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 1.39 2008/04/25 21:49:49 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, 0);
 176         if (ap->a_waitfor == MNT_WAIT)
 177                 hammer_wait_inode(ip);
 178         return (ip->error);
 179 }
 180
 181 /*
 182  * hammer_vop_read { vp, uio, ioflag, cred }
 183  */
 184 static
 185 int
 186 hammer_vop_read(struct vop_read_args *ap)
 187 {
 188         struct hammer_transaction trans;
 189         hammer_inode_t ip;
 190         off_t offset;
 191         struct buf *bp;
 192         struct uio *uio;
 193         int error;
 194         int n;
 195         int seqcount;
 196
 197         if (ap->a_vp->v_type != VREG)
 198                 return (EINVAL);
 199         ip = VTOI(ap->a_vp);
 200         error = 0;
 201         seqcount = ap->a_ioflag >> 16;
 202
 203         hammer_start_transaction(&trans, ip->hmp);
 204
 205         /*
 206          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 207          */
 208         uio = ap->a_uio;
 209         while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
 210                 offset = uio->uio_offset & HAMMER_BUFMASK;
 211 #if 0
 212                 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
 213                                      uio->uio_offset - offset, HAMMER_BUFSIZE,
 214                                      MAXBSIZE, seqcount, &bp);
 215 #endif
 216                 error = bread(ap->a_vp, uio->uio_offset - offset,
 217                               HAMMER_BUFSIZE, &bp);
 218                 if (error) {
 219                         brelse(bp);
 220                         break;
 221                 }
 222                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 223                 n = HAMMER_BUFSIZE - offset;
 224                 if (n > uio->uio_resid)
 225                         n = uio->uio_resid;
 226                 if (n > ip->ino_rec.ino_size - uio->uio_offset)
 227                         n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
 228                 error = uiomove((char *)bp->b_data + offset, n, uio);
 229                 if (error) {
 230                         bqrelse(bp);
 231                         break;
 232                 }
 233                 bqrelse(bp);
 234         }
 235         if ((ip->flags & HAMMER_INODE_RO) == 0 &&
 236             (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
 237                 ip->ino_rec.ino_atime = trans.time;
 238                 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
 239         }
 240         hammer_done_transaction(&trans);
 241         return (error);
 242 }
 243
 244 /*
 245  * hammer_vop_write { vp, uio, ioflag, cred }
 246  */
 247 static
 248 int
 249 hammer_vop_write(struct vop_write_args *ap)
 250 {
 251         struct hammer_transaction trans;
 252         struct hammer_inode *ip;
 253         struct uio *uio;
 254         off_t offset;
 255         struct buf *bp;
 256         int error;
 257         int n;
 258         int flags;
 259         int count;
 260
 261         if (ap->a_vp->v_type != VREG)
 262                 return (EINVAL);
 263         ip = VTOI(ap->a_vp);
 264         error = 0;
 265
 266         if (ip->flags & HAMMER_INODE_RO)
 267                 return (EROFS);
 268
 269         /*
 270          * Create a transaction to cover the operations we perform.
 271          */
 272         hammer_start_transaction(&trans, ip->hmp);
 273         uio = ap->a_uio;
 274
 275         /*
 276          * Check append mode
 277          */
 278         if (ap->a_ioflag & IO_APPEND)
 279                 uio->uio_offset = ip->ino_rec.ino_size;
 280
 281         /*
 282          * Check for illegal write offsets.  Valid range is 0...2^63-1
 283          */
 284         if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
 285                 hammer_done_transaction(&trans);
 286                 return (EFBIG);
 287         }
 288
 289         /*
 290          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 291          */
 292         count = 0;
 293         while (uio->uio_resid > 0) {
 294                 int fixsize = 0;
 295
 296                 /*
 297                  * Do not allow huge writes to deadlock the buffer cache
 298                  */
 299                 if ((++count & 15) == 0) {
 300                         vn_unlock(ap->a_vp);
 301                         if ((ap->a_ioflag & IO_NOBWILL) == 0)
 302                                 bwillwrite();
 303                         vn_lock(ap->a_vp, LK_EXCLUSIVE|LK_RETRY);
 304                 }
 305
 306                 offset = uio->uio_offset & HAMMER_BUFMASK;
 307                 n = HAMMER_BUFSIZE - offset;
 308                 if (n > uio->uio_resid)
 309                         n = uio->uio_resid;
 310                 if (uio->uio_offset + n > ip->ino_rec.ino_size) {
 311                         vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
 312                         fixsize = 1;
 313                 }
 314
 315                 if (uio->uio_segflg == UIO_NOCOPY) {
 316                         /*
 317                          * Issuing a write with the same data backing the
 318                          * buffer.  Instantiate the buffer to collect the
 319                          * backing vm pages, then read-in any missing bits.
 320                          *
 321                          * This case is used by vop_stdputpages().
 322                          */
 323                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 324                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 325                         if ((bp->b_flags & B_CACHE) == 0) {
 326                                 bqrelse(bp);
 327                                 error = bread(ap->a_vp,
 328                                               uio->uio_offset - offset,
 329                                               HAMMER_BUFSIZE, &bp);
 330                         }
 331                 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
 332                         /*
 333                          * entirely overwrite the buffer
 334                          */
 335                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 336                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 337                 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
 338                         /*
 339                          * XXX
 340                          */
 341                         bp = getblk(ap->a_vp, uio->uio_offset - offset,
 342                                     HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
 343                         vfs_bio_clrbuf(bp);
 344                 } else {
 345                         /*
 346                          * Partial overwrite, read in any missing bits then
 347                          * replace the portion being written.
 348                          */
 349                         error = bread(ap->a_vp, uio->uio_offset - offset,
 350                                       HAMMER_BUFSIZE, &bp);
 351                         if (error == 0)
 352                                 bheavy(bp);
 353                 }
 354                 if (error == 0)
 355                         error = uiomove((char *)bp->b_data + offset, n, uio);
 356
 357                 /*
 358                  * If we screwed up we have to undo any VM size changes we
 359                  * made.
 360                  */
 361                 if (error) {
 362                         brelse(bp);
 363                         if (fixsize) {
 364                                 vtruncbuf(ap->a_vp, ip->ino_rec.ino_size,
 365                                           HAMMER_BUFSIZE);
 366                         }
 367                         break;
 368                 }
 369                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 370                 hammer_lock_sh(&ip->lock);
 371                 if (ip->ino_rec.ino_size < uio->uio_offset) {
 372                         ip->ino_rec.ino_size = uio->uio_offset;
 373                         flags = HAMMER_INODE_RDIRTY;
 374                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
 375                 } else {
 376                         flags = 0;
 377                 }
 378                 ip->ino_rec.ino_mtime = trans.time;
 379                 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
 380                 hammer_modify_inode(&trans, ip, flags);
 381                 hammer_unlock(&ip->lock);
 382
 383                 if (ap->a_ioflag & IO_SYNC) {
 384                         bwrite(bp);
 385                 } else if (ap->a_ioflag & IO_DIRECT) {
 386                         bawrite(bp);
 387 #if 0
 388                 } else if ((ap->a_ioflag >> 16) == IO_SEQMAX &&
 389                            (uio->uio_offset & HAMMER_BUFMASK) == 0) {
 390                         /*
 391                          * XXX HAMMER can only fsync the whole inode,
 392                          * doing it on every buffer would be a bad idea.
