sys/vfs/hammer/hammer_vnops.c

   1 /*
   2  * Copyright (c) 2007-2008 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.47 2008/05/04 09:06:45 dillon Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/fcntl.h>
  41 #include <sys/namecache.h>
  42 #include <sys/vnode.h>
  43 #include <sys/lockf.h>
  44 #include <sys/event.h>
  45 #include <sys/stat.h>
  46 #include <sys/dirent.h>
  47 #include <vm/vm_extern.h>
  48 #include <vfs/fifofs/fifo.h>
  49 #include "hammer.h"
  50
  51 /*
  52  * USERFS VNOPS
  53  */
  54 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
  55 static int hammer_vop_fsync(struct vop_fsync_args *);
  56 static int hammer_vop_read(struct vop_read_args *);
  57 static int hammer_vop_write(struct vop_write_args *);
  58 static int hammer_vop_access(struct vop_access_args *);
  59 static int hammer_vop_advlock(struct vop_advlock_args *);
  60 static int hammer_vop_close(struct vop_close_args *);
  61 static int hammer_vop_ncreate(struct vop_ncreate_args *);
  62 static int hammer_vop_getattr(struct vop_getattr_args *);
  63 static int hammer_vop_nresolve(struct vop_nresolve_args *);
  64 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
  65 static int hammer_vop_nlink(struct vop_nlink_args *);
  66 static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
  67 static int hammer_vop_nmknod(struct vop_nmknod_args *);
  68 static int hammer_vop_open(struct vop_open_args *);
  69 static int hammer_vop_pathconf(struct vop_pathconf_args *);
  70 static int hammer_vop_print(struct vop_print_args *);
  71 static int hammer_vop_readdir(struct vop_readdir_args *);
  72 static int hammer_vop_readlink(struct vop_readlink_args *);
  73 static int hammer_vop_nremove(struct vop_nremove_args *);
  74 static int hammer_vop_nrename(struct vop_nrename_args *);
  75 static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
  76 static int hammer_vop_setattr(struct vop_setattr_args *);
  77 static int hammer_vop_strategy(struct vop_strategy_args *);
  78 static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
  79 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
  80 static int hammer_vop_ioctl(struct vop_ioctl_args *);
  81 static int hammer_vop_mountctl(struct vop_mountctl_args *);
  82
  83 static int hammer_vop_fifoclose (struct vop_close_args *);
  84 static int hammer_vop_fiforead (struct vop_read_args *);
  85 static int hammer_vop_fifowrite (struct vop_write_args *);
  86
  87 static int hammer_vop_specclose (struct vop_close_args *);
  88 static int hammer_vop_specread (struct vop_read_args *);
  89 static int hammer_vop_specwrite (struct vop_write_args *);
  90
  91 struct vop_ops hammer_vnode_vops = {
  92         .vop_default =          vop_defaultop,
  93         .vop_fsync =            hammer_vop_fsync,
  94         .vop_getpages =         vop_stdgetpages,
  95         .vop_putpages =         vop_stdputpages,
  96         .vop_read =             hammer_vop_read,
  97         .vop_write =            hammer_vop_write,
  98         .vop_access =           hammer_vop_access,
  99         .vop_advlock =          hammer_vop_advlock,
 100         .vop_close =            hammer_vop_close,
 101         .vop_ncreate =          hammer_vop_ncreate,
 102         .vop_getattr =          hammer_vop_getattr,
 103         .vop_inactive =         hammer_vop_inactive,
 104         .vop_reclaim =          hammer_vop_reclaim,
 105         .vop_nresolve =         hammer_vop_nresolve,
 106         .vop_nlookupdotdot =    hammer_vop_nlookupdotdot,
 107         .vop_nlink =            hammer_vop_nlink,
 108         .vop_nmkdir =           hammer_vop_nmkdir,
 109         .vop_nmknod =           hammer_vop_nmknod,
 110         .vop_open =             hammer_vop_open,
 111         .vop_pathconf =         hammer_vop_pathconf,
 112         .vop_print =            hammer_vop_print,
 113         .vop_readdir =          hammer_vop_readdir,
 114         .vop_readlink =         hammer_vop_readlink,
 115         .vop_nremove =          hammer_vop_nremove,
 116         .vop_nrename =          hammer_vop_nrename,
 117         .vop_nrmdir =           hammer_vop_nrmdir,
 118         .vop_setattr =          hammer_vop_setattr,
 119         .vop_strategy =         hammer_vop_strategy,
 120         .vop_nsymlink =         hammer_vop_nsymlink,
 121         .vop_nwhiteout =        hammer_vop_nwhiteout,
 122         .vop_ioctl =            hammer_vop_ioctl,
 123         .vop_mountctl =         hammer_vop_mountctl
 124 };
 125
 126 struct vop_ops hammer_spec_vops = {
 127         .vop_default =          spec_vnoperate,
 128         .vop_fsync =            hammer_vop_fsync,
 129         .vop_read =             hammer_vop_specread,
 130         .vop_write =            hammer_vop_specwrite,
 131         .vop_access =           hammer_vop_access,
 132         .vop_close =            hammer_vop_specclose,
 133         .vop_getattr =          hammer_vop_getattr,
 134         .vop_inactive =         hammer_vop_inactive,
 135         .vop_reclaim =          hammer_vop_reclaim,
 136         .vop_setattr =          hammer_vop_setattr
 137 };
 138
 139 struct vop_ops hammer_fifo_vops = {
 140         .vop_default =          fifo_vnoperate,
 141         .vop_fsync =            hammer_vop_fsync,
 142         .vop_read =             hammer_vop_fiforead,
 143         .vop_write =            hammer_vop_fifowrite,
 144         .vop_access =           hammer_vop_access,
 145         .vop_close =            hammer_vop_fifoclose,
 146         .vop_getattr =          hammer_vop_getattr,
 147         .vop_inactive =         hammer_vop_inactive,
 148         .vop_reclaim =          hammer_vop_reclaim,
 149         .vop_setattr =          hammer_vop_setattr
 150 };
 151
 152 static int hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
 153                            struct vnode *dvp, struct ucred *cred, int flags);
 154 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
 155 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
 156
 157 #if 0
 158 static
 159 int
 160 hammer_vop_vnoperate(struct vop_generic_args *)
 161 {
 162         return (VOCALL(&hammer_vnode_vops, ap));
 163 }
 164 #endif
 165
 166 /*
 167  * hammer_vop_fsync { vp, waitfor }
 168  */
 169 static
 170 int
 171 hammer_vop_fsync(struct vop_fsync_args *ap)
 172 {
 173         hammer_inode_t ip = VTOI(ap->a_vp);
 174
 175         hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
 176         vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL);
 177         if (ap->a_waitfor == MNT_WAIT)
 178                 hammer_wait_inode(ip);
 179         return (ip->error);
 180 }
 181
 182 /*
 183  * hammer_vop_read { vp, uio, ioflag, cred }
 184  */
 185 static
 186 int
 187 hammer_vop_read(struct vop_read_args *ap)
 188 {
 189         struct hammer_transaction trans;
 190         hammer_inode_t ip;
 191         off_t offset;
 192         struct buf *bp;
 193         struct uio *uio;
 194         int error;
 195         int n;
 196         int seqcount;
 197
 198         if (ap->a_vp->v_type != VREG)
 199                 return (EINVAL);
 200         ip = VTOI(ap->a_vp);
 201         error = 0;
 202         seqcount = ap->a_ioflag >> 16;
 203
 204         hammer_start_transaction(&trans, ip->hmp);
 205
 206         /*
 207          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 208          */
 209         uio = ap->a_uio;
 210         while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
 211                 offset = uio->uio_offset & HAMMER_BUFMASK;
 212 #if 0
 213                 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
 214                                      uio->uio_offset - offset, HAMMER_BUFSIZE,
 215                                      MAXBSIZE, seqcount, &bp);
 216 #endif
 217                 error = bread(ap->a_vp, uio->uio_offset - offset,
 218                               HAMMER_BUFSIZE, &bp);
 219                 if (error) {
 220                         brelse(bp);
 221                         break;
 222                 }
 223                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 224                 n = HAMMER_BUFSIZE - offset;
 225                 if (n > uio->uio_resid)
 226                         n = uio->uio_resid;
 227                 if (n > ip->ino_rec.ino_size - uio->uio_offset)
 228                         n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
 229                 error = uiomove((char *)bp->b_data + offset, n, uio);
 230                 if (error) {
 231                         bqrelse(bp);
 232                         break;
 233                 }
 234                 bqrelse(bp);
 235         }
 236         if ((ip->flags & HAMMER_INODE_RO) == 0 &&
 237             (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
 238                 ip->ino_rec.ino_atime = trans.time;
 239                 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
 240         }
 241         hammer_done_transaction(&trans);
 242         return (error);
 243 }
 244
 245 /*
 246  * hammer_vop_write { vp, uio, ioflag, cred }
 247  */
 248 static
 249 int
 250 hammer_vop_write(struct vop_write_args *ap)
 251 {
 252         struct hammer_transaction trans;
 253         struct hammer_inode *ip;
 254         struct uio *uio;
 255         off_t offset;
 256         struct buf *bp;
 257         int error;
 258         int n;
 259         int flags;
 260         int count;
 261
 262         if (ap->a_vp->v_type != VREG)
 263                 return (EINVAL);
 264         ip = VTOI(ap->a_vp);
 265         error = 0;
 266
 267         if (ip->flags & HAMMER_INODE_RO)
 268                 return (EROFS);
 269
 270         /*
 271          * Create a transaction to cover the operations we perform.
