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         if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
 185                           HAMMER2_INODE_MODIFIED |
 186                           HAMMER2_INODE_RESIZED)) &&
 187             (ip->flags & HAMMER2_INODE_ISDELETED) == 0) {
 188                 hammer2_inode_sideq_t *ipul;
 189
 190                 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
 191                 ipul->ip = ip;
 192
 193                 hammer2_spin_ex(&pmp->list_spin);
 194                 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
 195                         /* ref -> sideq */
 196                         atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ);
 197                         TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry);
 198                         hammer2_spin_unex(&pmp->list_spin);
 199                 } else {
 200                         hammer2_spin_unex(&pmp->list_spin);
 201                         kfree(ipul, pmp->minode);
 202                         hammer2_inode_drop(ip);         /* vp ref */
 203                 }
 204                 /* retain ref from vp for ipul */
 205         } else {
 206                 hammer2_inode_drop(ip);                 /* vp ref */
 207         }
 208
 209         /*
 210          * XXX handle background sync when ip dirty, kernel will no longer
 211          * notify us regarding this inode because there is no longer a
 212          * vnode attached to it.
 213          */
 214
 215         LOCKSTOP;
 216         return (0);
 217 }
 218
 219 static
 220 int
 221 hammer2_vop_fsync(struct vop_fsync_args *ap)
 222 {
 223         hammer2_inode_t *ip;
 224         struct vnode *vp;
 225
 226         LOCKSTART;
 227         vp = ap->a_vp;
 228         ip = VTOI(vp);
 229
 230 #if 0
 231         /* XXX can't do this yet */
 232         hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH);
 233         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 234 #endif
 235         hammer2_trans_init(ip->pmp, 0);
 236         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 237
 238         /*
 239          * Calling chain_flush here creates a lot of duplicative
 240          * COW operations due to non-optimal vnode ordering.
 241          *
 242          * Only do it for an actual fsync() syscall.  The other forms
 243          * which call this function will eventually call chain_flush
 244          * on the volume root as a catch-all, which is far more optimal.
 245          */
 246         hammer2_inode_lock(ip, 0);
 247         if (ip->flags & HAMMER2_INODE_MODIFIED)
 248                 hammer2_inode_chain_sync(ip);
 249         hammer2_inode_unlock(ip);
 250         hammer2_trans_done(ip->pmp);
 251
 252         LOCKSTOP;
 253         return (0);
 254 }
 255
 256 static
 257 int
 258 hammer2_vop_access(struct vop_access_args *ap)
 259 {
 260         hammer2_inode_t *ip = VTOI(ap->a_vp);
 261         uid_t uid;
 262         gid_t gid;
 263         int error;
 264
 265         LOCKSTART;
 266         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 267         uid = hammer2_to_unix_xid(&ip->meta.uid);
 268         gid = hammer2_to_unix_xid(&ip->meta.gid);
 269         error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
 270         hammer2_inode_unlock(ip);
 271
 272         LOCKSTOP;
 273         return (error);
 274 }
 275
 276 static
 277 int
 278 hammer2_vop_getattr(struct vop_getattr_args *ap)
 279 {
 280         hammer2_pfs_t *pmp;
 281         hammer2_inode_t *ip;
 282         struct vnode *vp;
 283         struct vattr *vap;
 284         hammer2_chain_t *chain;
 285         int i;
 286
 287         LOCKSTART;
 288         vp = ap->a_vp;
 289         vap = ap->a_vap;
 290
 291         ip = VTOI(vp);
 292         pmp = ip->pmp;
 293
 294         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 295
 296         vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
 297         vap->va_fileid = ip->meta.inum;
 298         vap->va_mode = ip->meta.mode;
 299         vap->va_nlink = ip->meta.nlinks;
 300         vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
 301         vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
 302         vap->va_rmajor = 0;
 303         vap->va_rminor = 0;
 304         vap->va_size = ip->meta.size;   /* protected by shared lock */
 305         vap->va_blocksize = HAMMER2_PBUFSIZE;
 306         vap->va_flags = ip->meta.uflags;
 307         hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
 308         hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
 309         hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
 310         vap->va_gen = 1;
 311         vap->va_bytes = 0;
 312         if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) {
 313                 /*
 314                  * Can't really calculate directory use sans the files under
 315                  * it, just assume one block for now.
 316                  */
 317                 vap->va_bytes += HAMMER2_INODE_BYTES;
 318         } else {
 319                 for (i = 0; i < ip->cluster.nchains; ++i) {
 320                         if ((chain = ip->cluster.array[i].chain) != NULL) {
 321                                 if (vap->va_bytes < chain->bref.data_count)
 322                                         vap->va_bytes = chain->bref.data_count;
 323                         }
 324                 }
 325         }
 326         vap->va_type = hammer2_get_vtype(ip->meta.type);
 327         vap->va_filerev = 0;
 328         vap->va_uid_uuid = ip->meta.uid;
 329         vap->va_gid_uuid = ip->meta.gid;
 330         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 331                           VA_FSID_UUID_VALID;
 332
 333         hammer2_inode_unlock(ip);
 334
 335         LOCKSTOP;
 336         return (0);
 337 }
 338
 339 static
 340 int
 341 hammer2_vop_setattr(struct vop_setattr_args *ap)
 342 {
 343         hammer2_inode_t *ip;
 344         struct vnode *vp;
 345         struct vattr *vap;
 346         int error;
 347         int kflags = 0;
 348         uint64_t ctime;
 349
 350         LOCKSTART;
 351         vp = ap->a_vp;
 352         vap = ap->a_vap;
 353         hammer2_update_time(&ctime);
 354
 355         ip = VTOI(vp);
 356
 357         if (ip->pmp->ronly) {
 358                 LOCKSTOP;
 359                 return(EROFS);
 360         }
 361
 362         hammer2_pfs_memory_wait(ip->pmp);
 363         hammer2_trans_init(ip->pmp, 0);
 364         hammer2_inode_lock(ip, 0);
 365         error = 0;
 366
 367         if (vap->va_flags != VNOVAL) {
 368                 u_int32_t flags;
 369
 370                 flags = ip->meta.uflags;
 371                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
 372                                      hammer2_to_unix_xid(&ip->meta.uid),
 373                                      ap->a_cred);
 374                 if (error == 0) {
 375                         if (ip->meta.uflags != flags) {
 376                                 hammer2_inode_modify(ip);
 377                                 ip->meta.uflags = flags;
 378                                 ip->meta.ctime = ctime;
 379                                 kflags |= NOTE_ATTRIB;
 380                         }
 381                         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 382                                 error = 0;
 383                                 goto done;
 384                         }
 385                 }
 386                 goto done;
 387         }
 388         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 389                 error = EPERM;
 390                 goto done;
 391         }
 392         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
 393                 mode_t cur_mode = ip->meta.mode;
 394                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 395                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 396                 uuid_t uuid_uid;
 397                 uuid_t uuid_gid;
 398
 399                 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
 400                                          ap->a_cred,
 401                                          &cur_uid, &cur_gid, &cur_mode);
 402                 if (error == 0) {
 403                         hammer2_guid_to_uuid(&uuid_uid, cur_uid);
 404                         hammer2_guid_to_uuid(&uuid_gid, cur_gid);
 405                         if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
 406                             bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
 407                             ip->meta.mode != cur_mode
 408                         ) {
 409                                 hammer2_inode_modify(ip);
 410                                 ip->meta.uid = uuid_uid;
 411                                 ip->meta.gid = uuid_gid;
 412                                 ip->meta.mode = cur_mode;
 413                                 ip->meta.ctime = ctime;
 414                         }
 415                         kflags |= NOTE_ATTRIB;
 416                 }
 417         }
 418
 419         /*
 420          * Resize the file
 421          */
 422         if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
 423                 switch(vp->v_type) {
 424                 case VREG:
 425                         if (vap->va_size == ip->meta.size)
 426                                 break;
 427                         if (vap->va_size < ip->meta.size) {
 428                                 hammer2_truncate_file(ip, vap->va_size);
 429                         } else {
 430                                 hammer2_extend_file(ip, vap->va_size);
 431                         }
 432                         hammer2_inode_modify(ip);
 433                         ip->meta.mtime = ctime;
 434                         break;
 435                 default:
 436                         error = EINVAL;
 437                         goto done;
 438                 }
 439         }
 440 #if 0
 441         /* atime not supported */
 442         if (vap->va_atime.tv_sec != VNOVAL) {
 443                 hammer2_inode_modify(ip);
 444                 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
 445                 kflags |= NOTE_ATTRIB;
 446         }
 447 #endif
 448         if (vap->va_mode != (mode_t)VNOVAL) {
 449                 mode_t cur_mode = ip->meta.mode;
 450                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 451                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 452
 453                 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
 454                                          cur_uid, cur_gid, &cur_mode);
 455                 if (error == 0 && ip->meta.mode != cur_mode) {
 456                         hammer2_inode_modify(ip);
 457                         ip->meta.mode = cur_mode;
 458                         ip->meta.ctime = ctime;
 459                         kflags |= NOTE_ATTRIB;
 460                 }
 461         }
 462
 463         if (vap->va_mtime.tv_sec != VNOVAL) {
 464                 hammer2_inode_modify(ip);
 465                 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
 466                 kflags |= NOTE_ATTRIB;
 467         }
 468
 469 done:
 470         /*
 471          * If a truncation occurred we must call inode_fsync() now in order
 472          * to trim the related data chains, otherwise a later expansion can
 473          * cause havoc.
