_setvp(ap->a_nch, *ap->a_vpp);
 886                 }
 887         }
 888         return (error);
 889 }
 890
 891 /*
 892  * hammer_vop_open { vp, mode, cred, fp }
 893  */
 894 static
 895 int
 896 hammer_vop_open(struct vop_open_args *ap)
 897 {
 898         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
 899                 return (EROFS);
 900
 901         return(vop_stdopen(ap));
 902 }
 903
 904 /*
 905  * hammer_vop_pathconf { vp, name, retval }
 906  */
 907 static
 908 int
 909 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 910 {
 911         return EOPNOTSUPP;
 912 }
 913
 914 /*
 915  * hammer_vop_print { vp }
 916  */
 917 static
 918 int
 919 hammer_vop_print(struct vop_print_args *ap)
 920 {
 921         return EOPNOTSUPP;
 922 }
 923
 924 /*
 925  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 926  */
 927 static
 928 int
 929 hammer_vop_readdir(struct vop_readdir_args *ap)
 930 {
 931         struct hammer_cursor cursor;
 932         struct hammer_inode *ip;
 933         struct uio *uio;
 934         hammer_record_ondisk_t rec;
 935         hammer_base_elm_t base;
 936         int error;
 937         int cookie_index;
 938         int ncookies;
 939         off_t *cookies;
 940         off_t saveoff;
 941         int r;
 942
 943         ip = VTOI(ap->a_vp);
 944         uio = ap->a_uio;
 945         saveoff = uio->uio_offset;
 946
 947         if (ap->a_ncookies) {
 948                 ncookies = uio->uio_resid / 16 + 1;
 949                 if (ncookies > 1024)
 950                         ncookies = 1024;
 951                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 952                 cookie_index = 0;
 953         } else {
 954                 ncookies = -1;
 955                 cookies = NULL;
 956                 cookie_index = 0;
 957         }
 958
 959         /*
 960          * Handle artificial entries
 961          */
 962         error = 0;
 963         if (saveoff == 0) {
 964                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
 965                 if (r)
 966                         goto done;
 967                 if (cookies)
 968                         cookies[cookie_index] = saveoff;
 969                 ++saveoff;
 970                 ++cookie_index;
 971                 if (cookie_index == ncookies)
 972                         goto done;
 973         }
 974         if (saveoff == 1) {
 975                 if (ip->ino_data.parent_obj_id) {
 976                         r = vop_write_dirent(&error, uio,
 977                                              ip->ino_data.parent_obj_id,
 978                                              DT_DIR, 2, "..");
 979                 } else {
 980                         r = vop_write_dirent(&error, uio,
 981                                              ip->obj_id, DT_DIR, 2, "..");
 982                 }
 983                 if (r)
 984                         goto done;
 985                 if (cookies)
 986                         cookies[cookie_index] = saveoff;
 987                 ++saveoff;
 988                 ++cookie_index;
 989                 if (cookie_index == ncookies)
 990                         goto done;
 991         }
 992
 993         /*
 994          * Key range (begin and end inclusive) to scan.  Directory keys
 995          * directly translate to a 64 bit 'seek' position.