 272          */
 273         hammer_start_transaction(&trans, ip->hmp);
 274         uio = ap->a_uio;
 275
 276         /*
 277          * Check append mode
 278          */
 279         if (ap->a_ioflag & IO_APPEND)
 280                 uio->uio_offset = ip->ino_rec.ino_size;
 281
 282         /*
 283          * Check for illegal write offsets.  Valid range is 0...2^63-1
 284          */
 285         if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
 286                 hammer_done_transaction(&trans);
 287                 return (EFBIG);
 288         }
 289
 290         /*
 291          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 292          */
 293         count = 0;
 294         while (uio->uio_resid > 0) {
 295                 int fixsize = 0;
 296
 297                 /*
 298                  * Do not allow huge writes to deadlock the buffer cache
 299                  */
 300                 if ((++count & 15) == 0) {
 301                         vn_unlock(ap->a_vp);
 302                         if ((ap->a_ioflag & IO_NOBWILL) == 0)
 303                                 bwillwrite();
 304                         vn_lock(ap->a_vp, LK_EXCLUSIVE|LK_RETRY);
 305                 }
 306
 307                 offset = uio->uio_offset & HAMMER_BUFMASK;
 308                 n = HAMMER_BUFSIZE - offset;
 309                 if (n > uio->uio_resid)
 310                         n = uio->uio_resid;
 311                 if (uio->uio_offset + n > ip->ino_rec.ino_size) {
 312                         vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
 313                         fixsize = 1;
 314                 }
 315
 316                 if (uio->uio_segflg == UIO_NOCOPY) {
 317                         /*
 318                          * Issuing a write with the same data backing the
 319                          * buffer.  Instantiate the buffer to collect the
 320                          * backing vm pages, then read-in any missing bits.
 321                          *
 322                          * This case is used by vop_stdputpages().
 323                          */
 324                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 325                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 326                         if ((bp->b_flags & B_CACHE) == 0) {
 327                                 bqrelse(bp);
 328                                 error = bread(ap->a_vp,
 329                                               uio->uio_offset - offset,
 330                                               HAMMER_BUFSIZE, &bp);
 331                         }
 332                 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
 333                         /*
 334                          * Even though we are entirely overwriting the buffer
 335                          * we may still have to zero it out to avoid a
 336                          * mmap/write visibility issue.
 337                          */
 338                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 339                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 340                         if ((bp->b_flags & B_CACHE) == 0)
 341                                 vfs_bio_clrbuf(bp);
 342                 } else if (uio->uio_offset - offset >= ip->ino_rec.ino_size) {
 343                         /*
 344                          * If the base offset of the buffer is beyond the
 345                          * file EOF, we don't have to issue a read.
 346                          */
 347                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 348                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 349                         vfs_bio_clrbuf(bp);
 350                 } else {
 351                         /*
 352                          * Partial overwrite, read in any missing bits then
 353                          * replace the portion being written.
 354                          */
 355                         error = bread(ap->a_vp, uio->uio_offset - offset,
 356                                       HAMMER_BUFSIZE, &bp);
 357                         if (error == 0)
 358                                 bheavy(bp);
 359                 }
 360                 if (error == 0)
 361                         error = uiomove((char *)bp->b_data + offset, n, uio);
 362
 363                 /*
 364                  * If we screwed up we have to undo any VM size changes we
 365                  * made.
 366                  */
 367                 if (error) {
 368                         brelse(bp);
 369                         if (fixsize) {
 370                                 vtruncbuf(ap->a_vp, ip->ino_rec.ino_size,
 371                                           HAMMER_BUFSIZE);
 372                         }
 373                         break;
 374                 }
 375                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 376                 if (ip->ino_rec.ino_size < uio->uio_offset) {
 377                         ip->ino_rec.ino_size = uio->uio_offset;
 378                         flags = HAMMER_INODE_RDIRTY;
 379                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
 380                 } else {
 381                         flags = 0;
 382                 }
 383                 ip->ino_rec.ino_mtime = trans.time;
 384                 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
 385                 hammer_modify_inode(&trans, ip, flags);
 386
 387                 if (ap->a_ioflag & IO_SYNC) {
 388                         bwrite(bp);
 389                 } else if (ap->a_ioflag & IO_DIRECT) {
 390                         bawrite(bp);
 391 #if 0
 392                 } else if ((ap->a_ioflag >> 16) == IO_SEQMAX &&
 393                            (uio->uio_offset & HAMMER_BUFMASK) == 0) {
 394                         /*
 395                          * XXX HAMMER can only fsync the whole inode,
 396                          * doing it on every buffer would be a bad idea.
 397                          */
 398                         /*
 399                          * If seqcount indicates sequential operation and
 400                          * we just finished filling a buffer, push it out
 401                          * now to prevent the buffer cache from becoming
 402                          * too full, which would trigger non-optimal
 403                          * flushes.
 404                          */
 405                         bdwrite(bp);
 406 #endif
 407                 } else {
 408                         bdwrite(bp);
 409                 }
 410         }
 411         hammer_done_transaction(&trans);
 412         return (error);
 413 }
 414
 415 /*
 416  * hammer_vop_access { vp, mode, cred }
 417  */
 418 static
 419 int
 420 hammer_vop_access(struct vop_access_args *ap)
 421 {
 422         struct hammer_inode *ip = VTOI(ap->a_vp);
 423         uid_t uid;
 424         gid_t gid;
 425         int error;
 426
 427         uid = hammer_to_unix_xid(&ip->ino_data.uid);
 428         gid = hammer_to_unix_xid(&ip->ino_data.gid);
 429
 430         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
 431                                   ip->ino_data.uflags);
 432         return (error);
 433 }
 434
 435 /*
 436  * hammer_vop_advlock { vp, id, op, fl, flags }
 437  */
 438 static
 439 int
 440 hammer_vop_advlock(struct vop_advlock_args *ap)
 441 {
 442         struct hammer_inode *ip = VTOI(ap->a_vp);
 443
 444         return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
 445 }
 446
 447 /*
 448  * hammer_vop_close { vp, fflag }
 449  */
 450 static
 451 int
 452 hammer_vop_close(struct vop_close_args *ap)
 453 {
 454         return (vop_stdclose(ap));
 455 }
 456
 457 /*
 458  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
 459  *
 460  * The operating system has already ensured that the directory entry
 461  * does not exist and done all appropriate namespace locking.
 462  */
 463 static
 464 int
 465 hammer_vop_ncreate(struct vop_ncreate_args *ap)
 466 {
 467         struct hammer_transaction trans;
 468         struct hammer_inode *dip;
 469         struct hammer_inode *nip;
 470         struct nchandle *nch;
 471         int error;
 472
 473         nch = ap->a_nch;
 474         dip = VTOI(ap->a_dvp);
 475
 476         if (dip->flags & HAMMER_INODE_RO)
 477                 return (EROFS);
 478
 479         /*
 480          * Create a transaction to cover the operations we perform.
 481          */
 482         hammer_start_transaction(&trans, dip->hmp);
 483
 484         /*
 485          * Create a new filesystem object of the requested type.  The
 486          * returned inode will be referenced and shared-locked to prevent
 487          * it from being moved to the flusher.
 488          */
 489
 490         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 491         if (error) {
 492                 kprintf("hammer_create_inode error %d\n", error);
 493                 hammer_done_transaction(&trans);
 494                 *ap->a_vpp = NULL;
 495                 return (error);
 496         }
 497
 498         /*
 499          * Add the new filesystem object to the directory.  This will also
 500          * bump the inode's link count.
 501          */
 502         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 503         if (error)
 504                 kprintf("hammer_ip_add_directory error %d\n", error);
 505
 506         /*
 507          * Finish up.