 393                          */
 394                         /*
 395                          * If seqcount indicates sequential operation and
 396                          * we just finished filling a buffer, push it out
 397                          * now to prevent the buffer cache from becoming
 398                          * too full, which would trigger non-optimal
 399                          * flushes.
 400                          */
 401                         bdwrite(bp);
 402 #endif
 403                 } else {
 404                         bdwrite(bp);
 405                 }
 406         }
 407         hammer_done_transaction(&trans);
 408         return (error);
 409 }
 410
 411 /*
 412  * hammer_vop_access { vp, mode, cred }
 413  */
 414 static
 415 int
 416 hammer_vop_access(struct vop_access_args *ap)
 417 {
 418         struct hammer_inode *ip = VTOI(ap->a_vp);
 419         uid_t uid;
 420         gid_t gid;
 421         int error;
 422
 423         uid = hammer_to_unix_xid(&ip->ino_data.uid);
 424         gid = hammer_to_unix_xid(&ip->ino_data.gid);
 425
 426         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
 427                                   ip->ino_data.uflags);
 428         return (error);
 429 }
 430
 431 /*
 432  * hammer_vop_advlock { vp, id, op, fl, flags }
 433  */
 434 static
 435 int
 436 hammer_vop_advlock(struct vop_advlock_args *ap)
 437 {
 438         struct hammer_inode *ip = VTOI(ap->a_vp);
 439
 440         return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
 441 }
 442
 443 /*
 444  * hammer_vop_close { vp, fflag }
 445  */
 446 static
 447 int
 448 hammer_vop_close(struct vop_close_args *ap)
 449 {
 450         return (vop_stdclose(ap));
 451 }
 452
 453 /*
 454  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
 455  *
 456  * The operating system has already ensured that the directory entry
 457  * does not exist and done all appropriate namespace locking.
 458  */
 459 static
 460 int
 461 hammer_vop_ncreate(struct vop_ncreate_args *ap)
 462 {
 463         struct hammer_transaction trans;
 464         struct hammer_inode *dip;
 465         struct hammer_inode *nip;
 466         struct nchandle *nch;
 467         int error;
 468
 469         nch = ap->a_nch;
 470         dip = VTOI(ap->a_dvp);
 471
 472         if (dip->flags & HAMMER_INODE_RO)
 473                 return (EROFS);
 474
 475         /*
 476          * Create a transaction to cover the operations we perform.
 477          */
 478         hammer_start_transaction(&trans, dip->hmp);
 479
 480         /*
 481          * Create a new filesystem object of the requested type.  The
 482          * returned inode will be referenced and shared-locked to prevent
 483          * it from being moved to the flusher.
 484          */
 485
 486         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 487         if (error)
 488                 kprintf("hammer_create_inode error %d\n", error);
 489         if (error) {
 490                 hammer_done_transaction(&trans);
 491                 *ap->a_vpp = NULL;
 492                 return (error);
 493         }
 494         hammer_lock_sh(&dip->lock);
 495
 496         /*
 497          * Add the new filesystem object to the directory.  This will also
 498          * bump the inode's link count.
 499          */
 500         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 501         if (error)
 502                 kprintf("hammer_ip_add_directory error %d\n", error);
 503         hammer_unlock(&dip->lock);
 504         hammer_unlock(&nip->lock);
 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, LK_EXCLUSIVE, 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, LK_EXCLUSIVE, &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]);
 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         if (error == 0)
 681                 error = hammer_ip_first(&cursor, dip);
 682
 683         rec = NULL;
 684         obj_id = 0;
 685
 686         while (error == 0) {
 687                 error = hammer_ip_resolve_data(&cursor);
 688                 if (error)
 689                         break;
 690                 rec = cursor.record;
 691                 if (nlen == rec->entry.base.data_len &&
 692                     bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
 693                         obj_id = rec->entry.obj_id;
 694                         break;
 695                 }
 696                 error = hammer_ip_next(&cursor);
 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, LK_EXCLUSIVE, &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, LK_EXCLUSIVE, 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         hammer_lock_sh(&ip->lock);
 813         hammer_lock_sh(&dip->lock);
 814         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 815         hammer_unlock(&dip->lock);
 816         hammer_unlock(&ip->lock);
 817
 818         /*
 819          * Finish up.
 820          */
 821         if (error == 0) {
 822                 cache_setunresolved(nch);
 823                 cache_setvp(nch, ap->a_vp);
 824         }
 825         hammer_done_transaction(&trans);
 826         return (error);
 827 }
 828
 829 /*
 830  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 831  *
 832  * The operating system has already ensured that the directory entry
 833  * does not exist and done all appropriate namespace locking.
 834  */
 835 static
 836 int
 837 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 838 {
 839         struct hammer_transaction trans;
 840         struct hammer_inode *dip;
 841         struct hammer_inode *nip;
 842         struct nchandle *nch;
 843         int error;
 844
 845         nch = ap->a_nch;
 846         dip = VTOI(ap->a_dvp);
 847
 848         if (dip->flags & HAMMER_INODE_RO)
 849                 return (EROFS);
 850
 851         /*
 852          * Create a transaction to cover the operations we perform.
 853          */
 854         hammer_start_transaction(&trans, dip->hmp);
 855
 856         /*
 857          * Create a new filesystem object of the requested type.  The
 858          * returned inode will be referenced but not locked.
 859          */
 860         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 861         if (error) {
 862                 kprintf("hammer_mkdir error %d\n", error);
 863                 hammer_done_transaction(&trans);
 864                 *ap->a_vpp = NULL;
 865                 return (error);
 866         }
 867         /*
 868          * Add the new filesystem object to the directory.  This will also
 869          * bump the inode's link count.
 870          */
 871         hammer_lock_sh(&dip->lock);
 872         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 873         hammer_unlock(&dip->lock);
 874         hammer_unlock(&nip->lock);
 875         if (error)
 876                 kprintf("hammer_mkdir (add) error %d\n", error);
 877
 878         /*
 879          * Finish up.
