sys/vfs/hammer/hammer_vnops.c

   1 /*
   2  * Copyright (c) 2007 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.14 2007/12/31 05:44:33 dillon Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/fcntl.h>
  41 #include <sys/namecache.h>
  42 #include <sys/vnode.h>
  43 #include <sys/lockf.h>
  44 #include <sys/event.h>
  45 #include <sys/stat.h>
  46 #include <sys/dirent.h>
  47 #include <vm/vm_extern.h>
  48 #include <vfs/fifofs/fifo.h>
  49 #include "hammer.h"
  50
  51 /*
  52  * USERFS VNOPS
  53  */
  54 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
  55 static int hammer_vop_fsync(struct vop_fsync_args *);
  56 static int hammer_vop_read(struct vop_read_args *);
  57 static int hammer_vop_write(struct vop_write_args *);
  58 static int hammer_vop_access(struct vop_access_args *);
  59 static int hammer_vop_advlock(struct vop_advlock_args *);
  60 static int hammer_vop_close(struct vop_close_args *);
  61 static int hammer_vop_ncreate(struct vop_ncreate_args *);
  62 static int hammer_vop_getattr(struct vop_getattr_args *);
  63 static int hammer_vop_nresolve(struct vop_nresolve_args *);
  64 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
  65 static int hammer_vop_nlink(struct vop_nlink_args *);
  66 static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
  67 static int hammer_vop_nmknod(struct vop_nmknod_args *);
  68 static int hammer_vop_open(struct vop_open_args *);
  69 static int hammer_vop_pathconf(struct vop_pathconf_args *);
  70 static int hammer_vop_print(struct vop_print_args *);
  71 static int hammer_vop_readdir(struct vop_readdir_args *);
  72 static int hammer_vop_readlink(struct vop_readlink_args *);
  73 static int hammer_vop_nremove(struct vop_nremove_args *);
  74 static int hammer_vop_nrename(struct vop_nrename_args *);
  75 static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
  76 static int hammer_vop_setattr(struct vop_setattr_args *);
  77 static int hammer_vop_strategy(struct vop_strategy_args *);
  78 static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
  79 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
  80
  81 static int hammer_vop_fifoclose (struct vop_close_args *);
  82 static int hammer_vop_fiforead (struct vop_read_args *);
  83 static int hammer_vop_fifowrite (struct vop_write_args *);
  84
  85 static int hammer_vop_specclose (struct vop_close_args *);
  86 static int hammer_vop_specread (struct vop_read_args *);
  87 static int hammer_vop_specwrite (struct vop_write_args *);
  88
  89 struct vop_ops hammer_vnode_vops = {
  90         .vop_default =          vop_defaultop,
  91         .vop_fsync =            hammer_vop_fsync,
  92         .vop_getpages =         vop_stdgetpages,
  93         .vop_putpages =         vop_stdputpages,
  94         .vop_read =             hammer_vop_read,
  95         .vop_write =            hammer_vop_write,
  96         .vop_access =           hammer_vop_access,
  97         .vop_advlock =          hammer_vop_advlock,
  98         .vop_close =            hammer_vop_close,
  99         .vop_ncreate =          hammer_vop_ncreate,
 100         .vop_getattr =          hammer_vop_getattr,
 101         .vop_inactive =         hammer_vop_inactive,
 102         .vop_reclaim =          hammer_vop_reclaim,
 103         .vop_nresolve =         hammer_vop_nresolve,
 104         .vop_nlookupdotdot =    hammer_vop_nlookupdotdot,
 105         .vop_nlink =            hammer_vop_nlink,
 106         .vop_nmkdir =           hammer_vop_nmkdir,
 107         .vop_nmknod =           hammer_vop_nmknod,
 108         .vop_open =             hammer_vop_open,
 109         .vop_pathconf =         hammer_vop_pathconf,
 110         .vop_print =            hammer_vop_print,
 111         .vop_readdir =          hammer_vop_readdir,
 112         .vop_readlink =         hammer_vop_readlink,
 113         .vop_nremove =          hammer_vop_nremove,
 114         .vop_nrename =          hammer_vop_nrename,
 115         .vop_nrmdir =           hammer_vop_nrmdir,
 116         .vop_setattr =          hammer_vop_setattr,
 117         .vop_strategy =         hammer_vop_strategy,
 118         .vop_nsymlink =         hammer_vop_nsymlink,
 119         .vop_nwhiteout =        hammer_vop_nwhiteout
 120 };
 121
 122 struct vop_ops hammer_spec_vops = {
 123         .vop_default =          spec_vnoperate,
 124         .vop_fsync =            hammer_vop_fsync,
 125         .vop_read =             hammer_vop_specread,
 126         .vop_write =            hammer_vop_specwrite,
 127         .vop_access =           hammer_vop_access,
 128         .vop_close =            hammer_vop_specclose,
 129         .vop_getattr =          hammer_vop_getattr,
 130         .vop_inactive =         hammer_vop_inactive,
 131         .vop_reclaim =          hammer_vop_reclaim,
 132         .vop_setattr =          hammer_vop_setattr
 133 };
 134
 135 struct vop_ops hammer_fifo_vops = {
 136         .vop_default =          fifo_vnoperate,
 137         .vop_fsync =            hammer_vop_fsync,
 138         .vop_read =             hammer_vop_fiforead,
 139         .vop_write =            hammer_vop_fifowrite,
 140         .vop_access =           hammer_vop_access,
 141         .vop_close =            hammer_vop_fifoclose,
 142         .vop_getattr =          hammer_vop_getattr,
 143         .vop_inactive =         hammer_vop_inactive,
 144         .vop_reclaim =          hammer_vop_reclaim,
 145         .vop_setattr =          hammer_vop_setattr
 146 };
 147
 148 static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp,
 149                            struct ucred *cred, int flags);
 150 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
 151 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
 152
 153 #if 0
 154 static
 155 int
 156 hammer_vop_vnoperate(struct vop_generic_args *)
 157 {
 158         return (VOCALL(&hammer_vnode_vops, ap));
 159 }
 160 #endif
 161
 162 /*
 163  * hammer_vop_fsync { vp, waitfor }
 164  */
 165 static
 166 int
 167 hammer_vop_fsync(struct vop_fsync_args *ap)
 168 {
 169         hammer_inode_t ip;
 170         int error;
 171
 172         ip = VTOI(ap->a_vp);
 173         error = hammer_sync_inode(ip, ap->a_waitfor, 0);
 174         return (error);
 175 }
 176
 177 /*
 178  * hammer_vop_read { vp, uio, ioflag, cred }
 179  */
 180 static
 181 int
 182 hammer_vop_read(struct vop_read_args *ap)
 183 {
 184         struct hammer_transaction trans;
 185         hammer_inode_t ip;
 186         off_t offset;
 187         struct buf *bp;
 188         struct uio *uio;
 189         int error;
 190         int n;
 191         int seqcount;
 192
 193         if (ap->a_vp->v_type != VREG)
 194                 return (EINVAL);
 195         ip = VTOI(ap->a_vp);
 196         error = 0;
 197         seqcount = ap->a_ioflag >> 16;
 198
 199         hammer_start_transaction(&trans, ip->hmp);
 200
 201         /*
 202          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 203          */
 204         uio = ap->a_uio;
 205         while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
 206                 offset = uio->uio_offset & HAMMER_BUFMASK;
 207 #if 0
 208                 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
 209                                      uio->uio_offset - offset, HAMMER_BUFSIZE,
 210                                      MAXBSIZE, seqcount, &bp);
 211 #endif
 212                 error = bread(ap->a_vp, uio->uio_offset - offset,
 213                               HAMMER_BUFSIZE, &bp);
 214                 if (error) {
 215                         brelse(bp);
 216                         break;
 217                 }
 218                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 219                 n = HAMMER_BUFSIZE - offset;
 220                 if (n > uio->uio_resid)
 221                         n = uio->uio_resid;
 222                 if (n > ip->ino_rec.ino_size - uio->uio_offset)
 223                         n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
 224                 error = uiomove((char *)bp->b_data + offset, n, uio);
 225                 if (error) {
 226                         bqrelse(bp);
 227                         break;
 228                 }
 229                 ip->ino_rec.ino_atime = trans.tid;
 230                 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
 231                 bqrelse(bp);
 232         }
 233         hammer_commit_transaction(&trans);
 234         return (error);
 235 }
 236
 237 /*
 238  * hammer_vop_write { vp, uio, ioflag, cred }
 239  */
 240 static
 241 int
 242 hammer_vop_write(struct vop_write_args *ap)
 243 {
 244         struct hammer_transaction trans;
 245         struct hammer_inode *ip;
 246         struct uio *uio;
 247         off_t offset;
 248         struct buf *bp;
 249         int error;
 250         int n;
 251         int flags;
 252
 253         if (ap->a_vp->v_type != VREG)
 254                 return (EINVAL);
 255         ip = VTOI(ap->a_vp);
 256         error = 0;
 257
 258         /*
 259          * Create a transaction to cover the operations we perform.