 508          */
 509         if (error) {
 510                 hammer_rel_inode(nip, 0);
 511                 hammer_done_transaction(&trans);
 512                 *ap->a_vpp = NULL;
 513         } else {
 514                 error = hammer_get_vnode(nip, ap->a_vpp);
 515                 hammer_done_transaction(&trans);
 516                 hammer_rel_inode(nip, 0);
 517                 if (error == 0) {
 518                         cache_setunresolved(ap->a_nch);
 519                         cache_setvp(ap->a_nch, *ap->a_vpp);
 520                 }
 521         }
 522         return (error);
 523 }
 524
 525 /*
 526  * hammer_vop_getattr { vp, vap }
 527  */
 528 static
 529 int
 530 hammer_vop_getattr(struct vop_getattr_args *ap)
 531 {
 532         struct hammer_inode *ip = VTOI(ap->a_vp);
 533         struct vattr *vap = ap->a_vap;
 534
 535 #if 0
 536         if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
 537             (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
 538             ip->obj_asof == XXX
 539         ) {
 540                 /* LAZYMOD XXX */
 541         }
 542         hammer_itimes(ap->a_vp);
 543 #endif
 544
 545         vap->va_fsid = ip->hmp->fsid_udev;
 546         vap->va_fileid = ip->ino_rec.base.base.obj_id;
 547         vap->va_mode = ip->ino_data.mode;
 548         vap->va_nlink = ip->ino_rec.ino_nlinks;
 549         vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
 550         vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
 551         vap->va_rmajor = 0;
 552         vap->va_rminor = 0;
 553         vap->va_size = ip->ino_rec.ino_size;
 554         hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
 555         hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
 556         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
 557         vap->va_flags = ip->ino_data.uflags;
 558         vap->va_gen = 1;        /* hammer inums are unique for all time */
 559         vap->va_blocksize = HAMMER_BUFSIZE;
 560         vap->va_bytes = (ip->ino_rec.ino_size + 63) & ~63;
 561         vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
 562         vap->va_filerev = 0;    /* XXX */
 563         /* mtime uniquely identifies any adjustments made to the file */
 564         vap->va_fsmid = ip->ino_rec.ino_mtime;
 565         vap->va_uid_uuid = ip->ino_data.uid;
 566         vap->va_gid_uuid = ip->ino_data.gid;
 567         vap->va_fsid_uuid = ip->hmp->fsid;
 568         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 569                           VA_FSID_UUID_VALID;
 570
 571         switch (ip->ino_rec.base.base.obj_type) {
 572         case HAMMER_OBJTYPE_CDEV:
 573         case HAMMER_OBJTYPE_BDEV:
 574                 vap->va_rmajor = ip->ino_data.rmajor;
 575                 vap->va_rminor = ip->ino_data.rminor;
 576                 break;
 577         default:
 578                 break;
 579         }
 580
 581         return(0);
 582 }
 583
 584 /*
 585  * hammer_vop_nresolve { nch, dvp, cred }
 586  *
 587  * Locate the requested directory entry.
 588  */
 589 static
 590 int
 591 hammer_vop_nresolve(struct vop_nresolve_args *ap)
 592 {
 593         struct hammer_transaction trans;
 594         struct namecache *ncp;
 595         hammer_inode_t dip;
 596         hammer_inode_t ip;
 597         hammer_tid_t asof;
 598         struct hammer_cursor cursor;
 599         union hammer_record_ondisk *rec;
 600         struct vnode *vp;
 601         int64_t namekey;
 602         int error;
 603         int i;
 604         int nlen;
 605         int flags;
 606         u_int64_t obj_id;
 607
 608         /*
 609          * Misc initialization, plus handle as-of name extensions.  Look for
 610          * the '@@' extension.  Note that as-of files and directories cannot
 611          * be modified.
 612          */
 613         dip = VTOI(ap->a_dvp);
 614         ncp = ap->a_nch->ncp;
 615         asof = dip->obj_asof;
 616         nlen = ncp->nc_nlen;
 617         flags = dip->flags;
 618
 619         hammer_simple_transaction(&trans, dip->hmp);
 620
 621         for (i = 0; i < nlen; ++i) {
 622                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
 623                         asof = hammer_str_to_tid(ncp->nc_name + i + 2);
 624                         flags |= HAMMER_INODE_RO;
 625                         break;
 626                 }
 627         }
 628         nlen = i;
 629
 630         /*
 631          * If there is no path component the time extension is relative to
 632          * dip.
 633          */
 634         if (nlen == 0) {
 635                 ip = hammer_get_inode(&trans, &dip->cache[1], dip->obj_id,
 636                                       asof, flags, &error);
 637                 if (error == 0) {
 638                         error = hammer_get_vnode(ip, &vp);
 639                         hammer_rel_inode(ip, 0);
 640                 } else {
 641                         vp = NULL;
 642                 }
 643                 if (error == 0) {
 644                         vn_unlock(vp);
 645                         cache_setvp(ap->a_nch, vp);
 646                         vrele(vp);
 647                 }
 648                 goto done;
 649         }
 650
 651         /*
 652          * Calculate the namekey and setup the key range for the scan.  This
 653          * works kinda like a chained hash table where the lower 32 bits
 654          * of the namekey synthesize the chain.
 655          *
 656          * The key range is inclusive of both key_beg and key_end.
 657          */
 658         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
 659
 660         error = hammer_init_cursor(&trans, &cursor, &dip->cache[0], dip);
 661         cursor.key_beg.obj_id = dip->obj_id;
 662         cursor.key_beg.key = namekey;
 663         cursor.key_beg.create_tid = 0;
 664         cursor.key_beg.delete_tid = 0;
 665         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 666         cursor.key_beg.obj_type = 0;
 667
 668         cursor.key_end = cursor.key_beg;
 669         cursor.key_end.key |= 0xFFFFFFFFULL;
 670         cursor.asof = asof;
 671         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
 672
 673         /*
 674          * Scan all matching records (the chain), locate the one matching
 675          * the requested path component.
 676          *
 677          * The hammer_ip_*() functions merge in-memory records with on-disk
 678          * records for the purposes of the search.
 679          */
 680         obj_id = 0;
 681
 682         if (error == 0) {
 683                 rec = NULL;
 684                 error = hammer_ip_first(&cursor);
 685                 while (error == 0) {
 686                         error = hammer_ip_resolve_data(&cursor);
 687                         if (error)
 688                                 break;
 689                         rec = cursor.record;
 690                         if (nlen == rec->entry.base.data_len &&
 691                             bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
 692                                 obj_id = rec->entry.obj_id;
 693                                 break;
 694                         }
 695                         error = hammer_ip_next(&cursor);
 696                 }
 697         }
 698         hammer_done_cursor(&cursor);
 699         if (error == 0) {
 700                 ip = hammer_get_inode(&trans, &dip->cache[1],
 701                                       obj_id, asof, flags, &error);
 702                 if (error == 0) {
 703                         error = hammer_get_vnode(ip, &vp);
 704                         hammer_rel_inode(ip, 0);
 705                 } else {
 706                         vp = NULL;
 707                 }
 708                 if (error == 0) {
 709                         vn_unlock(vp);
 710                         cache_setvp(ap->a_nch, vp);
 711                         vrele(vp);
 712                 }
 713         } else if (error == ENOENT) {
 714                 cache_setvp(ap->a_nch, NULL);
 715         }
 716 done:
 717         hammer_done_transaction(&trans);
 718         return (error);
 719 }
 720
 721 /*
 722  * hammer_vop_nlookupdotdot { dvp, vpp, cred }
 723  *
 724  * Locate the parent directory of a directory vnode.
 725  *
 726  * dvp is referenced but not locked.  *vpp must be returned referenced and
 727  * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
 728  * at the root, instead it could indicate that the directory we were in was
 729  * removed.
 730  *
 731  * NOTE: as-of sequences are not linked into the directory structure.  If
 732  * we are at the root with a different asof then the mount point, reload
 733  * the same directory with the mount point's asof.   I'm not sure what this
 734  * will do to NFS.  We encode ASOF stamps in NFS file handles so it might not
 735  * get confused, but it hasn't been tested.