 880          */
 881         if (error) {
 882                 hammer_rel_inode(nip, 0);
 883                 *ap->a_vpp = NULL;
 884         } else {
 885                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 886                 hammer_rel_inode(nip, 0);
 887                 if (error == 0) {
 888                         cache_setunresolved(ap->a_nch);
 889                         cache_setvp(ap->a_nch, *ap->a_vpp);
 890                 }
 891         }
 892         hammer_done_transaction(&trans);
 893         return (error);
 894 }
 895
 896 /*
 897  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 898  *
 899  * The operating system has already ensured that the directory entry
 900  * does not exist and done all appropriate namespace locking.
 901  */
 902 static
 903 int
 904 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 905 {
 906         struct hammer_transaction trans;
 907         struct hammer_inode *dip;
 908         struct hammer_inode *nip;
 909         struct nchandle *nch;
 910         int error;
 911
 912         nch = ap->a_nch;
 913         dip = VTOI(ap->a_dvp);
 914
 915         if (dip->flags & HAMMER_INODE_RO)
 916                 return (EROFS);
 917
 918         /*
 919          * Create a transaction to cover the operations we perform.
 920          */
 921         hammer_start_transaction(&trans, dip->hmp);
 922
 923         /*
 924          * Create a new filesystem object of the requested type.  The
 925          * returned inode will be referenced but not locked.
 926          */
 927         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 928         if (error) {
 929                 hammer_done_transaction(&trans);
 930                 *ap->a_vpp = NULL;
 931                 return (error);
 932         }
 933
 934         /*
 935          * Add the new filesystem object to the directory.  This will also
 936          * bump the inode's link count.
 937          */
 938         hammer_lock_sh(&dip->lock);
 939         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 940         hammer_unlock(&dip->lock);
 941         hammer_unlock(&nip->lock);
 942
 943         /*
 944          * Finish up.
 945          */
 946         if (error) {
 947                 hammer_rel_inode(nip, 0);
 948                 *ap->a_vpp = NULL;
 949         } else {
 950                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 951                 hammer_rel_inode(nip, 0);
 952                 if (error == 0) {
 953                         cache_setunresolved(ap->a_nch);
 954                         cache_setvp(ap->a_nch, *ap->a_vpp);
 955                 }
 956         }
 957         hammer_done_transaction(&trans);
 958         return (error);
 959 }
 960
 961 /*
 962  * hammer_vop_open { vp, mode, cred, fp }
 963  */
 964 static
 965 int
 966 hammer_vop_open(struct vop_open_args *ap)
 967 {
 968         if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
 969                 return (EROFS);
 970
 971         return(vop_stdopen(ap));
 972 }
 973
 974 /*
 975  * hammer_vop_pathconf { vp, name, retval }
 976  */
 977 static
 978 int
 979 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 980 {
 981         return EOPNOTSUPP;
 982 }
 983
 984 /*
 985  * hammer_vop_print { vp }
 986  */
 987 static
 988 int
 989 hammer_vop_print(struct vop_print_args *ap)
 990 {
 991         return EOPNOTSUPP;
 992 }
 993
 994 /*
 995  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 996  */
 997 static
 998 int
 999 hammer_vop_readdir(struct vop_readdir_args *ap)
1000 {
1001         struct hammer_transaction trans;
1002         struct hammer_cursor cursor;
1003         struct hammer_inode *ip;
1004         struct uio *uio;
1005         hammer_record_ondisk_t rec;
1006         hammer_base_elm_t base;
1007         int error;
1008         int cookie_index;
1009         int ncookies;
1010         off_t *cookies;
1011         off_t saveoff;
1012         int r;
1013
1014         ip = VTOI(ap->a_vp);
1015         uio = ap->a_uio;
1016         saveoff = uio->uio_offset;
1017
1018         if (ap->a_ncookies) {
1019                 ncookies = uio->uio_resid / 16 + 1;
1020                 if (ncookies > 1024)
1021                         ncookies = 1024;
1022                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1023                 cookie_index = 0;
1024         } else {
1025                 ncookies = -1;
1026                 cookies = NULL;
1027                 cookie_index = 0;
1028         }
1029
1030         hammer_simple_transaction(&trans, ip->hmp);
1031
1032         /*
1033          * Handle artificial entries
1034          */
1035         error = 0;
1036         if (saveoff == 0) {
1037                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1038                 if (r)
1039                         goto done;
1040                 if (cookies)
1041                         cookies[cookie_index] = saveoff;
1042                 ++saveoff;
1043                 ++cookie_index;
1044                 if (cookie_index == ncookies)
1045                         goto done;
1046         }
1047         if (saveoff == 1) {
1048                 if (ip->ino_data.parent_obj_id) {
1049                         r = vop_write_dirent(&error, uio,
1050                                              ip->ino_data.parent_obj_id,
1051                                              DT_DIR, 2, "..");
1052                 } else {
1053                         r = vop_write_dirent(&error, uio,
1054                                              ip->obj_id, DT_DIR, 2, "..");
1055                 }
1056                 if (r)
1057                         goto done;
1058                 if (cookies)
1059                         cookies[cookie_index] = saveoff;
1060                 ++saveoff;
1061                 ++cookie_index;
1062                 if (cookie_index == ncookies)
1063                         goto done;
1064         }
1065
1066         /*
1067          * Key range (begin and end inclusive) to scan.  Directory keys
1068          * directly translate to a 64 bit 'seek' position.
