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 1.26 2008/02/05 07:58:43 dillon Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/fcntl.h>
  41 #include <sys/namecache.h>
  42 #include <sys/vnode.h>
  43 #include <sys/lockf.h>
  44 #include <sys/event.h>
  45 #include <sys/stat.h>
  46 #include <sys/dirent.h>
  47 #include <vm/vm_extern.h>
  48 #include <vfs/fifofs/fifo.h>
  49 #include "hammer.h"
  50
  51 /*
  52  * USERFS VNOPS
  53  */
  54 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
  55 static int hammer_vop_fsync(struct vop_fsync_args *);
  56 static int hammer_vop_read(struct vop_read_args *);
  57 static int hammer_vop_write(struct vop_write_args *);
  58 static int hammer_vop_access(struct vop_access_args *);
  59 static int hammer_vop_advlock(struct vop_advlock_args *);
  60 static int hammer_vop_close(struct vop_close_args *);
  61 static int hammer_vop_ncreate(struct vop_ncreate_args *);
  62 static int hammer_vop_getattr(struct vop_getattr_args *);
  63 static int hammer_vop_nresolve(struct vop_nresolve_args *);
  64 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
  65 static int hammer_vop_nlink(struct vop_nlink_args *);
  66 static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
  67 static int hammer_vop_nmknod(struct vop_nmknod_args *);
  68 static int hammer_vop_open(struct vop_open_args *);
  69 static int hammer_vop_pathconf(struct vop_pathconf_args *);
  70 static int hammer_vop_print(struct vop_print_args *);
  71 static int hammer_vop_readdir(struct vop_readdir_args *);
  72 static int hammer_vop_readlink(struct vop_readlink_args *);
  73 static int hammer_vop_nremove(struct vop_nremove_args *);
  74 static int hammer_vop_nrename(struct vop_nrename_args *);
  75 static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
  76 static int hammer_vop_setattr(struct vop_setattr_args *);
  77 static int hammer_vop_strategy(struct vop_strategy_args *);
  78 static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
  79 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
  80 static int hammer_vop_ioctl(struct vop_ioctl_args *);
  81
  82 static int hammer_vop_fifoclose (struct vop_close_args *);
  83 static int hammer_vop_fiforead (struct vop_read_args *);
  84 static int hammer_vop_fifowrite (struct vop_write_args *);
  85
  86 static int hammer_vop_specclose (struct vop_close_args *);
  87 static int hammer_vop_specread (struct vop_read_args *);
  88 static int hammer_vop_specwrite (struct vop_write_args *);
  89
  90 struct vop_ops hammer_vnode_vops = {
  91         .vop_default =          vop_defaultop,
  92         .vop_fsync =            hammer_vop_fsync,
  93         .vop_getpages =         vop_stdgetpages,
  94         .vop_putpages =         vop_stdputpages,
  95         .vop_read =             hammer_vop_read,
  96         .vop_write =            hammer_vop_write,
  97         .vop_access =           hammer_vop_access,
  98         .vop_advlock =          hammer_vop_advlock,
  99         .vop_close =            hammer_vop_close,
 100         .vop_ncreate =          hammer_vop_ncreate,
 101         .vop_getattr =          hammer_vop_getattr,
 102         .vop_inactive =         hammer_vop_inactive,
 103         .vop_reclaim =          hammer_vop_reclaim,
 104         .vop_nresolve =         hammer_vop_nresolve,
 105         .vop_nlookupdotdot =    hammer_vop_nlookupdotdot,
 106         .vop_nlink =            hammer_vop_nlink,
 107         .vop_nmkdir =           hammer_vop_nmkdir,
 108         .vop_nmknod =           hammer_vop_nmknod,
 109         .vop_open =             hammer_vop_open,
 110         .vop_pathconf =         hammer_vop_pathconf,
 111         .vop_print =            hammer_vop_print,
 112         .vop_readdir =          hammer_vop_readdir,
 113         .vop_readlink =         hammer_vop_readlink,
 114         .vop_nremove =          hammer_vop_nremove,
 115         .vop_nrename =          hammer_vop_nrename,
 116         .vop_nrmdir =           hammer_vop_nrmdir,
 117         .vop_setattr =          hammer_vop_setattr,
 118         .vop_strategy =         hammer_vop_strategy,
 119         .vop_nsymlink =         hammer_vop_nsymlink,
 120         .vop_nwhiteout =        hammer_vop_nwhiteout,
 121         .vop_ioctl =            hammer_vop_ioctl
 122 };
 123
 124 struct vop_ops hammer_spec_vops = {
 125         .vop_default =          spec_vnoperate,
 126         .vop_fsync =            hammer_vop_fsync,
 127         .vop_read =             hammer_vop_specread,
 128         .vop_write =            hammer_vop_specwrite,
 129         .vop_access =           hammer_vop_access,
 130         .vop_close =            hammer_vop_specclose,
 131         .vop_getattr =          hammer_vop_getattr,
 132         .vop_inactive =         hammer_vop_inactive,
 133         .vop_reclaim =          hammer_vop_reclaim,
 134         .vop_setattr =          hammer_vop_setattr
 135 };
 136
 137 struct vop_ops hammer_fifo_vops = {
 138         .vop_default =          fifo_vnoperate,
 139         .vop_fsync =            hammer_vop_fsync,
 140         .vop_read =             hammer_vop_fiforead,
 141         .vop_write =            hammer_vop_fifowrite,
 142         .vop_access =           hammer_vop_access,
 143         .vop_close =            hammer_vop_fifoclose,
 144         .vop_getattr =          hammer_vop_getattr,
 145         .vop_inactive =         hammer_vop_inactive,
 146         .vop_reclaim =          hammer_vop_reclaim,
 147         .vop_setattr =          hammer_vop_setattr
 148 };
 149
 150 static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp,
 151                            struct ucred *cred, int flags);
 152 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
 153 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
 154
 155 #if 0
 156 static
 157 int
 158 hammer_vop_vnoperate(struct vop_generic_args *)
 159 {
 160         return (VOCALL(&hammer_vnode_vops, ap));
 161 }
 162 #endif
 163
 164 /*
 165  * hammer_vop_fsync { vp, waitfor }
 166  */
 167 static
 168 int
 169 hammer_vop_fsync(struct vop_fsync_args *ap)
 170 {
 171         hammer_inode_t ip;
 172         int error;
 173
 174         ip = VTOI(ap->a_vp);
 175         error = hammer_sync_inode(ip, ap->a_waitfor, 0);
 176         return (error);
 177 }
 178
 179 /*
 180  * hammer_vop_read { vp, uio, ioflag, cred }
 181  */
 182 static
 183 int
 184 hammer_vop_read(struct vop_read_args *ap)
 185 {
 186         struct hammer_transaction trans;
 187         hammer_inode_t ip;
 188         off_t offset;
 189         struct buf *bp;
 190         struct uio *uio;
 191         int error;
 192         int n;
 193         int seqcount;
 194
 195         if (ap->a_vp->v_type != VREG)
 196                 return (EINVAL);
 197         ip = VTOI(ap->a_vp);
 198         error = 0;
 199         seqcount = ap->a_ioflag >> 16;
 200
 201         hammer_start_transaction(&trans, ip->hmp);
 202
 203         /*
 204          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 205          */
 206         uio = ap->a_uio;
 207         while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
 208                 offset = uio->uio_offset & HAMMER_BUFMASK;
 209 #if 0
 210                 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
 211                                      uio->uio_offset - offset, HAMMER_BUFSIZE,
 212                                      MAXBSIZE, seqcount, &bp);
 213 #endif
 214                 error = bread(ap->a_vp, uio->uio_offset - offset,
 215                               HAMMER_BUFSIZE, &bp);
 216                 if (error) {
 217                         brelse(bp);
 218                         break;
 219                 }
 220                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 221                 n = HAMMER_BUFSIZE - offset;
 222                 if (n > uio->uio_resid)
 223                         n = uio->uio_resid;
 224                 if (n > ip->ino_rec.ino_size - uio->uio_offset)
 225                         n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
 226                 error = uiomove((char *)bp->b_data + offset, n, uio);
 227                 if (error) {
 228                         bqrelse(bp);
 229                         break;
 230                 }
 231                 if ((ip->flags & HAMMER_INODE_RO) == 0) {
 232                         ip->ino_rec.ino_atime = trans.tid;
 233                         hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
 234                 }
 235                 bqrelse(bp);
 236         }
 237         hammer_commit_transaction(&trans);
 238         return (error);
 239 }
 240
 241 /*
 242  * hammer_vop_write { vp, uio, ioflag, cred }
 243  */
 244 static
 245 int
 246 hammer_vop_write(struct vop_write_args *ap)
 247 {
 248         struct hammer_transaction trans;
 249         struct hammer_inode *ip;
 250         struct uio *uio;
 251         off_t offset;
 252         struct buf *bp;
 253         int error;
 254         int n;
 255         int flags;
 256
 257         if (ap->a_vp->v_type != VREG)
 258                 return (EINVAL);
 259         ip = VTOI(ap->a_vp);
 260         error = 0;
 261
 262         if (ip->flags & HAMMER_INODE_RO)
 263                 return (EROFS);
 264
 265         /*
 266          * Create a transaction to cover the operations we perform.