 260          */
 261         hammer_start_transaction(&trans, ip->hmp);
 262         uio = ap->a_uio;
 263
 264         /*
 265          * Check append mode
 266          */
 267         if (ap->a_ioflag & IO_APPEND)
 268                 uio->uio_offset = ip->ino_rec.ino_size;
 269
 270         /*
 271          * Check for illegal write offsets.  Valid range is 0...2^63-1
 272          */
 273         if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
 274                 hammer_commit_transaction(&trans);
 275                 return (EFBIG);
 276         }
 277
 278         /*
 279          * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
 280          */
 281         while (uio->uio_resid > 0) {
 282                 offset = uio->uio_offset & HAMMER_BUFMASK;
 283                 if (uio->uio_segflg == UIO_NOCOPY) {
 284                         /*
 285                          * Issuing a write with the same data backing the
 286                          * buffer.  Instantiate the buffer to collect the
 287                          * backing vm pages, then read-in any missing bits.
 288                          *
 289                          * This case is used by vop_stdputpages().
 290                          */
 291                         bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
 292                                     0, 0);
 293                         if ((bp->b_flags & B_CACHE) == 0) {
 294                                 bqrelse(bp);
 295                                 error = bread(ap->a_vp,
 296                                               uio->uio_offset - offset,
 297                                               HAMMER_BUFSIZE, &bp);
 298                                 if (error) {
 299                                         brelse(bp);
 300                                         break;
 301                                 }
 302                         }
 303                 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
 304                         /*
 305                          * entirely overwrite the buffer
 306                          */
 307                         bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
 308                                     0, 0);
 309                 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
 310                         /*
 311                          * XXX
 312                          */
 313                         bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
 314                                     0, 0);
 315                         vfs_bio_clrbuf(bp);
 316                 } else {
 317                         /*
 318                          * Partial overwrite, read in any missing bits then
 319                          * replace the portion being written.
 320                          */
 321                         error = bread(ap->a_vp, uio->uio_offset - offset,
 322                                       HAMMER_BUFSIZE, &bp);
 323                         if (error) {
 324                                 brelse(bp);
 325                                 break;
 326                         }
 327                 }
 328                 n = HAMMER_BUFSIZE - offset;
 329                 if (n > uio->uio_resid)
 330                         n = uio->uio_resid;
 331                 error = uiomove((char *)bp->b_data + offset, n, uio);
 332                 if (error) {
 333                         brelse(bp);
 334                         break;
 335                 }
 336                 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
 337                 if (ip->ino_rec.ino_size < uio->uio_offset) {
 338                         ip->ino_rec.ino_size = uio->uio_offset;
 339                         ip->ino_rec.ino_mtime = trans.tid;
 340                         flags = HAMMER_INODE_RDIRTY | HAMMER_INODE_ITIMES |
 341                                 HAMMER_INODE_TID;
 342                         vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
 343                 } else {
 344                         flags = HAMMER_INODE_TID;
 345                 }
 346                 hammer_modify_inode(&trans, ip, flags);
 347                 if (ap->a_ioflag & IO_SYNC) {
 348                         bwrite(bp);
 349                 } else if (ap->a_ioflag & IO_DIRECT) {
 350                         bawrite(bp);
 351                 } else {
 352                         bdwrite(bp);
 353                 }
 354         }
 355         if (error)
 356                 hammer_abort_transaction(&trans);
 357         else
 358                 hammer_commit_transaction(&trans);
 359         return (error);
 360 }
 361
 362 /*
 363  * hammer_vop_access { vp, mode, cred }
 364  */
 365 static
 366 int
 367 hammer_vop_access(struct vop_access_args *ap)
 368 {
 369         struct hammer_inode *ip = VTOI(ap->a_vp);
 370         uid_t uid;
 371         gid_t gid;
 372         int error;
 373
 374         uid = hammer_to_unix_xid(&ip->ino_data.uid);
 375         gid = hammer_to_unix_xid(&ip->ino_data.gid);
 376
 377         error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
 378                                   ip->ino_data.uflags);
 379         return (error);
 380 }
 381
 382 /*
 383  * hammer_vop_advlock { vp, id, op, fl, flags }
 384  */
 385 static
 386 int
 387 hammer_vop_advlock(struct vop_advlock_args *ap)
 388 {
 389         struct hammer_inode *ip = VTOI(ap->a_vp);
 390
 391         return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
 392 }
 393
 394 /*
 395  * hammer_vop_close { vp, fflag }
 396  */
 397 static
 398 int
 399 hammer_vop_close(struct vop_close_args *ap)
 400 {
 401         return (vop_stdclose(ap));
 402 }
 403
 404 /*
 405  * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
 406  *
 407  * The operating system has already ensured that the directory entry
 408  * does not exist and done all appropriate namespace locking.
 409  */
 410 static
 411 int
 412 hammer_vop_ncreate(struct vop_ncreate_args *ap)
 413 {
 414         struct hammer_transaction trans;
 415         struct hammer_inode *dip;
 416         struct hammer_inode *nip;
 417         struct nchandle *nch;
 418         int error;
 419
 420         nch = ap->a_nch;
 421         dip = VTOI(ap->a_dvp);
 422
 423         /*
 424          * Create a transaction to cover the operations we perform.