1069          */
1070         hammer_init_cursor(&trans, &cursor, &ip->cache[0]);
1071         cursor.key_beg.obj_id = ip->obj_id;
1072         cursor.key_beg.create_tid = 0;
1073         cursor.key_beg.delete_tid = 0;
1074         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1075         cursor.key_beg.obj_type = 0;
1076         cursor.key_beg.key = saveoff;
1077
1078         cursor.key_end = cursor.key_beg;
1079         cursor.key_end.key = HAMMER_MAX_KEY;
1080         cursor.asof = ip->obj_asof;
1081         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1082
1083         error = hammer_ip_first(&cursor, ip);
1084
1085         while (error == 0) {
1086                 error = hammer_ip_resolve_record_and_data(&cursor);
1087                 if (error)
1088                         break;
1089                 rec = cursor.record;
1090                 base = &rec->base.base;
1091                 saveoff = base->key;
1092
1093                 if (base->obj_id != ip->obj_id)
1094                         panic("readdir: bad record at %p", cursor.node);
1095
1096                 r = vop_write_dirent(
1097                              &error, uio, rec->entry.obj_id,
1098                              hammer_get_dtype(rec->entry.base.base.obj_type),
1099                              rec->entry.base.data_len,
1100                              (void *)cursor.data);
1101                 if (r)
1102                         break;
1103                 ++saveoff;
1104                 if (cookies)
1105                         cookies[cookie_index] = base->key;
1106                 ++cookie_index;
1107                 if (cookie_index == ncookies)
1108                         break;
1109                 error = hammer_ip_next(&cursor);
1110         }
1111         hammer_done_cursor(&cursor);
1112
1113 done:
1114         hammer_done_transaction(&trans);
1115
1116         if (ap->a_eofflag)
1117                 *ap->a_eofflag = (error == ENOENT);
1118         uio->uio_offset = saveoff;
1119         if (error && cookie_index == 0) {
1120                 if (error == ENOENT)
1121                         error = 0;
1122                 if (cookies) {
1123                         kfree(cookies, M_TEMP);
1124                         *ap->a_ncookies = 0;
1125                         *ap->a_cookies = NULL;
1126                 }
1127         } else {
1128                 if (error == ENOENT)
1129                         error = 0;
1130                 if (cookies) {
1131                         *ap->a_ncookies = cookie_index;
1132                         *ap->a_cookies = cookies;
1133                 }
1134         }
1135         return(error);
1136 }
1137
1138 /*
1139  * hammer_vop_readlink { vp, uio, cred }
1140  */
1141 static
1142 int
1143 hammer_vop_readlink(struct vop_readlink_args *ap)
1144 {
1145         struct hammer_transaction trans;
1146         struct hammer_cursor cursor;
1147         struct hammer_inode *ip;
1148         int error;
1149
1150         ip = VTOI(ap->a_vp);
1151
1152         hammer_simple_transaction(&trans, ip->hmp);
1153
1154         hammer_init_cursor(&trans, &cursor, &ip->cache[0]);
1155
1156         /*
1157          * Key range (begin and end inclusive) to scan.  Directory keys
1158          * directly translate to a 64 bit 'seek' position.
1159          */
1160         cursor.key_beg.obj_id = ip->obj_id;
1161         cursor.key_beg.create_tid = 0;
1162         cursor.key_beg.delete_tid = 0;
1163         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1164         cursor.key_beg.obj_type = 0;
1165         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1166         cursor.asof = ip->obj_asof;
1167         cursor.flags |= HAMMER_CURSOR_ASOF;
1168
1169         error = hammer_ip_lookup(&cursor, ip);
1170         if (error == 0) {
1171                 error = hammer_ip_resolve_data(&cursor);
1172                 if (error == 0) {
1173                         error = uiomove((char *)cursor.data,
1174                                         cursor.record->base.data_len,
1175                                         ap->a_uio);
1176                 }
1177         }
1178         hammer_done_cursor(&cursor);
1179         hammer_done_transaction(&trans);
1180         return(error);
1181 }
1182
1183 /*
1184  * hammer_vop_nremove { nch, dvp, cred }
1185  */
1186 static
1187 int
1188 hammer_vop_nremove(struct vop_nremove_args *ap)
1189 {
1190         struct hammer_transaction trans;
1191         int error;
1192
1193         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1194         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1195         hammer_done_transaction(&trans);
1196
1197         return (error);
1198 }
1199
1200 /*
1201  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1202  */
1203 static
1204 int
1205 hammer_vop_nrename(struct vop_nrename_args *ap)
1206 {
1207         struct hammer_transaction trans;
1208         struct namecache *fncp;
1209         struct namecache *tncp;
1210         struct hammer_inode *fdip;
1211         struct hammer_inode *tdip;
1212         struct hammer_inode *ip;
1213         struct hammer_cursor cursor;
1214         union hammer_record_ondisk *rec;
1215         int64_t namekey;
1216         int error;
1217
1218         fdip = VTOI(ap->a_fdvp);
1219         tdip = VTOI(ap->a_tdvp);
1220         fncp = ap->a_fnch->ncp;
1221         tncp = ap->a_tnch->ncp;
1222         ip = VTOI(fncp->nc_vp);
1223         KKASSERT(ip != NULL);
1224
1225         if (fdip->flags & HAMMER_INODE_RO)
1226                 return (EROFS);
1227         if (tdip->flags & HAMMER_INODE_RO)
1228                 return (EROFS);
1229         if (ip->flags & HAMMER_INODE_RO)
1230                 return (EROFS);
1231
1232         hammer_start_transaction(&trans, fdip->hmp);
1233
1234         hammer_lock_sh(&ip->lock);
1235         if (fdip->obj_id < tdip->obj_id) {
1236                 hammer_lock_sh(&fdip->lock);
1237                 hammer_lock_sh(&tdip->lock);
1238         } else {
1239                 hammer_lock_sh(&tdip->lock);
1240                 hammer_lock_sh(&fdip->lock);
1241         }
1242
1243         /*
1244          * Remove tncp from the target directory and then link ip as
1245          * tncp. XXX pass trans to dounlink
1246          *
1247          * Force the inode sync-time to match the transaction so it is
1248          * in-sync with the creation of the target directory entry.
1249          */
1250         error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1251         if (error == 0 || error == ENOENT) {
1252                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1253                 if (error == 0) {
1254                         ip->ino_data.parent_obj_id = tdip->obj_id;
1255                         hammer_modify_inode(&trans, ip, HAMMER_INODE_DDIRTY);
1256                 }
1257         }
1258         if (error)
1259                 goto failed; /* XXX */
1260
1261         /*
1262          * Locate the record in the originating directory and remove it.
1263          *
1264          * Calculate the namekey and setup the key range for the scan.  This
1265          * works kinda like a chained hash table where the lower 32 bits
1266          * of the namekey synthesize the chain.
1267          *
1268          * The key range is inclusive of both key_beg and key_end.
1269          */
1270         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1271 retry:
1272         hammer_init_cursor(&trans, &cursor, &fdip->cache[0]);
1273         cursor.key_beg.obj_id = fdip->obj_id;
1274         cursor.key_beg.key = namekey;
1275         cursor.key_beg.create_tid = 0;
1276         cursor.key_beg.delete_tid = 0;
1277         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1278         cursor.key_beg.obj_type = 0;
1279
1280         cursor.key_end = cursor.key_beg;
1281         cursor.key_end.key |= 0xFFFFFFFFULL;
1282         cursor.asof = fdip->obj_asof;
1283         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1284
1285         /*
1286          * Scan all matching records (the chain), locate the one matching
1287          * the requested path component.
1288          *
1289          * The hammer_ip_*() functions merge in-memory records with on-disk
1290          * records for the purposes of the search.
1291          */
1292         error = hammer_ip_first(&cursor, fdip);
1293         while (error == 0) {
1294                 if (hammer_ip_resolve_data(&cursor) != 0)
1295                         break;
1296                 rec = cursor.record;
1297                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1298                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1299                         break;
1300                 }
1301                 error = hammer_ip_next(&cursor);
1302         }
1303
1304         /*
1305          * If all is ok we have to get the inode so we can adjust nlinks.
1306          *
1307          * WARNING: hammer_ip_del_directory() may have to terminate the
1308          * cursor to avoid a recursion.  It's ok to call hammer_done_cursor()
1309          * twice.