 267          */
 268         hammer_start_transaction(&trans, ip->hmp);
 269         uio = ap->a_uio;
 270
 271         /*
 272          * Check append mode
 273          */
 274         if (ap->a_ioflag & IO_APPEND)
 275                 uio->uio_offset = ip->ino_rec.ino_size;
 276
 277         /*
 278          * Check for illegal write offsets.  Valid range is 0...2^63-1
 279          */
 280         if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
 281                 hammer_commit_transaction(&trans);
 282                 return (EFBIG);
 283         }
 284
 285         /*
 286          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 287          */
 288         while (uio->uio_resid > 0) {
 289                 int fixsize = 0;
 290
 291                 offset = uio->uio_offset & HAMMER_BUFMASK;
 292                 n = HAMMER_BUFSIZE - offset;
 293                 if (n > uio->uio_resid)
 294                         n = uio->uio_resid;
 295                 if (uio->uio_offset + n > ip->ino_rec.ino_size) {
 296                         vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
 297                         fixsize = 1;
 298                 }
 299
 300                 if (uio->uio_segflg == UIO_NOCOPY) {
 301                         /*
 302                          * Issuing a write with the same data backing the
 303                          * buffer.  Instantiate the buffer to collect the
 304                          * backing vm pages, then read-in any missing bits.
 305                          *
 306                          * This case is used by vop_stdputpages().
 307                          */
 308                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 309                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 310                         if ((bp->b_flags & B_CACHE) == 0) {
 311                                 bqrelse(bp);
 312                                 error = bread(ap->a_vp,
 313                                               uio->uio_offset - offset,
 314                                               HAMMER_BUFSIZE, &bp);
 315                         }
 316                 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
 317                         /*
 318                          * entirely overwrite the buffer
 319                          */
 320                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 321                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 322                 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
 323                         /*
 324                          * XXX
 325                          */
 326                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 327                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 328                         vfs_bio_clrbuf(bp);
 329                 } else {
 330                         /*
 331                          * Partial overwrite, read in any missing bits then
 332                          * replace the portion being written.
 333                          */
 334                         error = bread(ap->a_vp, uio->uio_offset - offset,
 335                                       HAMMER_BUFSIZE, &bp);
 336                         if (error == 0)
 337                                 bheavy(bp);
 338                 }
 339                 if (error == 0)
 340                         error = uiomove((char *)bp->b_data + offset, n, uio);
 341
 342                 /*
 343                  * If we screwed up we have to undo any VM size changes we
 344                  * made.
 345                  */
 346                 if (error) {
 347                         brelse(bp);
 348                         if (fixsize) {
 349                                 vtruncbuf(ap->a_vp, ip->ino_rec.ino_size,
 350                                           HAMMER_BUFSIZE);
 351                         }
 352                         break;
 353                 }
 354                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 355                 if (ip->ino_rec.ino_size < uio->uio_offset) {
 356                         ip->ino_rec.ino_size = uio->uio_offset;
 357                         flags = HAMMER_INODE_RDIRTY;
 358                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
 359                 } else {
 360                         flags = 0;
 361                 }
 362                 ip->ino_rec.ino_mtime = trans.tid;
 363                 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
 364                 hammer_modify_inode(&trans, ip, flags);
 365
 366                 /*
 367                  * The file write must be tagged with the same TID as the
 368                  * inode, for consistency in case the inode changed size.
 369                  * This guarantees the on-disk data records will have a
 370                  * TID <= the inode TID representing the size change.
 371                  *
 372                  * If a prior write has not yet flushed, retain its TID.
 373                  */
 374                 if (bp->b_tid == 0)
 375                         bp->b_tid = ip->last_tid;
 376
 377                 if (ap->a_ioflag & IO_SYNC) {
 378                         bwrite(bp);
 379                 } else if (ap->a_ioflag & IO_DIRECT) {
 380                         bawrite(bp);
 381                 } else {
 382                         bdwrite(bp);
 383                 }
 384         }
 385         if (error)
 386                 hammer_abort_transaction(&trans);
 387         else
 388                 hammer_commit_transaction(&trans);
 389         return (error);
 390 }
 391
 392 /*
 393  * hammer_vop_access { vp, mode, cred }
 394  */
 395 static
 396 int
 397 hammer_vop_access(struct vop_access_args *ap)
 398 {
 399         struct hammer_inode *ip = VTOI(ap->a_vp);
 400         uid_t uid;
 401         gid_t gid;
 402         int error;
 403
 404         uid = hammer_to_unix_xid(&ip->ino_data.uid);
 405         gid = hammer_to_unix_xid(&ip->ino_data.gid);
 406
 407         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
 408                                   ip->ino_data.uflags);
 409         return (error);
 410 }
 411
 412 /*
 413  * hammer_vop_advlock { vp, id, op, fl, flags }
 414  */
 415 static
 416 int
 417 hammer_vop_advlock(struct vop_advlock_args *ap)
 418 {
 419         struct hammer_inode *ip = VTOI(ap->a_vp);
 420
 421         return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
 422 }
 423
 424 /*
 425  * hammer_vop_close { vp, fflag }
 426  */
 427 static
 428 int
 429 hammer_vop_close(struct vop_close_args *ap)
 430 {
 431         return (vop_stdclose(ap));
 432 }
 433
 434 /*
 435  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
 436  *
 437  * The operating system has already ensured that the directory entry
 438  * does not exist and done all appropriate namespace locking.
 439  */
 440 static
 441 int
 442 hammer_vop_ncreate(struct vop_ncreate_args *ap)
 443 {
 444         struct hammer_transaction trans;
 445         struct hammer_inode *dip;
 446         struct hammer_inode *nip;
 447         struct nchandle *nch;
 448         int error;
 449
 450         nch = ap->a_nch;
 451         dip = VTOI(ap->a_dvp);
 452
 453         if (dip->flags & HAMMER_INODE_RO)
 454                 return (EROFS);
 455
 456         /*
 457          * Create a transaction to cover the operations we perform.
 458          */
 459         hammer_start_transaction(&trans, dip->hmp);
 460
 461         /*
 462          * Create a new filesystem object of the requested type.  The
 463          * returned inode will be referenced but not locked.