 474          *
 475          * If an extend occured that changed the DIRECTDATA state, we must
 476          * call inode_fsync now in order to prepare the inode's indirect
 477          * block table.
 478          */
 479         if (ip->flags & HAMMER2_INODE_RESIZED)
 480                 hammer2_inode_chain_sync(ip);
 481
 482         /*
 483          * Cleanup.
 484          */
 485         hammer2_inode_unlock(ip);
 486         hammer2_trans_done(ip->pmp);
 487         hammer2_knote(ip->vp, kflags);
 488
 489         LOCKSTOP;
 490         return (error);
 491 }
 492
 493 static
 494 int
 495 hammer2_vop_readdir(struct vop_readdir_args *ap)
 496 {
 497         hammer2_xop_readdir_t *xop;
 498         hammer2_blockref_t bref;
 499         hammer2_inode_t *ip;
 500         hammer2_tid_t inum;
 501         hammer2_key_t lkey;
 502         struct uio *uio;
 503         off_t *cookies;
 504         off_t saveoff;
 505         int cookie_index;
 506         int ncookies;
 507         int error;
 508         int eofflag;
 509         int dtype;
 510         int r;
 511
 512         LOCKSTART;
 513         ip = VTOI(ap->a_vp);
 514         uio = ap->a_uio;
 515         saveoff = uio->uio_offset;
 516         eofflag = 0;
 517         error = 0;
 518
 519         /*
 520          * Setup cookies directory entry cookies if requested
 521          */
 522         if (ap->a_ncookies) {
 523                 ncookies = uio->uio_resid / 16 + 1;
 524                 if (ncookies > 1024)
 525                         ncookies = 1024;
 526                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 527         } else {
 528                 ncookies = -1;
 529                 cookies = NULL;
 530         }
 531         cookie_index = 0;
 532
 533         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 534
 535         /*
 536          * Handle artificial entries.  To ensure that only positive 64 bit
 537          * quantities are returned to userland we always strip off bit 63.
 538          * The hash code is designed such that codes 0x0000-0x7FFF are not
 539          * used, allowing us to use these codes for articial entries.
 540          *
 541          * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
 542          * allow '..' to cross the mount point into (e.g.) the super-root.
 543          */
 544         if (saveoff == 0) {
 545                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 546                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
 547                 if (r)
 548                         goto done;
 549                 if (cookies)
 550                         cookies[cookie_index] = saveoff;
 551                 ++saveoff;
 552                 ++cookie_index;
 553                 if (cookie_index == ncookies)
 554                         goto done;
 555         }
 556
 557         if (saveoff == 1) {
 558                 /*
 559                  * Be careful with lockorder when accessing ".."
 560                  *
 561                  * (ip is the current dir. xip is the parent dir).
 562                  */
 563                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 564                 if (ip->pip && ip != ip->pmp->iroot)
 565                         inum = ip->pip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 566                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
 567                 if (r)
 568                         goto done;
 569                 if (cookies)
 570                         cookies[cookie_index] = saveoff;
 571                 ++saveoff;
 572                 ++cookie_index;
 573                 if (cookie_index == ncookies)
 574                         goto done;
 575         }
 576
 577         lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
 578         if (hammer2_debug & 0x0020)
 579                 kprintf("readdir: lkey %016jx\n", lkey);
 580         if (error)
 581                 goto done;
 582
 583         /*
 584          * Use XOP for cluster scan.
 585          *
 586          * parent is the inode cluster, already locked for us.  Don't
 587          * double lock shared locks as this will screw up upgrades.
 588          */
 589         xop = hammer2_xop_alloc(ip, 0);
 590         xop->lkey = lkey;
 591         hammer2_xop_start(&xop->head, hammer2_xop_readdir);
 592
 593         for (;;) {
 594                 const hammer2_inode_data_t *ripdata;
 595
 596                 error = hammer2_xop_collect(&xop->head, 0);
 597                 if (error)
 598                         break;
 599                 if (cookie_index == ncookies)
 600                         break;
 601                 if (hammer2_debug & 0x0020)
 602                 kprintf("cluster chain %p %p\n",
 603                         xop->head.cluster.focus,
 604                         (xop->head.cluster.focus ?
 605                          xop->head.cluster.focus->data : (void *)-1));
 606                 ripdata = &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
 607                 hammer2_cluster_bref(&xop->head.cluster, &bref);
 608                 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
 609                         dtype = hammer2_get_dtype(ripdata);
 610                         saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 611                         r = vop_write_dirent(&error, uio,
 612                                              ripdata->meta.inum &
 613                                               HAMMER2_DIRHASH_USERMSK,
 614                                              dtype,
 615                                              ripdata->meta.name_len,
 616                                              ripdata->filename);
 617                         if (r)
 618                                 break;
 619                         if (cookies)
 620                                 cookies[cookie_index] = saveoff;
 621                         ++cookie_index;
 622                 } else {
 623                         /* XXX chain error */
 624                         kprintf("bad chain type readdir %d\n", bref.type);
 625                 }
 626         }
 627         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
 628         if (error == ENOENT) {
 629                 error = 0;
 630                 eofflag = 1;
 631                 saveoff = (hammer2_key_t)-1;
 632         } else {
 633                 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 634         }
 635 done:
 636         hammer2_inode_unlock(ip);
 637         if (ap->a_eofflag)
 638                 *ap->a_eofflag = eofflag;
 639         if (hammer2_debug & 0x0020)
 640                 kprintf("readdir: done at %016jx\n", saveoff);
 641         uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
 642         if (error && cookie_index == 0) {
 643                 if (cookies) {
 644                         kfree(cookies, M_TEMP);
 645                         *ap->a_ncookies = 0;
 646                         *ap->a_cookies = NULL;
 647                 }
 648         } else {
 649                 if (cookies) {
 650                         *ap->a_ncookies = cookie_index;
 651                         *ap->a_cookies = cookies;
 652                 }
 653         }
 654         LOCKSTOP;
 655         return (error);
 656 }
 657
 658 /*
 659  * hammer2_vop_readlink { vp, uio, cred }
 660  */
 661 static
 662 int
 663 hammer2_vop_readlink(struct vop_readlink_args *ap)
 664 {
 665         struct vnode *vp;
 666         hammer2_inode_t *ip;
 667         int error;
 668
 669         vp = ap->a_vp;
 670         if (vp->v_type != VLNK)
 671                 return (EINVAL);
 672         ip = VTOI(vp);
 673
 674         error = hammer2_read_file(ip, ap->a_uio, 0);
 675         return (error);
 676 }
 677
 678 static
 679 int
 680 hammer2_vop_read(struct vop_read_args *ap)
 681 {
 682         struct vnode *vp;
 683         hammer2_inode_t *ip;
 684         struct uio *uio;
 685         int error;
 686         int seqcount;
 687         int bigread;
 688
 689         /*
 690          * Read operations supported on this vnode?