1310          */
1311         if (error == 0)
1312                 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1313
1314         /*
1315          * XXX A deadlock here will break rename's atomicy for the purposes
1316          * of crash recovery.
1317          */
1318         if (error == EDEADLK) {
1319                 hammer_unlock(&ip->lock);
1320                 hammer_unlock(&fdip->lock);
1321                 hammer_unlock(&tdip->lock);
1322                 hammer_done_cursor(&cursor);
1323                 hammer_lock_sh(&ip->lock);
1324                 if (fdip->obj_id < tdip->obj_id) {
1325                         hammer_lock_sh(&fdip->lock);
1326                         hammer_lock_sh(&tdip->lock);
1327                 } else {
1328                         hammer_lock_sh(&tdip->lock);
1329                         hammer_lock_sh(&fdip->lock);
1330                 }
1331                 goto retry;
1332         }
1333
1334         /*
1335          * Cleanup and tell the kernel that the rename succeeded.
1336          */
1337         hammer_done_cursor(&cursor);
1338         if (error == 0)
1339                 cache_rename(ap->a_fnch, ap->a_tnch);
1340
1341 failed:
1342         hammer_unlock(&ip->lock);
1343         hammer_unlock(&fdip->lock);
1344         hammer_unlock(&tdip->lock);
1345         hammer_done_transaction(&trans);
1346         return (error);
1347 }
1348
1349 /*
1350  * hammer_vop_nrmdir { nch, dvp, cred }
1351  */
1352 static
1353 int
1354 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1355 {
1356         struct hammer_transaction trans;
1357         int error;
1358
1359         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1360         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1361         hammer_done_transaction(&trans);
1362
1363         return (error);
1364 }
1365
1366 /*
1367  * hammer_vop_setattr { vp, vap, cred }
1368  */
1369 static
1370 int
1371 hammer_vop_setattr(struct vop_setattr_args *ap)
1372 {
1373         struct hammer_transaction trans;
1374         struct vattr *vap;
1375         struct hammer_inode *ip;
1376         int modflags;
1377         int error;
1378         int truncating;
1379         off_t aligned_size;
1380         u_int32_t flags;
1381         uuid_t uuid;
1382
1383         vap = ap->a_vap;
1384         ip = ap->a_vp->v_data;
1385         modflags = 0;
1386
1387         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1388                 return(EROFS);
1389         if (ip->flags & HAMMER_INODE_RO)
1390                 return (EROFS);
1391
1392         hammer_start_transaction(&trans, ip->hmp);
1393         hammer_lock_sh(&ip->lock);
1394         error = 0;
1395
1396         if (vap->va_flags != VNOVAL) {
1397                 flags = ip->ino_data.uflags;
1398                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1399                                          hammer_to_unix_xid(&ip->ino_data.uid),
1400                                          ap->a_cred);
1401                 if (error == 0) {
1402                         if (ip->ino_data.uflags != flags) {
1403                                 ip->ino_data.uflags = flags;
1404                                 modflags |= HAMMER_INODE_DDIRTY;
1405                         }
1406                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1407                                 error = 0;
1408                                 goto done;
1409                         }
1410                 }
1411                 goto done;
1412         }
1413         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1414                 error = EPERM;
1415                 goto done;
1416         }
1417         if (vap->va_uid != (uid_t)VNOVAL) {
1418                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1419                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1420                         ip->ino_data.uid = uuid;
1421                         modflags |= HAMMER_INODE_DDIRTY;
1422                 }
1423         }
1424         if (vap->va_gid != (uid_t)VNOVAL) {
1425                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1426                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1427                         ip->ino_data.gid = uuid;
1428                         modflags |= HAMMER_INODE_DDIRTY;
1429                 }
1430         }
1431         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1432                 switch(ap->a_vp->v_type) {
1433                 case VREG:
1434                         if (vap->va_size == ip->ino_rec.ino_size)
1435                                 break;
1436                         /*
1437                          * XXX break atomicy, we can deadlock the backend
1438                          * if we do not release the lock.  Probably not a
1439                          * big deal here.
1440                          */
1441                         hammer_unlock(&ip->lock);
1442                         if (vap->va_size < ip->ino_rec.ino_size) {
1443                                 vtruncbuf(ap->a_vp, vap->va_size,
1444                                           HAMMER_BUFSIZE);
1445                                 truncating = 1;
1446                         } else {
1447                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1448                                 truncating = 0;
1449                         }
1450                         hammer_lock_sh(&ip->lock);
1451                         ip->ino_rec.ino_size = vap->va_size;
1452                         modflags |= HAMMER_INODE_RDIRTY;
1453                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1454                                        ~HAMMER_BUFMASK64;
1455
1456                         /*
1457                          * on-media truncation is cached in the inode until
1458                          * the inode is synchronized.
1459                          */
1460                         if (truncating) {
1461                                 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1462                                         ip->flags |= HAMMER_INODE_TRUNCATED;
1463                                         ip->trunc_off = vap->va_size;
1464                                 } else if (ip->trunc_off > vap->va_size) {
1465                                         ip->trunc_off = vap->va_size;
1466                                 }
1467                         }
1468
1469                         /*
1470                          * If truncating we have to clean out a portion of
1471                          * the last block on-disk.  We do this in the
1472                          * front-end buffer cache.