 464          */
 465
 466         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 467         if (error)
 468                 kprintf("hammer_create_inode error %d\n", error);
 469         if (error) {
 470                 hammer_abort_transaction(&trans);
 471                 *ap->a_vpp = NULL;
 472                 return (error);
 473         }
 474
 475         /*
 476          * Add the new filesystem object to the directory.  This will also
 477          * bump the inode's link count.
 478          */
 479         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 480         if (error)
 481                 kprintf("hammer_ip_add_directory error %d\n", error);
 482
 483         /*
 484          * Finish up.
 485          */
 486         if (error) {
 487                 hammer_rel_inode(nip, 0);
 488                 hammer_abort_transaction(&trans);
 489                 *ap->a_vpp = NULL;
 490         } else {
 491                 hammer_commit_transaction(&trans);
 492                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 493                 hammer_rel_inode(nip, 0);
 494                 if (error == 0) {
 495                         cache_setunresolved(ap->a_nch);
 496                         cache_setvp(ap->a_nch, *ap->a_vpp);
 497                 }
 498         }
 499         return (error);
 500 }
 501
 502 /*
 503  * hammer_vop_getattr { vp, vap }
 504  */
 505 static
 506 int
 507 hammer_vop_getattr(struct vop_getattr_args *ap)
 508 {
 509         struct hammer_inode *ip = VTOI(ap->a_vp);
 510         struct vattr *vap = ap->a_vap;
 511
 512 #if 0
 513         if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
 514             (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
 515             ip->obj_asof == XXX
 516         ) {
 517                 /* LAZYMOD XXX */
 518         }
 519         hammer_itimes(ap->a_vp);
 520 #endif
 521
 522         vap->va_fsid = ip->hmp->fsid_udev;
 523         vap->va_fileid = ip->ino_rec.base.base.obj_id;
 524         vap->va_mode = ip->ino_data.mode;
 525         vap->va_nlink = ip->ino_rec.ino_nlinks;
 526         vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
 527         vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
 528         vap->va_rmajor = 0;
 529         vap->va_rminor = 0;
 530         vap->va_size = ip->ino_rec.ino_size;
 531         hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
 532         hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
 533         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
 534         vap->va_flags = ip->ino_data.uflags;
 535         vap->va_gen = 1;        /* hammer inums are unique for all time */
 536         vap->va_blocksize = 32768; /* XXX - extract from root volume */
 537         vap->va_bytes = ip->ino_rec.ino_size;
 538         vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
 539         vap->va_filerev = 0;    /* XXX */
 540         /* mtime uniquely identifies any adjustments made to the file */
 541         vap->va_fsmid = ip->ino_rec.ino_mtime;
 542         vap->va_uid_uuid = ip->ino_data.uid;
 543         vap->va_gid_uuid = ip->ino_data.gid;
 544         vap->va_fsid_uuid = ip->hmp->fsid;
 545         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 546                           VA_FSID_UUID_VALID;
 547
 548         switch (ip->ino_rec.base.base.obj_type) {
 549         case HAMMER_OBJTYPE_CDEV:
 550         case HAMMER_OBJTYPE_BDEV:
 551                 vap->va_rmajor = ip->ino_data.rmajor;
 552                 vap->va_rminor = ip->ino_data.rminor;
 553                 break;
 554         default:
 555                 break;
 556         }
 557
 558         return(0);
 559 }
 560
 561 /*
 562  * hammer_vop_nresolve { nch, dvp, cred }
 563  *
 564  * Locate the requested directory entry.
 565  */
 566 static
 567 int
 568 hammer_vop_nresolve(struct vop_nresolve_args *ap)
 569 {
 570         struct namecache *ncp;
 571         hammer_inode_t dip;
 572         hammer_inode_t ip;
 573         hammer_tid_t asof;
 574         struct hammer_cursor cursor;
 575         union hammer_record_ondisk *rec;
 576         struct vnode *vp;
 577         int64_t namekey;
 578         int error;
 579         int i;
 580         int nlen;
 581         int flags;
 582         u_int64_t obj_id;
 583
 584         /*
 585          * Misc initialization, plus handle as-of name extensions.  Look for
 586          * the '@@' extension.  Note that as-of files and directories cannot
 587          * be modified.
 588          */
 589         dip = VTOI(ap->a_dvp);
 590         ncp = ap->a_nch->ncp;
 591         asof = dip->obj_asof;
 592         nlen = ncp->nc_nlen;
 593         flags = dip->flags;
 594
 595         for (i = 0; i < nlen; ++i) {
 596                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
 597                         asof = hammer_str_to_tid(ncp->nc_name + i + 2);
 598                         flags |= HAMMER_INODE_RO;
 599                         break;
 600                 }
 601         }
 602         nlen = i;
 603
 604         /*
 605          * If there is no path component the time extension is relative to
 606          * dip.
 607          */
 608         if (nlen == 0) {
 609                 ip = hammer_get_inode(dip->hmp, &dip->cache[1], dip->obj_id,
 610                                       asof, flags, &error);
 611                 if (error == 0) {
 612                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 613                         hammer_rel_inode(ip, 0);
 614                 } else {
 615                         vp = NULL;
 616                 }
 617                 if (error == 0) {
 618                         vn_unlock(vp);
 619                         cache_setvp(ap->a_nch, vp);
 620                         vrele(vp);
 621                 }
 622                 return(error);
 623         }
 624
 625         /*
 626          * Calculate the namekey and setup the key range for the scan.  This
 627          * works kinda like a chained hash table where the lower 32 bits
 628          * of the namekey synthesize the chain.
 629          *
 630          * The key range is inclusive of both key_beg and key_end.
 631          */
 632         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
 633
 634         error = hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
 635         cursor.key_beg.obj_id = dip->obj_id;
 636         cursor.key_beg.key = namekey;
 637         cursor.key_beg.create_tid = 0;
 638         cursor.key_beg.delete_tid = 0;
 639         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 640         cursor.key_beg.obj_type = 0;
 641
 642         cursor.key_end = cursor.key_beg;
 643         cursor.key_end.key |= 0xFFFFFFFFULL;
 644         cursor.asof = asof;
 645         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
 646
 647         /*
 648          * Scan all matching records (the chain), locate the one matching
 649          * the requested path component.
 650          *
 651          * The hammer_ip_*() functions merge in-memory records with on-disk
 652          * records for the purposes of the search.
 653          */
 654         if (error == 0)
 655                 error = hammer_ip_first(&cursor, dip);
 656
 657         rec = NULL;
 658         obj_id = 0;
 659
 660         while (error == 0) {
 661                 error = hammer_ip_resolve_data(&cursor);
 662                 if (error)
 663                         break;
 664                 rec = cursor.record;
 665                 if (nlen == rec->entry.base.data_len &&
 666                     bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
 667                         obj_id = rec->entry.obj_id;
 668                         break;
 669                 }
 670                 error = hammer_ip_next(&cursor);
 671         }
 672         hammer_done_cursor(&cursor);
 673         if (error == 0) {
 674                 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
 675                                       obj_id, asof, flags, &error);
 676                 if (error == 0) {
 677                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 678                         hammer_rel_inode(ip, 0);
 679                 } else {
 680                         vp = NULL;
 681                 }
 682                 if (error == 0) {
 683                         vn_unlock(vp);
 684                         cache_setvp(ap->a_nch, vp);
 685                         vrele(vp);
 686                 }
 687         } else if (error == ENOENT) {
 688                 cache_setvp(ap->a_nch, NULL);
 689         }
 690         return (error);
 691 }
 692
 693 /*
 694  * hammer_vop_nlookupdotdot { dvp, vpp, cred }
 695  *
 696  * Locate the parent directory of a directory vnode.