 736  */
 737 static
 738 int
 739 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
 740 {
 741         struct hammer_transaction trans;
 742         struct hammer_inode *dip;
 743         struct hammer_inode *ip;
 744         int64_t parent_obj_id;
 745         hammer_tid_t asof;
 746         int error;
 747
 748         dip = VTOI(ap->a_dvp);
 749         asof = dip->obj_asof;
 750         parent_obj_id = dip->ino_data.parent_obj_id;
 751
 752         if (parent_obj_id == 0) {
 753                 if (dip->obj_id == HAMMER_OBJID_ROOT &&
 754                    asof != dip->hmp->asof) {
 755                         parent_obj_id = dip->obj_id;
 756                         asof = dip->hmp->asof;
 757                         *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
 758                         ksnprintf(*ap->a_fakename, 19, "0x%016llx",
 759                                    dip->obj_asof);
 760                 } else {
 761                         *ap->a_vpp = NULL;
 762                         return ENOENT;
 763                 }
 764         }
 765
 766         hammer_simple_transaction(&trans, dip->hmp);
 767
 768         ip = hammer_get_inode(&trans, &dip->cache[1], parent_obj_id,
 769                               asof, dip->flags, &error);
 770         if (ip) {
 771                 error = hammer_get_vnode(ip, ap->a_vpp);
 772                 hammer_rel_inode(ip, 0);
 773         } else {
 774                 *ap->a_vpp = NULL;
 775         }
 776         hammer_done_transaction(&trans);
 777         return (error);
 778 }
 779
 780 /*
 781  * hammer_vop_nlink { nch, dvp, vp, cred }
 782  */
 783 static
 784 int
 785 hammer_vop_nlink(struct vop_nlink_args *ap)
 786 {
 787         struct hammer_transaction trans;
 788         struct hammer_inode *dip;
 789         struct hammer_inode *ip;
 790         struct nchandle *nch;
 791         int error;
 792
 793         nch = ap->a_nch;
 794         dip = VTOI(ap->a_dvp);
 795         ip = VTOI(ap->a_vp);
 796
 797         if (dip->flags & HAMMER_INODE_RO)
 798                 return (EROFS);
 799         if (ip->flags & HAMMER_INODE_RO)
 800                 return (EROFS);
 801
 802         /*
 803          * Create a transaction to cover the operations we perform.
 804          */
 805         hammer_start_transaction(&trans, dip->hmp);
 806
 807         /*
 808          * Add the filesystem object to the directory.  Note that neither
 809          * dip nor ip are referenced or locked, but their vnodes are
 810          * referenced.  This function will bump the inode's link count.
 811          */
 812         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 813
 814         /*
 815          * Finish up.
 816          */
 817         if (error == 0) {
 818                 cache_setunresolved(nch);
 819                 cache_setvp(nch, ap->a_vp);
 820         }
 821         hammer_done_transaction(&trans);
 822         return (error);
 823 }
 824
 825 /*
 826  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 827  *
 828  * The operating system has already ensured that the directory entry
 829  * does not exist and done all appropriate namespace locking.
 830  */
 831 static
 832 int
 833 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 834 {
 835         struct hammer_transaction trans;
 836         struct hammer_inode *dip;
 837         struct hammer_inode *nip;
 838         struct nchandle *nch;
 839         int error;
 840
 841         nch = ap->a_nch;
 842         dip = VTOI(ap->a_dvp);
 843
 844         if (dip->flags & HAMMER_INODE_RO)
 845                 return (EROFS);
 846
 847         /*
 848          * Create a transaction to cover the operations we perform.
 849          */
 850         hammer_start_transaction(&trans, dip->hmp);
 851
 852         /*
 853          * Create a new filesystem object of the requested type.  The
 854          * returned inode will be referenced but not locked.
 855          */
 856         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 857         if (error) {
 858                 kprintf("hammer_mkdir error %d\n", error);
 859                 hammer_done_transaction(&trans);
 860                 *ap->a_vpp = NULL;
 861                 return (error);
 862         }
 863         /*
 864          * Add the new filesystem object to the directory.  This will also
 865          * bump the inode's link count.
 866          */
 867         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 868         if (error)
 869                 kprintf("hammer_mkdir (add) error %d\n", error);
 870
 871         /*
 872          * Finish up.
 873          */
 874         if (error) {
 875                 hammer_rel_inode(nip, 0);
 876                 *ap->a_vpp = NULL;
 877         } else {
 878                 error = hammer_get_vnode(nip, ap->a_vpp);
 879                 hammer_rel_inode(nip, 0);
 880                 if (error == 0) {
 881                         cache_setunresolved(ap->a_nch);
 882                         cache_setvp(ap->a_nch, *ap->a_vpp);
 883                 }
 884         }
 885         hammer_done_transaction(&trans);
 886         return (error);
 887 }
 888
 889 /*
 890  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 891  *
 892  * The operating system has already ensured that the directory entry
 893  * does not exist and done all appropriate namespace locking.
 894  */
 895 static
 896 int
 897 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 898 {
 899         struct hammer_transaction trans;
 900         struct hammer_inode *dip;
 901         struct hammer_inode *nip;
 902         struct nchandle *nch;
 903         int error;
 904
 905         nch = ap->a_nch;
 906         dip = VTOI(ap->a_dvp);
 907
 908         if (dip->flags & HAMMER_INODE_RO)
 909                 return (EROFS);
 910
 911         /*
 912          * Create a transaction to cover the operations we perform.
 913          */
 914         hammer_start_transaction(&trans, dip->hmp);
 915
 916         /*
 917          * Create a new filesystem object of the requested type.  The
 918          * returned inode will be referenced but not locked.
 919          */
 920         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 921         if (error) {
 922                 hammer_done_transaction(&trans);
 923                 *ap->a_vpp = NULL;
 924                 return (error);
 925         }
 926
 927         /*
 928          * Add the new filesystem object to the directory.  This will also
 929          * bump the inode's link count.
 930          */
 931         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 932
 933         /*
 934          * Finish up.
 935          */
 936         if (error) {
 937                 hammer_rel_inode(nip, 0);
 938                 *ap->a_vpp = NULL;
 939         } else {
 940                 error = hammer_get_vnode(nip, ap->a_vpp);
 941                 hammer_rel_inode(nip, 0);
 942                 if (error == 0) {
 943                         cache_setunresolved(ap->a_nch);
 944                         cache_setvp(ap->a_nch, *ap->a_vpp);
 945                 }
 946         }
 947         hammer_done_transaction(&trans);
 948         return (error);
 949 }
 950
 951 /*
 952  * hammer_vop_open { vp, mode, cred, fp }
 953  */
 954 static
 955 int
 956 hammer_vop_open(struct vop_open_args *ap)
 957 {
 958         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
 959                 return (EROFS);
 960
 961         return(vop_stdopen(ap));
 962 }
 963
 964 /*
 965  * hammer_vop_pathconf { vp, name, retval }
 966  */
 967 static
 968 int
 969 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 970 {
 971         return EOPNOTSUPP;
 972 }
 973
 974 /*
 975  * hammer_vop_print { vp }
 976  */
 977 static
 978 int
 979 hammer_vop_print(struct vop_print_args *ap)
 980 {
 981         return EOPNOTSUPP;
 982 }
 983
 984 /*
 985  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 986  */
 987 static
 988 int
 989 hammer_vop_readdir(struct vop_readdir_args *ap)
 990 {
 991         struct hammer_transaction trans;
 992         struct hammer_cursor cursor;
 993         struct hammer_inode *ip;
 994         struct uio *uio;
 995         hammer_record_ondisk_t rec;
 996         hammer_base_elm_t base;
 997         int error;
 998         int cookie_index;
 999         int ncookies;
1000         off_t *cookies;
1001         off_t saveoff;
1002         int r;
1003
1004         ip = VTOI(ap->a_vp);
1005         uio = ap->a_uio;
1006         saveoff = uio->uio_offset;
1007
1008         if (ap->a_ncookies) {
1009                 ncookies = uio->uio_resid / 16 + 1;
1010                 if (ncookies > 1024)
1011                         ncookies = 1024;
1012                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1013                 cookie_index = 0;
1014         } else {
1015                 ncookies = -1;
1016                 cookies = NULL;
1017                 cookie_index = 0;
1018         }
1019
1020         hammer_simple_transaction(&trans, ip->hmp);
1021
1022         /*
1023          * Handle artificial entries
1024          */
1025         error = 0;
1026         if (saveoff == 0) {
1027                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1028                 if (r)
1029                         goto done;
1030                 if (cookies)
1031                         cookies[cookie_index] = saveoff;
1032                 ++saveoff;
1033                 ++cookie_index;
1034                 if (cookie_index == ncookies)
1035                         goto done;
1036         }
1037         if (saveoff == 1) {
1038                 if (ip->ino_data.parent_obj_id) {
1039                         r = vop_write_dirent(&error, uio,
1040                                              ip->ino_data.parent_obj_id,
1041                                              DT_DIR, 2, "..");
1042                 } else {
1043                         r = vop_write_dirent(&error, uio,
1044                                              ip->obj_id, DT_DIR, 2, "..");
1045                 }
1046                 if (r)
1047                         goto done;
1048                 if (cookies)
1049                         cookies[cookie_index] = saveoff;
1050                 ++saveoff;
1051                 ++cookie_index;
1052                 if (cookie_index == ncookies)
1053                         goto done;
1054         }
1055
1056         /*
1057          * Key range (begin and end inclusive) to scan.  Directory keys
1058          * directly translate to a 64 bit 'seek' position.