1473                          */
1474                         if (truncating && vap->va_size < aligned_size) {
1475                                 struct buf *bp;
1476                                 int offset;
1477
1478                                 offset = vap->va_size & HAMMER_BUFMASK;
1479                                 error = bread(ap->a_vp,
1480                                               aligned_size - HAMMER_BUFSIZE,
1481                                               HAMMER_BUFSIZE, &bp);
1482                                 if (error == 0) {
1483                                         bzero(bp->b_data + offset,
1484                                               HAMMER_BUFSIZE - offset);
1485                                         bdwrite(bp);
1486                                 } else {
1487                                         brelse(bp);
1488                                 }
1489                         }
1490                         break;
1491                 case VDATABASE:
1492                         if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1493                                 ip->flags |= HAMMER_INODE_TRUNCATED;
1494                                 ip->trunc_off = vap->va_size;
1495                         } else if (ip->trunc_off > vap->va_size) {
1496                                 ip->trunc_off = vap->va_size;
1497                         }
1498                         ip->ino_rec.ino_size = vap->va_size;
1499                         modflags |= HAMMER_INODE_RDIRTY;
1500                         break;
1501                 default:
1502                         error = EINVAL;
1503                         goto done;
1504                 }
1505                 break;
1506         }
1507         if (vap->va_atime.tv_sec != VNOVAL) {
1508                 ip->ino_rec.ino_atime =
1509                         hammer_timespec_to_transid(&vap->va_atime);
1510                 modflags |= HAMMER_INODE_ITIMES;
1511         }
1512         if (vap->va_mtime.tv_sec != VNOVAL) {
1513                 ip->ino_rec.ino_mtime =
1514                         hammer_timespec_to_transid(&vap->va_mtime);
1515                 modflags |= HAMMER_INODE_ITIMES;
1516         }
1517         if (vap->va_mode != (mode_t)VNOVAL) {
1518                 if (ip->ino_data.mode != vap->va_mode) {
1519                         ip->ino_data.mode = vap->va_mode;
1520                         modflags |= HAMMER_INODE_DDIRTY;
1521                 }
1522         }
1523 done:
1524         if (error == 0)
1525                 hammer_modify_inode(&trans, ip, modflags);
1526         hammer_unlock(&ip->lock);
1527         hammer_done_transaction(&trans);
1528         return (error);
1529 }
1530
1531 /*
1532  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1533  */
1534 static
1535 int
1536 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1537 {
1538         struct hammer_transaction trans;
1539         struct hammer_inode *dip;
1540         struct hammer_inode *nip;
1541         struct nchandle *nch;
1542         hammer_record_t record;
1543         int error;
1544         int bytes;
1545
1546         ap->a_vap->va_type = VLNK;
1547
1548         nch = ap->a_nch;
1549         dip = VTOI(ap->a_dvp);
1550
1551         if (dip->flags & HAMMER_INODE_RO)
1552                 return (EROFS);
1553
1554         /*
1555          * Create a transaction to cover the operations we perform.
1556          */
1557         hammer_start_transaction(&trans, dip->hmp);
1558
1559         /*
1560          * Create a new filesystem object of the requested type.  The
1561          * returned inode will be referenced but not locked.
1562          */
1563
1564         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1565         if (error) {
1566                 hammer_done_transaction(&trans);
1567                 *ap->a_vpp = NULL;
1568                 return (error);
1569         }
1570
1571         /*
1572          * Add the new filesystem object to the directory.  This will also
1573          * bump the inode's link count.
1574          */
1575         hammer_lock_sh(&dip->lock);
1576         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1577
1578         /*
1579          * Add a record representing the symlink.  symlink stores the link
1580          * as pure data, not a string, and is no \0 terminated.
1581          */
1582         if (error == 0) {
1583                 record = hammer_alloc_mem_record(nip);
1584                 bytes = strlen(ap->a_target);
1585
1586                 record->rec.base.base.key = HAMMER_FIXKEY_SYMLINK;
1587                 record->rec.base.base.rec_type = HAMMER_RECTYPE_FIX;
1588                 record->rec.base.data_len = bytes;
1589                 record->data = (void *)ap->a_target;
1590                 /* will be reallocated by routine below */
1591                 error = hammer_ip_add_record(&trans, record);
1592
1593                 /*
1594                  * Set the file size to the length of the link.
1595                  */
1596                 if (error == 0) {
1597                         nip->ino_rec.ino_size = bytes;
1598                         hammer_modify_inode(&trans, nip, HAMMER_INODE_RDIRTY);
1599                 }
1600         }
1601         hammer_unlock(&dip->lock);
1602         hammer_unlock(&nip->lock);
1603
1604         /*
1605          * Finish up.
1606          */
1607         if (error) {
1608                 hammer_rel_inode(nip, 0);
1609                 *ap->a_vpp = NULL;
1610         } else {
1611                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1612                 hammer_rel_inode(nip, 0);
1613                 if (error == 0) {
1614                         cache_setunresolved(ap->a_nch);
1615                         cache_setvp(ap->a_nch, *ap->a_vpp);
1616                 }
1617         }
1618         hammer_done_transaction(&trans);
1619         return (error);
1620 }
1621
1622 /*
1623  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1624  */
1625 static
1626 int
1627 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1628 {
1629         struct hammer_transaction trans;
1630         int error;
1631
1632         hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1633         error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1634                                 ap->a_cred, ap->a_flags);
1635         hammer_done_transaction(&trans);
1636
1637         return (error);
1638 }
1639
1640 /*
1641  * hammer_vop_ioctl { vp, command, data, fflag, cred }
1642  */
1643 static
1644 int
1645 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1646 {
1647         struct hammer_inode *ip = ap->a_vp->v_data;
1648
1649         return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1650                             ap->a_fflag, ap->a_cred));
1651 }
1652
1653 static
1654 int
1655 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1656 {
1657         struct mount *mp;
1658         int error;
1659
1660         mp = ap->a_head.a_ops->head.vv_mount;
1661
1662         switch(ap->a_op) {
1663         case MOUNTCTL_SET_EXPORT:
1664                 if (ap->a_ctllen != sizeof(struct export_args))
1665                         error = EINVAL;
1666                 error = hammer_vfs_export(mp, ap->a_op,
1667                                       (const struct export_args *)ap->a_ctl);
1668                 break;
1669         default:
1670                 error = journal_mountctl(ap);
1671                 break;
1672         }
1673         return(error);
1674 }
1675
1676 /*
1677  * hammer_vop_strategy { vp, bio }
1678  *
1679  * Strategy call, used for regular file read & write only.  Note that the
1680  * bp may represent a cluster.
1681  *
1682  * To simplify operation and allow better optimizations in the future,
1683  * this code does not make any assumptions with regards to buffer alignment
1684  * or size.
1685  */
1686 static
1687 int
1688 hammer_vop_strategy(struct vop_strategy_args *ap)
1689 {
1690         struct buf *bp;
1691         int error;
1692
1693         bp = ap->a_bio->bio_buf;
1694
1695         switch(bp->b_cmd) {
1696         case BUF_CMD_READ:
1697                 error = hammer_vop_strategy_read(ap);
1698                 break;
1699         case BUF_CMD_WRITE:
1700                 error = hammer_vop_strategy_write(ap);
1701                 break;
1702         default:
1703                 bp->b_error = error = EINVAL;
1704                 bp->b_flags |= B_ERROR;
1705                 biodone(ap->a_bio);
1706                 break;
1707         }
1708         return (error);
1709 }
1710
1711 /*
1712  * Read from a regular file.  Iterate the related records and fill in the
1713  * BIO/BUF.  Gaps are zero-filled.
1714  *
1715  * The support code in hammer_object.c should be used to deal with mixed
1716  * in-memory and on-disk records.
