sys/vfs/hammer2/hammer2_vnops.c

   1 /*
   2  * Copyright (c) 2011-2015 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@dragonflybsd.org>
   6  * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
   7  * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  *
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in
  17  *    the documentation and/or other materials provided with the
  18  *    distribution.
  19  * 3. Neither the name of The DragonFly Project nor the names of its
  20  *    contributors may be used to endorse or promote products derived
  21  *    from this software without specific, prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  26  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  27  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  28  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  29  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  30  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  31  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  32  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  33  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  */
  36 /*
  37  * Kernel Filesystem interface
  38  *
  39  * NOTE! local ipdata pointers must be reloaded on any modifying operation
  40  *       to the inode as its underlying chain may have changed.
  41  */
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/kernel.h>
  46 #include <sys/fcntl.h>
  47 #include <sys/buf.h>
  48 #include <sys/proc.h>
  49 #include <sys/namei.h>
  50 #include <sys/mount.h>
  51 #include <sys/vnode.h>
  52 #include <sys/mountctl.h>
  53 #include <sys/dirent.h>
  54 #include <sys/uio.h>
  55 #include <sys/objcache.h>
  56 #include <sys/event.h>
  57 #include <sys/file.h>
  58 #include <vfs/fifofs/fifo.h>
  59
  60 #include "hammer2.h"
  61
  62 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
  63                                 int seqcount);
  64 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
  65                                 int ioflag, int seqcount);
  66 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
  67 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);
  68
  69 struct objcache *cache_xops;
  70
  71 static __inline
  72 void
  73 hammer2_knote(struct vnode *vp, int flags)
  74 {
  75         if (flags)
  76                 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
  77 }
  78
  79 /*
  80  * Last reference to a vnode is going away but it is still cached.
  81  */
  82 static
  83 int
  84 hammer2_vop_inactive(struct vop_inactive_args *ap)
  85 {
  86         hammer2_inode_t *ip;
  87         struct vnode *vp;
  88
  89         LOCKSTART;
  90         vp = ap->a_vp;
  91         ip = VTOI(vp);
  92
  93         /*
  94          * Degenerate case
  95          */
  96         if (ip == NULL) {
  97                 vrecycle(vp);
  98                 LOCKSTOP;
  99                 return (0);
 100         }
 101
 102         /*
 103          * Check for deleted inodes and recycle immediately on the last
 104          * release.  Be sure to destroy any left-over buffer cache buffers
 105          * so we do not waste time trying to flush them.
 106          *
 107          * Note that deleting the file block chains under the inode chain
 108          * would just be a waste of energy, so don't do it.
 109          *
 110          * WARNING: nvtruncbuf() can only be safely called without the inode
 111          *          lock held due to the way our write thread works.
 112          */
 113         if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
 114                 hammer2_key_t lbase;
 115                 int nblksize;
 116
 117                 /*
 118                  * Detect updates to the embedded data which may be
 119                  * synchronized by the strategy code.  Simply mark the
 120                  * inode modified so it gets picked up by our normal flush.
 121                  */
 122                 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
 123                 nvtruncbuf(vp, 0, nblksize, 0, 0);
 124                 vrecycle(vp);
 125         }
 126         LOCKSTOP;
 127         return (0);
 128 }
 129
 130 /*
 131  * Reclaim a vnode so that it can be reused; after the inode is
 132  * disassociated, the filesystem must manage it alone.
 133  */
 134 static
 135 int
 136 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
 137 {
 138         hammer2_inode_t *ip;
 139         hammer2_pfs_t *pmp;
 140         struct vnode *vp;
 141
 142         LOCKSTART;
 143         vp = ap->a_vp;
 144         ip = VTOI(vp);
 145         if (ip == NULL) {
 146                 LOCKSTOP;
 147                 return(0);
 148         }
 149         pmp = ip->pmp;
 150
 151         /*
 152          * The final close of a deleted file or directory marks it for
 153          * destruction.  The DELETED flag allows the flusher to shortcut
 154          * any modified blocks still unflushed (that is, just ignore them).
 155          *
 156          * HAMMER2 usually does not try to optimize the freemap by returning
 157          * deleted blocks to it as it does not usually know how many snapshots
 158          * might be referencing portions of the file/dir.
 159          */
 160         vp->v_data = NULL;
 161         ip->vp = NULL;
 162
 163         /*
 164          * NOTE! We do not attempt to flush chains here, flushing is
 165          *       really fragile and could also deadlock.
 166          */
 167         vclrisdirty(vp);
 168
 169         /*
 170          * An unlinked inode may have been relinked to the ihidden directory.
 171          * This occurs if the inode was unlinked while open.  Reclamation of
 172          * these inodes requires processing we cannot safely do here so add
 173          * the inode to the sideq in that situation.
 174          *
 175          * A modified inode may require chain synchronization which will no
 176          * longer be driven by a sync or fsync without the vnode, also use
 177          * the sideq for that.
 178          *
 179          * A reclaim can occur at any time so we cannot safely start a
 180          * transaction to handle reclamation of unlinked files.  Instead,
 181          * the ip is left with a reference and placed on a linked list and
 182          * handled later on.
 183          */
 184
 185         if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
 186                           HAMMER2_INODE_MODIFIED |
 187                           HAMMER2_INODE_RESIZED)) &&
 188             (ip->flags & HAMMER2_INODE_ISDELETED) == 0) {
 189                 hammer2_inode_sideq_t *ipul;
 190
 191                 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
 192                 ipul->ip = ip;
 193
 194                 hammer2_spin_ex(&pmp->list_spin);
 195                 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
 196                         /* ref -> sideq */
 197                         atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ);
 198                         TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry);
 199                         hammer2_spin_unex(&pmp->list_spin);
 200                 } else {
 201                         hammer2_spin_unex(&pmp->list_spin);
 202                         kfree(ipul, pmp->minode);
 203                         hammer2_inode_drop(ip);         /* vp ref */
 204                 }
 205                 /* retain ref from vp for ipul */
 206         } else {
 207                 hammer2_inode_drop(ip);                 /* vp ref */
 208         }
 209
 210         /*
 211          * XXX handle background sync when ip dirty, kernel will no longer
 212          * notify us regarding this inode because there is no longer a
 213          * vnode attached to it.
 214          */
 215
 216         LOCKSTOP;
 217         return (0);
 218 }
 219
 220 static
 221 int
 222 hammer2_vop_fsync(struct vop_fsync_args *ap)
 223 {
 224         hammer2_inode_t *ip;
 225         struct vnode *vp;
 226
 227         LOCKSTART;
 228         vp = ap->a_vp;
 229         ip = VTOI(vp);
 230
 231 #if 0
 232         /* XXX can't do this yet */
 233         hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH);
 234         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 235 #endif
 236         hammer2_trans_init(ip->pmp, 0);
 237         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 238
 239         /*
 240          * Calling chain_flush here creates a lot of duplicative
 241          * COW operations due to non-optimal vnode ordering.
 242          *
 243          * Only do it for an actual fsync() syscall.  The other forms
 244          * which call this function will eventually call chain_flush
 245          * on the volume root as a catch-all, which is far more optimal.
 246          */
 247         hammer2_inode_lock(ip, 0);
 248         if (ip->flags & HAMMER2_INODE_MODIFIED)
 249                 hammer2_inode_chain_sync(ip);
 250         hammer2_inode_unlock(ip);
 251         hammer2_trans_done(ip->pmp);
 252
 253         LOCKSTOP;
 254         return (0);
 255 }
 256
 257 static
 258 int
 259 hammer2_vop_access(struct vop_access_args *ap)
 260 {
 261         hammer2_inode_t *ip = VTOI(ap->a_vp);
 262         uid_t uid;
 263         gid_t gid;
 264         int error;
 265
 266         LOCKSTART;
 267         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 268         uid = hammer2_to_unix_xid(&ip->meta.uid);
 269         gid = hammer2_to_unix_xid(&ip->meta.gid);
 270         error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
 271         hammer2_inode_unlock(ip);
 272
 273         LOCKSTOP;
 274         return (error);
 275 }
 276
 277 static
 278 int
 279 hammer2_vop_getattr(struct vop_getattr_args *ap)
 280 {
 281         hammer2_pfs_t *pmp;
 282         hammer2_inode_t *ip;
 283         struct vnode *vp;
 284         struct vattr *vap;
 285         hammer2_chain_t *chain;
 286         int i;
 287
 288         LOCKSTART;
 289         vp = ap->a_vp;
 290         vap = ap->a_vap;
 291
 292         ip = VTOI(vp);
 293         pmp = ip->pmp;
 294
 295         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 296
 297         vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
 298         vap->va_fileid = ip->meta.inum;
 299         vap->va_mode = ip->meta.mode;
 300         vap->va_nlink = ip->meta.nlinks;
 301         vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
 302         vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
 303         vap->va_rmajor = 0;
 304         vap->va_rminor = 0;
 305         vap->va_size = ip->meta.size;   /* protected by shared lock */
 306         vap->va_blocksize = HAMMER2_PBUFSIZE;
 307         vap->va_flags = ip->meta.uflags;
 308         hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
 309         hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
 310         hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
 311         vap->va_gen = 1;
 312         vap->va_bytes = 0;
 313         if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) {
 314                 /*
 315                  * Can't really calculate directory use sans the files under
 316                  * it, just assume one block for now.