1059          */
1060         hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1061         cursor.key_beg.obj_id = ip->obj_id;
1062         cursor.key_beg.create_tid = 0;
1063         cursor.key_beg.delete_tid = 0;
1064         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1065         cursor.key_beg.obj_type = 0;
1066         cursor.key_beg.key = saveoff;
1067
1068         cursor.key_end = cursor.key_beg;
1069         cursor.key_end.key = HAMMER_MAX_KEY;
1070         cursor.asof = ip->obj_asof;
1071         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1072
1073         error = hammer_ip_first(&cursor);
1074
1075         while (error == 0) {
1076                 error = hammer_ip_resolve_record_and_data(&cursor);
1077                 if (error)
1078                         break;
1079                 rec = cursor.record;
1080                 base = &rec->base.base;
1081                 saveoff = base->key;
1082
1083                 if (base->obj_id != ip->obj_id)
1084                         panic("readdir: bad record at %p", cursor.node);
1085
1086                 r = vop_write_dirent(
1087                              &error, uio, rec->entry.obj_id,
1088                              hammer_get_dtype(rec->entry.base.base.obj_type),
1089                              rec->entry.base.data_len,
1090                              (void *)cursor.data);
1091                 if (r)
1092                         break;
1093                 ++saveoff;
1094                 if (cookies)
1095                         cookies[cookie_index] = base->key;
1096                 ++cookie_index;
1097                 if (cookie_index == ncookies)
1098                         break;
1099                 error = hammer_ip_next(&cursor);
1100         }
1101         hammer_done_cursor(&cursor);
1102
1103 done:
1104         hammer_done_transaction(&trans);
1105
1106         if (ap->a_eofflag)
1107                 *ap->a_eofflag = (error == ENOENT);
1108         uio->uio_offset = saveoff;
1109         if (error && cookie_index == 0) {
1110                 if (error == ENOENT)
1111                         error = 0;
1112                 if (cookies) {
1113                         kfree(cookies, M_TEMP);
1114                         *ap->a_ncookies = 0;
1115                         *ap->a_cookies = NULL;
1116                 }
1117         } else {
1118                 if (error == ENOENT)
1119                         error = 0;
1120                 if (cookies) {
1121                         *ap->a_ncookies = cookie_index;
1122                         *ap->a_cookies = cookies;
1123                 }
1124         }
1125         return(error);
1126 }
1127
1128 /*
1129  * hammer_vop_readlink { vp, uio, cred }
1130  */
1131 static
1132 int
1133 hammer_vop_readlink(struct vop_readlink_args *ap)
1134 {
1135         struct hammer_transaction trans;
1136         struct hammer_cursor cursor;
1137         struct hammer_inode *ip;
1138         int error;
1139
1140         ip = VTOI(ap->a_vp);
1141
1142         hammer_simple_transaction(&trans, ip->hmp);
1143
1144         hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1145
1146         /*
1147          * Key range (begin and end inclusive) to scan.  Directory keys
1148          * directly translate to a 64 bit 'seek' position.
1149          */
1150         cursor.key_beg.obj_id = ip->obj_id;
1151         cursor.key_beg.create_tid = 0;
1152         cursor.key_beg.delete_tid = 0;
1153         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1154         cursor.key_beg.obj_type = 0;
1155         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1156         cursor.asof = ip->obj_asof;
1157         cursor.flags |= HAMMER_CURSOR_ASOF;
1158
1159         error = hammer_ip_lookup(&cursor, ip);
1160         if (error == 0) {
1161                 error = hammer_ip_resolve_data(&cursor);
1162                 if (error == 0) {
1163                         error = uiomove((char *)cursor.data,
1164                                         cursor.record->base.data_len,
1165                                         ap->a_uio);
1166                 }
1167         }
1168         hammer_done_cursor(&cursor);
1169         hammer_done_transaction(&trans);
1170         return(error);
1171 }
1172
1173 /*
1174  * hammer_vop_nremove { nch, dvp, cred }
1175  */
1176 static
1177 int
1178 hammer_vop_nremove(struct vop_nremove_args *ap)
1179 {
1180         struct hammer_transaction trans;
1181         int error;
1182
1183         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1184         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1185         hammer_done_transaction(&trans);
1186
1187         return (error);
1188 }
1189
1190 /*
1191  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1192  */
1193 static
1194 int
1195 hammer_vop_nrename(struct vop_nrename_args *ap)
1196 {
1197         struct hammer_transaction trans;
1198         struct namecache *fncp;
1199         struct namecache *tncp;
1200         struct hammer_inode *fdip;
1201         struct hammer_inode *tdip;
1202         struct hammer_inode *ip;
1203         struct hammer_cursor cursor;
1204         union hammer_record_ondisk *rec;
1205         int64_t namekey;
1206         int error;
1207
1208         fdip = VTOI(ap->a_fdvp);
1209         tdip = VTOI(ap->a_tdvp);
1210         fncp = ap->a_fnch->ncp;
1211         tncp = ap->a_tnch->ncp;
1212         ip = VTOI(fncp->nc_vp);
1213         KKASSERT(ip != NULL);
1214
1215         if (fdip->flags & HAMMER_INODE_RO)
1216                 return (EROFS);
1217         if (tdip->flags & HAMMER_INODE_RO)
1218                 return (EROFS);
1219         if (ip->flags & HAMMER_INODE_RO)
1220                 return (EROFS);
1221
1222         hammer_start_transaction(&trans, fdip->hmp);
1223
1224         /*
1225          * Remove tncp from the target directory and then link ip as
1226          * tncp. XXX pass trans to dounlink
1227          *
1228          * Force the inode sync-time to match the transaction so it is
1229          * in-sync with the creation of the target directory entry.
1230          */
1231         error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1232         if (error == 0 || error == ENOENT) {
1233                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1234                 if (error == 0) {
1235                         ip->ino_data.parent_obj_id = tdip->obj_id;
1236                         hammer_modify_inode(&trans, ip, HAMMER_INODE_DDIRTY);
1237                 }
1238         }
1239         if (error)
1240                 goto failed; /* XXX */
1241
1242         /*
1243          * Locate the record in the originating directory and remove it.
1244          *
1245          * Calculate the namekey and setup the key range for the scan.  This
1246          * works kinda like a chained hash table where the lower 32 bits
1247          * of the namekey synthesize the chain.
1248          *
1249          * The key range is inclusive of both key_beg and key_end.
1250          */
1251         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1252 retry:
1253         hammer_init_cursor(&trans, &cursor, &fdip->cache[0], fdip);
1254         cursor.key_beg.obj_id = fdip->obj_id;
1255         cursor.key_beg.key = namekey;
1256         cursor.key_beg.create_tid = 0;
1257         cursor.key_beg.delete_tid = 0;
1258         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1259         cursor.key_beg.obj_type = 0;
1260
1261         cursor.key_end = cursor.key_beg;
1262         cursor.key_end.key |= 0xFFFFFFFFULL;
1263         cursor.asof = fdip->obj_asof;
1264         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1265
1266         /*
1267          * Scan all matching records (the chain), locate the one matching
1268          * the requested path component.
1269          *
1270          * The hammer_ip_*() functions merge in-memory records with on-disk
1271          * records for the purposes of the search.
1272          */
1273         error = hammer_ip_first(&cursor);
1274         while (error == 0) {
1275                 if (hammer_ip_resolve_data(&cursor) != 0)
1276                         break;
1277                 rec = cursor.record;
1278                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1279                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1280                         break;
1281                 }
1282                 error = hammer_ip_next(&cursor);
1283         }
1284
1285         /*
1286          * If all is ok we have to get the inode so we can adjust nlinks.
1287          *
1288          * WARNING: hammer_ip_del_directory() may have to terminate the
1289          * cursor to avoid a recursion.  It's ok to call hammer_done_cursor()
1290          * twice.
1291          */
1292         if (error == 0)
1293                 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1294
1295         /*
1296          * XXX A deadlock here will break rename's atomicy for the purposes
1297          * of crash recovery.
1298          */
1299         if (error == EDEADLK) {
1300                 hammer_done_cursor(&cursor);
1301                 goto retry;
1302         }
1303
1304         /*
1305          * Cleanup and tell the kernel that the rename succeeded.