1717  *
1718  * XXX atime update
1719  */
1720 static
1721 int
1722 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1723 {
1724         struct hammer_transaction trans;
1725         struct hammer_inode *ip;
1726         struct hammer_cursor cursor;
1727         hammer_record_ondisk_t rec;
1728         hammer_base_elm_t base;
1729         struct bio *bio;
1730         struct buf *bp;
1731         int64_t rec_offset;
1732         int64_t ran_end;
1733         int64_t tmp64;
1734         int error;
1735         int boff;
1736         int roff;
1737         int n;
1738
1739         bio = ap->a_bio;
1740         bp = bio->bio_buf;
1741         ip = ap->a_vp->v_data;
1742
1743         hammer_simple_transaction(&trans, ip->hmp);
1744         hammer_init_cursor(&trans, &cursor, &ip->cache[0]);
1745
1746         /*
1747          * Key range (begin and end inclusive) to scan.  Note that the key's
1748          * stored in the actual records represent BASE+LEN, not BASE.  The
1749          * first record containing bio_offset will have a key > bio_offset.
1750          */
1751         cursor.key_beg.obj_id = ip->obj_id;
1752         cursor.key_beg.create_tid = 0;
1753         cursor.key_beg.delete_tid = 0;
1754         cursor.key_beg.obj_type = 0;
1755         cursor.key_beg.key = bio->bio_offset + 1;
1756         cursor.asof = ip->obj_asof;
1757         cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1758
1759         cursor.key_end = cursor.key_beg;
1760         KKASSERT(ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_REGFILE);
1761 #if 0
1762         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1763                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1764                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1765                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1766         } else
1767 #endif
1768         {
1769                 ran_end = bio->bio_offset + bp->b_bufsize;
1770                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1771                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1772                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1773                 if (tmp64 < ran_end)
1774                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1775                 else
1776                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1777         }
1778         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1779
1780         error = hammer_ip_first(&cursor, ip);
1781         boff = 0;
1782
1783         while (error == 0) {
1784                 error = hammer_ip_resolve_data(&cursor);
1785                 if (error)
1786                         break;
1787                 rec = cursor.record;
1788                 base = &rec->base.base;
1789
1790                 rec_offset = base->key - rec->data.base.data_len;
1791
1792                 /*
1793                  * Calculate the gap, if any, and zero-fill it.
1794                  */
1795                 n = (int)(rec_offset - (bio->bio_offset + boff));
1796                 if (n > 0) {
1797                         if (n > bp->b_bufsize - boff)
1798                                 n = bp->b_bufsize - boff;
1799                         bzero((char *)bp->b_data + boff, n);
1800                         boff += n;
1801                         n = 0;
1802                 }
1803
1804                 /*
1805                  * Calculate the data offset in the record and the number
1806                  * of bytes we can copy.
1807                  *
1808                  * Note there is a degenerate case here where boff may
1809                  * already be at bp->b_bufsize.
1810                  */
1811                 roff = -n;
1812                 rec_offset += roff;
1813                 n = rec->data.base.data_len - roff;
1814                 KKASSERT(n > 0);
1815                 if (n > bp->b_bufsize - boff)
1816                         n = bp->b_bufsize - boff;
1817
1818                 /*
1819                  * If we cached a truncation point on our front-end the
1820                  * on-disk version may still have physical records beyond
1821                  * that point.  Truncate visibility.
1822                  */
1823                 if (ip->trunc_off <= rec_offset)
1824                         n = 0;
1825                 else if (ip->trunc_off < rec_offset + n)
1826                         n = (int)(ip->trunc_off - rec_offset);
1827
1828                 /*
1829                  * Copy
1830                  */
1831                 if (n) {
1832                         bcopy((char *)cursor.data + roff,
1833                               (char *)bp->b_data + boff, n);
1834                         boff += n;
1835                 }
1836                 if (boff == bp->b_bufsize)
1837                         break;
1838                 error = hammer_ip_next(&cursor);
1839         }
1840         hammer_done_cursor(&cursor);
1841         hammer_done_transaction(&trans);
1842
1843         /*
1844          * There may have been a gap after the last record
1845          */
1846         if (error == ENOENT)
1847                 error = 0;
1848         if (error == 0 && boff != bp->b_bufsize) {
1849                 KKASSERT(boff < bp->b_bufsize);
1850                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1851                 /* boff = bp->b_bufsize; */
1852         }
1853         bp->b_resid = 0;
1854         bp->b_error = error;
1855         if (error)
1856                 bp->b_flags |= B_ERROR;
1857         biodone(ap->a_bio);
1858         return(error);
1859 }
1860
1861 /*
1862  * Write to a regular file.   Because this is a strategy call the OS is
1863  * trying to actually sync data to the media.   HAMMER can only flush
1864  * the entire inode (so the TID remains properly synchronized).
1865  *
1866  * Basically all we do here is place the bio on the inode's flush queue
1867  * and activate the flusher.
1868  */
1869 static
1870 int
1871 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1872 {
1873         hammer_inode_t ip;
1874         struct bio *bio;
1875         struct buf *bp;
1876
1877         bio = ap->a_bio;
1878         bp = bio->bio_buf;
1879         ip = ap->a_vp->v_data;
1880
1881         if (ip->flags & HAMMER_INODE_RO) {
1882                 bp->b_error = EROFS;
1883                 bp->b_flags |= B_ERROR;
1884                 biodone(ap->a_bio);
1885                 return(EROFS);
1886         }
1887
1888         /*
1889          * If the inode is being flushed we cannot re-queue buffers
1890          * it may have already flushed, or it could result in duplicate
1891          * records in the database.
1892          */
1893         BUF_KERNPROC(bp);
1894         if (ip->flush_state == HAMMER_FST_FLUSH)
1895                 TAILQ_INSERT_TAIL(&ip->bio_alt_list, bio, bio_act);
1896         else
1897                 TAILQ_INSERT_TAIL(&ip->bio_list, bio, bio_act);
1898         hammer_modify_inode(NULL, ip, HAMMER_INODE_XDIRTY);
1899         hammer_flush_inode(ip, 0);
1900         return(0);
1901 }
1902
1903 /*
1904  * Backend code which actually performs the write to the media.  This
1905  * routine is typically called from the flusher.  The bio will be disposed
1906  * of (biodone'd) by this routine.
1907  *
1908  * Iterate the related records and mark for deletion.  If existing edge
1909  * records (left and right side) overlap our write they have to be marked
1910  * deleted and new records created, usually referencing a portion of the
1911  * original data.  Then add a record to represent the buffer.
1912  */
1913 int
1914 hammer_dowrite(hammer_transaction_t trans, hammer_inode_t ip, struct bio *bio)
1915 {
1916         struct buf *bp = bio->bio_buf;
1917         int error;
1918
1919         KKASSERT(ip->flush_state == HAMMER_FST_FLUSH);
1920
1921         /*
1922          * Delete any records overlapping our range.  This function will
1923          * (eventually) properly truncate partial overlaps.