 317                  */
 318                 vap->va_bytes += HAMMER2_INODE_BYTES;
 319         } else {
 320                 for (i = 0; i < ip->cluster.nchains; ++i) {
 321                         if ((chain = ip->cluster.array[i].chain) != NULL) {
 322                                 if (vap->va_bytes < chain->bref.data_count)
 323                                         vap->va_bytes = chain->bref.data_count;
 324                         }
 325                 }
 326         }
 327         vap->va_type = hammer2_get_vtype(ip->meta.type);
 328         vap->va_filerev = 0;
 329         vap->va_uid_uuid = ip->meta.uid;
 330         vap->va_gid_uuid = ip->meta.gid;
 331         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 332                           VA_FSID_UUID_VALID;
 333
 334         hammer2_inode_unlock(ip);
 335
 336         LOCKSTOP;
 337         return (0);
 338 }
 339
 340 static
 341 int
 342 hammer2_vop_setattr(struct vop_setattr_args *ap)
 343 {
 344         hammer2_inode_t *ip;
 345         struct vnode *vp;
 346         struct vattr *vap;
 347         int error;
 348         int kflags = 0;
 349         uint64_t ctime;
 350
 351         LOCKSTART;
 352         vp = ap->a_vp;
 353         vap = ap->a_vap;
 354         hammer2_update_time(&ctime);
 355
 356         ip = VTOI(vp);
 357
 358         if (ip->pmp->ronly) {
 359                 LOCKSTOP;
 360                 return(EROFS);
 361         }
 362
 363         hammer2_pfs_memory_wait(ip->pmp);
 364         hammer2_trans_init(ip->pmp, 0);
 365         hammer2_inode_lock(ip, 0);
 366         error = 0;
 367
 368         if (vap->va_flags != VNOVAL) {
 369                 uint32_t flags;
 370
 371                 flags = ip->meta.uflags;
 372                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
 373                                      hammer2_to_unix_xid(&ip->meta.uid),
 374                                      ap->a_cred);
 375                 if (error == 0) {
 376                         if (ip->meta.uflags != flags) {
 377                                 hammer2_inode_modify(ip);
 378                                 ip->meta.uflags = flags;
 379                                 ip->meta.ctime = ctime;
 380                                 kflags |= NOTE_ATTRIB;
 381                         }
 382                         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 383                                 error = 0;
 384                                 goto done;
 385                         }
 386                 }
 387                 goto done;
 388         }
 389         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 390                 error = EPERM;
 391                 goto done;
 392         }
 393         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
 394                 mode_t cur_mode = ip->meta.mode;
 395                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 396                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 397                 uuid_t uuid_uid;
 398                 uuid_t uuid_gid;
 399
 400                 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
 401                                          ap->a_cred,
 402                                          &cur_uid, &cur_gid, &cur_mode);
 403                 if (error == 0) {
 404                         hammer2_guid_to_uuid(&uuid_uid, cur_uid);
 405                         hammer2_guid_to_uuid(&uuid_gid, cur_gid);
 406                         if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
 407                             bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
 408                             ip->meta.mode != cur_mode
 409                         ) {
 410                                 hammer2_inode_modify(ip);
 411                                 ip->meta.uid = uuid_uid;
 412                                 ip->meta.gid = uuid_gid;
 413                                 ip->meta.mode = cur_mode;
 414                                 ip->meta.ctime = ctime;
 415                         }
 416                         kflags |= NOTE_ATTRIB;
 417                 }
 418         }
 419
 420         /*
 421          * Resize the file
 422          */
 423         if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
 424                 switch(vp->v_type) {
 425                 case VREG:
 426                         if (vap->va_size == ip->meta.size)
 427                                 break;
 428                         if (vap->va_size < ip->meta.size) {
 429                                 hammer2_mtx_ex(&ip->truncate_lock);
 430                                 hammer2_truncate_file(ip, vap->va_size);
 431                                 hammer2_mtx_unlock(&ip->truncate_lock);
 432                         } else {
 433                                 hammer2_extend_file(ip, vap->va_size);
 434                         }
 435                         hammer2_inode_modify(ip);
 436                         ip->meta.mtime = ctime;
 437                         break;
 438                 default:
 439                         error = EINVAL;
 440                         goto done;
 441                 }
 442         }
 443 #if 0
 444         /* atime not supported */
 445         if (vap->va_atime.tv_sec != VNOVAL) {
 446                 hammer2_inode_modify(ip);
 447                 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
 448                 kflags |= NOTE_ATTRIB;
 449         }
 450 #endif
 451         if (vap->va_mode != (mode_t)VNOVAL) {
 452                 mode_t cur_mode = ip->meta.mode;
 453                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 454                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 455
 456                 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
 457                                          cur_uid, cur_gid, &cur_mode);
 458                 if (error == 0 && ip->meta.mode != cur_mode) {
 459                         hammer2_inode_modify(ip);
 460                         ip->meta.mode = cur_mode;
 461                         ip->meta.ctime = ctime;
 462                         kflags |= NOTE_ATTRIB;
 463                 }
 464         }
 465
 466         if (vap->va_mtime.tv_sec != VNOVAL) {
 467                 hammer2_inode_modify(ip);
 468                 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
 469                 kflags |= NOTE_ATTRIB;
 470         }
 471
 472 done:
 473         /*
 474          * If a truncation occurred we must call inode_fsync() now in order
 475          * to trim the related data chains, otherwise a later expansion can
 476          * cause havoc.
 477          *
 478          * If an extend occured that changed the DIRECTDATA state, we must
 479          * call inode_fsync now in order to prepare the inode's indirect
 480          * block table.
 481          */
 482         if (ip->flags & HAMMER2_INODE_RESIZED)
 483                 hammer2_inode_chain_sync(ip);
 484
 485         /*
 486          * Cleanup.
 487          */
 488         hammer2_inode_unlock(ip);
 489         hammer2_trans_done(ip->pmp);
 490         hammer2_knote(ip->vp, kflags);
 491
 492         LOCKSTOP;
 493         return (error);
 494 }
 495
 496 static
 497 int
 498 hammer2_vop_readdir(struct vop_readdir_args *ap)
 499 {
 500         hammer2_xop_readdir_t *xop;
 501         hammer2_blockref_t bref;
 502         hammer2_inode_t *ip;
 503         hammer2_tid_t inum;
 504         hammer2_key_t lkey;
 505         struct uio *uio;
 506         off_t *cookies;
 507         off_t saveoff;
 508         int cookie_index;
 509         int ncookies;
 510         int error;
 511         int eofflag;
 512         int dtype;
 513         int r;
 514
 515         LOCKSTART;
 516         ip = VTOI(ap->a_vp);
 517         uio = ap->a_uio;
 518         saveoff = uio->uio_offset;
 519         eofflag = 0;
 520         error = 0;
 521
 522         /*
 523          * Setup cookies directory entry cookies if requested
 524          */
 525         if (ap->a_ncookies) {
 526                 ncookies = uio->uio_resid / 16 + 1;
 527                 if (ncookies > 1024)
 528                         ncookies = 1024;
 529                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 530         } else {
 531                 ncookies = -1;
 532                 cookies = NULL;
 533         }
 534         cookie_index = 0;
 535
 536         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 537
 538         /*
 539          * Handle artificial entries.  To ensure that only positive 64 bit
 540          * quantities are returned to userland we always strip off bit 63.
 541          * The hash code is designed such that codes 0x0000-0x7FFF are not
 542          * used, allowing us to use these codes for articial entries.
 543          *
 544          * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
 545          * allow '..' to cross the mount point into (e.g.) the super-root.
 546          */
 547         if (saveoff == 0) {
 548                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 549                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
 550                 if (r)
 551                         goto done;
 552                 if (cookies)
 553                         cookies[cookie_index] = saveoff;
 554                 ++saveoff;
 555                 ++cookie_index;
 556                 if (cookie_index == ncookies)
 557                         goto done;
 558         }
 559
 560         if (saveoff == 1) {
 561                 /*
 562                  * Be careful with lockorder when accessing ".."
 563                  *
 564                  * (ip is the current dir. xip is the parent dir).
 565                  */
 566                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 567                 if (ip->pip && ip != ip->pmp->iroot)
 568                         inum = ip->pip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 569                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
 570                 if (r)
 571                         goto done;
 572                 if (cookies)
 573                         cookies[cookie_index] = saveoff;
 574                 ++saveoff;
 575                 ++cookie_index;
 576                 if (cookie_index == ncookies)
 577                         goto done;
 578         }
 579
 580         lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
 581         if (hammer2_debug & 0x0020)
 582                 kprintf("readdir: lkey %016jx\n", lkey);
 583         if (error)
 584                 goto done;
 585
 586         /*
 587          * Use XOP for cluster scan.
 588          *
 589          * parent is the inode cluster, already locked for us.  Don't
 590          * double lock shared locks as this will screw up upgrades.