1306          */
1307         hammer_done_cursor(&cursor);
1308         if (error == 0)
1309                 cache_rename(ap->a_fnch, ap->a_tnch);
1310
1311 failed:
1312         hammer_done_transaction(&trans);
1313         return (error);
1314 }
1315
1316 /*
1317  * hammer_vop_nrmdir { nch, dvp, cred }
1318  */
1319 static
1320 int
1321 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1322 {
1323         struct hammer_transaction trans;
1324         int error;
1325
1326         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1327         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1328         hammer_done_transaction(&trans);
1329
1330         return (error);
1331 }
1332
1333 /*
1334  * hammer_vop_setattr { vp, vap, cred }
1335  */
1336 static
1337 int
1338 hammer_vop_setattr(struct vop_setattr_args *ap)
1339 {
1340         struct hammer_transaction trans;
1341         struct vattr *vap;
1342         struct hammer_inode *ip;
1343         int modflags;
1344         int error;
1345         int truncating;
1346         off_t aligned_size;
1347         u_int32_t flags;
1348         uuid_t uuid;
1349
1350         vap = ap->a_vap;
1351         ip = ap->a_vp->v_data;
1352         modflags = 0;
1353
1354         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1355                 return(EROFS);
1356         if (ip->flags & HAMMER_INODE_RO)
1357                 return (EROFS);
1358
1359         hammer_start_transaction(&trans, ip->hmp);
1360         error = 0;
1361
1362         if (vap->va_flags != VNOVAL) {
1363                 flags = ip->ino_data.uflags;
1364                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1365                                          hammer_to_unix_xid(&ip->ino_data.uid),
1366                                          ap->a_cred);
1367                 if (error == 0) {
1368                         if (ip->ino_data.uflags != flags) {
1369                                 ip->ino_data.uflags = flags;
1370                                 modflags |= HAMMER_INODE_DDIRTY;
1371                         }
1372                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1373                                 error = 0;
1374                                 goto done;
1375                         }
1376                 }
1377                 goto done;
1378         }
1379         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1380                 error = EPERM;
1381                 goto done;
1382         }
1383         if (vap->va_uid != (uid_t)VNOVAL) {
1384                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1385                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1386                         ip->ino_data.uid = uuid;
1387                         modflags |= HAMMER_INODE_DDIRTY;
1388                 }
1389         }
1390         if (vap->va_gid != (uid_t)VNOVAL) {
1391                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1392                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1393                         ip->ino_data.gid = uuid;
1394                         modflags |= HAMMER_INODE_DDIRTY;
1395                 }
1396         }
1397         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1398                 switch(ap->a_vp->v_type) {
1399                 case VREG:
1400                         if (vap->va_size == ip->ino_rec.ino_size)
1401                                 break;
1402                         /*
1403                          * XXX break atomicy, we can deadlock the backend
1404                          * if we do not release the lock.  Probably not a
1405                          * big deal here.
1406                          */
1407                         if (vap->va_size < ip->ino_rec.ino_size) {
1408                                 vtruncbuf(ap->a_vp, vap->va_size,
1409                                           HAMMER_BUFSIZE);
1410                                 truncating = 1;
1411                         } else {
1412                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1413                                 truncating = 0;
1414                         }
1415                         ip->ino_rec.ino_size = vap->va_size;
1416                         modflags |= HAMMER_INODE_RDIRTY;
1417                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1418                                        ~HAMMER_BUFMASK64;
1419
1420                         /*
1421                          * on-media truncation is cached in the inode until
1422                          * the inode is synchronized.
1423                          */
1424                         if (truncating) {
1425                                 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1426                                         ip->flags |= HAMMER_INODE_TRUNCATED;
1427                                         ip->trunc_off = vap->va_size;
1428                                 } else if (ip->trunc_off > vap->va_size) {
1429                                         ip->trunc_off = vap->va_size;
1430                                 }
1431                         }
1432
1433                         /*
1434                          * If truncating we have to clean out a portion of
1435                          * the last block on-disk.  We do this in the
1436                          * front-end buffer cache.
1437                          */
1438                         if (truncating && vap->va_size < aligned_size) {
1439                                 struct buf *bp;
1440                                 int offset;
1441
1442                                 offset = vap->va_size & HAMMER_BUFMASK;
1443                                 error = bread(ap->a_vp,
1444                                               aligned_size - HAMMER_BUFSIZE,
1445                                               HAMMER_BUFSIZE, &bp);
1446                                 if (error == 0) {
1447                                         bzero(bp->b_data + offset,
1448                                               HAMMER_BUFSIZE - offset);
1449                                         bdwrite(bp);
1450                                 } else {
1451                                         brelse(bp);
1452                                 }
1453                         }
1454                         break;
1455                 case VDATABASE:
1456                         if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1457                                 ip->flags |= HAMMER_INODE_TRUNCATED;
1458                                 ip->trunc_off = vap->va_size;
1459                         } else if (ip->trunc_off > vap->va_size) {
1460                                 ip->trunc_off = vap->va_size;
1461                         }
1462                         ip->ino_rec.ino_size = vap->va_size;
1463                         modflags |= HAMMER_INODE_RDIRTY;
1464                         break;
1465                 default:
1466                         error = EINVAL;
1467                         goto done;
1468                 }
1469                 break;
1470         }
1471         if (vap->va_atime.tv_sec != VNOVAL) {
1472                 ip->ino_rec.ino_atime =
1473                         hammer_timespec_to_transid(&vap->va_atime);
1474                 modflags |= HAMMER_INODE_ITIMES;
1475         }
1476         if (vap->va_mtime.tv_sec != VNOVAL) {
1477                 ip->ino_rec.ino_mtime =
1478                         hammer_timespec_to_transid(&vap->va_mtime);
1479                 modflags |= HAMMER_INODE_ITIMES;
1480         }
1481         if (vap->va_mode != (mode_t)VNOVAL) {
1482                 if (ip->ino_data.mode != vap->va_mode) {
1483                         ip->ino_data.mode = vap->va_mode;
1484                         modflags |= HAMMER_INODE_DDIRTY;
1485                 }
1486         }
1487 done:
1488         if (error == 0)
1489                 hammer_modify_inode(&trans, ip, modflags);
1490         hammer_done_transaction(&trans);
1491         return (error);
1492 }
1493
1494 /*
1495  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1496  */
1497 static
1498 int
1499 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1500 {
1501         struct hammer_transaction trans;
1502         struct hammer_inode *dip;
1503         struct hammer_inode *nip;
1504         struct nchandle *nch;
1505         hammer_record_t record;
1506         int error;
1507         int bytes;
1508
1509         ap->a_vap->va_type = VLNK;
1510
1511         nch = ap->a_nch;
1512         dip = VTOI(ap->a_dvp);
1513
1514         if (dip->flags & HAMMER_INODE_RO)
1515                 return (EROFS);
1516
1517         /*
1518          * Create a transaction to cover the operations we perform.
1519          */
1520         hammer_start_transaction(&trans, dip->hmp);
1521
1522         /*
1523          * Create a new filesystem object of the requested type.  The
1524          * returned inode will be referenced but not locked.
1525          */
1526
1527         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1528         if (error) {
1529                 hammer_done_transaction(&trans);
1530                 *ap->a_vpp = NULL;
1531                 return (error);
1532         }
1533
1534         /*
1535          * Add a record representing the symlink.  symlink stores the link
1536          * as pure data, not a string, and is no \0 terminated.
1537          */
1538         if (error == 0) {
1539                 record = hammer_alloc_mem_record(nip);
1540                 record->type = HAMMER_MEM_RECORD_GENERAL;
1541                 bytes = strlen(ap->a_target);
1542
1543                 record->rec.base.base.key = HAMMER_FIXKEY_SYMLINK;
1544                 record->rec.base.base.rec_type = HAMMER_RECTYPE_FIX;
1545                 record->rec.base.data_len = bytes;
1546                 record->data = (void *)ap->a_target;
1547                 /* will be reallocated by routine below */
1548                 error = hammer_ip_add_record(&trans, record);
1549
1550                 /*
1551                  * Set the file size to the length of the link.
1552                  */
1553                 if (error == 0) {
1554                         nip->ino_rec.ino_size = bytes;
1555                         hammer_modify_inode(&trans, nip, HAMMER_INODE_RDIRTY);
1556                 }
1557         }
1558         if (error == 0)
1559                 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1560
1561         /*
1562          * Finish up.