 697  *
 698  * dvp is referenced but not locked.  *vpp must be returned referenced and
 699  * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
 700  * at the root, instead it could indicate that the directory we were in was
 701  * removed.
 702  */
 703 static
 704 int
 705 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
 706 {
 707         struct hammer_inode *dip;
 708         struct hammer_inode *ip;
 709         u_int64_t parent_obj_id;
 710         int error;
 711
 712         dip = VTOI(ap->a_dvp);
 713         if ((parent_obj_id = dip->ino_data.parent_obj_id) == 0) {
 714                 *ap->a_vpp = NULL;
 715                 return ENOENT;
 716         }
 717
 718         ip = hammer_get_inode(dip->hmp, &dip->cache[1], parent_obj_id,
 719                               dip->obj_asof, dip->flags, &error);
 720         if (ip == NULL) {
 721                 *ap->a_vpp = NULL;
 722                 return(error);
 723         }
 724         error = hammer_get_vnode(ip, LK_EXCLUSIVE, ap->a_vpp);
 725         hammer_rel_inode(ip, 0);
 726         return (error);
 727 }
 728
 729 /*
 730  * hammer_vop_nlink { nch, dvp, vp, cred }
 731  */
 732 static
 733 int
 734 hammer_vop_nlink(struct vop_nlink_args *ap)
 735 {
 736         struct hammer_transaction trans;
 737         struct hammer_inode *dip;
 738         struct hammer_inode *ip;
 739         struct nchandle *nch;
 740         int error;
 741
 742         nch = ap->a_nch;
 743         dip = VTOI(ap->a_dvp);
 744         ip = VTOI(ap->a_vp);
 745
 746         if (dip->flags & HAMMER_INODE_RO)
 747                 return (EROFS);
 748         if (ip->flags & HAMMER_INODE_RO)
 749                 return (EROFS);
 750
 751         /*
 752          * Create a transaction to cover the operations we perform.
 753          */
 754         hammer_start_transaction(&trans, dip->hmp);
 755
 756         /*
 757          * Add the filesystem object to the directory.  Note that neither
 758          * dip nor ip are referenced or locked, but their vnodes are
 759          * referenced.  This function will bump the inode's link count.
 760          */
 761         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 762
 763         /*
 764          * Finish up.
 765          */
 766         if (error) {
 767                 hammer_abort_transaction(&trans);
 768         } else {
 769                 cache_setunresolved(nch);
 770                 cache_setvp(nch, ap->a_vp);
 771                 hammer_commit_transaction(&trans);
 772         }
 773         return (error);
 774 }
 775
 776 /*
 777  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 778  *
 779  * The operating system has already ensured that the directory entry
 780  * does not exist and done all appropriate namespace locking.
 781  */
 782 static
 783 int
 784 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 785 {
 786         struct hammer_transaction trans;
 787         struct hammer_inode *dip;
 788         struct hammer_inode *nip;
 789         struct nchandle *nch;
 790         int error;
 791
 792         nch = ap->a_nch;
 793         dip = VTOI(ap->a_dvp);
 794
 795         if (dip->flags & HAMMER_INODE_RO)
 796                 return (EROFS);
 797
 798         /*
 799          * Create a transaction to cover the operations we perform.
 800          */
 801         hammer_start_transaction(&trans, dip->hmp);
 802
 803         /*
 804          * Create a new filesystem object of the requested type.  The
 805          * returned inode will be referenced but not locked.
 806          */
 807         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 808         if (error)
 809                 kprintf("hammer_mkdir error %d\n", error);
 810         if (error) {
 811                 hammer_abort_transaction(&trans);
 812                 *ap->a_vpp = NULL;
 813                 return (error);
 814         }
 815
 816         /*
 817          * Add the new filesystem object to the directory.  This will also
 818          * bump the inode's link count.
 819          */
 820         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 821         if (error)
 822                 kprintf("hammer_mkdir (add) error %d\n", error);
 823
 824         /*
 825          * Finish up.
 826          */
 827         if (error) {
 828                 hammer_rel_inode(nip, 0);
 829                 hammer_abort_transaction(&trans);
 830                 *ap->a_vpp = NULL;
 831         } else {
 832                 hammer_commit_transaction(&trans);
 833                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 834                 hammer_rel_inode(nip, 0);
 835                 if (error == 0) {
 836                         cache_setunresolved(ap->a_nch);
 837                         cache_setvp(ap->a_nch, *ap->a_vpp);
 838                 }
 839         }
 840         return (error);
 841 }
 842
 843 /*
 844  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 845  *
 846  * The operating system has already ensured that the directory entry
 847  * does not exist and done all appropriate namespace locking.
 848  */
 849 static
 850 int
 851 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 852 {
 853         struct hammer_transaction trans;
 854         struct hammer_inode *dip;
 855         struct hammer_inode *nip;
 856         struct nchandle *nch;
 857         int error;
 858
 859         nch = ap->a_nch;
 860         dip = VTOI(ap->a_dvp);
 861
 862         if (dip->flags & HAMMER_INODE_RO)
 863                 return (EROFS);
 864
 865         /*
 866          * Create a transaction to cover the operations we perform.
 867          */
 868         hammer_start_transaction(&trans, dip->hmp);
 869
 870         /*
 871          * Create a new filesystem object of the requested type.  The
 872          * returned inode will be referenced but not locked.
 873          */
 874         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 875         if (error) {
 876                 hammer_abort_transaction(&trans);
 877                 *ap->a_vpp = NULL;
 878                 return (error);
 879         }
 880
 881         /*
 882          * Add the new filesystem object to the directory.  This will also
 883          * bump the inode's link count.
 884          */
 885         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 886
 887         /*
 888          * Finish up.
 889          */
 890         if (error) {
 891                 hammer_rel_inode(nip, 0);
 892                 hammer_abort_transaction(&trans);
 893                 *ap->a_vpp = NULL;
 894         } else {
 895                 hammer_commit_transaction(&trans);
 896                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 897                 hammer_rel_inode(nip, 0);
 898                 if (error == 0) {
 899                         cache_setunresolved(ap->a_nch);
 900                         cache_setvp(ap->a_nch, *ap->a_vpp);
 901                 }
 902         }
 903         return (error);
 904 }
 905
 906 /*
 907  * hammer_vop_open { vp, mode, cred, fp }
 908  */
 909 static
 910 int
 911 hammer_vop_open(struct vop_open_args *ap)
 912 {
 913         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
 914                 return (EROFS);
 915
 916         return(vop_stdopen(ap));
 917 }
 918
 919 /*
 920  * hammer_vop_pathconf { vp, name, retval }
 921  */
 922 static
 923 int
 924 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 925 {
 926         return EOPNOTSUPP;
 927 }
 928
 929 /*
 930  * hammer_vop_print { vp }
 931  */
 932 static
 933 int
 934 hammer_vop_print(struct vop_print_args *ap)
 935 {
 936         return EOPNOTSUPP;
 937 }
 938
 939 /*
 940  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 941  */
 942 static
 943 int
 944 hammer_vop_readdir(struct vop_readdir_args *ap)
 945 {
 946         struct hammer_cursor cursor;
 947         struct hammer_inode *ip;
 948         struct uio *uio;
 949         hammer_record_ondisk_t rec;
 950         hammer_base_elm_t base;
 951         int error;
 952         int cookie_index;
 953         int ncookies;
 954         off_t *cookies;
 955         off_t saveoff;
 956         int r;
 957
 958         ip = VTOI(ap->a_vp);
 959         uio = ap->a_uio;
 960         saveoff = uio->uio_offset;
 961
 962         if (ap->a_ncookies) {
 963                 ncookies = uio->uio_resid / 16 + 1;
 964                 if (ncookies > 1024)
 965                         ncookies = 1024;
 966                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 967                 cookie_index = 0;
 968         } else {
 969                 ncookies = -1;
 970                 cookies = NULL;
 971                 cookie_index = 0;
 972         }
 973
 974         /*
 975          * Handle artificial entries
 976          */
 977         error = 0;
 978         if (saveoff == 0) {
 979                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
 980                 if (r)
 981                         goto done;
 982                 if (cookies)
 983                         cookies[cookie_index] = saveoff;
 984                 ++saveoff;
 985                 ++cookie_index;
 986                 if (cookie_index == ncookies)
 987                         goto done;
 988         }
 989         if (saveoff == 1) {
 990                 if (ip->ino_data.parent_obj_id) {
 991                         r = vop_write_dirent(&error, uio,
 992                                              ip->ino_data.parent_obj_id,
 993                                              DT_DIR, 2, "..");
 994                 } else {
 995                         r = vop_write_dirent(&error, uio,
 996                                              ip->obj_id, DT_DIR, 2, "..");
 997                 }
 998                 if (r)
 999                         goto done;
1000                 if (cookies)
1001                         cookies[cookie_index] = saveoff;
1002                 ++saveoff;
1003                 ++cookie_index;
1004                 if (cookie_index == ncookies)
1005                         goto done;
1006         }
1007
1008         /*
1009          * Key range (begin and end inclusive) to scan.  Directory keys
1010          * directly translate to a 64 bit 'seek' position.