 996          */
 997         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
 998         cursor.key_beg.obj_id = ip->obj_id;
 999         cursor.key_beg.create_tid = 0;
1000         cursor.key_beg.delete_tid = 0;
1001         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1002         cursor.key_beg.obj_type = 0;
1003         cursor.key_beg.key = saveoff;
1004
1005         cursor.key_end = cursor.key_beg;
1006         cursor.key_end.key = HAMMER_MAX_KEY;
1007         cursor.asof = ip->obj_asof;
1008         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1009
1010         error = hammer_ip_first(&cursor, ip);
1011
1012         while (error == 0) {
1013                 error = hammer_ip_resolve_data(&cursor);
1014                 if (error)
1015                         break;
1016                 rec = cursor.record;
1017                 base = &rec->base.base;
1018                 saveoff = base->key;
1019
1020                 if (base->obj_id != ip->obj_id)
1021                         panic("readdir: bad record at %p", cursor.node);
1022
1023                 r = vop_write_dirent(
1024                              &error, uio, rec->entry.obj_id,
1025                              hammer_get_dtype(rec->entry.base.base.obj_type),
1026                              rec->entry.base.data_len,
1027                              (void *)cursor.data);
1028                 if (r)
1029                         break;
1030                 ++saveoff;
1031                 if (cookies)
1032                         cookies[cookie_index] = base->key;
1033                 ++cookie_index;
1034                 if (cookie_index == ncookies)
1035                         break;
1036                 error = hammer_ip_next(&cursor);
1037         }
1038         hammer_done_cursor(&cursor);
1039
1040 done:
1041         if (ap->a_eofflag)
1042                 *ap->a_eofflag = (error == ENOENT);
1043         uio->uio_offset = saveoff;
1044         if (error && cookie_index == 0) {
1045                 if (error == ENOENT)
1046                         error = 0;
1047                 if (cookies) {
1048                         kfree(cookies, M_TEMP);
1049                         *ap->a_ncookies = 0;
1050                         *ap->a_cookies = NULL;
1051                 }
1052         } else {
1053                 if (error == ENOENT)
1054                         error = 0;
1055                 if (cookies) {
1056                         *ap->a_ncookies = cookie_index;
1057                         *ap->a_cookies = cookies;
1058                 }
1059         }
1060         return(error);
1061 }
1062
1063 /*
1064  * hammer_vop_readlink { vp, uio, cred }
1065  */
1066 static
1067 int
1068 hammer_vop_readlink(struct vop_readlink_args *ap)
1069 {
1070         struct hammer_cursor cursor;
1071         struct hammer_inode *ip;
1072         int error;
1073
1074         ip = VTOI(ap->a_vp);
1075         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1076
1077         /*
1078          * Key range (begin and end inclusive) to scan.  Directory keys
1079          * directly translate to a 64 bit 'seek' position.
1080          */
1081         cursor.key_beg.obj_id = ip->obj_id;
1082         cursor.key_beg.create_tid = 0;
1083         cursor.key_beg.delete_tid = 0;
1084         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1085         cursor.key_beg.obj_type = 0;
1086         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1087         cursor.asof = ip->obj_asof;
1088         cursor.flags |= HAMMER_CURSOR_ASOF;
1089
1090         error = hammer_ip_lookup(&cursor, ip);
1091         if (error == 0) {
1092                 error = hammer_ip_resolve_data(&cursor);
1093                 if (error == 0) {
1094                         error = uiomove((char *)cursor.data,
1095                                         cursor.record->generic.base.data_len,
1096                                         ap->a_uio);
1097                 }
1098         }
1099         hammer_done_cursor(&cursor);
1100         return(error);
1101 }
1102
1103 /*
1104  * hammer_vop_nremove { nch, dvp, cred }
1105  */
1106 static
1107 int
1108 hammer_vop_nremove(struct vop_nremove_args *ap)
1109 {
1110         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1111 }
1112
1113 /*
1114  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1115  */
1116 static
1117 int
1118 hammer_vop_nrename(struct vop_nrename_args *ap)
1119 {
1120         struct hammer_transaction trans;
1121         struct namecache *fncp;
1122         struct namecache *tncp;
1123         struct hammer_inode *fdip;
1124         struct hammer_inode *tdip;
1125         struct hammer_inode *ip;
1126         struct hammer_cursor cursor;
1127         union hammer_record_ondisk *rec;
1128         int64_t namekey;
1129         int error;
1130
1131         fdip = VTOI(ap->a_fdvp);
1132         tdip = VTOI(ap->a_tdvp);
1133         fncp = ap->a_fnch->ncp;
1134         tncp = ap->a_tnch->ncp;
1135         ip = VTOI(fncp->nc_vp);
1136         KKASSERT(ip != NULL);
1137
1138         if (fdip->flags & HAMMER_INODE_RO)
1139                 return (EROFS);
1140         if (tdip->flags & HAMMER_INODE_RO)
1141                 return (EROFS);
1142         if (ip->flags & HAMMER_INODE_RO)
1143                 return (EROFS);
1144
1145         hammer_start_transaction(&trans, fdip->hmp);
1146
1147         /*
1148          * Remove tncp from the target directory and then link ip as
1149          * tncp. XXX pass trans to dounlink
1150          */
1151         error = hammer_dounlink(ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1152         if (error == 0 || error == ENOENT)
1153                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1154         if (error)
1155                 goto failed; /* XXX */
1156
1157         /*
1158          * Locate the record in the originating directory and remove it.