 691          */
 692         vp = ap->a_vp;
 693         if (vp->v_type != VREG)
 694                 return (EINVAL);
 695
 696         /*
 697          * Misc
 698          */
 699         ip = VTOI(vp);
 700         uio = ap->a_uio;
 701         error = 0;
 702
 703         seqcount = ap->a_ioflag >> 16;
 704         bigread = (uio->uio_resid > 100 * 1024 * 1024);
 705
 706         error = hammer2_read_file(ip, uio, seqcount);
 707         return (error);
 708 }
 709
 710 static
 711 int
 712 hammer2_vop_write(struct vop_write_args *ap)
 713 {
 714         hammer2_inode_t *ip;
 715         thread_t td;
 716         struct vnode *vp;
 717         struct uio *uio;
 718         int error;
 719         int seqcount;
 720
 721         /*
 722          * Read operations supported on this vnode?
 723          */
 724         vp = ap->a_vp;
 725         if (vp->v_type != VREG)
 726                 return (EINVAL);
 727
 728         /*
 729          * Misc
 730          */
 731         ip = VTOI(vp);
 732         uio = ap->a_uio;
 733         error = 0;
 734         if (ip->pmp->ronly) {
 735                 return (EROFS);
 736         }
 737
 738         seqcount = ap->a_ioflag >> 16;
 739
 740         /*
 741          * Check resource limit
 742          */
 743         if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
 744             uio->uio_offset + uio->uio_resid >
 745              td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 746                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 747                 return (EFBIG);
 748         }
 749
 750         /*
 751          * The transaction interlocks against flushes initiations
 752          * (note: but will run concurrently with the actual flush).
 753          */
 754         hammer2_trans_init(ip->pmp, 0);
 755         error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
 756         hammer2_trans_done(ip->pmp);
 757
 758         return (error);
 759 }
 760
 761 /*
 762  * Perform read operations on a file or symlink given an UNLOCKED
 763  * inode and uio.
 764  *
 765  * The passed ip is not locked.
 766  */
 767 static
 768 int
 769 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
 770 {
 771         hammer2_off_t size;
 772         struct buf *bp;
 773         int error;
 774
 775         error = 0;
 776
 777         /*
 778          * UIO read loop.
 779          *
 780          * WARNING! Assumes that the kernel interlocks size changes at the
 781          *          vnode level.
 782          */
 783         hammer2_mtx_sh(&ip->lock);
 784         size = ip->meta.size;
 785         hammer2_mtx_unlock(&ip->lock);
 786
 787         while (uio->uio_resid > 0 && uio->uio_offset < size) {
 788                 hammer2_key_t lbase;
 789                 hammer2_key_t leof;
 790                 int lblksize;
 791                 int loff;
 792                 int n;
 793
 794                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 795                                                 &lbase, &leof);
 796
 797                 error = cluster_read(ip->vp, leof, lbase, lblksize,
 798                                      uio->uio_resid, seqcount * BKVASIZE,
 799                                      &bp);
 800
 801                 if (error)
 802                         break;
 803                 loff = (int)(uio->uio_offset - lbase);
 804                 n = lblksize - loff;
 805                 if (n > uio->uio_resid)
 806                         n = uio->uio_resid;
 807                 if (n > size - uio->uio_offset)
 808                         n = (int)(size - uio->uio_offset);
 809                 bp->b_flags |= B_AGE;
 810                 uiomove((char *)bp->b_data + loff, n, uio);
 811                 bqrelse(bp);
 812         }
 813         return (error);
 814 }
 815
 816 /*
 817  * Write to the file represented by the inode via the logical buffer cache.
 818  * The inode may represent a regular file or a symlink.
 819  *
 820  * The inode must not be locked.
 821  */
 822 static
 823 int
 824 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
 825                    int ioflag, int seqcount)
 826 {
 827         hammer2_key_t old_eof;
 828         hammer2_key_t new_eof;
 829         struct buf *bp;
 830         int kflags;
 831         int error;
 832         int modified;
 833
 834         /*
 835          * Setup if append
 836          *
 837          * WARNING! Assumes that the kernel interlocks size changes at the
 838          *          vnode level.
 839          */
 840         hammer2_mtx_ex(&ip->lock);
 841         if (ioflag & IO_APPEND)
 842                 uio->uio_offset = ip->meta.size;
 843         old_eof = ip->meta.size;
 844
 845         /*
 846          * Extend the file if necessary.  If the write fails at some point
 847          * we will truncate it back down to cover as much as we were able
 848          * to write.
 849          *
 850          * Doing this now makes it easier to calculate buffer sizes in
 851          * the loop.
 852          */
 853         kflags = 0;
 854         error = 0;
 855         modified = 0;
 856
 857         if (uio->uio_offset + uio->uio_resid > old_eof) {
 858                 new_eof = uio->uio_offset + uio->uio_resid;
 859                 modified = 1;
 860                 hammer2_extend_file(ip, new_eof);
 861                 kflags |= NOTE_EXTEND;
 862         } else {
 863                 new_eof = old_eof;
 864         }
 865         hammer2_mtx_unlock(&ip->lock);
 866
 867         /*
 868          * UIO write loop
 869          */
 870         while (uio->uio_resid > 0) {
 871                 hammer2_key_t lbase;
 872                 int trivial;
 873                 int endofblk;
 874                 int lblksize;
 875                 int loff;
 876                 int n;
 877
 878                 /*
 879                  * Don't allow the buffer build to blow out the buffer
 880                  * cache.
 881                  */
 882                 if ((ioflag & IO_RECURSE) == 0)
 883                         bwillwrite(HAMMER2_PBUFSIZE);
 884
 885                 /*
 886                  * This nominally tells us how much we can cluster and
 887                  * what the logical buffer size needs to be.  Currently
 888                  * we don't try to cluster the write and just handle one
 889                  * block at a time.
 890                  */
 891                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 892                                                 &lbase, NULL);
 893                 loff = (int)(uio->uio_offset - lbase);
 894
 895                 KKASSERT(lblksize <= 65536);
 896
 897                 /*
 898                  * Calculate bytes to copy this transfer and whether the
 899                  * copy completely covers the buffer or not.
 900                  */
 901                 trivial = 0;
 902                 n = lblksize - loff;
 903                 if (n > uio->uio_resid) {
 904                         n = uio->uio_resid;
 905                         if (loff == lbase && uio->uio_offset + n == new_eof)
 906                                 trivial = 1;
 907                         endofblk = 0;
 908                 } else {
 909                         if (loff == 0)
 910                                 trivial = 1;
 911                         endofblk = 1;
 912                 }
 913
 914                 /*
 915                  * Get the buffer
 916                  */
 917                 if (uio->uio_segflg == UIO_NOCOPY) {
 918                         /*
 919                          * Issuing a write with the same data backing the
 920                          * buffer.  Instantiate the buffer to collect the
 921                          * backing vm pages, then read-in any missing bits.
 922                          *
 923                          * This case is used by vop_stdputpages().
 924                          */
 925                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 926                         if ((bp->b_flags & B_CACHE) == 0) {
 927                                 bqrelse(bp);
 928                                 error = bread(ip->vp, lbase, lblksize, &bp);
 929                         }
 930                 } else if (trivial) {
 931                         /*
 932                          * Even though we are entirely overwriting the buffer
 933                          * we may still have to zero it out to avoid a
 934                          * mmap/write visibility issue.
 935                          */
 936                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 937                         if ((bp->b_flags & B_CACHE) == 0)
 938                                 vfs_bio_clrbuf(bp);
 939                 } else {
 940                         /*
 941                          * Partial overwrite, read in any missing bits then
 942                          * replace the portion being written.
 943                          *
 944                          * (The strategy code will detect zero-fill physical
 945                          * blocks for this case).