1011          */
1012         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1013         cursor.key_beg.obj_id = ip->obj_id;
1014         cursor.key_beg.create_tid = 0;
1015         cursor.key_beg.delete_tid = 0;
1016         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1017         cursor.key_beg.obj_type = 0;
1018         cursor.key_beg.key = saveoff;
1019
1020         cursor.key_end = cursor.key_beg;
1021         cursor.key_end.key = HAMMER_MAX_KEY;
1022         cursor.asof = ip->obj_asof;
1023         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1024
1025         error = hammer_ip_first(&cursor, ip);
1026
1027         while (error == 0) {
1028                 error = hammer_ip_resolve_data(&cursor);
1029                 if (error)
1030                         break;
1031                 rec = cursor.record;
1032                 base = &rec->base.base;
1033                 saveoff = base->key;
1034
1035                 if (base->obj_id != ip->obj_id)
1036                         panic("readdir: bad record at %p", cursor.node);
1037
1038                 r = vop_write_dirent(
1039                              &error, uio, rec->entry.obj_id,
1040                              hammer_get_dtype(rec->entry.base.base.obj_type),
1041                              rec->entry.base.data_len,
1042                              (void *)cursor.data);
1043                 if (r)
1044                         break;
1045                 ++saveoff;
1046                 if (cookies)
1047                         cookies[cookie_index] = base->key;
1048                 ++cookie_index;
1049                 if (cookie_index == ncookies)
1050                         break;
1051                 error = hammer_ip_next(&cursor);
1052         }
1053         hammer_done_cursor(&cursor);
1054
1055 done:
1056         if (ap->a_eofflag)
1057                 *ap->a_eofflag = (error == ENOENT);
1058         uio->uio_offset = saveoff;
1059         if (error && cookie_index == 0) {
1060                 if (error == ENOENT)
1061                         error = 0;
1062                 if (cookies) {
1063                         kfree(cookies, M_TEMP);
1064                         *ap->a_ncookies = 0;
1065                         *ap->a_cookies = NULL;
1066                 }
1067         } else {
1068                 if (error == ENOENT)
1069                         error = 0;
1070                 if (cookies) {
1071                         *ap->a_ncookies = cookie_index;
1072                         *ap->a_cookies = cookies;
1073                 }
1074         }
1075         return(error);
1076 }
1077
1078 /*
1079  * hammer_vop_readlink { vp, uio, cred }
1080  */
1081 static
1082 int
1083 hammer_vop_readlink(struct vop_readlink_args *ap)
1084 {
1085         struct hammer_cursor cursor;
1086         struct hammer_inode *ip;
1087         int error;
1088
1089         ip = VTOI(ap->a_vp);
1090         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1091
1092         /*
1093          * Key range (begin and end inclusive) to scan.  Directory keys
1094          * directly translate to a 64 bit 'seek' position.
1095          */
1096         cursor.key_beg.obj_id = ip->obj_id;
1097         cursor.key_beg.create_tid = 0;
1098         cursor.key_beg.delete_tid = 0;
1099         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1100         cursor.key_beg.obj_type = 0;
1101         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1102         cursor.asof = ip->obj_asof;
1103         cursor.flags |= HAMMER_CURSOR_ASOF;
1104
1105         error = hammer_ip_lookup(&cursor, ip);
1106         if (error == 0) {
1107                 error = hammer_ip_resolve_data(&cursor);
1108                 if (error == 0) {
1109                         error = uiomove((char *)cursor.data,
1110                                         cursor.record->generic.base.data_len,
1111                                         ap->a_uio);
1112                 }
1113         }
1114         hammer_done_cursor(&cursor);
1115         return(error);
1116 }
1117
1118 /*
1119  * hammer_vop_nremove { nch, dvp, cred }
1120  */
1121 static
1122 int
1123 hammer_vop_nremove(struct vop_nremove_args *ap)
1124 {
1125         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1126 }
1127
1128 /*
1129  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1130  */
1131 static
1132 int
1133 hammer_vop_nrename(struct vop_nrename_args *ap)
1134 {
1135         struct hammer_transaction trans;
1136         struct namecache *fncp;
1137         struct namecache *tncp;
1138         struct hammer_inode *fdip;
1139         struct hammer_inode *tdip;
1140         struct hammer_inode *ip;
1141         struct hammer_cursor cursor;
1142         union hammer_record_ondisk *rec;
1143         int64_t namekey;
1144         int error;
1145
1146         fdip = VTOI(ap->a_fdvp);
1147         tdip = VTOI(ap->a_tdvp);
1148         fncp = ap->a_fnch->ncp;
1149         tncp = ap->a_tnch->ncp;
1150         ip = VTOI(fncp->nc_vp);
1151         KKASSERT(ip != NULL);
1152
1153         if (fdip->flags & HAMMER_INODE_RO)
1154                 return (EROFS);
1155         if (tdip->flags & HAMMER_INODE_RO)
1156                 return (EROFS);
1157         if (ip->flags & HAMMER_INODE_RO)
1158                 return (EROFS);
1159
1160         hammer_start_transaction(&trans, fdip->hmp);
1161
1162         /*
1163          * Remove tncp from the target directory and then link ip as
1164          * tncp. XXX pass trans to dounlink
1165          */
1166         error = hammer_dounlink(ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1167         if (error == 0 || error == ENOENT)
1168                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1169         if (error)
1170                 goto failed; /* XXX */
1171
1172         /*
1173          * Locate the record in the originating directory and remove it.
1174          *
1175          * Calculate the namekey and setup the key range for the scan.  This
1176          * works kinda like a chained hash table where the lower 32 bits
1177          * of the namekey synthesize the chain.
1178          *
1179          * The key range is inclusive of both key_beg and key_end.
1180          */
1181         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1182 retry:
1183         hammer_init_cursor_hmp(&cursor, &fdip->cache[0], fdip->hmp);
1184         cursor.key_beg.obj_id = fdip->obj_id;
1185         cursor.key_beg.key = namekey;
1186         cursor.key_beg.create_tid = 0;
1187         cursor.key_beg.delete_tid = 0;
1188         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1189         cursor.key_beg.obj_type = 0;
1190
1191         cursor.key_end = cursor.key_beg;
1192         cursor.key_end.key |= 0xFFFFFFFFULL;
1193         cursor.asof = fdip->obj_asof;
1194         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1195
1196         /*
1197          * Scan all matching records (the chain), locate the one matching
1198          * the requested path component.