1159          *
1160          * Calculate the namekey and setup the key range for the scan.  This
1161          * works kinda like a chained hash table where the lower 32 bits
1162          * of the namekey synthesize the chain.
1163          *
1164          * The key range is inclusive of both key_beg and key_end.
1165          */
1166         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1167 retry:
1168         hammer_init_cursor_hmp(&cursor, &fdip->cache[0], fdip->hmp);
1169         cursor.key_beg.obj_id = fdip->obj_id;
1170         cursor.key_beg.key = namekey;
1171         cursor.key_beg.create_tid = 0;
1172         cursor.key_beg.delete_tid = 0;
1173         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1174         cursor.key_beg.obj_type = 0;
1175
1176         cursor.key_end = cursor.key_beg;
1177         cursor.key_end.key |= 0xFFFFFFFFULL;
1178         cursor.asof = fdip->obj_asof;
1179         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1180
1181         /*
1182          * Scan all matching records (the chain), locate the one matching
1183          * the requested path component.
1184          *
1185          * The hammer_ip_*() functions merge in-memory records with on-disk
1186          * records for the purposes of the search.
1187          */
1188         error = hammer_ip_first(&cursor, fdip);
1189         while (error == 0) {
1190                 if (hammer_ip_resolve_data(&cursor) != 0)
1191                         break;
1192                 rec = cursor.record;
1193                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1194                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1195                         break;
1196                 }
1197                 error = hammer_ip_next(&cursor);
1198         }
1199
1200         /*
1201          * If all is ok we have to get the inode so we can adjust nlinks.
1202          *
1203          * WARNING: hammer_ip_del_directory() may have to terminate the
1204          * cursor to avoid a recursion.  It's ok to call hammer_done_cursor()
1205          * twice.
1206          */
1207         if (error == 0)
1208                 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1209         hammer_done_cursor(&cursor);
1210         if (error == 0)
1211                 cache_rename(ap->a_fnch, ap->a_tnch);
1212         if (error == EDEADLK)
1213                 goto retry;
1214 failed:
1215         if (error == 0) {
1216                 hammer_commit_transaction(&trans);
1217         } else {
1218                 hammer_abort_transaction(&trans);
1219         }
1220         return (error);
1221 }
1222
1223 /*
1224  * hammer_vop_nrmdir { nch, dvp, cred }
1225  */
1226 static
1227 int
1228 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1229 {
1230         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1231 }
1232
1233 /*
1234  * hammer_vop_setattr { vp, vap, cred }
1235  */
1236 static
1237 int
1238 hammer_vop_setattr(struct vop_setattr_args *ap)
1239 {
1240         struct hammer_transaction trans;
1241         struct hammer_cursor *spike = NULL;
1242         struct vattr *vap;
1243         struct hammer_inode *ip;
1244         int modflags;
1245         int error;
1246         int truncating;
1247         int64_t aligned_size;
1248         u_int32_t flags;
1249         uuid_t uuid;
1250
1251         vap = ap->a_vap;
1252         ip = ap->a_vp->v_data;
1253         modflags = 0;
1254
1255         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1256                 return(EROFS);
1257         if (ip->flags & HAMMER_INODE_RO)
1258                 return (EROFS);
1259
1260         hammer_start_transaction(&trans, ip->hmp);
1261         error = 0;
1262
1263         if (vap->va_flags != VNOVAL) {
1264                 flags = ip->ino_data.uflags;
1265                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1266                                          hammer_to_unix_xid(&ip->ino_data.uid),
1267                                          ap->a_cred);
1268                 if (error == 0) {
1269                         if (ip->ino_data.uflags != flags) {
1270                                 ip->ino_data.uflags = flags;
1271                                 modflags |= HAMMER_INODE_DDIRTY;
1272                         }
1273                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1274                                 error = 0;
1275                                 goto done;
1276                         }
1277                 }
1278                 goto done;
1279         }
1280         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1281                 error = EPERM;
1282                 goto done;
1283         }
1284         if (vap->va_uid != (uid_t)VNOVAL) {
1285                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1286                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1287                         ip->ino_data.uid = uuid;
1288                         modflags |= HAMMER_INODE_DDIRTY;
1289                 }
1290         }
1291         if (vap->va_gid != (uid_t)VNOVAL) {
1292                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1293                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1294                         ip->ino_data.gid = uuid;