 591          */
 592         xop = hammer2_xop_alloc(ip, 0);
 593         xop->lkey = lkey;
 594         hammer2_xop_start(&xop->head, hammer2_xop_readdir);
 595
 596         for (;;) {
 597                 const hammer2_inode_data_t *ripdata;
 598
 599                 error = hammer2_xop_collect(&xop->head, 0);
 600                 if (error)
 601                         break;
 602                 if (cookie_index == ncookies)
 603                         break;
 604                 if (hammer2_debug & 0x0020)
 605                 kprintf("cluster chain %p %p\n",
 606                         xop->head.cluster.focus,
 607                         (xop->head.cluster.focus ?
 608                          xop->head.cluster.focus->data : (void *)-1));
 609                 ripdata = &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
 610                 hammer2_cluster_bref(&xop->head.cluster, &bref);
 611                 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
 612                         dtype = hammer2_get_dtype(ripdata);
 613                         saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 614                         r = vop_write_dirent(&error, uio,
 615                                              ripdata->meta.inum &
 616                                               HAMMER2_DIRHASH_USERMSK,
 617                                              dtype,
 618                                              ripdata->meta.name_len,
 619                                              ripdata->filename);
 620                         if (r)
 621                                 break;
 622                         if (cookies)
 623                                 cookies[cookie_index] = saveoff;
 624                         ++cookie_index;
 625                 } else {
 626                         /* XXX chain error */
 627                         kprintf("bad chain type readdir %d\n", bref.type);
 628                 }
 629         }
 630         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
 631         if (error == ENOENT) {
 632                 error = 0;
 633                 eofflag = 1;
 634                 saveoff = (hammer2_key_t)-1;
 635         } else {
 636                 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 637         }
 638 done:
 639         hammer2_inode_unlock(ip);
 640         if (ap->a_eofflag)
 641                 *ap->a_eofflag = eofflag;
 642         if (hammer2_debug & 0x0020)
 643                 kprintf("readdir: done at %016jx\n", saveoff);
 644         uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
 645         if (error && cookie_index == 0) {
 646                 if (cookies) {
 647                         kfree(cookies, M_TEMP);
 648                         *ap->a_ncookies = 0;
 649                         *ap->a_cookies = NULL;
 650                 }
 651         } else {
 652                 if (cookies) {
 653                         *ap->a_ncookies = cookie_index;
 654                         *ap->a_cookies = cookies;
 655                 }
 656         }
 657         LOCKSTOP;
 658         return (error);
 659 }
 660
 661 /*
 662  * hammer2_vop_readlink { vp, uio, cred }
 663  */
 664 static
 665 int
 666 hammer2_vop_readlink(struct vop_readlink_args *ap)
 667 {
 668         struct vnode *vp;
 669         hammer2_inode_t *ip;
 670         int error;
 671
 672         vp = ap->a_vp;
 673         if (vp->v_type != VLNK)
 674                 return (EINVAL);
 675         ip = VTOI(vp);
 676
 677         error = hammer2_read_file(ip, ap->a_uio, 0);
 678         return (error);
 679 }
 680
 681 static
 682 int
 683 hammer2_vop_read(struct vop_read_args *ap)
 684 {
 685         struct vnode *vp;
 686         hammer2_inode_t *ip;
 687         struct uio *uio;
 688         int error;
 689         int seqcount;
 690         int bigread;
 691
 692         /*
 693          * Read operations supported on this vnode?
 694          */
 695         vp = ap->a_vp;
 696         if (vp->v_type != VREG)
 697                 return (EINVAL);
 698
 699         /*
 700          * Misc
 701          */
 702         ip = VTOI(vp);
 703         uio = ap->a_uio;
 704         error = 0;
 705
 706         seqcount = ap->a_ioflag >> 16;
 707         bigread = (uio->uio_resid > 100 * 1024 * 1024);
 708
 709         error = hammer2_read_file(ip, uio, seqcount);
 710         return (error);
 711 }
 712
 713 static
 714 int
 715 hammer2_vop_write(struct vop_write_args *ap)
 716 {
 717         hammer2_inode_t *ip;
 718         thread_t td;
 719         struct vnode *vp;
 720         struct uio *uio;
 721         int error;
 722         int seqcount;
 723
 724         /*
 725          * Read operations supported on this vnode?
 726          */
 727         vp = ap->a_vp;
 728         if (vp->v_type != VREG)
 729                 return (EINVAL);
 730
 731         /*
 732          * Misc
 733          */
 734         ip = VTOI(vp);
 735         uio = ap->a_uio;
 736         error = 0;
 737         if (ip->pmp->ronly) {
 738                 return (EROFS);
 739         }
 740
 741         seqcount = ap->a_ioflag >> 16;
 742
 743         /*
 744          * Check resource limit
 745          */
 746         if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
 747             uio->uio_offset + uio->uio_resid >
 748              td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 749                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 750                 return (EFBIG);
 751         }
 752
 753         /*
 754          * The transaction interlocks against flushes initiations
 755          * (note: but will run concurrently with the actual flush).
 756          */
 757         hammer2_trans_init(ip->pmp, 0);
 758         error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
 759         hammer2_trans_done(ip->pmp);
 760
 761         return (error);
 762 }
 763
 764 /*
 765  * Perform read operations on a file or symlink given an UNLOCKED
 766  * inode and uio.
 767  *
 768  * The passed ip is not locked.
 769  */
 770 static
 771 int
 772 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
 773 {
 774         hammer2_off_t size;
 775         struct buf *bp;
 776         int error;
 777
 778         error = 0;
 779
 780         /*
 781          * UIO read loop.
 782          *
 783          * WARNING! Assumes that the kernel interlocks size changes at the
 784          *          vnode level.
 785          */
 786         hammer2_mtx_sh(&ip->lock);
 787         hammer2_mtx_sh(&ip->truncate_lock);
 788         size = ip->meta.size;
 789         hammer2_mtx_unlock(&ip->lock);
 790
 791         while (uio->uio_resid > 0 && uio->uio_offset < size) {
 792                 hammer2_key_t lbase;
 793                 hammer2_key_t leof;
 794                 int lblksize;
 795                 int loff;
 796                 int n;
 797
 798                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 799                                                 &lbase, &leof);
 800
 801                 error = cluster_read(ip->vp, leof, lbase, lblksize,
 802                                      uio->uio_resid, seqcount * BKVASIZE,
 803                                      &bp);
 804
 805                 if (error)
 806                         break;
 807                 loff = (int)(uio->uio_offset - lbase);
 808                 n = lblksize - loff;
 809                 if (n > uio->uio_resid)
 810                         n = uio->uio_resid;
 811                 if (n > size - uio->uio_offset)
 812                         n = (int)(size - uio->uio_offset);
 813                 bp->b_flags |= B_AGE;
 814                 uiomove((char *)bp->b_data + loff, n, uio);
 815                 bqrelse(bp);
 816         }
 817         hammer2_mtx_unlock(&ip->truncate_lock);
 818
 819         return (error);
 820 }
 821
 822 /*
 823  * Write to the file represented by the inode via the logical buffer cache.
 824  * The inode may represent a regular file or a symlink.
 825  *
 826  * The inode must not be locked.
 827  */
 828 static
 829 int
 830 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
 831                    int ioflag, int seqcount)
 832 {
 833         hammer2_key_t old_eof;
 834         hammer2_key_t new_eof;
 835         struct buf *bp;
 836         int kflags;
 837         int error;
 838         int modified;
 839
 840         /*
 841          * Setup if append
 842          *
 843          * WARNING! Assumes that the kernel interlocks size changes at the
 844          *          vnode level.
 845          */
 846         hammer2_mtx_ex(&ip->lock);
 847         hammer2_mtx_sh(&ip->truncate_lock);
 848         if (ioflag & IO_APPEND)
 849                 uio->uio_offset = ip->meta.size;
 850         old_eof = ip->meta.size;
 851
 852         /*
 853          * Extend the file if necessary.  If the write fails at some point
 854          * we will truncate it back down to cover as much as we were able
 855          * to write.
 856          *
 857          * Doing this now makes it easier to calculate buffer sizes in
 858          * the loop.
 859          */
 860         kflags = 0;
 861         error = 0;
 862         modified = 0;
 863
 864         if (uio->uio_offset + uio->uio_resid > old_eof) {
 865                 new_eof = uio->uio_offset + uio->uio_resid;
 866                 modified = 1;
 867                 hammer2_extend_file(ip, new_eof);
 868                 kflags |= NOTE_EXTEND;
 869         } else {
 870                 new_eof = old_eof;
 871         }
 872         hammer2_mtx_unlock(&ip->lock);
 873
 874         /*
 875          * UIO write loop
 876          */
 877         while (uio->uio_resid > 0) {
 878                 hammer2_key_t lbase;
 879                 int trivial;
 880                 int endofblk;
 881                 int lblksize;
 882                 int loff;
 883                 int n;
 884
 885                 /*
 886                  * Don't allow the buffer build to blow out the buffer
 887                  * cache.
 888                  */
 889                 if ((ioflag & IO_RECURSE) == 0)
 890                         bwillwrite(HAMMER2_PBUFSIZE);
 891
 892                 /*
 893                  * This nominally tells us how much we can cluster and
 894                  * what the logical buffer size needs to be.  Currently
 895                  * we don't try to cluster the write and just handle one
 896                  * block at a time.