1199          *
1200          * The hammer_ip_*() functions merge in-memory records with on-disk
1201          * records for the purposes of the search.
1202          */
1203         error = hammer_ip_first(&cursor, fdip);
1204         while (error == 0) {
1205                 if (hammer_ip_resolve_data(&cursor) != 0)
1206                         break;
1207                 rec = cursor.record;
1208                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1209                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1210                         break;
1211                 }
1212                 error = hammer_ip_next(&cursor);
1213         }
1214
1215         /*
1216          * If all is ok we have to get the inode so we can adjust nlinks.
1217          *
1218          * WARNING: hammer_ip_del_directory() may have to terminate the
1219          * cursor to avoid a recursion.  It's ok to call hammer_done_cursor()
1220          * twice.
1221          */
1222         if (error == 0)
1223                 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1224         hammer_done_cursor(&cursor);
1225         if (error == 0)
1226                 cache_rename(ap->a_fnch, ap->a_tnch);
1227         if (error == EDEADLK)
1228                 goto retry;
1229 failed:
1230         if (error == 0) {
1231                 hammer_commit_transaction(&trans);
1232         } else {
1233                 hammer_abort_transaction(&trans);
1234         }
1235         return (error);
1236 }
1237
1238 /*
1239  * hammer_vop_nrmdir { nch, dvp, cred }
1240  */
1241 static
1242 int
1243 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1244 {
1245         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1246 }
1247
1248 /*
1249  * hammer_vop_setattr { vp, vap, cred }
1250  */
1251 static
1252 int
1253 hammer_vop_setattr(struct vop_setattr_args *ap)
1254 {
1255         struct hammer_transaction trans;
1256         struct hammer_cursor *spike = NULL;
1257         struct vattr *vap;
1258         struct hammer_inode *ip;
1259         int modflags;
1260         int error;
1261         int truncating;
1262         int64_t aligned_size;
1263         u_int32_t flags;
1264         uuid_t uuid;
1265
1266         vap = ap->a_vap;
1267         ip = ap->a_vp->v_data;
1268         modflags = 0;
1269
1270         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1271                 return(EROFS);
1272         if (ip->flags & HAMMER_INODE_RO)
1273                 return (EROFS);
1274
1275         hammer_start_transaction(&trans, ip->hmp);
1276         error = 0;
1277
1278         if (vap->va_flags != VNOVAL) {
1279                 flags = ip->ino_data.uflags;
1280                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1281                                          hammer_to_unix_xid(&ip->ino_data.uid),
1282                                          ap->a_cred);
1283                 if (error == 0) {
1284                         if (ip->ino_data.uflags != flags) {
1285                                 ip->ino_data.uflags = flags;
1286                                 modflags |= HAMMER_INODE_DDIRTY;
1287                         }
1288                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1289                                 error = 0;
1290                                 goto done;
1291                         }
1292                 }
1293                 goto done;
1294         }
1295         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1296                 error = EPERM;
1297                 goto done;
1298         }
1299         if (vap->va_uid != (uid_t)VNOVAL) {
1300                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1301                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1302                         ip->ino_data.uid = uuid;
1303                         modflags |= HAMMER_INODE_DDIRTY;
1304                 }
1305         }
1306         if (vap->va_gid != (uid_t)VNOVAL) {
1307                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1308                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1309                         ip->ino_data.gid = uuid;
1310                         modflags |= HAMMER_INODE_DDIRTY;
1311                 }
1312         }
1313         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1314                 switch(ap->a_vp->v_type) {
1315                 case VREG:
1316                         if (vap->va_size == ip->ino_rec.ino_size)
1317                                 break;
1318                         if (vap->va_size < ip->ino_rec.ino_size) {
1319                                 vtruncbuf(ap->a_vp, vap->va_size,
1320                                           HAMMER_BUFSIZE);
1321                                 truncating = 1;
1322                         } else {
1323                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1324                                 truncating = 0;
1325                         }
1326                         ip->ino_rec.ino_size = vap->va_size;
1327                         modflags |= HAMMER_INODE_RDIRTY;
1328                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1329                                         ~(int64_t)HAMMER_BUFMASK;
1330
1331                         if (truncating) {
1332                                 error = hammer_ip_delete_range(&trans, ip,
1333                                                     aligned_size,
1334                                                     0x7FFFFFFFFFFFFFFFLL,
1335                                                     &spike);
1336                         }
1337                         /*
1338                          * If truncating we have to clean out a portion of
1339                          * the last block on-disk.
1340                          */
1341                         if (truncating && error == 0 &&
1342                             vap->va_size < aligned_size) {
1343                                 struct buf *bp;
1344                                 int offset;
1345
1346                                 offset = vap->va_size & HAMMER_BUFMASK;
1347                                 error = bread(ap->a_vp,
1348                                               aligned_size - HAMMER_BUFSIZE,
1349                                               HAMMER_BUFSIZE, &bp);
1350                                 if (error == 0) {
1351                                         bzero(bp->b_data + offset,
1352                                               HAMMER_BUFSIZE - offset);
1353                                         bdwrite(bp);
1354                                 } else {
1355                                         brelse(bp);
1356                                 }
1357                         }
1358                         break;
1359                 case VDATABASE:
1360                         error = hammer_ip_delete_range(&trans, ip,
1361                                                     vap->va_size,
1362                                                     0x7FFFFFFFFFFFFFFFLL,
1363                                                     &spike);
1364                         ip->ino_rec.ino_size = vap->va_size;
1365                         modflags |= HAMMER_INODE_RDIRTY;
1366                         break;
1367                 default:
1368                         error = EINVAL;
1369                         goto done;
1370                 }
1371                 if (error == ENOSPC) {
1372                         error = hammer_spike(&spike);
1373                         if (error == 0)
1374                                 continue;
1375                 }
1376                 KKASSERT(spike == NULL);
1377                 break;
1378         }
1379         if (vap->va_atime.tv_sec != VNOVAL) {
1380                 ip->ino_rec.ino_atime =
1381                         hammer_timespec_to_transid(&vap->va_atime);
1382                 modflags |= HAMMER_INODE_ITIMES;
1383         }
1384         if (vap->va_mtime.tv_sec != VNOVAL) {
1385                 ip->ino_rec.ino_mtime =
1386                         hammer_timespec_to_transid(&vap->va_mtime);
1387                 modflags |= HAMMER_INODE_ITIMES;
1388         }
1389         if (vap->va_mode != (mode_t)VNOVAL) {
1390                 if (ip->ino_data.mode != vap->va_mode) {
1391                         ip->ino_data.mode = vap->va_mode;
1392                         modflags |= HAMMER_INODE_DDIRTY;
1393                 }
1394         }
1395 done:
1396         if (error) {
1397                 hammer_abort_transaction(&trans);
1398         } else {
1399                 hammer_modify_inode(&trans, ip, modflags);
1400                 hammer_commit_transaction(&trans);
1401         }
1402         return (error);
1403 }
1404
1405 /*
1406  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1407  */
1408 static
1409 int
1410 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1411 {
1412         struct hammer_transaction trans;
1413         struct hammer_inode *dip;
1414         struct hammer_inode *nip;
1415         struct nchandle *nch;
1416         hammer_record_t record;
1417         int error;
1418         int bytes;
1419
1420         ap->a_vap->va_type = VLNK;
1421
1422         nch = ap->a_nch;
1423         dip = VTOI(ap->a_dvp);
1424
1425         if (dip->flags & HAMMER_INODE_RO)
1426                 return (EROFS);
1427
1428         /*
1429          * Create a transaction to cover the operations we perform.