1295                         modflags |= HAMMER_INODE_DDIRTY;
1296                 }
1297         }
1298         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1299                 switch(ap->a_vp->v_type) {
1300                 case VREG:
1301                         if (vap->va_size == ip->ino_rec.ino_size)
1302                                 break;
1303                         if (vap->va_size < ip->ino_rec.ino_size) {
1304                                 vtruncbuf(ap->a_vp, vap->va_size,
1305                                           HAMMER_BUFSIZE);
1306                                 truncating = 1;
1307                         } else {
1308                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1309                                 truncating = 0;
1310                         }
1311                         ip->ino_rec.ino_size = vap->va_size;
1312                         modflags |= HAMMER_INODE_RDIRTY;
1313                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1314                                         ~(int64_t)HAMMER_BUFMASK;
1315
1316                         if (truncating) {
1317                                 error = hammer_ip_delete_range(&trans, ip,
1318                                                     aligned_size,
1319                                                     0x7FFFFFFFFFFFFFFFLL,
1320                                                     &spike);
1321                         }
1322                         /*
1323                          * If truncating we have to clean out a portion of
1324                          * the last block on-disk.
1325                          */
1326                         if (truncating && error == 0 &&
1327                             vap->va_size < aligned_size) {
1328                                 struct buf *bp;
1329                                 int offset;
1330
1331                                 offset = vap->va_size & HAMMER_BUFMASK;
1332                                 error = bread(ap->a_vp,
1333                                               aligned_size - HAMMER_BUFSIZE,
1334                                               HAMMER_BUFSIZE, &bp);
1335                                 if (error == 0) {
1336                                         bzero(bp->b_data + offset,
1337                                               HAMMER_BUFSIZE - offset);
1338                                         bdwrite(bp);
1339                                 } else {
1340                                         brelse(bp);
1341                                 }
1342                         }
1343                         break;
1344                 case VDATABASE:
1345                         error = hammer_ip_delete_range(&trans, ip,
1346                                                     vap->va_size,
1347                                                     0x7FFFFFFFFFFFFFFFLL,
1348                                                     &spike);
1349                         ip->ino_rec.ino_size = vap->va_size;
1350                         modflags |= HAMMER_INODE_RDIRTY;
1351                         break;
1352                 default:
1353                         error = EINVAL;
1354                         goto done;
1355                 }
1356                 if (error == ENOSPC) {
1357                         error = hammer_spike(&spike);
1358                         if (error == 0)
1359                                 continue;
1360                 }
1361                 KKASSERT(spike == NULL);
1362                 break;
1363         }
1364         if (vap->va_atime.tv_sec != VNOVAL) {
1365                 ip->ino_rec.ino_atime =
1366                         hammer_timespec_to_transid(&vap->va_atime);
1367                 modflags |= HAMMER_INODE_ITIMES;
1368         }
1369         if (vap->va_mtime.tv_sec != VNOVAL) {
1370                 ip->ino_rec.ino_mtime =
1371                         hammer_timespec_to_transid(&vap->va_mtime);
1372                 modflags |= HAMMER_INODE_ITIMES;
1373         }
1374         if (vap->va_mode != (mode_t)VNOVAL) {
1375                 if (ip->ino_data.mode != vap->va_mode) {
1376                         ip->ino_data.mode = vap->va_mode;
1377                         modflags |= HAMMER_INODE_DDIRTY;
1378                 }
1379         }
1380 done:
1381         if (error) {
1382                 hammer_abort_transaction(&trans);
1383         } else {
1384                 hammer_modify_inode(&trans, ip, modflags);
1385                 hammer_commit_transaction(&trans);
1386         }
1387         return (error);
1388 }
1389
1390 /*
1391  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1392  */
1393 static
1394 int
1395 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1396 {
1397         struct hammer_transaction trans;
1398         struct hammer_inode *dip;
1399         struct hammer_inode *nip;
1400         struct nchandle *nch;
1401         hammer_record_t record;
1402         int error;
1403         int bytes;
1404
1405         ap->a_vap->va_type = VLNK;
1406
1407         nch = ap->a_nch;
1408         dip = VTOI(ap->a_dvp);
1409
1410         if (dip->flags & HAMMER_INODE_RO)
1411                 return (EROFS);
1412
1413         /*
1414          * Create a transaction to cover the operations we perform.