1924          */
1925         if (ip->sync_ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1926                 error = hammer_ip_delete_range(trans, ip, bio->bio_offset,
1927                                                bio->bio_offset);
1928         } else {
1929                 error = hammer_ip_delete_range(trans, ip, bio->bio_offset,
1930                                                bio->bio_offset +
1931                                                 bp->b_bufsize - 1);
1932         }
1933
1934         /*
1935          * Add a single record to cover the write.  We can write a record
1936          * with only the actual file data - for example, a small 200 byte
1937          * file does not have to write out a 16K record.
1938          *
1939          * While the data size does not have to be aligned, we still do it
1940          * to reduce fragmentation in a future allocation model.
1941          */
1942         if (error == 0) {
1943                 int limit_size;
1944
1945                 if (ip->sync_ino_rec.ino_size - bio->bio_offset >
1946                     bp->b_bufsize) {
1947                             limit_size = bp->b_bufsize;
1948                 } else {
1949                         limit_size = (int)(ip->sync_ino_rec.ino_size -
1950                                            bio->bio_offset);
1951                         KKASSERT(limit_size >= 0);
1952                         limit_size = (limit_size + 63) & ~63;
1953                 }
1954                 error = hammer_ip_sync_data(trans, ip, bio->bio_offset,
1955                                             bp->b_data, limit_size);
1956         }
1957
1958         if (error) {
1959                 bp->b_resid = bp->b_bufsize;
1960                 bp->b_error = error;
1961                 bp->b_flags |= B_ERROR;
1962         } else {
1963                 bp->b_resid = 0;
1964         }
1965         biodone(bio);
1966         return(error);
1967 }
1968
1969 /*
1970  * dounlink - disconnect a directory entry
1971  *
1972  * XXX whiteout support not really in yet
1973  */
1974 static int
1975 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
1976                 struct vnode *dvp, struct ucred *cred, int flags)
1977 {
1978         struct namecache *ncp;
1979         hammer_inode_t dip;
1980         hammer_inode_t ip;
1981         hammer_record_ondisk_t rec;
1982         struct hammer_cursor cursor;
1983         int64_t namekey;
1984         int error;
1985
1986         /*
1987          * Calculate the namekey and setup the key range for the scan.  This
1988          * works kinda like a chained hash table where the lower 32 bits
1989          * of the namekey synthesize the chain.
1990          *
1991          * The key range is inclusive of both key_beg and key_end.
1992          */
1993         dip = VTOI(dvp);
1994         ncp = nch->ncp;
1995
1996         if (dip->flags & HAMMER_INODE_RO)
1997                 return (EROFS);
1998
1999         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
2000 retry:
2001         hammer_init_cursor(trans, &cursor, &dip->cache[0]);
2002         cursor.key_beg.obj_id = dip->obj_id;
2003         cursor.key_beg.key = namekey;
2004         cursor.key_beg.create_tid = 0;
2005         cursor.key_beg.delete_tid = 0;
2006         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
2007         cursor.key_beg.obj_type = 0;
2008
2009         cursor.key_end = cursor.key_beg;
2010         cursor.key_end.key |= 0xFFFFFFFFULL;
2011         cursor.asof = dip->obj_asof;
2012         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
2013
2014         /*
2015          * Scan all matching records (the chain), locate the one matching
2016          * the requested path component.  info->last_error contains the
2017          * error code on search termination and could be 0, ENOENT, or
2018          * something else.
2019          *
2020          * The hammer_ip_*() functions merge in-memory records with on-disk
2021          * records for the purposes of the search.
2022          */
2023         error = hammer_ip_first(&cursor, dip);
2024         while (error == 0) {
2025                 error = hammer_ip_resolve_data(&cursor);
2026                 if (error)
2027                         break;
2028                 rec = cursor.record;
2029                 if (ncp->nc_nlen == rec->entry.base.data_len &&
2030                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
2031                         break;
2032                 }
2033                 error = hammer_ip_next(&cursor);
2034         }
2035
2036         /*
2037          * If all is ok we have to get the inode so we can adjust nlinks.
2038          *
2039          * If the target is a directory, it must be empty.
2040          */
2041         if (error == 0) {
2042                 ip = hammer_get_inode(trans, &dip->cache[1],
2043                                       rec->entry.obj_id,
2044                                       dip->hmp->asof, 0, &error);
2045                 if (error == ENOENT) {
2046                         kprintf("obj_id %016llx\n", rec->entry.obj_id);
2047                         Debugger("ENOENT unlinking object that should exist");
2048                 }
2049                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
2050                                   HAMMER_OBJTYPE_DIRECTORY) {
2051                         error = hammer_ip_check_directory_empty(trans, ip);
2052                 }
2053                 /*
2054                  * WARNING: hammer_ip_del_directory() may have to terminate
2055                  * the cursor to avoid a lock recursion.  It's ok to call
2056                  * hammer_done_cursor() twice.
2057                  */
2058                 if (error == 0) {
2059                         hammer_lock_sh(&ip->lock);
2060                         hammer_lock_sh(&dip->lock);
2061                         error = hammer_ip_del_directory(trans, &cursor,
2062                                                         dip, ip);
2063                         hammer_unlock(&dip->lock);
2064                         hammer_unlock(&ip->lock);
2065                 }
2066                 if (error == 0) {
2067                         cache_setunresolved(nch);
2068                         cache_setvp(nch, NULL);
2069                         /* XXX locking */
2070                         if (ip->vp)
2071                                 cache_inval_vp(ip->vp, CINV_DESTROY);
2072                 }
2073                 hammer_rel_inode(ip, 0);
2074         }
2075         hammer_done_cursor(&cursor);
2076         if (error == EDEADLK)
2077                 goto retry;
2078
2079         return (error);
2080 }
2081
2082 /************************************************************************
2083  *                          FIFO AND SPECFS OPS                         *
2084  ************************************************************************
2085  *
2086  */
2087
2088 static int
2089 hammer_vop_fifoclose (struct vop_close_args *ap)
2090 {
2091         /* XXX update itimes */
2092         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2093 }
2094
2095 static int
2096 hammer_vop_fiforead (struct vop_read_args *ap)
2097 {
2098         int error;
2099
2100         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2101         /* XXX update access time */
2102         return (error);
2103 }
2104
2105 static int
2106 hammer_vop_fifowrite (struct vop_write_args *ap)
2107 {
2108         int error;
2109
2110         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2111         /* XXX update access time */
2112         return (error);
2113 }
2114
2115 static int
2116 hammer_vop_specclose (struct vop_close_args *ap)
2117 {
2118         /* XXX update itimes */
2119         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2120 }
2121
2122 static int
2123 hammer_vop_specread (struct vop_read_args *ap)
2124 {
2125         /* XXX update access time */
2126         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2127 }
2128
2129 static int
2130 hammer_vop_specwrite (struct vop_write_args *ap)
2131 {
2132         /* XXX update last change time */
2133         return (VOCALL(&spec_vnode_vops, &ap->a_head));
2134 }
2135