 425          */
 426         hammer_start_transaction(&trans, dip->hmp);
 427
 428         /*
 429          * Create a new filesystem object of the requested type.  The
 430          * returned inode will be referenced but not locked.
 431          */
 432
 433         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 434         if (error)
 435                 kprintf("hammer_create_inode error %d\n", error);
 436         if (error) {
 437                 hammer_abort_transaction(&trans);
 438                 *ap->a_vpp = NULL;
 439                 return (error);
 440         }
 441
 442         /*
 443          * Add the new filesystem object to the directory.  This will also
 444          * bump the inode's link count.
 445          */
 446         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 447         if (error)
 448                 kprintf("hammer_ip_add_directory error %d\n", error);
 449
 450         /*
 451          * Finish up.
 452          */
 453         if (error) {
 454                 hammer_rel_inode(nip, 0);
 455                 hammer_abort_transaction(&trans);
 456                 *ap->a_vpp = NULL;
 457         } else {
 458                 hammer_commit_transaction(&trans);
 459                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 460                 hammer_rel_inode(nip, 0);
 461                 if (error == 0) {
 462                         cache_setunresolved(ap->a_nch);
 463                         cache_setvp(ap->a_nch, *ap->a_vpp);
 464                 }
 465         }
 466         return (error);
 467 }
 468
 469 /*
 470  * hammer_vop_getattr { vp, vap }
 471  */
 472 static
 473 int
 474 hammer_vop_getattr(struct vop_getattr_args *ap)
 475 {
 476         struct hammer_inode *ip = VTOI(ap->a_vp);
 477         struct vattr *vap = ap->a_vap;
 478
 479 #if 0
 480         if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
 481             (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
 482             ip->obj_asof == XXX
 483         ) {
 484                 /* LAZYMOD XXX */
 485         }
 486         hammer_itimes(ap->a_vp);
 487 #endif
 488
 489         vap->va_fsid = ip->hmp->fsid_udev;
 490         vap->va_fileid = ip->ino_rec.base.base.obj_id;
 491         vap->va_mode = ip->ino_data.mode;
 492         vap->va_nlink = ip->ino_rec.ino_nlinks;
 493         vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
 494         vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
 495         vap->va_rmajor = 0;
 496         vap->va_rminor = 0;
 497         vap->va_size = ip->ino_rec.ino_size;
 498         hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
 499         hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
 500         hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
 501         vap->va_flags = ip->ino_data.uflags;
 502         vap->va_gen = 1;        /* hammer inums are unique for all time */
 503         vap->va_blocksize = 32768; /* XXX - extract from root volume */
 504         vap->va_bytes = ip->ino_rec.ino_size;
 505         vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
 506         vap->va_filerev = 0;    /* XXX */
 507         /* mtime uniquely identifies any adjustments made to the file */
 508         vap->va_fsmid = ip->ino_rec.ino_mtime;
 509         vap->va_uid_uuid = ip->ino_data.uid;
 510         vap->va_gid_uuid = ip->ino_data.gid;
 511         vap->va_fsid_uuid = ip->hmp->fsid;
 512         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 513                           VA_FSID_UUID_VALID;
 514
 515         switch (ip->ino_rec.base.base.obj_type) {
 516         case HAMMER_OBJTYPE_CDEV:
 517         case HAMMER_OBJTYPE_BDEV:
 518                 vap->va_rmajor = ip->ino_data.rmajor;
 519                 vap->va_rminor = ip->ino_data.rminor;
 520                 break;
 521         default:
 522                 break;
 523         }
 524
 525         return(0);
 526 }
 527
 528 /*
 529  * hammer_vop_nresolve { nch, dvp, cred }
 530  *
 531  * Locate the requested directory entry.
 532  */
 533 static
 534 int
 535 hammer_vop_nresolve(struct vop_nresolve_args *ap)
 536 {
 537         struct namecache *ncp;
 538         hammer_inode_t dip;
 539         hammer_inode_t ip;
 540         hammer_tid_t asof;
 541         struct hammer_cursor cursor;
 542         union hammer_record_ondisk *rec;
 543         struct vnode *vp;
 544         int64_t namekey;
 545         int error;
 546         int i;
 547         int nlen;
 548
 549         /*
 550          * Misc initialization, plus handle as-of name extensions.  Look for
 551          * the '@@' extension.  Note that as-of files and directories cannot
 552          * be modified.
 553          *
 554          *
 555          */
 556         dip = VTOI(ap->a_dvp);
 557         ncp = ap->a_nch->ncp;
 558         asof = dip->obj_asof;
 559         nlen = ncp->nc_nlen;
 560
 561         for (i = 0; i < nlen; ++i) {
 562                 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
 563                         asof = hammer_now_tid() -
 564                                strtoq(ncp->nc_name + i + 2, NULL, 0) *
 565                                1000000000LL;
 566                         kprintf("ASOF %016llx\n", asof);
 567                         break;
 568                 }
 569         }
 570         nlen = i;
 571
 572         /*
 573          * Calculate the namekey and setup the key range for the scan.  This
 574          * works kinda like a chained hash table where the lower 32 bits
 575          * of the namekey synthesize the chain.
 576          *
 577          * The key range is inclusive of both key_beg and key_end.
 578          */
 579         namekey = hammer_directory_namekey(ncp->nc_name, nlen);
 580
 581         hammer_init_cursor_ip(&cursor, dip);
 582         cursor.key_beg.obj_id = dip->obj_id;
 583         cursor.key_beg.key = namekey;
 584         cursor.key_beg.create_tid = asof;
 585         cursor.key_beg.delete_tid = 0;
 586         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 587         cursor.key_beg.obj_type = 0;
 588
 589         cursor.key_end = cursor.key_beg;
 590         cursor.key_end.key |= 0xFFFFFFFFULL;
 591         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
 592
 593         /*
 594          * Scan all matching records (the chain), locate the one matching
 595          * the requested path component.
 596          *
 597          * The hammer_ip_*() functions merge in-memory records with on-disk
 598          * records for the purposes of the search.
 599          */
 600         error = hammer_ip_first(&cursor, dip);
 601         rec = NULL;
 602         while (error == 0) {
 603                 error = hammer_ip_resolve_data(&cursor);
 604                 if (error)
 605                         break;
 606                 rec = cursor.record;
 607                 if (nlen == rec->entry.base.data_len &&
 608                     bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
 609                         break;
 610                 }
 611                 error = hammer_ip_next(&cursor);
 612         }
 613         if (error == 0) {
 614                 ip = hammer_get_inode(dip->hmp, rec->entry.obj_id,
 615                                       asof, &error);
 616                 if (error == 0) {
 617                         error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
 618                         hammer_rel_inode(ip, 0);
 619                 } else {
 620                         vp = NULL;
 621                 }
 622                 if (error == 0) {
 623                         vn_unlock(vp);
 624                         cache_setvp(ap->a_nch, vp);
 625                         vrele(vp);
 626                 }
 627         } else if (error == ENOENT) {
 628                 cache_setvp(ap->a_nch, NULL);
 629         }
 630         hammer_done_cursor(&cursor);
 631         return (error);
 632 }
 633
 634 /*
 635  * hammer_vop_nlookupdotdot { dvp, vpp, cred }
 636  *
 637  * Locate the parent directory of a directory vnode.