 946                          */
 947                         error = bread(ip->vp, lbase, lblksize, &bp);
 948                         if (error == 0)
 949                                 bheavy(bp);
 950                 }
 951
 952                 if (error) {
 953                         brelse(bp);
 954                         break;
 955                 }
 956
 957                 /*
 958                  * Ok, copy the data in
 959                  */
 960                 error = uiomove(bp->b_data + loff, n, uio);
 961                 kflags |= NOTE_WRITE;
 962                 modified = 1;
 963                 if (error) {
 964                         brelse(bp);
 965                         break;
 966                 }
 967
 968                 /*
 969                  * WARNING: Pageout daemon will issue UIO_NOCOPY writes
 970                  *          with IO_SYNC or IO_ASYNC set.  These writes
 971                  *          must be handled as the pageout daemon expects.
 972                  */
 973                 if (ioflag & IO_SYNC) {
 974                         bwrite(bp);
 975                 } else if ((ioflag & IO_DIRECT) && endofblk) {
 976                         bawrite(bp);
 977                 } else if (ioflag & IO_ASYNC) {
 978                         bawrite(bp);
 979                 } else {
 980                         bdwrite(bp);
 981                 }
 982         }
 983
 984         /*
 985          * Cleanup.  If we extended the file EOF but failed to write through
 986          * the entire write is a failure and we have to back-up.
 987          */
 988         if (error && new_eof != old_eof) {
 989                 hammer2_mtx_ex(&ip->lock);
 990                 hammer2_truncate_file(ip, old_eof);
 991                 if (ip->flags & HAMMER2_INODE_MODIFIED)
 992                         hammer2_inode_chain_sync(ip);
 993                 hammer2_mtx_unlock(&ip->lock);
 994         } else if (modified) {
 995                 hammer2_mtx_ex(&ip->lock);
 996                 hammer2_inode_modify(ip);
 997                 hammer2_update_time(&ip->meta.mtime);
 998                 if (ip->flags & HAMMER2_INODE_MODIFIED)
 999                         hammer2_inode_chain_sync(ip);
1000                 hammer2_mtx_unlock(&ip->lock);
1001                 hammer2_knote(ip->vp, kflags);
1002         }
1003         hammer2_trans_assert_strategy(ip->pmp);
1004
1005         return error;
1006 }
1007
1008 /*
1009  * Truncate the size of a file.  The inode must not be locked.
1010  *
1011  * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1012  * ensure that any on-media data beyond the new file EOF has been destroyed.
1013  *
1014  * WARNING: nvtruncbuf() can only be safely called without the inode lock
1015  *          held due to the way our write thread works.  If the truncation
1016  *          occurs in the middle of a buffer, nvtruncbuf() is responsible
1017  *          for dirtying that buffer and zeroing out trailing bytes.
1018  *
1019  * WARNING! Assumes that the kernel interlocks size changes at the
1020  *          vnode level.
1021  *
1022  * WARNING! Caller assumes responsibility for removing dead blocks
1023  *          if INODE_RESIZED is set.
1024  */
1025 static
1026 void
1027 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1028 {
1029         hammer2_key_t lbase;
1030         int nblksize;
1031
1032         LOCKSTART;
1033         hammer2_mtx_unlock(&ip->lock);
1034         if (ip->vp) {
1035                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1036                 nvtruncbuf(ip->vp, nsize,
1037                            nblksize, (int)nsize & (nblksize - 1),
1038                            0);
1039         }
1040         hammer2_mtx_ex(&ip->lock);
1041         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1042         ip->osize = ip->meta.size;
1043         ip->meta.size = nsize;
1044         atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED |
1045                                    HAMMER2_INODE_RESIZED);
1046         LOCKSTOP;
1047 }
1048
1049 /*
1050  * Extend the size of a file.  The inode must not be locked.
1051  *
1052  * Even though the file size is changing, we do not have to set the
1053  * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1054  * boundary.  When this occurs a hammer2_inode_chain_sync() is required
1055  * to prepare the inode cluster's indirect block table, otherwise
1056  * async execution of the strategy code will implode on us.
1057  *
1058  * WARNING! Assumes that the kernel interlocks size changes at the
1059  *          vnode level.
1060  *
1061  * WARNING! Caller assumes responsibility for transitioning out
1062  *          of the inode DIRECTDATA mode if INODE_RESIZED is set.
1063  */
1064 static
1065 void
1066 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1067 {
1068         hammer2_key_t lbase;
1069         hammer2_key_t osize;
1070         int oblksize;
1071         int nblksize;
1072
1073         LOCKSTART;
1074
1075         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1076         osize = ip->meta.size;
1077         ip->osize = osize;
1078         ip->meta.size = nsize;
1079         atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1080
1081         if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1082                 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1083                 hammer2_inode_chain_sync(ip);
1084         }
1085
1086         hammer2_mtx_unlock(&ip->lock);
1087         if (ip->vp) {
1088                 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1089                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1090                 nvextendbuf(ip->vp,
1091                             osize, nsize,
1092                             oblksize, nblksize,
1093                             -1, -1, 0);
1094         }
1095         hammer2_mtx_ex(&ip->lock);
1096
1097         LOCKSTOP;
1098 }
1099
1100 static
1101 int
1102 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1103 {
1104         hammer2_xop_nresolve_t *xop;
1105         hammer2_inode_t *ip;
1106         hammer2_inode_t *dip;
1107         struct namecache *ncp;
1108         struct vnode *vp;
1109         int error;
1110
1111         LOCKSTART;
1112         dip = VTOI(ap->a_dvp);
1113         xop = hammer2_xop_alloc(dip, 0);
1114
1115         ncp = ap->a_nch->ncp;
1116         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1117
1118         /*
1119          * Note: In DragonFly the kernel handles '.' and '..'.
1120          */
1121         hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1122         hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1123
1124         error = hammer2_xop_collect(&xop->head, 0);
1125         if (error) {
1126                 ip = NULL;
1127         } else {
1128                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1129         }
1130         hammer2_inode_unlock(dip);
1131
1132         /*
1133          * Acquire the related vnode
1134          *
1135          * NOTE: For error processing, only ENOENT resolves the namecache
1136          *       entry to NULL, otherwise we just return the error and
1137          *       leave the namecache unresolved.
1138          *
1139          * NOTE: multiple hammer2_inode structures can be aliased to the
1140          *       same chain element, for example for hardlinks.  This
1141          *       use case does not 'reattach' inode associations that
1142          *       might already exist, but always allocates a new one.
1143          *
1144          * WARNING: inode structure is locked exclusively via inode_get
1145          *          but chain was locked shared.  inode_unlock()
1146          *          will handle it properly.
1147          */
1148         if (ip) {
1149                 vp = hammer2_igetv(ip, &error);
1150                 if (error == 0) {
1151                         vn_unlock(vp);
1152                         cache_setvp(ap->a_nch, vp);
1153                 } else if (error == ENOENT) {
1154                         cache_setvp(ap->a_nch, NULL);
1155                 }
1156                 hammer2_inode_unlock(ip);
1157
1158                 /*
1159                  * The vp should not be released until after we've disposed
1160                  * of our locks, because it might cause vop_inactive() to
1161                  * be called.