 897                  */
 898                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 899                                                 &lbase, NULL);
 900                 loff = (int)(uio->uio_offset - lbase);
 901
 902                 KKASSERT(lblksize <= 65536);
 903
 904                 /*
 905                  * Calculate bytes to copy this transfer and whether the
 906                  * copy completely covers the buffer or not.
 907                  */
 908                 trivial = 0;
 909                 n = lblksize - loff;
 910                 if (n > uio->uio_resid) {
 911                         n = uio->uio_resid;
 912                         if (loff == lbase && uio->uio_offset + n == new_eof)
 913                                 trivial = 1;
 914                         endofblk = 0;
 915                 } else {
 916                         if (loff == 0)
 917                                 trivial = 1;
 918                         endofblk = 1;
 919                 }
 920
 921                 /*
 922                  * Get the buffer
 923                  */
 924                 if (uio->uio_segflg == UIO_NOCOPY) {
 925                         /*
 926                          * Issuing a write with the same data backing the
 927                          * buffer.  Instantiate the buffer to collect the
 928                          * backing vm pages, then read-in any missing bits.
 929                          *
 930                          * This case is used by vop_stdputpages().
 931                          */
 932                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 933                         if ((bp->b_flags & B_CACHE) == 0) {
 934                                 bqrelse(bp);
 935                                 error = bread(ip->vp, lbase, lblksize, &bp);
 936                         }
 937                 } else if (trivial) {
 938                         /*
 939                          * Even though we are entirely overwriting the buffer
 940                          * we may still have to zero it out to avoid a
 941                          * mmap/write visibility issue.
 942                          */
 943                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 944                         if ((bp->b_flags & B_CACHE) == 0)
 945                                 vfs_bio_clrbuf(bp);
 946                 } else {
 947                         /*
 948                          * Partial overwrite, read in any missing bits then
 949                          * replace the portion being written.
 950                          *
 951                          * (The strategy code will detect zero-fill physical
 952                          * blocks for this case).
 953                          */
 954                         error = bread(ip->vp, lbase, lblksize, &bp);
 955                         if (error == 0)
 956                                 bheavy(bp);
 957                 }
 958
 959                 if (error) {
 960                         brelse(bp);
 961                         break;
 962                 }
 963
 964                 /*
 965                  * Ok, copy the data in
 966                  */
 967                 error = uiomove(bp->b_data + loff, n, uio);
 968                 kflags |= NOTE_WRITE;
 969                 modified = 1;
 970                 if (error) {
 971                         brelse(bp);
 972                         break;
 973                 }
 974
 975                 /*
 976                  * WARNING: Pageout daemon will issue UIO_NOCOPY writes
 977                  *          with IO_SYNC or IO_ASYNC set.  These writes
 978                  *          must be handled as the pageout daemon expects.
 979                  *
 980                  * NOTE!    H2 relies on cluster_write() here because it
 981                  *          cannot preallocate disk blocks at the logical
 982                  *          level due to not knowing what the compression
 983                  *          size will be at this time.
 984                  *
 985                  *          We must use cluster_write() here and we depend
 986                  *          on the write-behind feature to flush buffers
 987                  *          appropriately.  If we let the buffer daemons do
 988                  *          it the block allocations will be all over the
 989                  *          map.
 990                  */
 991                 if (ioflag & IO_SYNC) {
 992                         bwrite(bp);
 993                 } else if ((ioflag & IO_DIRECT) && endofblk) {
 994                         bawrite(bp);
 995                 } else if (ioflag & IO_ASYNC) {
 996                         bawrite(bp);
 997                 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) {
 998                         bdwrite(bp);
 999                 } else {
1000                         bp->b_flags |= B_CLUSTEROK;
1001                         cluster_write(bp, new_eof, lblksize, seqcount);
1002                 }
1003         }
1004
1005         /*
1006          * Cleanup.  If we extended the file EOF but failed to write through
1007          * the entire write is a failure and we have to back-up.
1008          */
1009         if (error && new_eof != old_eof) {
1010                 hammer2_mtx_unlock(&ip->truncate_lock);
1011                 hammer2_mtx_ex(&ip->lock);
1012                 hammer2_mtx_ex(&ip->truncate_lock);
1013                 hammer2_truncate_file(ip, old_eof);
1014                 if (ip->flags & HAMMER2_INODE_MODIFIED)
1015                         hammer2_inode_chain_sync(ip);
1016                 hammer2_mtx_unlock(&ip->lock);
1017         } else if (modified) {
1018                 hammer2_mtx_ex(&ip->lock);
1019                 hammer2_inode_modify(ip);
1020                 hammer2_update_time(&ip->meta.mtime);
1021                 if (ip->flags & HAMMER2_INODE_MODIFIED)
1022                         hammer2_inode_chain_sync(ip);
1023                 hammer2_mtx_unlock(&ip->lock);
1024                 hammer2_knote(ip->vp, kflags);
1025         }
1026         hammer2_trans_assert_strategy(ip->pmp);
1027         hammer2_mtx_unlock(&ip->truncate_lock);
1028
1029         return error;
1030 }
1031
1032 /*
1033  * Truncate the size of a file.  The inode must not be locked.
1034  *
1035  * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1036  * ensure that any on-media data beyond the new file EOF has been destroyed.
1037  *
1038  * WARNING: nvtruncbuf() can only be safely called without the inode lock
1039  *          held due to the way our write thread works.  If the truncation
1040  *          occurs in the middle of a buffer, nvtruncbuf() is responsible
1041  *          for dirtying that buffer and zeroing out trailing bytes.
1042  *
1043  * WARNING! Assumes that the kernel interlocks size changes at the
1044  *          vnode level.
1045  *
1046  * WARNING! Caller assumes responsibility for removing dead blocks
1047  *          if INODE_RESIZED is set.
1048  */
1049 static
1050 void
1051 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1052 {
1053         hammer2_key_t lbase;
1054         int nblksize;
1055
1056         LOCKSTART;
1057         hammer2_mtx_unlock(&ip->lock);
1058         if (ip->vp) {
1059                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1060                 nvtruncbuf(ip->vp, nsize,
1061                            nblksize, (int)nsize & (nblksize - 1),
1062                            0);
1063         }
1064         hammer2_mtx_ex(&ip->lock);
1065         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1066         ip->osize = ip->meta.size;
1067         ip->meta.size = nsize;
1068         atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1069         hammer2_inode_modify(ip);
1070         LOCKSTOP;
1071 }
1072
1073 /*
1074  * Extend the size of a file.  The inode must not be locked.
1075  *
1076  * Even though the file size is changing, we do not have to set the
1077  * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1078  * boundary.  When this occurs a hammer2_inode_chain_sync() is required
1079  * to prepare the inode cluster's indirect block table, otherwise
1080  * async execution of the strategy code will implode on us.
1081  *
1082  * WARNING! Assumes that the kernel interlocks size changes at the
1083  *          vnode level.
1084  *
1085  * WARNING! Caller assumes responsibility for transitioning out
1086  *          of the inode DIRECTDATA mode if INODE_RESIZED is set.
1087  */
1088 static
1089 void
1090 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1091 {
1092         hammer2_key_t lbase;
1093         hammer2_key_t osize;
1094         int oblksize;
1095         int nblksize;
1096
1097         LOCKSTART;
1098
1099         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1100         hammer2_inode_modify(ip);
1101         osize = ip->meta.size;
1102         ip->osize = osize;
1103         ip->meta.size = nsize;
1104
1105         if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1106                 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1107                 hammer2_inode_chain_sync(ip);
1108         }
1109
1110         hammer2_mtx_unlock(&ip->lock);
1111         if (ip->vp) {
1112                 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1113                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1114                 nvextendbuf(ip->vp,
1115                             osize, nsize,
1116                             oblksize, nblksize,
1117                             -1, -1, 0);
1118         }
1119         hammer2_mtx_ex(&ip->lock);
1120
1121         LOCKSTOP;
1122 }
1123
1124 static
1125 int
1126 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1127 {
1128         hammer2_xop_nresolve_t *xop;
1129         hammer2_inode_t *ip;
1130         hammer2_inode_t *dip;
1131         struct namecache *ncp;
1132         struct vnode *vp;
1133         int error;
1134
1135         LOCKSTART;
1136         dip = VTOI(ap->a_dvp);
1137         xop = hammer2_xop_alloc(dip, 0);
1138
1139         ncp = ap->a_nch->ncp;
1140         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1141
1142         /*
1143          * Note: In DragonFly the kernel handles '.' and '..'.
1144          */
1145         hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1146         hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1147
1148         error = hammer2_xop_collect(&xop->head, 0);
1149         if (error) {
1150                 ip = NULL;
1151         } else {
1152                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1153         }
1154         hammer2_inode_unlock(dip);
1155
1156         /*
1157          * Acquire the related vnode
1158          *
1159          * NOTE: For error processing, only ENOENT resolves the namecache
1160          *       entry to NULL, otherwise we just return the error and
1161          *       leave the namecache unresolved.
1162          *
1163          * NOTE: multiple hammer2_inode structures can be aliased to the
1164          *       same chain element, for example for hardlinks.  This
1165          *       use case does not 'reattach' inode associations that
1166          *       might already exist, but always allocates a new one.