1415          */
1416         hammer_start_transaction(&trans, dip->hmp);
1417
1418         /*
1419          * Create a new filesystem object of the requested type.  The
1420          * returned inode will be referenced but not locked.
1421          */
1422
1423         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1424         if (error) {
1425                 hammer_abort_transaction(&trans);
1426                 *ap->a_vpp = NULL;
1427                 return (error);
1428         }
1429
1430         /*
1431          * Add the new filesystem object to the directory.  This will also
1432          * bump the inode's link count.
1433          */
1434         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1435
1436         /*
1437          * Add a record representing the symlink.  symlink stores the link
1438          * as pure data, not a string, and is no \0 terminated.
1439          */
1440         if (error == 0) {
1441                 record = hammer_alloc_mem_record(nip);
1442                 bytes = strlen(ap->a_target);
1443
1444                 record->rec.generic.base.base.key = HAMMER_FIXKEY_SYMLINK;
1445                 record->rec.generic.base.base.rec_type = HAMMER_RECTYPE_FIX;
1446                 record->rec.generic.base.data_len = bytes;
1447                 if (bytes <= sizeof(record->rec.generic.filler)) {
1448                         record->data = (void *)record->rec.generic.filler;
1449                         bcopy(ap->a_target, record->data, bytes);
1450                 } else {
1451                         record->data = (void *)ap->a_target;
1452                         /* will be reallocated by routine below */
1453                 }
1454                 error = hammer_ip_add_record(&trans, record);
1455         }
1456
1457         /*
1458          * Finish up.
1459          */
1460         if (error) {
1461                 hammer_rel_inode(nip, 0);
1462                 hammer_abort_transaction(&trans);
1463                 *ap->a_vpp = NULL;
1464         } else {
1465                 hammer_commit_transaction(&trans);
1466                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1467                 hammer_rel_inode(nip, 0);
1468                 if (error == 0) {
1469                         cache_setunresolved(ap->a_nch);
1470                         cache_setvp(ap->a_nch, *ap->a_vpp);
1471                 }
1472         }
1473         return (error);
1474 }
1475
1476 /*
1477  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1478  */
1479 static
1480 int
1481 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1482 {
1483         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1484 }
1485
1486 /*
1487  * hammer_vop_strategy { vp, bio }
1488  *
1489  * Strategy call, used for regular file read & write only.  Note that the
1490  * bp may represent a cluster.
1491  *
1492  * To simplify operation and allow better optimizations in the future,
1493  * this code does not make any assumptions with regards to buffer alignment
1494  * or size.