1162                  */
1163                 if (vp)
1164                         vrele(vp);
1165         } else {
1166                 error = ENOENT;
1167                 cache_setvp(ap->a_nch, NULL);
1168         }
1169         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1170         KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1171                 ("resolve error %d/%p ap %p\n",
1172                  error, ap->a_nch->ncp->nc_vp, ap));
1173         LOCKSTOP;
1174
1175         return error;
1176 }
1177
1178 static
1179 int
1180 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1181 {
1182         hammer2_inode_t *dip;
1183         hammer2_inode_t *ip;
1184         int error;
1185
1186         LOCKSTART;
1187         dip = VTOI(ap->a_dvp);
1188
1189         if ((ip = dip->pip) == NULL) {
1190                 *ap->a_vpp = NULL;
1191                 LOCKSTOP;
1192                 return ENOENT;
1193         }
1194         hammer2_inode_lock(ip, 0);
1195         *ap->a_vpp = hammer2_igetv(ip, &error);
1196         hammer2_inode_unlock(ip);
1197
1198         LOCKSTOP;
1199         return error;
1200 }
1201
1202 static
1203 int
1204 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1205 {
1206         hammer2_inode_t *dip;
1207         hammer2_inode_t *nip;
1208         struct namecache *ncp;
1209         const uint8_t *name;
1210         size_t name_len;
1211         int error;
1212
1213         LOCKSTART;
1214         dip = VTOI(ap->a_dvp);
1215         if (dip->pmp->ronly) {
1216                 LOCKSTOP;
1217                 return (EROFS);
1218         }
1219
1220         ncp = ap->a_nch->ncp;
1221         name = ncp->nc_name;
1222         name_len = ncp->nc_nlen;
1223
1224         hammer2_pfs_memory_wait(dip->pmp);
1225         hammer2_trans_init(dip->pmp, 0);
1226         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1227                                    name, name_len, 0,
1228                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1229                                    0, &error);
1230         if (error) {
1231                 KKASSERT(nip == NULL);
1232                 *ap->a_vpp = NULL;
1233         } else {
1234                 *ap->a_vpp = hammer2_igetv(nip, &error);
1235                 hammer2_inode_unlock(nip);
1236         }
1237         hammer2_trans_done(dip->pmp);
1238
1239         if (error == 0) {
1240                 cache_setunresolved(ap->a_nch);
1241                 cache_setvp(ap->a_nch, *ap->a_vpp);
1242         }
1243         LOCKSTOP;
1244         return error;
1245 }
1246
1247 static
1248 int
1249 hammer2_vop_open(struct vop_open_args *ap)
1250 {
1251         return vop_stdopen(ap);
1252 }
1253
1254 /*
1255  * hammer2_vop_advlock { vp, id, op, fl, flags }
1256  */
1257 static
1258 int
1259 hammer2_vop_advlock(struct vop_advlock_args *ap)
1260 {
1261         hammer2_inode_t *ip = VTOI(ap->a_vp);
1262         hammer2_off_t size;
1263
1264         size = ip->meta.size;
1265         return (lf_advlock(ap, &ip->advlock, size));
1266 }
1267
1268 static
1269 int
1270 hammer2_vop_close(struct vop_close_args *ap)
1271 {
1272         return vop_stdclose(ap);
1273 }
1274
1275 /*
1276  * hammer2_vop_nlink { nch, dvp, vp, cred }
1277  *
1278  * Create a hardlink from (vp) to {dvp, nch}.
1279  */
1280 static
1281 int
1282 hammer2_vop_nlink(struct vop_nlink_args *ap)
1283 {
1284         hammer2_xop_nlink_t *xop1;
1285         hammer2_inode_t *fdip;  /* target directory to create link in */
1286         hammer2_inode_t *tdip;  /* target directory to create link in */
1287         hammer2_inode_t *cdip;  /* common parent directory */
1288         hammer2_inode_t *ip;    /* inode we are hardlinking to */
1289         struct namecache *ncp;
1290         const uint8_t *name;
1291         size_t name_len;
1292         int nlink_locked;
1293         int error;
1294
1295         LOCKSTART;
1296         tdip = VTOI(ap->a_dvp);
1297         if (tdip->pmp->ronly) {
1298                 LOCKSTOP;
1299                 return (EROFS);
1300         }
1301
1302         ncp = ap->a_nch->ncp;
1303         name = ncp->nc_name;
1304         name_len = ncp->nc_nlen;
1305
1306         /*
1307          * ip represents the file being hardlinked.  The file could be a
1308          * normal file or a hardlink target if it has already been hardlinked.
1309          * If ip is a hardlinked target then ip->pip represents the location
1310          * of the hardlinked target, NOT the location of the hardlink pointer.
1311          *
1312          * Bump nlinks and potentially also create or move the hardlink
1313          * target in the parent directory common to (ip) and (tdip).  The
1314          * consolidation code can modify ip->cluster and ip->pip.  The
1315          * returned cluster is locked.
1316          */
1317         ip = VTOI(ap->a_vp);
1318         hammer2_pfs_memory_wait(ip->pmp);
1319         hammer2_trans_init(ip->pmp, 0);
1320
1321         /*
1322          * The common parent directory must be locked first to avoid deadlocks.
1323          * Also note that fdip and/or tdip might match cdip.
1324          *
1325          * WARNING!  The kernel's namecache locks are insufficient for
1326          *           protecting us from hardlink shifts, since unrelated
1327          *           rename() or link() calls on parent directories might
1328          *           cause a shift.  A PFS-wide lock is required for this
1329          *           situation.
1330          */
1331         if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY ||
1332             (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
1333                 lockmgr(&ip->pmp->lock_nlink, LK_EXCLUSIVE);
1334                 nlink_locked = 1;
1335         } else {
1336                 nlink_locked = 0;
1337         }
1338         fdip = ip->pip;
1339         cdip = hammer2_inode_common_parent(fdip, tdip);
1340         hammer2_inode_lock(cdip, 0);
1341         hammer2_inode_lock(fdip, 0);
1342         hammer2_inode_lock(tdip, 0);
1343         hammer2_inode_lock(ip, 0);
1344         error = 0;
1345
1346         /*
1347          * Dispatch xop_nlink unconditionally since we have to update nlinks.
1348          *
1349          * Otherwise we'd be able to avoid the XOP if the ip does not have
1350          * to be converted or moved.
1351          * If ip is not a hardlink target we must convert it to a hardlink.
1352          * If fdip != cdip we must shift the inode to cdip.
1353          *
1354          * XXX this and other nlink update usage should be passed top-down
1355          *     and not updated with a delta bottom-up.
1356          */
1357 #if 0
1358         if (fdip != cdip || (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE))
1359 #endif
1360         {
1361                 xop1 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1362                 hammer2_xop_setip2(&xop1->head, ip);
1363                 hammer2_xop_setip3(&xop1->head, cdip);
1364                 xop1->nlinks_delta = 1;
1365
1366                 hammer2_xop_start(&xop1->head, hammer2_xop_nlink);
1367                 error = hammer2_xop_collect(&xop1->head, 0);
1368                 hammer2_xop_retire(&xop1->head, HAMMER2_XOPMASK_VOP);
1369                 if (error == ENOENT)
1370                         error = 0;
1371         }
1372
1373         /*
1374          * Must synchronize original inode whos chains are now a hardlink
1375          * target.  We must match what the backend XOP did to the
1376          * chains.
1377          */
1378         if (error == 0 && (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1379                 hammer2_inode_modify(ip);
1380                 ip->meta.name_key = ip->meta.inum;
1381                 ip->meta.name_len = 18; /* "0x%016jx" */
1382         }
1383
1384         /*
1385          * Create the hardlink target and bump nlinks.
1386          */
1387         if (error == 0) {
1388                 hammer2_inode_create(tdip, NULL, NULL,
1389                                      name, name_len, 0,
1390                                      ip->meta.inum,
1391                                      HAMMER2_OBJTYPE_HARDLINK, ip->meta.type,
1392                                      0, &error);
1393                 hammer2_inode_modify(ip);
1394                 ++ip->meta.nlinks;
1395         }
1396         if (error == 0) {
1397                 cache_setunresolved(ap->a_nch);
1398                 cache_setvp(ap->a_nch, ap->a_vp);
1399         }
1400         hammer2_inode_unlock(ip);
1401         hammer2_inode_unlock(tdip);
1402         hammer2_inode_unlock(fdip);
1403         hammer2_inode_unlock(cdip);
1404         hammer2_inode_drop(cdip);
1405
1406         if (nlink_locked)
1407                 lockmgr(&ip->pmp->lock_nlink, LK_RELEASE);
1408         hammer2_trans_done(ip->pmp);
1409
1410         LOCKSTOP;
1411         return error;
1412 }
1413
1414 /*
1415  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1416  *
1417  * The operating system has already ensured that the directory entry
1418  * does not exist and done all appropriate namespace locking.