1167          *
1168          * WARNING: inode structure is locked exclusively via inode_get
1169          *          but chain was locked shared.  inode_unlock()
1170          *          will handle it properly.
1171          */
1172         if (ip) {
1173                 vp = hammer2_igetv(ip, &error);
1174                 if (error == 0) {
1175                         vn_unlock(vp);
1176                         cache_setvp(ap->a_nch, vp);
1177                 } else if (error == ENOENT) {
1178                         cache_setvp(ap->a_nch, NULL);
1179                 }
1180                 hammer2_inode_unlock(ip);
1181
1182                 /*
1183                  * The vp should not be released until after we've disposed
1184                  * of our locks, because it might cause vop_inactive() to
1185                  * be called.
1186                  */
1187                 if (vp)
1188                         vrele(vp);
1189         } else {
1190                 error = ENOENT;
1191                 cache_setvp(ap->a_nch, NULL);
1192         }
1193         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1194         KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1195                 ("resolve error %d/%p ap %p\n",
1196                  error, ap->a_nch->ncp->nc_vp, ap));
1197         LOCKSTOP;
1198
1199         return error;
1200 }
1201
1202 static
1203 int
1204 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1205 {
1206         hammer2_inode_t *dip;
1207         hammer2_inode_t *ip;
1208         int error;
1209
1210         LOCKSTART;
1211         dip = VTOI(ap->a_dvp);
1212
1213         if ((ip = dip->pip) == NULL) {
1214                 *ap->a_vpp = NULL;
1215                 LOCKSTOP;
1216                 return ENOENT;
1217         }
1218         hammer2_inode_lock(ip, 0);
1219         *ap->a_vpp = hammer2_igetv(ip, &error);
1220         hammer2_inode_unlock(ip);
1221
1222         LOCKSTOP;
1223         return error;
1224 }
1225
1226 static
1227 int
1228 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1229 {
1230         hammer2_inode_t *dip;
1231         hammer2_inode_t *nip;
1232         struct namecache *ncp;
1233         const uint8_t *name;
1234         size_t name_len;
1235         int error;
1236
1237         LOCKSTART;
1238         dip = VTOI(ap->a_dvp);
1239         if (dip->pmp->ronly) {
1240                 LOCKSTOP;
1241                 return (EROFS);
1242         }
1243
1244         ncp = ap->a_nch->ncp;
1245         name = ncp->nc_name;
1246         name_len = ncp->nc_nlen;
1247
1248         hammer2_pfs_memory_wait(dip->pmp);
1249         hammer2_trans_init(dip->pmp, 0);
1250         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1251                                    name, name_len, 0,
1252                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1253                                    0, &error);
1254         if (error) {
1255                 KKASSERT(nip == NULL);
1256                 *ap->a_vpp = NULL;
1257         } else {
1258                 *ap->a_vpp = hammer2_igetv(nip, &error);
1259                 hammer2_inode_unlock(nip);
1260         }
1261         hammer2_trans_done(dip->pmp);
1262
1263         if (error == 0) {
1264                 cache_setunresolved(ap->a_nch);
1265                 cache_setvp(ap->a_nch, *ap->a_vpp);
1266         }
1267         LOCKSTOP;
1268         return error;
1269 }
1270
1271 static
1272 int
1273 hammer2_vop_open(struct vop_open_args *ap)
1274 {
1275         return vop_stdopen(ap);
1276 }
1277
1278 /*
1279  * hammer2_vop_advlock { vp, id, op, fl, flags }
1280  */
1281 static
1282 int
1283 hammer2_vop_advlock(struct vop_advlock_args *ap)
1284 {
1285         hammer2_inode_t *ip = VTOI(ap->a_vp);
1286         hammer2_off_t size;
1287
1288         size = ip->meta.size;
1289         return (lf_advlock(ap, &ip->advlock, size));
1290 }
1291
1292 static
1293 int
1294 hammer2_vop_close(struct vop_close_args *ap)
1295 {
1296         return vop_stdclose(ap);
1297 }
1298
1299 /*
1300  * hammer2_vop_nlink { nch, dvp, vp, cred }
1301  *
1302  * Create a hardlink from (vp) to {dvp, nch}.
1303  */
1304 static
1305 int
1306 hammer2_vop_nlink(struct vop_nlink_args *ap)
1307 {
1308         hammer2_xop_nlink_t *xop1;
1309         hammer2_inode_t *fdip;  /* target directory to create link in */
1310         hammer2_inode_t *tdip;  /* target directory to create link in */
1311         hammer2_inode_t *cdip;  /* common parent directory */
1312         hammer2_inode_t *ip;    /* inode we are hardlinking to */
1313         struct namecache *ncp;
1314         const uint8_t *name;
1315         size_t name_len;
1316         int nlink_locked;
1317         int error;
1318
1319         LOCKSTART;
1320         tdip = VTOI(ap->a_dvp);
1321         if (tdip->pmp->ronly) {
1322                 LOCKSTOP;
1323                 return (EROFS);
1324         }
1325
1326         ncp = ap->a_nch->ncp;
1327         name = ncp->nc_name;
1328         name_len = ncp->nc_nlen;
1329
1330         /*
1331          * ip represents the file being hardlinked.  The file could be a
1332          * normal file or a hardlink target if it has already been hardlinked.
1333          * If ip is a hardlinked target then ip->pip represents the location
1334          * of the hardlinked target, NOT the location of the hardlink pointer.
1335          *
1336          * Bump nlinks and potentially also create or move the hardlink
1337          * target in the parent directory common to (ip) and (tdip).  The
1338          * consolidation code can modify ip->cluster and ip->pip.  The
1339          * returned cluster is locked.
1340          */
1341         ip = VTOI(ap->a_vp);
1342         hammer2_pfs_memory_wait(ip->pmp);
1343         hammer2_trans_init(ip->pmp, 0);
1344
1345         /*
1346          * The common parent directory must be locked first to avoid deadlocks.
1347          * Also note that fdip and/or tdip might match cdip.
1348          *
1349          * WARNING!  The kernel's namecache locks are insufficient for
1350          *           protecting us from hardlink shifts, since unrelated
1351          *           rename() or link() calls on parent directories might
1352          *           cause a shift.  A PFS-wide lock is required for this
1353          *           situation.
1354          */
1355         if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY ||
1356             (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
1357                 lockmgr(&ip->pmp->lock_nlink, LK_EXCLUSIVE);
1358                 nlink_locked = 1;
1359         } else {
1360                 nlink_locked = 0;
1361         }
1362         fdip = ip->pip;
1363         error = 0;
1364
1365         /*
1366          * Can return NULL and error == EXDEV if the common parent
1367          * crosses a directory with the xlink flag set.
1368          */
1369         cdip = hammer2_inode_common_parent(fdip, tdip, &error, 1);
1370         if (cdip)
1371                 hammer2_inode_lock(cdip, 0);
1372         hammer2_inode_lock(fdip, 0);
1373         hammer2_inode_lock(tdip, 0);
1374         hammer2_inode_lock(ip, 0);
1375
1376         /*
1377          * Dispatch xop_nlink unconditionally since we have to update nlinks.
1378          *
1379          * Otherwise we'd be able to avoid the XOP if the ip does not have
1380          * to be converted or moved.
1381          * If ip is not a hardlink target we must convert it to a hardlink.
1382          * If fdip != cdip we must shift the inode to cdip.
1383          *
1384          * XXX this and other nlink update usage should be passed top-down
1385          *     and not updated with a delta bottom-up.
1386          */
1387 #if 0
1388         if (fdip != cdip || (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE))
1389 #endif
1390         if (error == 0) {
1391                 xop1 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1392                 hammer2_xop_setip2(&xop1->head, ip);
1393                 hammer2_xop_setip3(&xop1->head, cdip);
1394                 xop1->nlinks_delta = 1;
1395
1396                 hammer2_xop_start(&xop1->head, hammer2_xop_nlink);
1397                 error = hammer2_xop_collect(&xop1->head, 0);
1398                 hammer2_xop_retire(&xop1->head, HAMMER2_XOPMASK_VOP);
1399                 if (error == ENOENT)
1400                         error = 0;
1401         }
1402
1403         /*
1404          * Must synchronize original inode whos chains are now a hardlink
1405          * target.  We must match what the backend XOP did to the
1406          * chains.
1407          */
1408         if (error == 0 && (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1409                 hammer2_inode_modify(ip);
1410                 ip->meta.name_key = ip->meta.inum;
1411                 ip->meta.name_len = 18; /* "0x%016jx" */
1412         }
1413
1414         /*
1415          * Create the hardlink target and bump nlinks.