1495  */
1496 static
1497 int
1498 hammer_vop_strategy(struct vop_strategy_args *ap)
1499 {
1500         struct buf *bp;
1501         int error;
1502
1503         bp = ap->a_bio->bio_buf;
1504
1505         switch(bp->b_cmd) {
1506         case BUF_CMD_READ:
1507                 error = hammer_vop_strategy_read(ap);
1508                 break;
1509         case BUF_CMD_WRITE:
1510                 error = hammer_vop_strategy_write(ap);
1511                 break;
1512         default:
1513                 error = EINVAL;
1514                 break;
1515         }
1516         bp->b_error = error;
1517         if (error)
1518                 bp->b_flags |= B_ERROR;
1519         biodone(ap->a_bio);
1520         return (error);
1521 }
1522
1523 /*
1524  * Read from a regular file.  Iterate the related records and fill in the
1525  * BIO/BUF.  Gaps are zero-filled.
1526  *
1527  * The support code in hammer_object.c should be used to deal with mixed
1528  * in-memory and on-disk records.
1529  *
1530  * XXX atime update
1531  */
1532 static
1533 int
1534 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1535 {
1536         struct hammer_inode *ip = ap->a_vp->v_data;
1537         struct hammer_cursor cursor;
1538         hammer_record_ondisk_t rec;
1539         hammer_base_elm_t base;
1540         struct bio *bio;
1541         struct buf *bp;
1542         int64_t rec_offset;
1543         int64_t ran_end;
1544         int64_t tmp64;
1545         int error;
1546         int boff;
1547         int roff;
1548         int n;
1549
1550         bio = ap->a_bio;
1551         bp = bio->bio_buf;
1552
1553         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1554
1555         /*
1556          * Key range (begin and end inclusive) to scan.  Note that the key's
1557          * stored in the actual records represent BASE+LEN, not BASE.  The
1558          * first record containing bio_offset will have a key > bio_offset.
1559          */
1560         cursor.key_beg.obj_id = ip->obj_id;
1561         cursor.key_beg.create_tid = 0;
1562         cursor.key_beg.delete_tid = 0;
1563         cursor.key_beg.obj_type = 0;
1564         cursor.key_beg.key = bio->bio_offset + 1;
1565         cursor.asof = ip->obj_asof;
1566         cursor.flags |= HAMMER_CURSOR_ASOF;
1567
1568         cursor.key_end = cursor.key_beg;
1569         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1570                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1571                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1572                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1573         } else {
1574                 ran_end = bio->bio_offset + bp->b_bufsize;
1575                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1576                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1577                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1578                 if (tmp64 < ran_end)
1579                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1580                 else
1581                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1582         }
1583         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1584
1585         error = hammer_ip_first(&cursor, ip);
1586         boff = 0;
1587
1588         while (error == 0) {
1589                 error = hammer_ip_resolve_data(&cursor);
1590                 if (error)
1591                         break;
1592                 rec = cursor.record;
1593                 base = &rec->base.base;
1594
1595                 rec_offset = base->key - rec->data.base.data_len;
1596
1597                 /*
1598                  * Calculate the gap, if any, and zero-fill it.
1599                  */
1600                 n = (int)(rec_offset - (bio->bio_offset + boff));
1601                 if (n > 0) {
1602                         if (n > bp->b_bufsize - boff)
1603                                 n = bp->b_bufsize - boff;
1604                         bzero((char *)bp->b_data + boff, n);
1605                         boff += n;
1606                         n = 0;
1607                 }
1608
1609                 /*
1610                  * Calculate the data offset in the record and the number
1611                  * of bytes we can copy.
1612                  *
1613                  * Note there is a degenerate case here where boff may
1614                  * already be at bp->b_bufsize.