1419  */
1420 static
1421 int
1422 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1423 {
1424         hammer2_inode_t *dip;
1425         hammer2_inode_t *nip;
1426         struct namecache *ncp;
1427         const uint8_t *name;
1428         size_t name_len;
1429         int error;
1430
1431         LOCKSTART;
1432         dip = VTOI(ap->a_dvp);
1433         if (dip->pmp->ronly) {
1434                 LOCKSTOP;
1435                 return (EROFS);
1436         }
1437
1438         ncp = ap->a_nch->ncp;
1439         name = ncp->nc_name;
1440         name_len = ncp->nc_nlen;
1441         hammer2_pfs_memory_wait(dip->pmp);
1442         hammer2_trans_init(dip->pmp, 0);
1443
1444         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1445                                    name, name_len, 0,
1446                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1447                                    0, &error);
1448         if (error) {
1449                 KKASSERT(nip == NULL);
1450                 *ap->a_vpp = NULL;
1451         } else {
1452                 *ap->a_vpp = hammer2_igetv(nip, &error);
1453                 hammer2_inode_unlock(nip);
1454         }
1455         hammer2_trans_done(dip->pmp);
1456
1457         if (error == 0) {
1458                 cache_setunresolved(ap->a_nch);
1459                 cache_setvp(ap->a_nch, *ap->a_vpp);
1460         }
1461         LOCKSTOP;
1462         return error;
1463 }
1464
1465 /*
1466  * Make a device node (typically a fifo)
1467  */
1468 static
1469 int
1470 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1471 {
1472         hammer2_inode_t *dip;
1473         hammer2_inode_t *nip;
1474         struct namecache *ncp;
1475         const uint8_t *name;
1476         size_t name_len;
1477         int error;
1478
1479         LOCKSTART;
1480         dip = VTOI(ap->a_dvp);
1481         if (dip->pmp->ronly) {
1482                 LOCKSTOP;
1483                 return (EROFS);
1484         }
1485
1486         ncp = ap->a_nch->ncp;
1487         name = ncp->nc_name;
1488         name_len = ncp->nc_nlen;
1489         hammer2_pfs_memory_wait(dip->pmp);
1490         hammer2_trans_init(dip->pmp, 0);
1491
1492         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1493                                    name, name_len, 0,
1494                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1495                                    0, &error);
1496         if (error) {
1497                 KKASSERT(nip == NULL);
1498                 *ap->a_vpp = NULL;
1499         } else {
1500                 *ap->a_vpp = hammer2_igetv(nip, &error);
1501                 hammer2_inode_unlock(nip);
1502         }
1503         hammer2_trans_done(dip->pmp);
1504
1505         if (error == 0) {
1506                 cache_setunresolved(ap->a_nch);
1507                 cache_setvp(ap->a_nch, *ap->a_vpp);
1508         }
1509         LOCKSTOP;
1510         return error;
1511 }
1512
1513 /*
1514  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1515  */
1516 static
1517 int
1518 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1519 {
1520         hammer2_inode_t *dip;
1521         hammer2_inode_t *nip;
1522         struct namecache *ncp;
1523         const uint8_t *name;
1524         size_t name_len;
1525         int error;
1526
1527         dip = VTOI(ap->a_dvp);
1528         if (dip->pmp->ronly)
1529                 return (EROFS);
1530
1531         ncp = ap->a_nch->ncp;
1532         name = ncp->nc_name;
1533         name_len = ncp->nc_nlen;
1534         hammer2_pfs_memory_wait(dip->pmp);
1535         hammer2_trans_init(dip->pmp, 0);
1536
1537         ap->a_vap->va_type = VLNK;      /* enforce type */
1538
1539         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1540                                    name, name_len, 0,
1541                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1542                                    0, &error);
1543         if (error) {
1544                 KKASSERT(nip == NULL);
1545                 *ap->a_vpp = NULL;
1546                 hammer2_trans_done(dip->pmp);
1547                 return error;
1548         }
1549         *ap->a_vpp = hammer2_igetv(nip, &error);
1550
1551         /*
1552          * Build the softlink (~like file data) and finalize the namecache.
1553          */
1554         if (error == 0) {
1555                 size_t bytes;
1556                 struct uio auio;
1557                 struct iovec aiov;
1558
1559                 bytes = strlen(ap->a_target);
1560
1561                 hammer2_inode_unlock(nip);
1562                 bzero(&auio, sizeof(auio));
1563                 bzero(&aiov, sizeof(aiov));
1564                 auio.uio_iov = &aiov;
1565                 auio.uio_segflg = UIO_SYSSPACE;
1566                 auio.uio_rw = UIO_WRITE;
1567                 auio.uio_resid = bytes;
1568                 auio.uio_iovcnt = 1;
1569                 auio.uio_td = curthread;
1570                 aiov.iov_base = ap->a_target;
1571                 aiov.iov_len = bytes;
1572                 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1573                 /* XXX handle error */
1574                 error = 0;
1575         } else {
1576                 hammer2_inode_unlock(nip);
1577         }
1578         hammer2_trans_done(dip->pmp);
1579
1580         /*
1581          * Finalize namecache
1582          */
1583         if (error == 0) {
1584                 cache_setunresolved(ap->a_nch);
1585                 cache_setvp(ap->a_nch, *ap->a_vpp);
1586                 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1587         }
1588         return error;
1589 }
1590
1591 /*
1592  * hammer2_vop_nremove { nch, dvp, cred }
1593  */
1594 static
1595 int
1596 hammer2_vop_nremove(struct vop_nremove_args *ap)
1597 {
1598         hammer2_xop_unlink_t *xop;
1599         hammer2_inode_t *dip;
1600         hammer2_inode_t *ip;
1601         struct namecache *ncp;
1602         int error;
1603         int isopen;
1604
1605         LOCKSTART;
1606         dip = VTOI(ap->a_dvp);
1607         if (dip->pmp->ronly) {
1608                 LOCKSTOP;
1609                 return(EROFS);
1610         }
1611
1612         ncp = ap->a_nch->ncp;
1613
1614         hammer2_pfs_memory_wait(dip->pmp);
1615         hammer2_trans_init(dip->pmp, 0);
1616         hammer2_inode_lock(dip, 0);
1617
1618         /*
1619          * The unlink XOP unlinks the path from the directory and
1620          * locates and returns the cluster associated with the real inode.
1621          * We have to handle nlinks here on the frontend.
1622          */
1623         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1624         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1625         isopen = cache_isopen(ap->a_nch);
1626         xop->isdir = 0;
1627         xop->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1628         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1629
1630         /*
1631          * Collect the real inode and adjust nlinks, destroy the real
1632          * inode if nlinks transitions to 0 and it was the real inode
1633          * (else it has already been removed).
1634          */
1635         error = hammer2_xop_collect(&xop->head, 0);
1636         hammer2_inode_unlock(dip);
1637
1638         if (error == 0) {
1639                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1640                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1641                 if (ip) {
1642                         hammer2_inode_unlink_finisher(ip, isopen);
1643                         hammer2_inode_unlock(ip);
1644                 }
1645         } else {
1646                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1647         }
1648
1649         hammer2_inode_run_sideq(dip->pmp);
1650         hammer2_trans_done(dip->pmp);
1651         if (error == 0)
1652                 cache_unlink(ap->a_nch);
1653         LOCKSTOP;
1654         return (error);
1655 }
1656
1657 /*
1658  * hammer2_vop_nrmdir { nch, dvp, cred }
1659  */
1660 static
1661 int
1662 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1663 {
1664         hammer2_xop_unlink_t *xop;
1665         hammer2_inode_t *dip;
1666         hammer2_inode_t *ip;
1667         struct namecache *ncp;
1668         int isopen;
1669         int error;
1670
1671         LOCKSTART;
1672         dip = VTOI(ap->a_dvp);
1673         if (dip->pmp->ronly) {
1674                 LOCKSTOP;
1675                 return(EROFS);
1676         }
1677
1678         hammer2_pfs_memory_wait(dip->pmp);
1679         hammer2_trans_init(dip->pmp, 0);
1680         hammer2_inode_lock(dip, 0);
1681
1682         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1683
1684         ncp = ap->a_nch->ncp;
1685         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1686         isopen = cache_isopen(ap->a_nch);
1687         xop->isdir = 1;
1688         xop->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1689         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1690
1691         /*
1692          * Collect the real inode and adjust nlinks, destroy the real
1693          * inode if nlinks transitions to 0 and it was the real inode
1694          * (else it has already been removed).