1416          */
1417         if (error == 0) {
1418                 hammer2_inode_create(tdip, NULL, NULL,
1419                                      name, name_len, 0,
1420                                      ip->meta.inum,
1421                                      HAMMER2_OBJTYPE_HARDLINK, ip->meta.type,
1422                                      0, &error);
1423                 hammer2_inode_modify(ip);
1424                 ++ip->meta.nlinks;
1425         }
1426         if (error == 0) {
1427                 cache_setunresolved(ap->a_nch);
1428                 cache_setvp(ap->a_nch, ap->a_vp);
1429         }
1430         hammer2_inode_unlock(ip);
1431         hammer2_inode_unlock(tdip);
1432         hammer2_inode_unlock(fdip);
1433         if (cdip) {
1434                 hammer2_inode_unlock(cdip);
1435                 hammer2_inode_drop(cdip);
1436         }
1437
1438         if (nlink_locked)
1439                 lockmgr(&ip->pmp->lock_nlink, LK_RELEASE);
1440         hammer2_trans_done(ip->pmp);
1441
1442         LOCKSTOP;
1443         return error;
1444 }
1445
1446 /*
1447  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1448  *
1449  * The operating system has already ensured that the directory entry
1450  * does not exist and done all appropriate namespace locking.
1451  */
1452 static
1453 int
1454 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1455 {
1456         hammer2_inode_t *dip;
1457         hammer2_inode_t *nip;
1458         struct namecache *ncp;
1459         const uint8_t *name;
1460         size_t name_len;
1461         int error;
1462
1463         LOCKSTART;
1464         dip = VTOI(ap->a_dvp);
1465         if (dip->pmp->ronly) {
1466                 LOCKSTOP;
1467                 return (EROFS);
1468         }
1469
1470         ncp = ap->a_nch->ncp;
1471         name = ncp->nc_name;
1472         name_len = ncp->nc_nlen;
1473         hammer2_pfs_memory_wait(dip->pmp);
1474         hammer2_trans_init(dip->pmp, 0);
1475
1476         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1477                                    name, name_len, 0,
1478                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1479                                    0, &error);
1480         if (error) {
1481                 KKASSERT(nip == NULL);
1482                 *ap->a_vpp = NULL;
1483         } else {
1484                 *ap->a_vpp = hammer2_igetv(nip, &error);
1485                 hammer2_inode_unlock(nip);
1486         }
1487         hammer2_trans_done(dip->pmp);
1488
1489         if (error == 0) {
1490                 cache_setunresolved(ap->a_nch);
1491                 cache_setvp(ap->a_nch, *ap->a_vpp);
1492         }
1493         LOCKSTOP;
1494         return error;
1495 }
1496
1497 /*
1498  * Make a device node (typically a fifo)
1499  */
1500 static
1501 int
1502 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1503 {
1504         hammer2_inode_t *dip;
1505         hammer2_inode_t *nip;
1506         struct namecache *ncp;
1507         const uint8_t *name;
1508         size_t name_len;
1509         int error;
1510
1511         LOCKSTART;
1512         dip = VTOI(ap->a_dvp);
1513         if (dip->pmp->ronly) {
1514                 LOCKSTOP;
1515                 return (EROFS);
1516         }
1517
1518         ncp = ap->a_nch->ncp;
1519         name = ncp->nc_name;
1520         name_len = ncp->nc_nlen;
1521         hammer2_pfs_memory_wait(dip->pmp);
1522         hammer2_trans_init(dip->pmp, 0);
1523
1524         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1525                                    name, name_len, 0,
1526                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1527                                    0, &error);
1528         if (error) {
1529                 KKASSERT(nip == NULL);
1530                 *ap->a_vpp = NULL;
1531         } else {
1532                 *ap->a_vpp = hammer2_igetv(nip, &error);
1533                 hammer2_inode_unlock(nip);
1534         }
1535         hammer2_trans_done(dip->pmp);
1536
1537         if (error == 0) {
1538                 cache_setunresolved(ap->a_nch);
1539                 cache_setvp(ap->a_nch, *ap->a_vpp);
1540         }
1541         LOCKSTOP;
1542         return error;
1543 }
1544
1545 /*
1546  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1547  */
1548 static
1549 int
1550 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1551 {
1552         hammer2_inode_t *dip;
1553         hammer2_inode_t *nip;
1554         struct namecache *ncp;
1555         const uint8_t *name;
1556         size_t name_len;
1557         int error;
1558
1559         dip = VTOI(ap->a_dvp);
1560         if (dip->pmp->ronly)
1561                 return (EROFS);
1562
1563         ncp = ap->a_nch->ncp;
1564         name = ncp->nc_name;
1565         name_len = ncp->nc_nlen;
1566         hammer2_pfs_memory_wait(dip->pmp);
1567         hammer2_trans_init(dip->pmp, 0);
1568
1569         ap->a_vap->va_type = VLNK;      /* enforce type */
1570
1571         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1572                                    name, name_len, 0,
1573                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1574                                    0, &error);
1575         if (error) {
1576                 KKASSERT(nip == NULL);
1577                 *ap->a_vpp = NULL;
1578                 hammer2_trans_done(dip->pmp);
1579                 return error;
1580         }
1581         *ap->a_vpp = hammer2_igetv(nip, &error);
1582
1583         /*
1584          * Build the softlink (~like file data) and finalize the namecache.
1585          */
1586         if (error == 0) {
1587                 size_t bytes;
1588                 struct uio auio;
1589                 struct iovec aiov;
1590
1591                 bytes = strlen(ap->a_target);
1592
1593                 hammer2_inode_unlock(nip);
1594                 bzero(&auio, sizeof(auio));
1595                 bzero(&aiov, sizeof(aiov));
1596                 auio.uio_iov = &aiov;
1597                 auio.uio_segflg = UIO_SYSSPACE;
1598                 auio.uio_rw = UIO_WRITE;
1599                 auio.uio_resid = bytes;
1600                 auio.uio_iovcnt = 1;
1601                 auio.uio_td = curthread;
1602                 aiov.iov_base = ap->a_target;
1603                 aiov.iov_len = bytes;
1604                 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1605                 /* XXX handle error */
1606                 error = 0;
1607         } else {
1608                 hammer2_inode_unlock(nip);
1609         }
1610         hammer2_trans_done(dip->pmp);
1611
1612         /*
1613          * Finalize namecache
1614          */
1615         if (error == 0) {
1616                 cache_setunresolved(ap->a_nch);
1617                 cache_setvp(ap->a_nch, *ap->a_vpp);
1618                 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1619         }
1620         return error;
1621 }
1622
1623 /*
1624  * hammer2_vop_nremove { nch, dvp, cred }
1625  */
1626 static
1627 int
1628 hammer2_vop_nremove(struct vop_nremove_args *ap)
1629 {
1630         hammer2_xop_unlink_t *xop;
1631         hammer2_inode_t *dip;
1632         hammer2_inode_t *ip;
1633         struct namecache *ncp;
1634         int error;
1635         int isopen;
1636
1637         LOCKSTART;
1638         dip = VTOI(ap->a_dvp);
1639         if (dip->pmp->ronly) {
1640                 LOCKSTOP;
1641                 return(EROFS);
1642         }
1643
1644         ncp = ap->a_nch->ncp;
1645
1646         hammer2_pfs_memory_wait(dip->pmp);
1647         hammer2_trans_init(dip->pmp, 0);
1648         hammer2_inode_lock(dip, 0);
1649
1650         /*
1651          * The unlink XOP unlinks the path from the directory and
1652          * locates and returns the cluster associated with the real inode.
1653          * We have to handle nlinks here on the frontend.
1654          */
1655         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1656         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1657         isopen = cache_isopen(ap->a_nch);
1658         xop->isdir = 0;
1659         xop->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1660         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1661
1662         /*
1663          * Collect the real inode and adjust nlinks, destroy the real
1664          * inode if nlinks transitions to 0 and it was the real inode
1665          * (else it has already been removed).
1666          */
1667         error = hammer2_xop_collect(&xop->head, 0);
1668         hammer2_inode_unlock(dip);
1669
1670         if (error == 0) {
1671                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1672                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1673                 if (ip) {
1674                         hammer2_inode_unlink_finisher(ip, isopen);
1675                         hammer2_inode_unlock(ip);
1676                 }
1677         } else {
1678                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1679         }
1680
1681         hammer2_inode_run_sideq(dip->pmp);
1682         hammer2_trans_done(dip->pmp);
1683         if (error == 0)
1684                 cache_unlink(ap->a_nch);
1685         LOCKSTOP;
1686         return (error);
1687 }
1688
1689 /*
1690  * hammer2_vop_nrmdir { nch, dvp, cred }
1691  */
1692 static
1693 int
1694 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1695 {
1696         hammer2_xop_unlink_t *xop;
1697         hammer2_inode_t *dip;
1698         hammer2_inode_t *ip;
1699         struct namecache *ncp;
1700         int isopen;
1701         int error;
1702
1703         LOCKSTART;
1704         dip = VTOI(ap->a_dvp);
1705         if (dip->pmp->ronly) {
1706                 LOCKSTOP;
1707                 return(EROFS);
1708         }
1709
1710         hammer2_pfs_memory_wait(dip->pmp);
1711         hammer2_trans_init(dip->pmp, 0);
1712         hammer2_inode_lock(dip, 0);
1713
1714         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1715
1716         ncp = ap->a_nch->ncp;
1717         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1718         isopen = cache_isopen(ap->a_nch);
1719         xop->isdir = 1;
1720         xop->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1721         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1722
1723         /*
1724          * Collect the real inode and adjust nlinks, destroy the real
1725          * inode if nlinks transitions to 0 and it was the real inode
1726          * (else it has already been removed).