 638  *
 639  * dvp is referenced but not locked.  *vpp must be returned referenced and
 640  * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
 641  * at the root, instead it could indicate that the directory we were in was
 642  * removed.
 643  */
 644 static
 645 int
 646 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
 647 {
 648         struct hammer_inode *dip;
 649         u_int64_t parent_obj_id;
 650
 651         dip = VTOI(ap->a_dvp);
 652         if ((parent_obj_id = dip->ino_data.parent_obj_id) == 0) {
 653                 *ap->a_vpp = NULL;
 654                 return ENOENT;
 655         }
 656         return(hammer_vfs_vget(dip->hmp->mp, parent_obj_id, ap->a_vpp));
 657 }
 658
 659 /*
 660  * hammer_vop_nlink { nch, dvp, vp, cred }
 661  */
 662 static
 663 int
 664 hammer_vop_nlink(struct vop_nlink_args *ap)
 665 {
 666         struct hammer_transaction trans;
 667         struct hammer_inode *dip;
 668         struct hammer_inode *ip;
 669         struct nchandle *nch;
 670         int error;
 671
 672         nch = ap->a_nch;
 673         dip = VTOI(ap->a_dvp);
 674         ip = VTOI(ap->a_vp);
 675
 676         /*
 677          * Create a transaction to cover the operations we perform.
 678          */
 679         hammer_start_transaction(&trans, dip->hmp);
 680
 681         /*
 682          * Add the filesystem object to the directory.  Note that neither
 683          * dip nor ip are referenced or locked, but their vnodes are
 684          * referenced.  This function will bump the inode's link count.
 685          */
 686         error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
 687
 688         /*
 689          * Finish up.
 690          */
 691         if (error) {
 692                 hammer_abort_transaction(&trans);
 693         } else {
 694                 cache_setunresolved(nch);
 695                 cache_setvp(nch, ap->a_vp);
 696                 hammer_commit_transaction(&trans);
 697         }
 698         return (error);
 699 }
 700
 701 /*
 702  * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
 703  *
 704  * The operating system has already ensured that the directory entry
 705  * does not exist and done all appropriate namespace locking.
 706  */
 707 static
 708 int
 709 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
 710 {
 711         struct hammer_transaction trans;
 712         struct hammer_inode *dip;
 713         struct hammer_inode *nip;
 714         struct nchandle *nch;
 715         int error;
 716
 717         nch = ap->a_nch;
 718         dip = VTOI(ap->a_dvp);
 719
 720         /*
 721          * Create a transaction to cover the operations we perform.
 722          */
 723         hammer_start_transaction(&trans, dip->hmp);
 724
 725         /*
 726          * Create a new filesystem object of the requested type.  The
 727          * returned inode will be referenced but not locked.
 728          */
 729         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 730         if (error)
 731                 kprintf("hammer_mkdir error %d\n", error);
 732         if (error) {
 733                 hammer_abort_transaction(&trans);
 734                 *ap->a_vpp = NULL;
 735                 return (error);
 736         }
 737
 738         /*
 739          * Add the new filesystem object to the directory.  This will also
 740          * bump the inode's link count.
 741          */
 742         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 743         if (error)
 744                 kprintf("hammer_mkdir (add) error %d\n", error);
 745
 746         /*
 747          * Finish up.
 748          */
 749         if (error) {
 750                 hammer_rel_inode(nip, 0);
 751                 hammer_abort_transaction(&trans);
 752                 *ap->a_vpp = NULL;
 753         } else {
 754                 hammer_commit_transaction(&trans);
 755                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 756                 hammer_rel_inode(nip, 0);
 757                 if (error == 0) {
 758                         cache_setunresolved(ap->a_nch);
 759                         cache_setvp(ap->a_nch, *ap->a_vpp);
 760                 }
 761         }
 762         return (error);
 763 }
 764
 765 /*
 766  * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
 767  *
 768  * The operating system has already ensured that the directory entry
 769  * does not exist and done all appropriate namespace locking.
 770  */
 771 static
 772 int
 773 hammer_vop_nmknod(struct vop_nmknod_args *ap)
 774 {
 775         struct hammer_transaction trans;
 776         struct hammer_inode *dip;
 777         struct hammer_inode *nip;
 778         struct nchandle *nch;
 779         int error;
 780
 781         nch = ap->a_nch;
 782         dip = VTOI(ap->a_dvp);
 783
 784         /*
 785          * Create a transaction to cover the operations we perform.
 786          */
 787         hammer_start_transaction(&trans, dip->hmp);
 788
 789         /*
 790          * Create a new filesystem object of the requested type.  The
 791          * returned inode will be referenced but not locked.
 792          */
 793         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
 794         if (error) {
 795                 hammer_abort_transaction(&trans);
 796                 *ap->a_vpp = NULL;
 797                 return (error);
 798         }
 799
 800         /*
 801          * Add the new filesystem object to the directory.  This will also
 802          * bump the inode's link count.
 803          */
 804         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
 805
 806         /*
 807          * Finish up.
 808          */
 809         if (error) {
 810                 hammer_rel_inode(nip, 0);
 811                 hammer_abort_transaction(&trans);
 812                 *ap->a_vpp = NULL;
 813         } else {
 814                 hammer_commit_transaction(&trans);
 815                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
 816                 hammer_rel_inode(nip, 0);
 817                 if (error == 0) {
 818                         cache_setunresolved(ap->a_nch);
 819                         cache_setvp(ap->a_nch, *ap->a_vpp);
 820                 }
 821         }
 822         return (error);
 823 }
 824
 825 /*
 826  * hammer_vop_open { vp, mode, cred, fp }
 827  */
 828 static
 829 int
 830 hammer_vop_open(struct vop_open_args *ap)
 831 {
 832         return(vop_stdopen(ap));
 833 }
 834
 835 /*
 836  * hammer_vop_pathconf { vp, name, retval }
 837  */
 838 static
 839 int
 840 hammer_vop_pathconf(struct vop_pathconf_args *ap)
 841 {
 842         return EOPNOTSUPP;
 843 }
 844
 845 /*
 846  * hammer_vop_print { vp }
 847  */
 848 static
 849 int
 850 hammer_vop_print(struct vop_print_args *ap)
 851 {
 852         return EOPNOTSUPP;
 853 }
 854
 855 /*
 856  * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 857  */
 858 static
 859 int
 860 hammer_vop_readdir(struct vop_readdir_args *ap)
 861 {
 862         struct hammer_cursor cursor;
 863         struct hammer_inode *ip;
 864         struct uio *uio;
 865         hammer_record_ondisk_t rec;
 866         hammer_base_elm_t base;
 867         int error;
 868         int cookie_index;
 869         int ncookies;
 870         off_t *cookies;
 871         off_t saveoff;
 872         int r;
 873
 874         ip = VTOI(ap->a_vp);
 875         uio = ap->a_uio;
 876         saveoff = uio->uio_offset;
 877
 878         if (ap->a_ncookies) {
 879                 ncookies = uio->uio_resid / 16 + 1;
 880                 if (ncookies > 1024)
 881                         ncookies = 1024;
 882                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 883                 cookie_index = 0;
 884         } else {
 885                 ncookies = -1;
 886                 cookies = NULL;
 887                 cookie_index = 0;
 888         }
 889
 890         /*
 891          * Handle artificial entries
 892          */
 893         error = 0;
 894         if (saveoff == 0) {
 895                 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
 896                 if (r)
 897                         goto done;
 898                 if (cookies)
 899                         cookies[cookie_index] = saveoff;
 900                 ++saveoff;
 901                 ++cookie_index;
 902                 if (cookie_index == ncookies)
 903                         goto done;
 904         }
 905         if (saveoff == 1) {
 906                 if (ip->ino_data.parent_obj_id) {
 907                         r = vop_write_dirent(&error, uio,
 908                                              ip->ino_data.parent_obj_id,
 909                                              DT_DIR, 2, "..");
 910                 } else {
 911                         r = vop_write_dirent(&error, uio,
 912                                              ip->obj_id, DT_DIR, 2, "..");
 913                 }
 914                 if (r)
 915                         goto done;
 916                 if (cookies)
 917                         cookies[cookie_index] = saveoff;
 918                 ++saveoff;
 919                 ++cookie_index;
 920                 if (cookie_index == ncookies)
 921                         goto done;
 922         }
 923
 924         /*
 925          * Key range (begin and end inclusive) to scan.  Directory keys
 926          * directly translate to a 64 bit 'seek' position.