1430          */
1431         hammer_start_transaction(&trans, dip->hmp);
1432
1433         /*
1434          * Create a new filesystem object of the requested type.  The
1435          * returned inode will be referenced but not locked.
1436          */
1437
1438         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1439         if (error) {
1440                 hammer_abort_transaction(&trans);
1441                 *ap->a_vpp = NULL;
1442                 return (error);
1443         }
1444
1445         /*
1446          * Add the new filesystem object to the directory.  This will also
1447          * bump the inode's link count.
1448          */
1449         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1450
1451         /*
1452          * Add a record representing the symlink.  symlink stores the link
1453          * as pure data, not a string, and is no \0 terminated.
1454          */
1455         if (error == 0) {
1456                 record = hammer_alloc_mem_record(nip);
1457                 bytes = strlen(ap->a_target);
1458
1459                 record->rec.generic.base.base.key = HAMMER_FIXKEY_SYMLINK;
1460                 record->rec.generic.base.base.rec_type = HAMMER_RECTYPE_FIX;
1461                 record->rec.generic.base.data_len = bytes;
1462                 if (bytes <= sizeof(record->rec.generic.filler)) {
1463                         record->data = (void *)record->rec.generic.filler;
1464                         bcopy(ap->a_target, record->data, bytes);
1465                 } else {
1466                         record->data = (void *)ap->a_target;
1467                         /* will be reallocated by routine below */
1468                 }
1469                 error = hammer_ip_add_record(&trans, record);
1470         }
1471
1472         /*
1473          * Finish up.
1474          */
1475         if (error) {
1476                 hammer_rel_inode(nip, 0);
1477                 hammer_abort_transaction(&trans);
1478                 *ap->a_vpp = NULL;
1479         } else {
1480                 hammer_commit_transaction(&trans);
1481                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1482                 hammer_rel_inode(nip, 0);
1483                 if (error == 0) {
1484                         cache_setunresolved(ap->a_nch);
1485                         cache_setvp(ap->a_nch, *ap->a_vpp);
1486                 }
1487         }
1488         return (error);
1489 }
1490
1491 /*
1492  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1493  */
1494 static
1495 int
1496 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1497 {
1498         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1499 }
1500
1501 /*
1502  * hammer_vop_ioctl { vp, command, data, fflag, cred }
1503  */
1504 static
1505 int
1506 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1507 {
1508         struct hammer_inode *ip = ap->a_vp->v_data;
1509
1510         return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1511                             ap->a_fflag, ap->a_cred));
1512 }
1513
1514 /*
1515  * hammer_vop_strategy { vp, bio }
1516  *
1517  * Strategy call, used for regular file read & write only.  Note that the
1518  * bp may represent a cluster.
1519  *
1520  * To simplify operation and allow better optimizations in the future,
1521  * this code does not make any assumptions with regards to buffer alignment
1522  * or size.
1523  */
1524 static
1525 int
1526 hammer_vop_strategy(struct vop_strategy_args *ap)
1527 {
1528         struct buf *bp;
1529         int error;
1530
1531         bp = ap->a_bio->bio_buf;
1532
1533         switch(bp->b_cmd) {
1534         case BUF_CMD_READ:
1535                 error = hammer_vop_strategy_read(ap);
1536                 break;
1537         case BUF_CMD_WRITE:
1538                 error = hammer_vop_strategy_write(ap);
1539                 break;
1540         default:
1541                 error = EINVAL;
1542                 break;
1543         }
1544         bp->b_error = error;
1545         if (error)
1546                 bp->b_flags |= B_ERROR;
1547         biodone(ap->a_bio);
1548         return (error);
1549 }
1550
1551 /*
1552  * Read from a regular file.  Iterate the related records and fill in the
1553  * BIO/BUF.  Gaps are zero-filled.
1554  *
1555  * The support code in hammer_object.c should be used to deal with mixed
1556  * in-memory and on-disk records.
1557  *
1558  * XXX atime update
1559  */
1560 static
1561 int
1562 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1563 {
1564         struct hammer_inode *ip = ap->a_vp->v_data;
1565         struct hammer_cursor cursor;
1566         hammer_record_ondisk_t rec;
1567         hammer_base_elm_t base;
1568         struct bio *bio;
1569         struct buf *bp;
1570         int64_t rec_offset;
1571         int64_t ran_end;
1572         int64_t tmp64;
1573         int error;
1574         int boff;
1575         int roff;
1576         int n;
1577
1578         bio = ap->a_bio;
1579         bp = bio->bio_buf;
1580
1581         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1582
1583         /*
1584          * Key range (begin and end inclusive) to scan.  Note that the key's
1585          * stored in the actual records represent BASE+LEN, not BASE.  The
1586          * first record containing bio_offset will have a key > bio_offset.
1587          */
1588         cursor.key_beg.obj_id = ip->obj_id;
1589         cursor.key_beg.create_tid = 0;
1590         cursor.key_beg.delete_tid = 0;
1591         cursor.key_beg.obj_type = 0;
1592         cursor.key_beg.key = bio->bio_offset + 1;
1593         cursor.asof = ip->obj_asof;
1594         cursor.flags |= HAMMER_CURSOR_ASOF;
1595
1596         cursor.key_end = cursor.key_beg;
1597         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1598                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1599                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1600                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1601         } else {
1602                 ran_end = bio->bio_offset + bp->b_bufsize;
1603                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1604                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1605                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1606                 if (tmp64 < ran_end)
1607                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1608                 else
1609                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1610         }
1611         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1612
1613         error = hammer_ip_first(&cursor, ip);
1614         boff = 0;
1615
1616         while (error == 0) {
1617                 error = hammer_ip_resolve_data(&cursor);
1618                 if (error)
1619                         break;
1620                 rec = cursor.record;
1621                 base = &rec->base.base;
1622
1623                 rec_offset = base->key - rec->data.base.data_len;
1624
1625                 /*
1626                  * Calculate the gap, if any, and zero-fill it.
1627                  */
1628                 n = (int)(rec_offset - (bio->bio_offset + boff));
1629                 if (n > 0) {
1630                         if (n > bp->b_bufsize - boff)
1631                                 n = bp->b_bufsize - boff;
1632                         bzero((char *)bp->b_data + boff, n);
1633                         boff += n;
1634                         n = 0;
1635                 }
1636
1637                 /*
1638                  * Calculate the data offset in the record and the number
1639                  * of bytes we can copy.
1640                  *
1641                  * Note there is a degenerate case here where boff may
1642                  * already be at bp->b_bufsize.
1643                  */
1644                 roff = -n;
1645                 n = rec->data.base.data_len - roff;
1646                 KKASSERT(n > 0);
1647                 if (n > bp->b_bufsize - boff)
1648                         n = bp->b_bufsize - boff;
1649                 bcopy((char *)cursor.data + roff, (char *)bp->b_data + boff, n);
1650                 boff += n;
1651                 if (boff == bp->b_bufsize)
1652                         break;
1653                 error = hammer_ip_next(&cursor);
1654         }
1655         hammer_done_cursor(&cursor);
1656
1657         /*
1658          * There may have been a gap after the last record
1659          */
1660         if (error == ENOENT)
1661                 error = 0;
1662         if (error == 0 && boff != bp->b_bufsize) {
1663                 KKASSERT(boff < bp->b_bufsize);
1664                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1665                 /* boff = bp->b_bufsize; */
1666         }
1667         bp->b_resid = 0;
1668         return(error);
1669 }
1670
1671 /*
1672  * Write to a regular file.  Iterate the related records and mark for
1673  * deletion.  If existing edge records (left and right side) overlap our
1674  * write they have to be marked deleted and new records created, usually
1675  * referencing a portion of the original data.  Then add a record to
1676  * represent the buffer.