1563          */
1564         if (error) {
1565                 hammer_rel_inode(nip, 0);
1566                 *ap->a_vpp = NULL;
1567         } else {
1568                 error = hammer_get_vnode(nip, ap->a_vpp);
1569                 hammer_rel_inode(nip, 0);
1570                 if (error == 0) {
1571                         cache_setunresolved(ap->a_nch);
1572                         cache_setvp(ap->a_nch, *ap->a_vpp);
1573                 }
1574         }
1575         hammer_done_transaction(&trans);
1576         return (error);
1577 }
1578
1579 /*
1580  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1581  */
1582 static
1583 int
1584 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1585 {
1586         struct hammer_transaction trans;
1587         int error;
1588
1589         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1590         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1591                                 ap->a_cred, ap->a_flags);
1592         hammer_done_transaction(&trans);
1593
1594         return (error);
1595 }
1596
1597 /*
1598  * hammer_vop_ioctl { vp, command, data, fflag, cred }
1599  */
1600 static
1601 int
1602 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1603 {
1604         struct hammer_inode *ip = ap->a_vp->v_data;
1605
1606         return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1607                             ap->a_fflag, ap->a_cred));
1608 }
1609
1610 static
1611 int
1612 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1613 {
1614         struct mount *mp;
1615         int error;
1616
1617         mp = ap->a_head.a_ops->head.vv_mount;
1618
1619         switch(ap->a_op) {
1620         case MOUNTCTL_SET_EXPORT:
1621                 if (ap->a_ctllen != sizeof(struct export_args))
1622                         error = EINVAL;
1623                 error = hammer_vfs_export(mp, ap->a_op,
1624                                       (const struct export_args *)ap->a_ctl);
1625                 break;
1626         default:
1627                 error = journal_mountctl(ap);
1628                 break;
1629         }
1630         return(error);
1631 }
1632
1633 /*
1634  * hammer_vop_strategy { vp, bio }
1635  *
1636  * Strategy call, used for regular file read & write only.  Note that the
1637  * bp may represent a cluster.
1638  *
1639  * To simplify operation and allow better optimizations in the future,
1640  * this code does not make any assumptions with regards to buffer alignment
1641  * or size.
1642  */
1643 static
1644 int
1645 hammer_vop_strategy(struct vop_strategy_args *ap)
1646 {
1647         struct buf *bp;
1648         int error;
1649
1650         bp = ap->a_bio->bio_buf;
1651
1652         switch(bp->b_cmd) {
1653         case BUF_CMD_READ:
1654                 error = hammer_vop_strategy_read(ap);
1655                 break;
1656         case BUF_CMD_WRITE:
1657                 error = hammer_vop_strategy_write(ap);
1658                 break;
1659         default:
1660                 bp->b_error = error = EINVAL;
1661                 bp->b_flags |= B_ERROR;
1662                 biodone(ap->a_bio);
1663                 break;
1664         }
1665         return (error);
1666 }
1667
1668 /*
1669  * Read from a regular file.  Iterate the related records and fill in the
1670  * BIO/BUF.  Gaps are zero-filled.
1671  *
1672  * The support code in hammer_object.c should be used to deal with mixed
1673  * in-memory and on-disk records.
1674  *
1675  * XXX atime update
1676  */
1677 static
1678 int
1679 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1680 {
1681         struct hammer_transaction trans;
1682         struct hammer_inode *ip;
1683         struct hammer_cursor cursor;
1684         hammer_record_ondisk_t rec;
1685         hammer_base_elm_t base;
1686         struct bio *bio;
1687         struct buf *bp;
1688         int64_t rec_offset;
1689         int64_t ran_end;
1690         int64_t tmp64;
1691         int error;
1692         int boff;
1693         int roff;
1694         int n;
1695
1696         bio = ap->a_bio;
1697         bp = bio->bio_buf;
1698         ip = ap->a_vp->v_data;
1699
1700         hammer_simple_transaction(&trans, ip->hmp);
1701         hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1702
1703         /*
1704          * Key range (begin and end inclusive) to scan.  Note that the key's
1705          * stored in the actual records represent BASE+LEN, not BASE.  The
1706          * first record containing bio_offset will have a key > bio_offset.
1707          */
1708         cursor.key_beg.obj_id = ip->obj_id;
1709         cursor.key_beg.create_tid = 0;
1710         cursor.key_beg.delete_tid = 0;
1711         cursor.key_beg.obj_type = 0;
1712         cursor.key_beg.key = bio->bio_offset + 1;
1713         cursor.asof = ip->obj_asof;
1714         cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1715
1716         cursor.key_end = cursor.key_beg;
1717         KKASSERT(ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_REGFILE);
1718 #if 0
1719         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1720                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1721                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1722                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1723         } else
1724 #endif
1725         {
1726                 ran_end = bio->bio_offset + bp->b_bufsize;
1727                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1728                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1729                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1730                 if (tmp64 < ran_end)
1731                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1732                 else
1733                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1734         }
1735         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1736
1737         error = hammer_ip_first(&cursor);
1738         boff = 0;
1739
1740         while (error == 0) {
1741                 error = hammer_ip_resolve_data(&cursor);
1742                 if (error)
1743                         break;
1744                 rec = cursor.record;
1745                 base = &rec->base.base;
1746
1747                 rec_offset = base->key - rec->data.base.data_len;
1748
1749                 /*
1750                  * Calculate the gap, if any, and zero-fill it.
1751                  */
1752                 n = (int)(rec_offset - (bio->bio_offset + boff));
1753                 if (n > 0) {
1754                         if (n > bp->b_bufsize - boff)
1755                                 n = bp->b_bufsize - boff;
1756                         bzero((char *)bp->b_data + boff, n);
1757                         boff += n;
1758                         n = 0;
1759                 }
1760
1761                 /*
1762                  * Calculate the data offset in the record and the number
1763                  * of bytes we can copy.
1764                  *
1765                  * Note there is a degenerate case here where boff may
1766                  * already be at bp->b_bufsize.
1767                  */
1768                 roff = -n;
1769                 rec_offset += roff;
1770                 n = rec->data.base.data_len - roff;
1771                 KKASSERT(n > 0);
1772                 if (n > bp->b_bufsize - boff)
1773                         n = bp->b_bufsize - boff;
1774
1775                 /*
1776                  * If we cached a truncation point on our front-end the
1777                  * on-disk version may still have physical records beyond
1778                  * that point.  Truncate visibility.
1779                  */
1780                 if (ip->trunc_off <= rec_offset)
1781                         n = 0;
1782                 else if (ip->trunc_off < rec_offset + n)
1783                         n = (int)(ip->trunc_off - rec_offset);
1784
1785                 /*
1786                  * Copy
1787                  */
1788                 if (n) {
1789                         bcopy((char *)cursor.data + roff,
1790                               (char *)bp->b_data + boff, n);
1791                         boff += n;
1792                 }
1793                 if (boff == bp->b_bufsize)
1794                         break;
1795                 error = hammer_ip_next(&cursor);
1796         }
1797         hammer_done_cursor(&cursor);
1798         hammer_done_transaction(&trans);
1799
1800         /*
1801          * There may have been a gap after the last record
1802          */
1803         if (error == ENOENT)
1804                 error = 0;
1805         if (error == 0 && boff != bp->b_bufsize) {
1806                 KKASSERT(boff < bp->b_bufsize);
1807                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1808                 /* boff = bp->b_bufsize; */
1809         }
1810         bp->b_resid = 0;
1811         bp->b_error = error;
1812         if (error)
1813                 bp->b_flags |= B_ERROR;
1814         biodone(ap->a_bio);
1815         return(error);
1816 }
1817
1818 /*
1819  * Write to a regular file.   Because this is a strategy call the OS is
1820  * trying to actually sync data to the media.   HAMMER can only flush
1821  * the entire inode (so the TID remains properly synchronized).
1822  *
1823  * Basically all we do here is place the bio on the inode's flush queue
1824  * and activate the flusher.
1825  */
1826 static
1827 int
1828 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1829 {
1830         hammer_inode_t ip;
1831         struct bio *bio;
1832         struct buf *bp;
1833
1834         bio = ap->a_bio;
1835         bp = bio->bio_buf;
1836         ip = ap->a_vp->v_data;
1837
1838         if (ip->flags & HAMMER_INODE_RO) {
1839                 bp->b_error = EROFS;
1840                 bp->b_flags |= B_ERROR;
1841                 biodone(ap->a_bio);
1842                 return(EROFS);
1843         }
1844
1845         /*
1846          * If the inode is being flushed we cannot re-queue buffers
1847          * it may have already flushed, or it could result in duplicate
1848          * records in the database.
1849          */
1850         BUF_KERNPROC(bp);
1851         if (ip->flags & HAMMER_INODE_WRITE_ALT)
1852                 TAILQ_INSERT_TAIL(&ip->bio_alt_list, bio, bio_act);
1853         else
1854                 TAILQ_INSERT_TAIL(&ip->bio_list, bio, bio_act);
1855         ++hammer_bio_count;
1856         hammer_modify_inode(NULL, ip, HAMMER_INODE_BUFS);
1857
1858         hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1859 #if 0
1860         /*
1861          * XXX
1862          *
1863          * If the write was not part of an integrated flush operation then
1864          * signal a flush.