1695          */
1696         error = hammer2_xop_collect(&xop->head, 0);
1697         hammer2_inode_unlock(dip);
1698
1699         if (error == 0) {
1700                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1701                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1702                 if (ip) {
1703                         hammer2_inode_unlink_finisher(ip, isopen);
1704                         hammer2_inode_unlock(ip);
1705                 }
1706         } else {
1707                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1708         }
1709         hammer2_inode_run_sideq(dip->pmp);
1710         hammer2_trans_done(dip->pmp);
1711         if (error == 0)
1712                 cache_unlink(ap->a_nch);
1713         LOCKSTOP;
1714         return (error);
1715 }
1716
1717 /*
1718  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1719  */
1720 static
1721 int
1722 hammer2_vop_nrename(struct vop_nrename_args *ap)
1723 {
1724         struct namecache *fncp;
1725         struct namecache *tncp;
1726         hammer2_inode_t *cdip;
1727         hammer2_inode_t *fdip;
1728         hammer2_inode_t *tdip;
1729         hammer2_inode_t *ip;
1730         const uint8_t *fname;
1731         size_t fname_len;
1732         const uint8_t *tname;
1733         size_t tname_len;
1734         int error;
1735         int tnch_error;
1736         int nlink_locked;
1737         hammer2_key_t tlhc;
1738
1739         if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1740                 return(EXDEV);
1741         if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1742                 return(EXDEV);
1743
1744         fdip = VTOI(ap->a_fdvp);        /* source directory */
1745         tdip = VTOI(ap->a_tdvp);        /* target directory */
1746
1747         if (fdip->pmp->ronly)
1748                 return(EROFS);
1749
1750         LOCKSTART;
1751         fncp = ap->a_fnch->ncp;         /* entry name in source */
1752         fname = fncp->nc_name;
1753         fname_len = fncp->nc_nlen;
1754
1755         tncp = ap->a_tnch->ncp;         /* entry name in target */
1756         tname = tncp->nc_name;
1757         tname_len = tncp->nc_nlen;
1758
1759         hammer2_pfs_memory_wait(tdip->pmp);
1760         hammer2_trans_init(tdip->pmp, 0);
1761
1762         /*
1763          * ip is the inode being renamed.  If this is a hardlink then
1764          * ip represents the actual file and not the hardlink marker.
1765          */
1766         ip = VTOI(fncp->nc_vp);
1767
1768         /*
1769          * The common parent directory must be locked first to avoid deadlocks.
1770          * Also note that fdip and/or tdip might match cdip.
1771          *
1772          * WARNING!  The kernel's namecache locks are insufficient for
1773          *           protecting us from hardlink shifts, since unrelated
1774          *           rename() or link() calls on parent directories might
1775          *           cause a shift.  A PFS-wide lock is required for this
1776          *           situation.
1777          */
1778         if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY ||
1779             (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
1780                 lockmgr(&ip->pmp->lock_nlink, LK_EXCLUSIVE);
1781                 nlink_locked = 1;
1782         } else {
1783                 nlink_locked = 0;
1784         }
1785
1786         cdip = hammer2_inode_common_parent(ip->pip, tdip);
1787         hammer2_inode_lock(cdip, 0);
1788         hammer2_inode_lock(fdip, 0);
1789         hammer2_inode_lock(tdip, 0);
1790         hammer2_inode_ref(ip);          /* extra ref */
1791         error = 0;
1792
1793         /*
1794          * If ip is a hardlink target and fdip != cdip we must shift the
1795          * inode to cdip.
1796          */
1797         hammer2_inode_lock(ip, 0);
1798         if (fdip != cdip &&
1799             (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
1800                 hammer2_xop_nlink_t *xop1;
1801
1802                 xop1 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1803                 hammer2_xop_setip2(&xop1->head, ip);
1804                 hammer2_xop_setip3(&xop1->head, cdip);
1805                 xop1->nlinks_delta = 0;
1806
1807                 hammer2_xop_start(&xop1->head, hammer2_xop_nlink);
1808                 error = hammer2_xop_collect(&xop1->head, 0);
1809                 hammer2_xop_retire(&xop1->head, HAMMER2_XOPMASK_VOP);
1810         }
1811         /* hammer2_inode_unlock(ip); */
1812
1813         /*
1814          * Delete the target namespace.
1815          */
1816         {
1817                 hammer2_xop_unlink_t *xop2;
1818                 hammer2_inode_t *tip;
1819                 int isopen;
1820
1821                 /*
1822                  * The unlink XOP unlinks the path from the directory and
1823                  * locates and returns the cluster associated with the real
1824                  * inode.  We have to handle nlinks here on the frontend.
1825                  */
1826                 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1827                 hammer2_xop_setname(&xop2->head, tname, tname_len);
1828                 isopen = cache_isopen(ap->a_tnch);
1829                 xop2->isdir = -1;
1830                 xop2->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1831                 hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
1832
1833                 /*
1834                  * Collect the real inode and adjust nlinks, destroy the real
1835                  * inode if nlinks transitions to 0 and it was the real inode
1836                  * (else it has already been removed).
1837                  */
1838                 tnch_error = hammer2_xop_collect(&xop2->head, 0);
1839                 /* hammer2_inode_unlock(tdip); */
1840
1841                 if (tnch_error == 0) {
1842                         tip = hammer2_inode_get(tdip->pmp, NULL,
1843                                                 &xop2->head.cluster, -1);
1844                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1845                         if (tip) {
1846                                 hammer2_inode_unlink_finisher(tip, isopen);
1847                                 hammer2_inode_unlock(tip);
1848                         }
1849                 } else {
1850                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1851                 }
1852                 /* hammer2_inode_lock(tdip, 0); */
1853
1854                 if (tnch_error && tnch_error != ENOENT) {
1855                         error = tnch_error;
1856                         goto done2;
1857                 }
1858         }
1859
1860         /*
1861          * Resolve the collision space for (tdip, tname, tname_len)
1862          *
1863          * tdip must be held exclusively locked to prevent races.
1864          */
1865         {
1866                 hammer2_xop_scanlhc_t *sxop;
1867                 hammer2_tid_t lhcbase;
1868
1869                 tlhc = hammer2_dirhash(tname, tname_len);
1870                 lhcbase = tlhc;
1871                 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1872                 sxop->lhc = tlhc;
1873                 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
1874                 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
1875                         if (tlhc != sxop->head.cluster.focus->bref.key)
1876                                 break;
1877                         ++tlhc;
1878                 }
1879                 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
1880
1881                 if (error) {
1882                         if (error != ENOENT)
1883                                 goto done2;
1884                         ++tlhc;
1885                         error = 0;
1886                 }
1887                 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
1888                         error = ENOSPC;
1889                         goto done2;
1890                 }
1891         }
1892
1893         /*
1894          * Everything is setup, do the rename.
1895          *
1896          * We have to synchronize ip->meta to the underlying operation.
1897          *
1898          * NOTE: To avoid deadlocks we cannot lock (ip) while we are
1899          *       unlinking elements from their directories.  Locking
1900          *       the nlinks field does not lock the whole inode.
1901          */
1902         /* hammer2_inode_lock(ip, 0); */
1903         if (error == 0) {
1904                 hammer2_xop_nrename_t *xop4;
1905
1906                 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1907                 xop4->lhc = tlhc;
1908                 xop4->ip_key = ip->meta.name_key;
1909                 hammer2_xop_setip2(&xop4->head, ip);
1910                 hammer2_xop_setip3(&xop4->head, tdip);
1911                 hammer2_xop_setname(&xop4->head, fname, fname_len);
1912                 hammer2_xop_setname2(&xop4->head, tname, tname_len);
1913                 hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
1914
1915                 error = hammer2_xop_collect(&xop4->head, 0);
1916                 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
1917
1918                 if (error == ENOENT)
1919                         error = 0;
1920                 if (error == 0 &&
1921                     (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1922                         hammer2_inode_modify(ip);
1923                         ip->meta.name_len = tname_len;
1924                         ip->meta.name_key = tlhc;
1925
1926                 }
1927         }
1928
1929         /*
1930          * Fixup ip->pip if we were renaming the actual file and not a
1931          * hardlink pointer.