 927          */
 928         hammer_init_cursor_ip(&cursor, ip);
 929         cursor.key_beg.obj_id = ip->obj_id;
 930         cursor.key_beg.create_tid = ip->obj_asof;
 931         cursor.key_beg.delete_tid = 0;
 932         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
 933         cursor.key_beg.obj_type = 0;
 934         cursor.key_beg.key = saveoff;
 935
 936         cursor.key_end = cursor.key_beg;
 937         cursor.key_end.key = HAMMER_MAX_KEY;
 938         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
 939
 940         error = hammer_ip_first(&cursor, ip);
 941
 942         while (error == 0) {
 943                 error = hammer_ip_resolve_data(&cursor);
 944                 if (error)
 945                         break;
 946                 rec = cursor.record;
 947                 base = &rec->base.base;
 948                 saveoff = base->key;
 949
 950                 if (base->obj_id != ip->obj_id)
 951                         panic("readdir: bad record at %p", cursor.node);
 952
 953                 r = vop_write_dirent(
 954                              &error, uio, rec->entry.obj_id,
 955                              hammer_get_dtype(rec->entry.base.base.obj_type),
 956                              rec->entry.base.data_len,
 957                              (void *)cursor.data);
 958                 if (r)
 959                         break;
 960                 ++saveoff;
 961                 if (cookies)
 962                         cookies[cookie_index] = base->key;
 963                 ++cookie_index;
 964                 if (cookie_index == ncookies)
 965                         break;
 966                 error = hammer_ip_next(&cursor);
 967         }
 968         hammer_done_cursor(&cursor);
 969
 970 done:
 971         if (ap->a_eofflag)
 972                 *ap->a_eofflag = (error == ENOENT);
 973         uio->uio_offset = saveoff;
 974         if (error && cookie_index == 0) {
 975                 if (error == ENOENT)
 976                         error = 0;
 977                 if (cookies) {
 978                         kfree(cookies, M_TEMP);
 979                         *ap->a_ncookies = 0;
 980                         *ap->a_cookies = NULL;
 981                 }
 982         } else {
 983                 if (error == ENOENT)
 984                         error = 0;
 985                 if (cookies) {
 986                         *ap->a_ncookies = cookie_index;
 987                         *ap->a_cookies = cookies;
 988                 }
 989         }
 990         return(error);
 991 }
 992
 993 /*
 994  * hammer_vop_readlink { vp, uio, cred }
 995  */
 996 static
 997 int
 998 hammer_vop_readlink(struct vop_readlink_args *ap)
 999 {
1000         struct hammer_cursor cursor;
1001         struct hammer_inode *ip;
1002         int error;
1003
1004         ip = VTOI(ap->a_vp);
1005         hammer_init_cursor_ip(&cursor, ip);
1006
1007         /*
1008          * Key range (begin and end inclusive) to scan.  Directory keys
1009          * directly translate to a 64 bit 'seek' position.
1010          */
1011         cursor.key_beg.obj_id = ip->obj_id;
1012         cursor.key_beg.create_tid = ip->obj_asof;
1013         cursor.key_beg.delete_tid = 0;
1014         cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1015         cursor.key_beg.obj_type = 0;
1016         cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1017
1018         error = hammer_ip_lookup(&cursor, ip);
1019         if (error == 0) {
1020                 error = hammer_ip_resolve_data(&cursor);
1021                 if (error == 0) {
1022                         error = uiomove((char *)cursor.data,
1023                                         cursor.record->generic.base.data_len,
1024                                         ap->a_uio);
1025                 }
1026         }
1027         hammer_done_cursor(&cursor);
1028         return(error);
1029 }
1030
1031 /*
1032  * hammer_vop_nremove { nch, dvp, cred }
1033  */
1034 static
1035 int
1036 hammer_vop_nremove(struct vop_nremove_args *ap)
1037 {
1038         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1039 }
1040
1041 /*
1042  * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1043  */
1044 static
1045 int
1046 hammer_vop_nrename(struct vop_nrename_args *ap)
1047 {
1048         struct hammer_transaction trans;
1049         struct namecache *fncp;
1050         struct namecache *tncp;
1051         struct hammer_inode *fdip;
1052         struct hammer_inode *tdip;
1053         struct hammer_inode *ip;
1054         struct hammer_cursor cursor;
1055         union hammer_record_ondisk *rec;
1056         int64_t namekey;
1057         int error;
1058
1059         fdip = VTOI(ap->a_fdvp);
1060         tdip = VTOI(ap->a_tdvp);
1061         fncp = ap->a_fnch->ncp;
1062         tncp = ap->a_tnch->ncp;
1063         ip = VTOI(fncp->nc_vp);
1064         KKASSERT(ip != NULL);
1065         hammer_start_transaction(&trans, fdip->hmp);
1066
1067         /*
1068          * Remove tncp from the target directory and then link ip as
1069          * tncp. XXX pass trans to dounlink
1070          */
1071         error = hammer_dounlink(ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1072         if (error == 0 || error == ENOENT)
1073                 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1074         if (error)
1075                 goto failed; /* XXX */
1076
1077         /*
1078          * Locate the record in the originating directory and remove it.
1079          *
1080          * Calculate the namekey and setup the key range for the scan.  This
1081          * works kinda like a chained hash table where the lower 32 bits
1082          * of the namekey synthesize the chain.
1083          *
1084          * The key range is inclusive of both key_beg and key_end.