1727          */
1728         error = hammer2_xop_collect(&xop->head, 0);
1729         hammer2_inode_unlock(dip);
1730
1731         if (error == 0) {
1732                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1733                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1734                 if (ip) {
1735                         hammer2_inode_unlink_finisher(ip, isopen);
1736                         hammer2_inode_unlock(ip);
1737                 }
1738         } else {
1739                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1740         }
1741         hammer2_inode_run_sideq(dip->pmp);
1742         hammer2_trans_done(dip->pmp);
1743         if (error == 0)
1744                 cache_unlink(ap->a_nch);
1745         LOCKSTOP;
1746         return (error);
1747 }
1748
1749 /*
1750  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1751  */
1752 static
1753 int
1754 hammer2_vop_nrename(struct vop_nrename_args *ap)
1755 {
1756         struct namecache *fncp;
1757         struct namecache *tncp;
1758         hammer2_inode_t *cdip;
1759         hammer2_inode_t *fdip;
1760         hammer2_inode_t *tdip;
1761         hammer2_inode_t *ip;
1762         const uint8_t *fname;
1763         size_t fname_len;
1764         const uint8_t *tname;
1765         size_t tname_len;
1766         int error;
1767         int tnch_error;
1768         int nlink_locked;
1769         hammer2_key_t tlhc;
1770
1771         if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1772                 return(EXDEV);
1773         if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1774                 return(EXDEV);
1775
1776         fdip = VTOI(ap->a_fdvp);        /* source directory */
1777         tdip = VTOI(ap->a_tdvp);        /* target directory */
1778
1779         if (fdip->pmp->ronly)
1780                 return(EROFS);
1781
1782         LOCKSTART;
1783         fncp = ap->a_fnch->ncp;         /* entry name in source */
1784         fname = fncp->nc_name;
1785         fname_len = fncp->nc_nlen;
1786
1787         tncp = ap->a_tnch->ncp;         /* entry name in target */
1788         tname = tncp->nc_name;
1789         tname_len = tncp->nc_nlen;
1790
1791         hammer2_pfs_memory_wait(tdip->pmp);
1792         hammer2_trans_init(tdip->pmp, 0);
1793
1794         /*
1795          * ip is the inode being renamed.  If this is a hardlink then
1796          * ip represents the actual file and not the hardlink marker.
1797          */
1798         ip = VTOI(fncp->nc_vp);
1799
1800         /*
1801          * The common parent directory must be locked first to avoid deadlocks.
1802          * Also note that fdip and/or tdip might match cdip.
1803          *
1804          * WARNING!  The kernel's namecache locks are insufficient for
1805          *           protecting us from hardlink shifts, since unrelated
1806          *           rename() or link() calls on parent directories might
1807          *           cause a shift.  A PFS-wide lock is required for this
1808          *           situation.
1809          */
1810         if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY ||
1811             (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
1812                 lockmgr(&ip->pmp->lock_nlink, LK_EXCLUSIVE);
1813                 nlink_locked = 1;
1814         } else {
1815                 nlink_locked = 0;
1816         }
1817
1818         /*
1819          * Can return NULL and error == EXDEV if the common parent
1820          * crosses a directory with the xlink flag set.
1821          */
1822         error = 0;
1823         cdip = hammer2_inode_common_parent(ip->pip, tdip, &error, 0);
1824         if (cdip == NULL) {
1825                 tnch_error = error;
1826                 goto done3;
1827         }
1828         hammer2_inode_lock(cdip, 0);
1829         hammer2_inode_lock(fdip, 0);
1830         hammer2_inode_lock(tdip, 0);
1831         hammer2_inode_ref(ip);          /* extra ref */
1832
1833         /*
1834          * If ip is a hardlink target and fdip != cdip we must shift the
1835          * inode to cdip.
1836          */
1837         hammer2_inode_lock(ip, 0);
1838
1839         if (fdip != cdip &&
1840             (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
1841                 hammer2_xop_nlink_t *xop1;
1842
1843                 xop1 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1844                 hammer2_xop_setip2(&xop1->head, ip);
1845                 hammer2_xop_setip3(&xop1->head, cdip);
1846                 xop1->nlinks_delta = 0;
1847
1848                 hammer2_xop_start(&xop1->head, hammer2_xop_nlink);
1849                 error = hammer2_xop_collect(&xop1->head, 0);
1850                 hammer2_xop_retire(&xop1->head, HAMMER2_XOPMASK_VOP);
1851         }
1852         /* hammer2_inode_unlock(ip); */
1853
1854         /*
1855          * Delete the target namespace.
1856          */
1857         {
1858                 hammer2_xop_unlink_t *xop2;
1859                 hammer2_inode_t *tip;
1860                 int isopen;
1861
1862                 /*
1863                  * The unlink XOP unlinks the path from the directory and
1864                  * locates and returns the cluster associated with the real
1865                  * inode.  We have to handle nlinks here on the frontend.
1866                  */
1867                 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1868                 hammer2_xop_setname(&xop2->head, tname, tname_len);
1869                 isopen = cache_isopen(ap->a_tnch);
1870                 xop2->isdir = -1;
1871                 xop2->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1872                 hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
1873
1874                 /*
1875                  * Collect the real inode and adjust nlinks, destroy the real
1876                  * inode if nlinks transitions to 0 and it was the real inode
1877                  * (else it has already been removed).
1878                  */
1879                 tnch_error = hammer2_xop_collect(&xop2->head, 0);
1880                 /* hammer2_inode_unlock(tdip); */
1881
1882                 if (tnch_error == 0) {
1883                         tip = hammer2_inode_get(tdip->pmp, NULL,
1884                                                 &xop2->head.cluster, -1);
1885                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1886                         if (tip) {
1887                                 hammer2_inode_unlink_finisher(tip, isopen);
1888                                 hammer2_inode_unlock(tip);
1889                         }
1890                 } else {
1891                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1892                 }
1893                 /* hammer2_inode_lock(tdip, 0); */
1894
1895                 if (tnch_error && tnch_error != ENOENT) {
1896                         error = tnch_error;
1897                         goto done2;
1898                 }
1899         }
1900
1901         /*
1902          * Resolve the collision space for (tdip, tname, tname_len)
1903          *
1904          * tdip must be held exclusively locked to prevent races.
1905          */
1906         {
1907                 hammer2_xop_scanlhc_t *sxop;
1908                 hammer2_tid_t lhcbase;
1909
1910                 tlhc = hammer2_dirhash(tname, tname_len);
1911                 lhcbase = tlhc;
1912                 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1913                 sxop->lhc = tlhc;
1914                 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
1915                 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
1916                         if (tlhc != sxop->head.cluster.focus->bref.key)
1917                                 break;
1918                         ++tlhc;
1919                 }
1920                 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
1921
1922                 if (error) {
1923                         if (error != ENOENT)
1924                                 goto done2;
1925                         ++tlhc;
1926                         error = 0;
1927                 }
1928                 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
1929                         error = ENOSPC;
1930                         goto done2;
1931                 }
1932         }
1933
1934         /*
1935          * Everything is setup, do the rename.
1936          *
1937          * We have to synchronize ip->meta to the underlying operation.
1938          *
1939          * NOTE: To avoid deadlocks we cannot lock (ip) while we are
1940          *       unlinking elements from their directories.  Locking
1941          *       the nlinks field does not lock the whole inode.
1942          */
1943         /* hammer2_inode_lock(ip, 0); */
1944         if (error == 0) {
1945                 hammer2_xop_nrename_t *xop4;
1946
1947                 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1948                 xop4->lhc = tlhc;
1949                 xop4->ip_key = ip->meta.name_key;
1950                 hammer2_xop_setip2(&xop4->head, ip);
1951                 hammer2_xop_setip3(&xop4->head, tdip);
1952                 hammer2_xop_setname(&xop4->head, fname, fname_len);
1953                 hammer2_xop_setname2(&xop4->head, tname, tname_len);
1954                 hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
1955
1956                 error = hammer2_xop_collect(&xop4->head, 0);
1957                 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
1958
1959                 if (error == ENOENT)
1960                         error = 0;
1961                 if (error == 0 &&
1962                     (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1963                         hammer2_inode_modify(ip);
1964                         ip->meta.name_len = tname_len;
1965                         ip->meta.name_key = tlhc;
1966
1967                 }
1968         }
1969
1970         /*
1971          * Fixup ip->pip if we were renaming the actual file and not a
1972          * hardlink pointer.
1973          */
1974         if (error == 0 && (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1975                 hammer2_inode_t *opip;
1976
1977                 if (ip->pip != tdip) {
1978                         hammer2_inode_ref(tdip);
1979                         opip = ip->pip;
1980                         ip->pip = tdip;
1981                         if (opip)
1982                                 hammer2_inode_drop(opip);
1983                 }
1984         }
1985 done2:
1986         hammer2_inode_unlock(ip);
1987         hammer2_inode_unlock(tdip);
1988         hammer2_inode_unlock(fdip);
1989         hammer2_inode_unlock(cdip);
1990         hammer2_inode_drop(ip);
1991         hammer2_inode_drop(cdip);
1992 done3:
1993         hammer2_inode_run_sideq(fdip->pmp);
1994
1995         if (nlink_locked)
1996                 lockmgr(&ip->pmp->lock_nlink, LK_RELEASE);
1997         hammer2_trans_done(tdip->pmp);
1998
1999         /*
2000          * Issue the namecache update after unlocking all the internal
2001          * hammer structures, otherwise we might deadlock.