1865          */
1866         if (ip->flush_state != HAMMER_FST_FLUSH ||
1867             (ip->flags & HAMMER_INODE_WRITE_ALT)) {
1868                 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1869         }
1870 #endif
1871         return(0);
1872 }
1873
1874 /*
1875  * Backend code which actually performs the write to the media.  This
1876  * routine is typically called from the flusher.  The bio will be disposed
1877  * of (biodone'd) by this routine.
1878  *
1879  * Iterate the related records and mark for deletion.  If existing edge
1880  * records (left and right side) overlap our write they have to be marked
1881  * deleted and new records created, usually referencing a portion of the
1882  * original data.  Then add a record to represent the buffer.
1883  */
1884 int
1885 hammer_dowrite(hammer_cursor_t cursor, hammer_inode_t ip, struct bio *bio)
1886 {
1887         struct buf *bp = bio->bio_buf;
1888         int error;
1889
1890         KKASSERT(ip->flush_state == HAMMER_FST_FLUSH);
1891
1892         /*
1893          * If the inode is going or gone, just throw away any frontend
1894          * buffers.
1895          */
1896         if (ip->flags & HAMMER_INODE_DELETED) {
1897                 bp->b_resid = 0;
1898                 biodone(bio);
1899         }
1900
1901         /*
1902          * Delete any records overlapping our range.  This function will
1903          * (eventually) properly truncate partial overlaps.
1904          */
1905         if (ip->sync_ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1906                 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
1907                                                bio->bio_offset);
1908         } else {
1909                 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
1910                                                bio->bio_offset +
1911                                                 bp->b_bufsize - 1);
1912         }
1913
1914         /*
1915          * Add a single record to cover the write.  We can write a record
1916          * with only the actual file data - for example, a small 200 byte
1917          * file does not have to write out a 16K record.
1918          *
1919          * While the data size does not have to be aligned, we still do it
1920          * to reduce fragmentation in a future allocation model.
1921          */
1922         if (error == 0) {
1923                 int limit_size;
1924
1925                 if (ip->sync_ino_rec.ino_size - bio->bio_offset >
1926                     bp->b_bufsize) {
1927                             limit_size = bp->b_bufsize;
1928                 } else {
1929                         limit_size = (int)(ip->sync_ino_rec.ino_size -
1930                                            bio->bio_offset);
1931                         KKASSERT(limit_size >= 0);
1932                         limit_size = (limit_size + 63) & ~63;
1933                 }
1934                 if (limit_size) {
1935                         error = hammer_ip_sync_data(cursor, ip, bio->bio_offset,
1936                                                     bp->b_data, limit_size);
1937                 }
1938         }
1939         if (error)
1940                 Debugger("hammer_dowrite: error");
1941
1942         if (error) {
1943                 bp->b_resid = bp->b_bufsize;
1944                 bp->b_error = error;
1945                 bp->b_flags |= B_ERROR;
1946         } else {
1947                 bp->b_resid = 0;
1948         }
1949         biodone(bio);
1950         --hammer_bio_count;
1951         return(error);
1952 }
1953
1954 /*
1955  * dounlink - disconnect a directory entry
1956  *
1957  * XXX whiteout support not really in yet
1958  */
1959 static int
1960 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
1961                 struct vnode *dvp, struct ucred *cred, int flags)
1962 {
1963         struct namecache *ncp;
1964         hammer_inode_t dip;
1965         hammer_inode_t ip;
1966         hammer_record_ondisk_t rec;
1967         struct hammer_cursor cursor;
1968         int64_t namekey;
1969         int error;
1970
1971         /*
1972          * Calculate the namekey and setup the key range for the scan.  This
1973          * works kinda like a chained hash table where the lower 32 bits
1974          * of the namekey synthesize the chain.
1975          *
1976          * The key range is inclusive of both key_beg and key_end.
1977          */
1978         dip = VTOI(dvp);
1979         ncp = nch->ncp;
1980
1981         if (dip->flags & HAMMER_INODE_RO)
1982                 return (EROFS);
1983
1984         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1985 retry:
1986         hammer_init_cursor(trans, &cursor, &dip->cache[0], dip);
1987         cursor.key_beg.obj_id = dip->obj_id;
1988         cursor.key_beg.key = namekey;
1989         cursor.key_beg.create_tid = 0;
1990         cursor.key_beg.delete_tid = 0;
1991         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1992         cursor.key_beg.obj_type = 0;
1993
1994         cursor.key_end = cursor.key_beg;
1995         cursor.key_end.key |= 0xFFFFFFFFULL;
1996         cursor.asof = dip->obj_asof;
1997         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1998
1999         /*
2000          * Scan all matching records (the chain), locate the one matching
2001          * the requested path component.  info->last_error contains the
2002          * error code on search termination and could be 0, ENOENT, or
2003          * something else.
2004          *
2005          * The hammer_ip_*() functions merge in-memory records with on-disk
2006          * records for the purposes of the search.
2007          */
2008         rec = NULL;
2009         error = hammer_ip_first(&cursor);
2010
2011         while (error == 0) {
2012                 error = hammer_ip_resolve_data(&cursor);
2013                 if (error)
2014                         break;
2015                 rec = cursor.record;
2016                 if (ncp->nc_nlen == rec->entry.base.data_len &&
2017                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
2018                         break;
2019                 }
2020                 error = hammer_ip_next(&cursor);
2021         }
2022
2023         /*
2024          * If all is ok we have to get the inode so we can adjust nlinks.
2025          *
2026          * If the target is a directory, it must be empty.
2027          */
2028         if (error == 0) {
2029                 ip = hammer_get_inode(trans, &dip->cache[1],
2030                                       rec->entry.obj_id,
2031                                       dip->hmp->asof, 0, &error);
2032                 if (error == ENOENT) {
2033                         kprintf("obj_id %016llx\n", rec->entry.obj_id);
2034                         Debugger("ENOENT unlinking object that should exist");
2035                 }
2036
2037                 /*
2038                  * If we are trying to remove a directory the directory must
2039                  * be empty.
2040                  *
2041                  * WARNING: hammer_ip_check_directory_empty() may have to
2042                  * terminate the cursor to avoid a deadlock.  It is ok to
2043                  * call hammer_done_cursor() twice.
2044                  */
2045                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
2046                                   HAMMER_OBJTYPE_DIRECTORY) {
2047                         error = hammer_ip_check_directory_empty(trans, &cursor,
2048                                                                 ip);
2049                 }
2050
2051                 /*
2052                  * Delete the directory entry.
2053                  *
2054                  * WARNING: hammer_ip_del_directory() may have to terminate
2055                  * the cursor to avoid a deadlock.  It is ok to call
2056                  * hammer_done_cursor() twice.
2057                  */
2058                 if (error == 0) {
2059                         error = hammer_ip_del_directory(trans, &cursor,
2060                                                         dip, ip);
2061                 }
2062                 if (error == 0) {
2063                         cache_setunresolved(nch);
2064                         cache_setvp(nch, NULL);
2065                         /* XXX locking */
2066                         if (ip->vp)
2067                                 cache_inval_vp(ip->vp, CINV_DESTROY);
2068                 }
2069                 hammer_rel_inode(ip, 0);
2070         }
2071         hammer_done_cursor(&cursor);
2072         if (error == EDEADLK)
2073                 goto retry;
2074
2075         return (error);
2076 }
2077
2078 /************************************************************************
2079  *                          FIFO AND SPECFS OPS                         *
2080  ************************************************************************
2081  *
2082  */
2083
2084 static int
2085 hammer_vop_fifoclose (struct vop_close_args *ap)
2086 {
2087         /* XXX update itimes */
2088         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2089 }
2090
2091 static int
2092 hammer_vop_fiforead (struct vop_read_args *ap)
2093 {
2094         int error;
2095
2096         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2097         /* XXX update access time */
2098         return (error);
2099 }
2100
2101 static int
2102 hammer_vop_fifowrite (struct vop_write_args *ap)
2103 {
2104         int error;
2105
2106         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2107         /* XXX update access time */
2108         return (error);
2109 }
2110
2111 static int
2112 hammer_vop_specclose (struct vop_close_args *ap)
2113 {
2114         /* XXX update itimes */
2115         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2116 }
2117
2118 static int
2119 hammer_vop_specread (struct vop_read_args *ap)
2120 {
2121         /* XXX update access time */
2122         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2123 }
2124
2125 static int
2126 hammer_vop_specwrite (struct vop_write_args *ap)
2127 {
2128         /* XXX update last change time */
2129         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2130 }
2131