1615                  */
1616                 roff = -n;
1617                 n = rec->data.base.data_len - roff;
1618                 KKASSERT(n > 0);
1619                 if (n > bp->b_bufsize - boff)
1620                         n = bp->b_bufsize - boff;
1621                 bcopy((char *)cursor.data + roff, (char *)bp->b_data + boff, n);
1622                 boff += n;
1623                 if (boff == bp->b_bufsize)
1624                         break;
1625                 error = hammer_ip_next(&cursor);
1626         }
1627         hammer_done_cursor(&cursor);
1628
1629         /*
1630          * There may have been a gap after the last record
1631          */
1632         if (error == ENOENT)
1633                 error = 0;
1634         if (error == 0 && boff != bp->b_bufsize) {
1635                 KKASSERT(boff < bp->b_bufsize);
1636                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1637                 /* boff = bp->b_bufsize; */
1638         }
1639         bp->b_resid = 0;
1640         return(error);
1641 }
1642
1643 /*
1644  * Write to a regular file.  Iterate the related records and mark for
1645  * deletion.  If existing edge records (left and right side) overlap our
1646  * write they have to be marked deleted and new records created, usually
1647  * referencing a portion of the original data.  Then add a record to
1648  * represent the buffer.
1649  *
1650  * The support code in hammer_object.c should be used to deal with mixed
1651  * in-memory and on-disk records.
1652  */
1653 static
1654 int
1655 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1656 {
1657         struct hammer_transaction trans;
1658         struct hammer_cursor *spike = NULL;
1659         hammer_inode_t ip;
1660         struct bio *bio;
1661         struct buf *bp;
1662         int error;
1663
1664         bio = ap->a_bio;
1665         bp = bio->bio_buf;
1666         ip = ap->a_vp->v_data;
1667
1668         if (ip->flags & HAMMER_INODE_RO)
1669                 return (EROFS);
1670
1671         hammer_start_transaction(&trans, ip->hmp);
1672
1673 retry:
1674         /*
1675          * Delete any records overlapping our range.  This function will
1676          * (eventually) properly truncate partial overlaps.
1677          */
1678         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1679                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1680                                                bio->bio_offset, &spike);
1681         } else {
1682                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1683                                                bio->bio_offset +
1684                                                 bp->b_bufsize - 1,
1685                                                &spike);
1686         }
1687
1688         /*
1689          * Add a single record to cover the write
1690          */
1691         if (error == 0) {
1692                 error = hammer_ip_sync_data(&trans, ip, bio->bio_offset,
1693                                             bp->b_data, bp->b_bufsize,
1694                                             &spike);
1695         }
1696
1697         /*
1698          * If we ran out of space the spike structure will be filled in
1699          * and we must call hammer_spike with it, then retry.
1700          */
1701         if (error == ENOSPC) {
1702                 error = hammer_spike(&spike);
1703                 if (error == 0)
1704                         goto retry;
1705         }
1706         KKASSERT(spike == NULL);
1707
1708         /*
1709          * If an error occured abort the transaction
1710          */
1711         if (error) {
1712                 /* XXX undo deletion */
1713                 hammer_abort_transaction(&trans);
1714                 bp->b_resid = bp->b_bufsize;
1715         } else {
1716                 hammer_commit_transaction(&trans);
1717                 bp->b_resid = 0;
1718         }
1719         return(error);
1720 }
1721
1722 /*
1723  * dounlink - disconnect a directory entry
1724  *
1725  * XXX whiteout support not really in yet
1726  */
1727 static int
1728 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1729                 int flags)
1730 {
1731         struct hammer_transaction trans;
1732         struct namecache *ncp;
1733         hammer_inode_t dip;
1734         hammer_inode_t ip;
1735         hammer_record_ondisk_t rec;
1736         struct hammer_cursor cursor;
1737         int64_t namekey;
1738         int error;
1739
1740         /*
1741          * Calculate the namekey and setup the key range for the scan.  This
1742          * works kinda like a chained hash table where the lower 32 bits
1743          * of the namekey synthesize the chain.
1744          *
1745          * The key range is inclusive of both key_beg and key_end.