2002          */
2003         if (tnch_error == 0) {
2004                 cache_unlink(ap->a_tnch);
2005                 cache_setunresolved(ap->a_tnch);
2006         }
2007         if (error == 0)
2008                 cache_rename(ap->a_fnch, ap->a_tnch);
2009
2010         LOCKSTOP;
2011         return (error);
2012 }
2013
2014 /*
2015  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2016  */
2017 static
2018 int
2019 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2020 {
2021         hammer2_inode_t *ip;
2022         int error;
2023
2024         LOCKSTART;
2025         ip = VTOI(ap->a_vp);
2026
2027         error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2028                               ap->a_fflag, ap->a_cred);
2029         LOCKSTOP;
2030         return (error);
2031 }
2032
2033 static
2034 int
2035 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2036 {
2037         struct mount *mp;
2038         hammer2_pfs_t *pmp;
2039         int rc;
2040
2041         LOCKSTART;
2042         switch (ap->a_op) {
2043         case (MOUNTCTL_SET_EXPORT):
2044                 mp = ap->a_head.a_ops->head.vv_mount;
2045                 pmp = MPTOPMP(mp);
2046
2047                 if (ap->a_ctllen != sizeof(struct export_args))
2048                         rc = (EINVAL);
2049                 else
2050                         rc = vfs_export(mp, &pmp->export,
2051                                         (const struct export_args *)ap->a_ctl);
2052                 break;
2053         default:
2054                 rc = vop_stdmountctl(ap);
2055                 break;
2056         }
2057         LOCKSTOP;
2058         return (rc);
2059 }
2060
2061 /*
2062  * KQFILTER
2063  */
2064 static void filt_hammer2detach(struct knote *kn);
2065 static int filt_hammer2read(struct knote *kn, long hint);
2066 static int filt_hammer2write(struct knote *kn, long hint);
2067 static int filt_hammer2vnode(struct knote *kn, long hint);
2068
2069 static struct filterops hammer2read_filtops =
2070         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2071           NULL, filt_hammer2detach, filt_hammer2read };
2072 static struct filterops hammer2write_filtops =
2073         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2074           NULL, filt_hammer2detach, filt_hammer2write };
2075 static struct filterops hammer2vnode_filtops =
2076         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2077           NULL, filt_hammer2detach, filt_hammer2vnode };
2078
2079 static
2080 int
2081 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2082 {
2083         struct vnode *vp = ap->a_vp;
2084         struct knote *kn = ap->a_kn;
2085
2086         switch (kn->kn_filter) {
2087         case EVFILT_READ:
2088                 kn->kn_fop = &hammer2read_filtops;
2089                 break;
2090         case EVFILT_WRITE:
2091                 kn->kn_fop = &hammer2write_filtops;
2092                 break;
2093         case EVFILT_VNODE:
2094                 kn->kn_fop = &hammer2vnode_filtops;
2095                 break;
2096         default:
2097                 return (EOPNOTSUPP);
2098         }
2099
2100         kn->kn_hook = (caddr_t)vp;
2101
2102         knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2103
2104         return(0);
2105 }
2106
2107 static void
2108 filt_hammer2detach(struct knote *kn)
2109 {
2110         struct vnode *vp = (void *)kn->kn_hook;
2111
2112         knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2113 }
2114
2115 static int
2116 filt_hammer2read(struct knote *kn, long hint)
2117 {
2118         struct vnode *vp = (void *)kn->kn_hook;
2119         hammer2_inode_t *ip = VTOI(vp);
2120         off_t off;
2121
2122         if (hint == NOTE_REVOKE) {
2123                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2124                 return(1);
2125         }
2126         off = ip->meta.size - kn->kn_fp->f_offset;
2127         kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2128         if (kn->kn_sfflags & NOTE_OLDAPI)
2129                 return(1);
2130         return (kn->kn_data != 0);
2131 }
2132
2133
2134 static int
2135 filt_hammer2write(struct knote *kn, long hint)
2136 {
2137         if (hint == NOTE_REVOKE)
2138                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2139         kn->kn_data = 0;
2140         return (1);
2141 }
2142
2143 static int
2144 filt_hammer2vnode(struct knote *kn, long hint)
2145 {
2146         if (kn->kn_sfflags & hint)
2147                 kn->kn_fflags |= hint;
2148         if (hint == NOTE_REVOKE) {
2149                 kn->kn_flags |= (EV_EOF | EV_NODATA);
2150                 return (1);
2151         }
2152         return (kn->kn_fflags != 0);
2153 }
2154
2155 /*
2156  * FIFO VOPS
2157  */
2158 static
2159 int
2160 hammer2_vop_markatime(struct vop_markatime_args *ap)
2161 {
2162         hammer2_inode_t *ip;
2163         struct vnode *vp;
2164
2165         vp = ap->a_vp;
2166         ip = VTOI(vp);
2167
2168         if (ip->pmp->ronly)
2169                 return(EROFS);
2170         return(0);
2171 }
2172
2173 static
2174 int
2175 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2176 {
2177         int error;
2178
2179         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2180         if (error)
2181                 error = hammer2_vop_kqfilter(ap);
2182         return(error);
2183 }
2184
2185 /*
2186  * VOPS vector
2187  */
2188 struct vop_ops hammer2_vnode_vops = {
2189         .vop_default    = vop_defaultop,
2190         .vop_fsync      = hammer2_vop_fsync,
2191         .vop_getpages   = vop_stdgetpages,
2192         .vop_putpages   = vop_stdputpages,
2193         .vop_access     = hammer2_vop_access,
2194         .vop_advlock    = hammer2_vop_advlock,
2195         .vop_close      = hammer2_vop_close,
2196         .vop_nlink      = hammer2_vop_nlink,
2197         .vop_ncreate    = hammer2_vop_ncreate,
2198         .vop_nsymlink   = hammer2_vop_nsymlink,
2199         .vop_nremove    = hammer2_vop_nremove,
2200         .vop_nrmdir     = hammer2_vop_nrmdir,
2201         .vop_nrename    = hammer2_vop_nrename,
2202         .vop_getattr    = hammer2_vop_getattr,
2203         .vop_setattr    = hammer2_vop_setattr,
2204         .vop_readdir    = hammer2_vop_readdir,
2205         .vop_readlink   = hammer2_vop_readlink,
2206         .vop_getpages   = vop_stdgetpages,
2207         .vop_putpages   = vop_stdputpages,
2208         .vop_read       = hammer2_vop_read,
2209         .vop_write      = hammer2_vop_write,
2210         .vop_open       = hammer2_vop_open,
2211         .vop_inactive   = hammer2_vop_inactive,
2212         .vop_reclaim    = hammer2_vop_reclaim,
2213         .vop_nresolve   = hammer2_vop_nresolve,
2214         .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2215         .vop_nmkdir     = hammer2_vop_nmkdir,
2216         .vop_nmknod     = hammer2_vop_nmknod,
2217         .vop_ioctl      = hammer2_vop_ioctl,
2218         .vop_mountctl   = hammer2_vop_mountctl,
2219         .vop_bmap       = hammer2_vop_bmap,
2220         .vop_strategy   = hammer2_vop_strategy,
2221         .vop_kqfilter   = hammer2_vop_kqfilter
2222 };
2223
2224 struct vop_ops hammer2_spec_vops = {
2225         .vop_default =          vop_defaultop,
2226         .vop_fsync =            hammer2_vop_fsync,
2227         .vop_read =             vop_stdnoread,
2228         .vop_write =            vop_stdnowrite,
2229         .vop_access =           hammer2_vop_access,
2230         .vop_close =            hammer2_vop_close,
2231         .vop_markatime =        hammer2_vop_markatime,
2232         .vop_getattr =          hammer2_vop_getattr,
2233         .vop_inactive =         hammer2_vop_inactive,
2234         .vop_reclaim =          hammer2_vop_reclaim,
2235         .vop_setattr =          hammer2_vop_setattr
2236 };
2237
2238 struct vop_ops hammer2_fifo_vops = {
2239         .vop_default =          fifo_vnoperate,
2240         .vop_fsync =            hammer2_vop_fsync,
2241 #if 0
2242         .vop_read =             hammer2_vop_fiforead,
2243         .vop_write =            hammer2_vop_fifowrite,
2244 #endif
2245         .vop_access =           hammer2_vop_access,
2246 #if 0
2247         .vop_close =            hammer2_vop_fifoclose,
2248 #endif
2249         .vop_markatime =        hammer2_vop_markatime,
2250         .vop_getattr =          hammer2_vop_getattr,
2251         .vop_inactive =         hammer2_vop_inactive,
2252         .vop_reclaim =          hammer2_vop_reclaim,
2253         .vop_setattr =          hammer2_vop_setattr,
2254         .vop_kqfilter =         hammer2_vop_fifokqfilter
2255 };
2256