1085          */
1086         namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1087
1088         hammer_init_cursor_ip(&cursor, fdip);
1089         cursor.key_beg.obj_id = fdip->obj_id;
1090         cursor.key_beg.key = namekey;
1091         cursor.key_beg.create_tid = fdip->obj_asof;
1092         cursor.key_beg.delete_tid = 0;
1093         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1094         cursor.key_beg.obj_type = 0;
1095
1096         cursor.key_end = cursor.key_beg;
1097         cursor.key_end.key |= 0xFFFFFFFFULL;
1098         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1099
1100         /*
1101          * Scan all matching records (the chain), locate the one matching
1102          * the requested path component.
1103          *
1104          * The hammer_ip_*() functions merge in-memory records with on-disk
1105          * records for the purposes of the search.
1106          */
1107         error = hammer_ip_first(&cursor, fdip);
1108         while (error == 0) {
1109                 if (hammer_ip_resolve_data(&cursor) != 0)
1110                         break;
1111                 rec = cursor.record;
1112                 if (fncp->nc_nlen == rec->entry.base.data_len &&
1113                     bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1114                         break;
1115                 }
1116                 error = hammer_ip_next(&cursor);
1117         }
1118
1119         /*
1120          * If all is ok we have to get the inode so we can adjust nlinks.
1121          */
1122         if (error)
1123                 goto done;
1124         error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1125
1126         if (error == 0) {
1127                 cache_rename(ap->a_fnch, ap->a_tnch);
1128                 cache_setvp(ap->a_tnch, ip->vp);
1129         }
1130 done:
1131         hammer_done_cursor(&cursor);
1132 failed:
1133         if (error == 0) {
1134                 hammer_commit_transaction(&trans);
1135         } else {
1136                 hammer_abort_transaction(&trans);
1137         }
1138         return (error);
1139 }
1140
1141 /*
1142  * hammer_vop_nrmdir { nch, dvp, cred }
1143  */
1144 static
1145 int
1146 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1147 {
1148         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1149 }
1150
1151 /*
1152  * hammer_vop_setattr { vp, vap, cred }
1153  */
1154 static
1155 int
1156 hammer_vop_setattr(struct vop_setattr_args *ap)
1157 {
1158         struct hammer_transaction trans;
1159         struct hammer_cursor *spike = NULL;
1160         struct vattr *vap;
1161         struct hammer_inode *ip;
1162         int modflags;
1163         int error;
1164         int64_t aligned_size;
1165         u_int32_t flags;
1166         uuid_t uuid;
1167
1168         vap = ap->a_vap;
1169         ip = ap->a_vp->v_data;
1170         modflags = 0;
1171
1172         if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1173                 return(EROFS);
1174
1175         hammer_start_transaction(&trans, ip->hmp);
1176         error = 0;
1177
1178         if (vap->va_flags != VNOVAL) {
1179                 flags = ip->ino_data.uflags;
1180                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1181                                          hammer_to_unix_xid(&ip->ino_data.uid),
1182                                          ap->a_cred);
1183                 if (error == 0) {
1184                         if (ip->ino_data.uflags != flags) {
1185                                 ip->ino_data.uflags = flags;
1186                                 modflags |= HAMMER_INODE_DDIRTY;
1187                         }
1188                         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1189                                 error = 0;
1190                                 goto done;
1191                         }
1192                 }
1193                 goto done;
1194         }
1195         if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1196                 error = EPERM;
1197                 goto done;
1198         }
1199         if (vap->va_uid != (uid_t)VNOVAL) {
1200                 hammer_guid_to_uuid(&uuid, vap->va_uid);
1201                 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1202                         ip->ino_data.uid = uuid;
1203                         modflags |= HAMMER_INODE_DDIRTY;
1204                 }
1205         }
1206         if (vap->va_gid != (uid_t)VNOVAL) {
1207                 hammer_guid_to_uuid(&uuid, vap->va_gid);
1208                 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1209                         ip->ino_data.gid = uuid;
1210                         modflags |= HAMMER_INODE_DDIRTY;
1211                 }
1212         }
1213         while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1214                 switch(ap->a_vp->v_type) {
1215                 case VREG:
1216                         if (vap->va_size < ip->ino_rec.ino_size) {
1217                                 vtruncbuf(ap->a_vp, vap->va_size,
1218                                           HAMMER_BUFSIZE);
1219                         } else if (vap->va_size > ip->ino_rec.ino_size) {
1220                                 vnode_pager_setsize(ap->a_vp, vap->va_size);
1221                         }
1222                         aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1223                                         ~(int64_t)HAMMER_BUFMASK;
1224                         error = hammer_ip_delete_range(&trans, ip,
1225                                                     aligned_size,
1226                                                     0x7FFFFFFFFFFFFFFFLL,
1227                                                     &spike);
1228                         ip->ino_rec.ino_size = vap->va_size;
1229                         modflags |= HAMMER_INODE_RDIRTY;
1230                         break;
1231                 case VDATABASE:
1232                         error = hammer_ip_delete_range(&trans, ip,
1233                                                     vap->va_size,
1234                                                     0x7FFFFFFFFFFFFFFFLL,
1235                                                     &spike);
1236                         ip->ino_rec.ino_size = vap->va_size;
1237                         modflags |= HAMMER_INODE_RDIRTY;
1238                         break;
1239                 default:
1240                         error = EINVAL;
1241                         goto done;
1242                 }
1243                 if (error == ENOSPC) {
1244                         error = hammer_spike(&spike);
1245                         if (error == 0)
1246                                 continue;
1247                 }
1248                 KKASSERT(spike == NULL);
1249                 break;
1250         }
1251         if (vap->va_atime.tv_sec != VNOVAL) {
1252                 ip->ino_rec.ino_atime =
1253                         hammer_timespec_to_transid(&vap->va_atime);
1254                 modflags |= HAMMER_INODE_ITIMES;
1255         }
1256         if (vap->va_mtime.tv_sec != VNOVAL) {
1257                 ip->ino_rec.ino_mtime =
1258                         hammer_timespec_to_transid(&vap->va_mtime);
1259                 modflags |= HAMMER_INODE_ITIMES;
1260         }
1261         if (vap->va_mode != (mode_t)VNOVAL) {
1262                 if (ip->ino_data.mode != vap->va_mode) {
1263                         ip->ino_data.mode = vap->va_mode;
1264                         modflags |= HAMMER_INODE_DDIRTY;
1265                 }
1266         }
1267 done:
1268         if (error) {
1269                 hammer_abort_transaction(&trans);
1270         } else {
1271                 if (modflags & (HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY))
1272                         modflags |= HAMMER_INODE_TID;
1273                 hammer_modify_inode(&trans, ip, modflags);
1274                 hammer_commit_transaction(&trans);
1275         }
1276         return (error);
1277 }
1278
1279 /*
1280  * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1281  */
1282 static
1283 int
1284 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1285 {
1286         struct hammer_transaction trans;
1287         struct hammer_inode *dip;
1288         struct hammer_inode *nip;
1289         struct nchandle *nch;
1290         hammer_record_t record;
1291         int error;
1292         int bytes;
1293
1294         ap->a_vap->va_type = VLNK;
1295
1296         nch = ap->a_nch;
1297         dip = VTOI(ap->a_dvp);
1298
1299         /*
1300          * Create a transaction to cover the operations we perform.