1746          */
1747         dip = VTOI(dvp);
1748         ncp = nch->ncp;
1749
1750         if (dip->flags & HAMMER_INODE_RO)
1751                 return (EROFS);
1752
1753         hammer_start_transaction(&trans, dip->hmp);
1754
1755         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1756 retry:
1757         hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
1758         cursor.key_beg.obj_id = dip->obj_id;
1759         cursor.key_beg.key = namekey;
1760         cursor.key_beg.create_tid = 0;
1761         cursor.key_beg.delete_tid = 0;
1762         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1763         cursor.key_beg.obj_type = 0;
1764
1765         cursor.key_end = cursor.key_beg;
1766         cursor.key_end.key |= 0xFFFFFFFFULL;
1767         cursor.asof = dip->obj_asof;
1768         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1769
1770         /*
1771          * Scan all matching records (the chain), locate the one matching
1772          * the requested path component.  info->last_error contains the
1773          * error code on search termination and could be 0, ENOENT, or
1774          * something else.
1775          *
1776          * The hammer_ip_*() functions merge in-memory records with on-disk
1777          * records for the purposes of the search.
1778          */
1779         error = hammer_ip_first(&cursor, dip);
1780         while (error == 0) {
1781                 error = hammer_ip_resolve_data(&cursor);
1782                 if (error)
1783                         break;
1784                 rec = cursor.record;
1785                 if (ncp->nc_nlen == rec->entry.base.data_len &&
1786                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1787                         break;
1788                 }
1789                 error = hammer_ip_next(&cursor);
1790         }
1791
1792         /*
1793          * If all is ok we have to get the inode so we can adjust nlinks.
1794          *
1795          * If the target is a directory, it must be empty.
1796          */
1797         if (error == 0) {
1798                 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
1799                                       rec->entry.obj_id,
1800                                       dip->hmp->asof, 0, &error);
1801                 if (error == ENOENT) {
1802                         kprintf("obj_id %016llx\n", rec->entry.obj_id);
1803                         Debugger("ENOENT unlinking object that should exist, cont to sync");
1804                         hammer_sync_hmp(dip->hmp, MNT_NOWAIT);
1805                         Debugger("ENOENT - sync done");
1806                 }
1807                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
1808                                   HAMMER_OBJTYPE_DIRECTORY) {
1809                         error = hammer_ip_check_directory_empty(&trans, ip);
1810                 }
1811                 /*
1812                  * WARNING: hammer_ip_del_directory() may have to terminate
1813                  * the cursor to avoid a lock recursion.  It's ok to call
1814                  * hammer_done_cursor() twice.
1815                  */
1816                 if (error == 0)
1817                         error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1818                 if (error == 0) {
1819                         cache_setunresolved(nch);
1820                         cache_setvp(nch, NULL);
1821                         /* XXX locking */
1822                         if (ip->vp)
1823                                 cache_inval_vp(ip->vp, CINV_DESTROY);
1824                 }
1825                 hammer_rel_inode(ip, 0);
1826         }
1827         hammer_done_cursor(&cursor);
1828         if (error == EDEADLK)
1829                 goto retry;
1830
1831         if (error == 0)
1832                 hammer_commit_transaction(&trans);
1833         else
1834                 hammer_abort_transaction(&trans);
1835         return (error);
1836 }
1837
1838 /************************************************************************
1839  *                          FIFO AND SPECFS OPS                         *
1840  ************************************************************************
1841  *
1842  */
1843
1844 static int
1845 hammer_vop_fifoclose (struct vop_close_args *ap)
1846 {
1847         /* XXX update itimes */
1848         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1849 }
1850
1851 static int
1852 hammer_vop_fiforead (struct vop_read_args *ap)
1853 {
1854         int error;
1855
1856         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1857         /* XXX update access time */
1858         return (error);
1859 }
1860
1861 static int
1862 hammer_vop_fifowrite (struct vop_write_args *ap)
1863 {
1864         int error;
1865
1866         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1867         /* XXX update access time */
1868         return (error);
1869 }
1870
1871 static int
1872 hammer_vop_specclose (struct vop_close_args *ap)
1873 {
1874         /* XXX update itimes */
1875         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1876 }
1877
1878 static int
1879 hammer_vop_specread (struct vop_read_args *ap)
1880 {
1881         /* XXX update access time */
1882         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1883 }
1884
1885 static int
1886 hammer_vop_specwrite (struct vop_write_args *ap)
1887 {
1888         /* XXX update last change time */
1889         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1890 }
1891