1677  *
1678  * The support code in hammer_object.c should be used to deal with mixed
1679  * in-memory and on-disk records.
1680  */
1681 static
1682 int
1683 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1684 {
1685         struct hammer_transaction trans;
1686         struct hammer_cursor *spike = NULL;
1687         hammer_inode_t ip;
1688         struct bio *bio;
1689         struct buf *bp;
1690         int error;
1691
1692         bio = ap->a_bio;
1693         bp = bio->bio_buf;
1694         ip = ap->a_vp->v_data;
1695
1696         if (ip->flags & HAMMER_INODE_RO)
1697                 return (EROFS);
1698
1699         /*
1700          * Start a transaction using the TID stored with the bp.
1701          */
1702         KKASSERT(bp->b_tid != 0);
1703         hammer_start_transaction_tid(&trans, ip->hmp, bp->b_tid);
1704
1705 retry:
1706         /*
1707          * Delete any records overlapping our range.  This function will
1708          * (eventually) properly truncate partial overlaps.
1709          */
1710         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1711                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1712                                                bio->bio_offset, &spike);
1713         } else {
1714                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1715                                                bio->bio_offset +
1716                                                 bp->b_bufsize - 1,
1717                                                &spike);
1718         }
1719
1720         /*
1721          * Add a single record to cover the write
1722          */
1723         if (error == 0) {
1724                 error = hammer_ip_sync_data(&trans, ip, bio->bio_offset,
1725                                             bp->b_data, bp->b_bufsize,
1726                                             &spike);
1727         }
1728
1729         /*
1730          * If we ran out of space the spike structure will be filled in
1731          * and we must call hammer_spike with it, then retry.
1732          */
1733         if (error == ENOSPC) {
1734                 error = hammer_spike(&spike);
1735                 if (error == 0)
1736                         goto retry;
1737         }
1738         KKASSERT(spike == NULL);
1739
1740         /*
1741          * If an error occured abort the transaction
1742          */
1743         if (error) {
1744                 /* XXX undo deletion */
1745                 hammer_abort_transaction(&trans);
1746                 bp->b_resid = bp->b_bufsize;
1747         } else {
1748                 hammer_commit_transaction(&trans);
1749                 bp->b_resid = 0;
1750                 bp->b_tid = 0;
1751         }
1752         return(error);
1753 }
1754
1755 /*
1756  * dounlink - disconnect a directory entry
1757  *
1758  * XXX whiteout support not really in yet
1759  */
1760 static int
1761 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1762                 int flags)
1763 {
1764         struct hammer_transaction trans;
1765         struct namecache *ncp;
1766         hammer_inode_t dip;
1767         hammer_inode_t ip;
1768         hammer_record_ondisk_t rec;
1769         struct hammer_cursor cursor;
1770         int64_t namekey;
1771         int error;
1772
1773         /*
1774          * Calculate the namekey and setup the key range for the scan.  This
1775          * works kinda like a chained hash table where the lower 32 bits
1776          * of the namekey synthesize the chain.
1777          *
1778          * The key range is inclusive of both key_beg and key_end.
1779          */
1780         dip = VTOI(dvp);
1781         ncp = nch->ncp;
1782
1783         if (dip->flags & HAMMER_INODE_RO)
1784                 return (EROFS);
1785
1786         hammer_start_transaction(&trans, dip->hmp);
1787
1788         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1789 retry:
1790         hammer_init_cursor_hmp(&cursor, &dip->cache[0], dip->hmp);
1791         cursor.key_beg.obj_id = dip->obj_id;
1792         cursor.key_beg.key = namekey;
1793         cursor.key_beg.create_tid = 0;
1794         cursor.key_beg.delete_tid = 0;
1795         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1796         cursor.key_beg.obj_type = 0;
1797
1798         cursor.key_end = cursor.key_beg;
1799         cursor.key_end.key |= 0xFFFFFFFFULL;
1800         cursor.asof = dip->obj_asof;
1801         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1802
1803         /*
1804          * Scan all matching records (the chain), locate the one matching
1805          * the requested path component.  info->last_error contains the
1806          * error code on search termination and could be 0, ENOENT, or
1807          * something else.
1808          *
1809          * The hammer_ip_*() functions merge in-memory records with on-disk
1810          * records for the purposes of the search.
1811          */
1812         error = hammer_ip_first(&cursor, dip);
1813         while (error == 0) {
1814                 error = hammer_ip_resolve_data(&cursor);
1815                 if (error)
1816                         break;
1817                 rec = cursor.record;
1818                 if (ncp->nc_nlen == rec->entry.base.data_len &&
1819                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1820                         break;
1821                 }
1822                 error = hammer_ip_next(&cursor);
1823         }
1824
1825         /*
1826          * If all is ok we have to get the inode so we can adjust nlinks.
1827          *
1828          * If the target is a directory, it must be empty.
1829          */
1830         if (error == 0) {
1831                 ip = hammer_get_inode(dip->hmp, &dip->cache[1],
1832                                       rec->entry.obj_id,
1833                                       dip->hmp->asof, 0, &error);
1834                 if (error == ENOENT) {
1835                         kprintf("obj_id %016llx\n", rec->entry.obj_id);
1836                         Debugger("ENOENT unlinking object that should exist, cont to sync");
1837                         hammer_sync_hmp(dip->hmp, MNT_NOWAIT);
1838                         Debugger("ENOENT - sync done");
1839                 }
1840                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
1841                                   HAMMER_OBJTYPE_DIRECTORY) {
1842                         error = hammer_ip_check_directory_empty(&trans, ip);
1843                 }
1844                 /*
1845                  * WARNING: hammer_ip_del_directory() may have to terminate
1846                  * the cursor to avoid a lock recursion.  It's ok to call
1847                  * hammer_done_cursor() twice.
1848                  */
1849                 if (error == 0)
1850                         error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1851                 if (error == 0) {
1852                         cache_setunresolved(nch);
1853                         cache_setvp(nch, NULL);
1854                         /* XXX locking */
1855                         if (ip->vp)
1856                                 cache_inval_vp(ip->vp, CINV_DESTROY);
1857                 }
1858                 hammer_rel_inode(ip, 0);
1859         }
1860         hammer_done_cursor(&cursor);
1861         if (error == EDEADLK)
1862                 goto retry;
1863
1864         if (error == 0)
1865                 hammer_commit_transaction(&trans);
1866         else
1867                 hammer_abort_transaction(&trans);
1868         return (error);
1869 }
1870
1871 /************************************************************************
1872  *                          FIFO AND SPECFS OPS                         *
1873  ************************************************************************
1874  *
1875  */
1876
1877 static int
1878 hammer_vop_fifoclose (struct vop_close_args *ap)
1879 {
1880         /* XXX update itimes */
1881         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1882 }
1883
1884 static int
1885 hammer_vop_fiforead (struct vop_read_args *ap)
1886 {
1887         int error;
1888
1889         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1890         /* XXX update access time */
1891         return (error);
1892 }
1893
1894 static int
1895 hammer_vop_fifowrite (struct vop_write_args *ap)
1896 {
1897         int error;
1898
1899         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1900         /* XXX update access time */
1901         return (error);
1902 }
1903
1904 static int
1905 hammer_vop_specclose (struct vop_close_args *ap)
1906 {
1907         /* XXX update itimes */
1908         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1909 }
1910
1911 static int
1912 hammer_vop_specread (struct vop_read_args *ap)
1913 {
1914         /* XXX update access time */
1915         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1916 }
1917
1918 static int
1919 hammer_vop_specwrite (struct vop_write_args *ap)
1920 {
1921         /* XXX update last change time */
1922         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1923 }
1924