1301          */
1302         hammer_start_transaction(&trans, dip->hmp);
1303
1304         /*
1305          * Create a new filesystem object of the requested type.  The
1306          * returned inode will be referenced but not locked.
1307          */
1308
1309         error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1310         if (error) {
1311                 hammer_abort_transaction(&trans);
1312                 *ap->a_vpp = NULL;
1313                 return (error);
1314         }
1315
1316         /*
1317          * Add the new filesystem object to the directory.  This will also
1318          * bump the inode's link count.
1319          */
1320         error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1321
1322         /*
1323          * Add a record representing the symlink.  symlink stores the link
1324          * as pure data, not a string, and is no \0 terminated.
1325          */
1326         if (error == 0) {
1327                 record = hammer_alloc_mem_record(nip);
1328                 bytes = strlen(ap->a_target);
1329
1330                 record->rec.generic.base.base.key = HAMMER_FIXKEY_SYMLINK;
1331                 record->rec.generic.base.base.rec_type = HAMMER_RECTYPE_FIX;
1332                 record->rec.generic.base.data_len = bytes;
1333                 if (bytes <= sizeof(record->rec.generic.filler)) {
1334                         record->data = (void *)record->rec.generic.filler;
1335                         bcopy(ap->a_target, record->data, bytes);
1336                 } else {
1337                         record->data = (void *)ap->a_target;
1338                         /* will be reallocated by routine below */
1339                 }
1340                 error = hammer_ip_add_record(&trans, record);
1341         }
1342
1343         /*
1344          * Finish up.
1345          */
1346         if (error) {
1347                 hammer_rel_inode(nip, 0);
1348                 hammer_abort_transaction(&trans);
1349                 *ap->a_vpp = NULL;
1350         } else {
1351                 hammer_commit_transaction(&trans);
1352                 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1353                 hammer_rel_inode(nip, 0);
1354                 if (error == 0) {
1355                         cache_setunresolved(ap->a_nch);
1356                         cache_setvp(ap->a_nch, *ap->a_vpp);
1357                 }
1358         }
1359         return (error);
1360 }
1361
1362 /*
1363  * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1364  */
1365 static
1366 int
1367 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1368 {
1369         return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1370 }
1371
1372 /*
1373  * hammer_vop_strategy { vp, bio }
1374  *
1375  * Strategy call, used for regular file read & write only.  Note that the
1376  * bp may represent a cluster.
1377  *
1378  * To simplify operation and allow better optimizations in the future,
1379  * this code does not make any assumptions with regards to buffer alignment
1380  * or size.
1381  */
1382 static
1383 int
1384 hammer_vop_strategy(struct vop_strategy_args *ap)
1385 {
1386         struct buf *bp;
1387         int error;
1388
1389         bp = ap->a_bio->bio_buf;
1390
1391         switch(bp->b_cmd) {
1392         case BUF_CMD_READ:
1393                 error = hammer_vop_strategy_read(ap);
1394                 break;
1395         case BUF_CMD_WRITE:
1396                 error = hammer_vop_strategy_write(ap);
1397                 break;
1398         default:
1399                 error = EINVAL;
1400                 break;
1401         }
1402         bp->b_error = error;
1403         if (error)
1404                 bp->b_flags |= B_ERROR;
1405         biodone(ap->a_bio);
1406         return (error);
1407 }
1408
1409 /*
1410  * Read from a regular file.  Iterate the related records and fill in the
1411  * BIO/BUF.  Gaps are zero-filled.
1412  *
1413  * The support code in hammer_object.c should be used to deal with mixed
1414  * in-memory and on-disk records.
1415  *
1416  * XXX atime update
1417  */
1418 static
1419 int
1420 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1421 {
1422         struct hammer_inode *ip = ap->a_vp->v_data;
1423         struct hammer_cursor cursor;
1424         hammer_record_ondisk_t rec;
1425         hammer_base_elm_t base;
1426         struct bio *bio;
1427         struct buf *bp;
1428         int64_t rec_offset;
1429         int64_t ran_end;
1430         int64_t tmp64;
1431         int error;
1432         int boff;
1433         int roff;
1434         int n;
1435
1436         bio = ap->a_bio;
1437         bp = bio->bio_buf;
1438
1439         hammer_init_cursor_ip(&cursor, ip);
1440
1441         /*
1442          * Key range (begin and end inclusive) to scan.  Note that the key's
1443          * stored in the actual records represent BASE+LEN, not BASE.  The
1444          * first record containing bio_offset will have a key > bio_offset.
1445          */
1446         cursor.key_beg.obj_id = ip->obj_id;
1447         cursor.key_beg.create_tid = ip->obj_asof;
1448         cursor.key_beg.delete_tid = 0;
1449         cursor.key_beg.obj_type = 0;
1450         cursor.key_beg.key = bio->bio_offset + 1;
1451
1452         cursor.key_end = cursor.key_beg;
1453         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1454                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1455                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1456                 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1457         } else {
1458                 ran_end = bio->bio_offset + bp->b_bufsize;
1459                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1460                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1461                 tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
1462                 if (tmp64 < ran_end)
1463                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1464                 else
1465                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1466         }
1467         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1468
1469         error = hammer_ip_first(&cursor, ip);
1470         boff = 0;
1471
1472         while (error == 0) {
1473                 error = hammer_ip_resolve_data(&cursor);
1474                 if (error)
1475                         break;
1476                 rec = cursor.record;
1477                 base = &rec->base.base;
1478
1479                 rec_offset = base->key - rec->data.base.data_len;
1480
1481                 /*
1482                  * Calculate the gap, if any, and zero-fill it.
1483                  */
1484                 n = (int)(rec_offset - (bio->bio_offset + boff));
1485                 if (n > 0) {
1486                         if (n > bp->b_bufsize - boff)
1487                                 n = bp->b_bufsize - boff;
1488                         kprintf("zfill %d bytes\n", n);
1489                         bzero((char *)bp->b_data + boff, n);
1490                         boff += n;
1491                         n = 0;
1492                 }
1493
1494                 /*
1495                  * Calculate the data offset in the record and the number
1496                  * of bytes we can copy.
1497                  *
1498                  * Note there is a degenerate case here where boff may
1499                  * already be at bp->b_bufsize.
1500                  */
1501                 roff = -n;
1502                 n = rec->data.base.data_len - roff;
1503                 KKASSERT(n > 0);
1504                 if (n > bp->b_bufsize - boff)
1505                         n = bp->b_bufsize - boff;
1506                 bcopy((char *)cursor.data + roff, (char *)bp->b_data + boff, n);
1507                 boff += n;
1508                 if (boff == bp->b_bufsize)
1509                         break;
1510                 error = hammer_ip_next(&cursor);
1511         }
1512         hammer_done_cursor(&cursor);
1513
1514         /*
1515          * There may have been a gap after the last record
1516          */
1517         if (error == ENOENT)
1518                 error = 0;
1519         if (error == 0 && boff != bp->b_bufsize) {
1520                 KKASSERT(boff < bp->b_bufsize);
1521                 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1522                 /* boff = bp->b_bufsize; */
1523         }
1524         bp->b_resid = 0;
1525         return(error);
1526 }
1527
1528 /*
1529  * Write to a regular file.  Iterate the related records and mark for
1530  * deletion.  If existing edge records (left and right side) overlap our
1531  * write they have to be marked deleted and new records created, usually
1532  * referencing a portion of the original data.  Then add a record to
1533  * represent the buffer.
1534  *
1535  * The support code in hammer_object.c should be used to deal with mixed
1536  * in-memory and on-disk records.
1537  */
1538 static
1539 int
1540 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1541 {
1542         struct hammer_transaction trans;
1543         struct hammer_cursor *spike = NULL;
1544         hammer_inode_t ip;
1545         struct bio *bio;
1546         struct buf *bp;
1547         int error;
1548
1549         bio = ap->a_bio;
1550         bp = bio->bio_buf;
1551         ip = ap->a_vp->v_data;
1552         hammer_start_transaction(&trans, ip->hmp);
1553
1554 retry:
1555         /*
1556          * Delete any records overlapping our range.  This function will
1557          * (eventually) properly truncate partial overlaps.
1558          */
1559         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1560                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1561                                                bio->bio_offset, &spike);
1562         } else {
1563                 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1564                                                bio->bio_offset +
1565                                                 bp->b_bufsize - 1,
1566                                                &spike);
1567         }
1568
1569         /*
1570          * Add a single record to cover the write
1571          */
1572         if (error == 0) {
1573                 error = hammer_ip_sync_data(&trans, ip, bio->bio_offset,
1574                                             bp->b_data, bp->b_bufsize,
1575                                             &spike);
1576         }
1577
1578         /*
1579          * If we ran out of space the spike structure will be filled in
1580          * and we must call hammer_spike with it, then retry.
1581          */
1582         if (error == ENOSPC) {
1583                 error = hammer_spike(&spike);
1584                 if (error == 0)
1585                         goto retry;
1586         }
1587         KKASSERT(spike == NULL);
1588
1589         /*
1590          * If an error occured abort the transaction
1591          */
1592         if (error) {
1593                 /* XXX undo deletion */
1594                 hammer_abort_transaction(&trans);
1595                 bp->b_resid = bp->b_bufsize;
1596         } else {
1597                 hammer_commit_transaction(&trans);
1598                 bp->b_resid = 0;
1599         }
1600         return(error);
1601 }
1602
1603 /*
1604  * dounlink - disconnect a directory entry
1605  *
1606  * XXX whiteout support not really in yet
1607  */
1608 static int
1609 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1610                 int flags)
1611 {
1612         struct hammer_transaction trans;
1613         struct namecache *ncp;
1614         hammer_inode_t dip;
1615         hammer_inode_t ip;
1616         hammer_record_ondisk_t rec;
1617         struct hammer_cursor cursor;
1618         int64_t namekey;
1619         int error;
1620
1621         /*
1622          * Calculate the namekey and setup the key range for the scan.  This
1623          * works kinda like a chained hash table where the lower 32 bits
1624          * of the namekey synthesize the chain.
1625          *
1626          * The key range is inclusive of both key_beg and key_end.
1627          */
1628         dip = VTOI(dvp);
1629         ncp = nch->ncp;
1630         namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1631
1632         hammer_init_cursor_ip(&cursor, dip);
1633         cursor.key_beg.obj_id = dip->obj_id;
1634         cursor.key_beg.key = namekey;
1635         cursor.key_beg.create_tid = dip->obj_asof;
1636         cursor.key_beg.delete_tid = 0;
1637         cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1638         cursor.key_beg.obj_type = 0;
1639
1640         cursor.key_end = cursor.key_beg;
1641         cursor.key_end.key |= 0xFFFFFFFFULL;
1642         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1643
1644         hammer_start_transaction(&trans, dip->hmp);
1645
1646         /*
1647          * Scan all matching records (the chain), locate the one matching
1648          * the requested path component.  info->last_error contains the
1649          * error code on search termination and could be 0, ENOENT, or
1650          * something else.
1651          *
1652          * The hammer_ip_*() functions merge in-memory records with on-disk
1653          * records for the purposes of the search.
1654          */
1655         error = hammer_ip_first(&cursor, dip);
1656         while (error == 0) {
1657                 error = hammer_ip_resolve_data(&cursor);
1658                 if (error)
1659                         break;
1660                 rec = cursor.record;
1661                 if (ncp->nc_nlen == rec->entry.base.data_len &&
1662                     bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1663                         break;
1664                 }
1665                 error = hammer_ip_next(&cursor);
1666         }
1667
1668         /*
1669          * If all is ok we have to get the inode so we can adjust nlinks.
1670          *
1671          * If the target is a directory, it must be empty.
1672          */
1673         if (error == 0) {
1674                 ip = hammer_get_inode(dip->hmp, rec->entry.obj_id,
1675                                       dip->hmp->asof, &error);
1676                 if (error == 0 && ip->ino_rec.base.base.obj_type ==
1677                                   HAMMER_OBJTYPE_DIRECTORY) {
1678                         error = hammer_ip_check_directory_empty(&trans, ip);
1679                 }
1680                 if (error == 0)
1681                         error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1682                 if (error == 0) {
1683                         cache_setunresolved(nch);
1684                         cache_setvp(nch, NULL);
1685                         /* XXX locking */
1686                         if (ip->vp)
1687                                 cache_inval_vp(ip->vp, CINV_DESTROY);
1688                 }
1689                 hammer_rel_inode(ip, 0);
1690         }
1691
1692         if (error == 0)
1693                 hammer_commit_transaction(&trans);
1694         else
1695                 hammer_abort_transaction(&trans);
1696         hammer_done_cursor(&cursor);
1697         return (error);
1698 }
1699
1700 /************************************************************************
1701  *                          FIFO AND SPECFS OPS                         *
1702  ************************************************************************
1703  *
1704  */
1705
1706 static int
1707 hammer_vop_fifoclose (struct vop_close_args *ap)
1708 {
1709         /* XXX update itimes */
1710         return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1711 }
1712
1713 static int
1714 hammer_vop_fiforead (struct vop_read_args *ap)
1715 {
1716         int error;
1717
1718         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1719         /* XXX update access time */
1720         return (error);
1721 }
1722
1723 static int
1724 hammer_vop_fifowrite (struct vop_write_args *ap)
1725 {
1726         int error;
1727
1728         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1729         /* XXX update access time */
1730         return (error);
1731 }
1732
1733 static int
1734 hammer_vop_specclose (struct vop_close_args *ap)
1735 {
1736         /* XXX update itimes */
1737         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1738 }
1739
1740 static int
1741 hammer_vop_specread (struct vop_read_args *ap)
1742 {
1743         /* XXX update access time */
1744         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1745 }
1746
1747 static int
1748 hammer_vop_specwrite (struct vop_write_args *ap)
1749 {
1750         /* XXX update last change time */
1751         return (VOCALL(&spec_vnode_vops, &ap->a_head));
1752 }
1753