1932          */
1933         if (error == 0 && (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1934                 hammer2_inode_t *opip;
1935
1936                 if (ip->pip != tdip) {
1937                         hammer2_inode_ref(tdip);
1938                         opip = ip->pip;
1939                         ip->pip = tdip;
1940                         if (opip)
1941                                 hammer2_inode_drop(opip);
1942                 }
1943         }
1944 done2:
1945         hammer2_inode_unlock(ip);
1946         hammer2_inode_unlock(tdip);
1947         hammer2_inode_unlock(fdip);
1948         hammer2_inode_unlock(cdip);
1949         hammer2_inode_drop(ip);
1950         hammer2_inode_drop(cdip);
1951         hammer2_inode_run_sideq(fdip->pmp);
1952
1953         if (nlink_locked)
1954                 lockmgr(&ip->pmp->lock_nlink, LK_RELEASE);
1955         hammer2_trans_done(tdip->pmp);
1956
1957         /*
1958          * Issue the namecache update after unlocking all the internal
1959          * hammer structures, otherwise we might deadlock.
1960          */
1961         if (tnch_error == 0) {
1962                 cache_unlink(ap->a_tnch);
1963                 cache_setunresolved(ap->a_tnch);
1964         }
1965         if (error == 0)
1966                 cache_rename(ap->a_fnch, ap->a_tnch);
1967
1968         LOCKSTOP;
1969         return (error);
1970 }
1971
1972 /*
1973  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
1974  */
1975 static
1976 int
1977 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
1978 {
1979         hammer2_inode_t *ip;
1980         int error;
1981
1982         LOCKSTART;
1983         ip = VTOI(ap->a_vp);
1984
1985         error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
1986                               ap->a_fflag, ap->a_cred);
1987         LOCKSTOP;
1988         return (error);
1989 }
1990
1991 static
1992 int
1993 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
1994 {
1995         struct mount *mp;
1996         hammer2_pfs_t *pmp;
1997         int rc;
1998
1999         LOCKSTART;
2000         switch (ap->a_op) {
2001         case (MOUNTCTL_SET_EXPORT):
2002                 mp = ap->a_head.a_ops->head.vv_mount;
2003                 pmp = MPTOPMP(mp);
2004
2005                 if (ap->a_ctllen != sizeof(struct export_args))
2006                         rc = (EINVAL);
2007                 else
2008                         rc = vfs_export(mp, &pmp->export,
2009                                         (const struct export_args *)ap->a_ctl);
2010                 break;
2011         default:
2012                 rc = vop_stdmountctl(ap);
2013                 break;
2014         }
2015         LOCKSTOP;
2016         return (rc);
2017 }
2018
2019 /*
2020  * KQFILTER
2021  */
2022 static void filt_hammer2detach(struct knote *kn);
2023 static int filt_hammer2read(struct knote *kn, long hint);
2024 static int filt_hammer2write(struct knote *kn, long hint);
2025 static int filt_hammer2vnode(struct knote *kn, long hint);
2026
2027 static struct filterops hammer2read_filtops =
2028         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2029           NULL, filt_hammer2detach, filt_hammer2read };
2030 static struct filterops hammer2write_filtops =
2031         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2032           NULL, filt_hammer2detach, filt_hammer2write };
2033 static struct filterops hammer2vnode_filtops =
2034         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2035           NULL, filt_hammer2detach, filt_hammer2vnode };
2036
2037 static
2038 int
2039 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2040 {
2041         struct vnode *vp = ap->a_vp;
2042         struct knote *kn = ap->a_kn;
2043
2044         switch (kn->kn_filter) {
2045         case EVFILT_READ:
2046                 kn->kn_fop = &hammer2read_filtops;
2047                 break;
2048         case EVFILT_WRITE:
2049                 kn->kn_fop = &hammer2write_filtops;
2050                 break;
2051         case EVFILT_VNODE:
2052                 kn->kn_fop = &hammer2vnode_filtops;
2053                 break;
2054         default:
2055                 return (EOPNOTSUPP);
2056         }
2057
2058         kn->kn_hook = (caddr_t)vp;
2059
2060         knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2061
2062         return(0);
2063 }
2064
2065 static void
2066 filt_hammer2detach(struct knote *kn)
2067 {
2068         struct vnode *vp = (void *)kn->kn_hook;
2069
2070         knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2071 }
2072
2073 static int
2074 filt_hammer2read(struct knote *kn, long hint)
2075 {
2076         struct vnode *vp = (void *)kn->kn_hook;
2077         hammer2_inode_t *ip = VTOI(vp);
2078         off_t off;
2079
2080         if (hint == NOTE_REVOKE) {
2081                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2082                 return(1);
2083         }
2084         off = ip->meta.size - kn->kn_fp->f_offset;
2085         kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2086         if (kn->kn_sfflags & NOTE_OLDAPI)
2087                 return(1);
2088         return (kn->kn_data != 0);
2089 }
2090
2091
2092 static int
2093 filt_hammer2write(struct knote *kn, long hint)
2094 {
2095         if (hint == NOTE_REVOKE)
2096                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2097         kn->kn_data = 0;
2098         return (1);
2099 }
2100
2101 static int
2102 filt_hammer2vnode(struct knote *kn, long hint)
2103 {
2104         if (kn->kn_sfflags & hint)
2105                 kn->kn_fflags |= hint;
2106         if (hint == NOTE_REVOKE) {
2107                 kn->kn_flags |= (EV_EOF | EV_NODATA);
2108                 return (1);
2109         }
2110         return (kn->kn_fflags != 0);
2111 }
2112
2113 /*
2114  * FIFO VOPS
2115  */
2116 static
2117 int
2118 hammer2_vop_markatime(struct vop_markatime_args *ap)
2119 {
2120         hammer2_inode_t *ip;
2121         struct vnode *vp;
2122
2123         vp = ap->a_vp;
2124         ip = VTOI(vp);
2125
2126         if (ip->pmp->ronly)
2127                 return(EROFS);
2128         return(0);
2129 }
2130
2131 static
2132 int
2133 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2134 {
2135         int error;
2136
2137         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2138         if (error)
2139                 error = hammer2_vop_kqfilter(ap);
2140         return(error);
2141 }
2142
2143 /*
2144  * VOPS vector
2145  */
2146 struct vop_ops hammer2_vnode_vops = {
2147         .vop_default    = vop_defaultop,
2148         .vop_fsync      = hammer2_vop_fsync,
2149         .vop_getpages   = vop_stdgetpages,
2150         .vop_putpages   = vop_stdputpages,
2151         .vop_access     = hammer2_vop_access,
2152         .vop_advlock    = hammer2_vop_advlock,
2153         .vop_close      = hammer2_vop_close,
2154         .vop_nlink      = hammer2_vop_nlink,
2155         .vop_ncreate    = hammer2_vop_ncreate,
2156         .vop_nsymlink   = hammer2_vop_nsymlink,
2157         .vop_nremove    = hammer2_vop_nremove,
2158         .vop_nrmdir     = hammer2_vop_nrmdir,
2159         .vop_nrename    = hammer2_vop_nrename,
2160         .vop_getattr    = hammer2_vop_getattr,
2161         .vop_setattr    = hammer2_vop_setattr,
2162         .vop_readdir    = hammer2_vop_readdir,
2163         .vop_readlink   = hammer2_vop_readlink,
2164         .vop_getpages   = vop_stdgetpages,
2165         .vop_putpages   = vop_stdputpages,
2166         .vop_read       = hammer2_vop_read,
2167         .vop_write      = hammer2_vop_write,
2168         .vop_open       = hammer2_vop_open,
2169         .vop_inactive   = hammer2_vop_inactive,
2170         .vop_reclaim    = hammer2_vop_reclaim,
2171         .vop_nresolve   = hammer2_vop_nresolve,
2172         .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2173         .vop_nmkdir     = hammer2_vop_nmkdir,
2174         .vop_nmknod     = hammer2_vop_nmknod,
2175         .vop_ioctl      = hammer2_vop_ioctl,
2176         .vop_mountctl   = hammer2_vop_mountctl,
2177         .vop_bmap       = hammer2_vop_bmap,
2178         .vop_strategy   = hammer2_vop_strategy,
2179         .vop_kqfilter   = hammer2_vop_kqfilter
2180 };
2181
2182 struct vop_ops hammer2_spec_vops = {
2183         .vop_default =          vop_defaultop,
2184         .vop_fsync =            hammer2_vop_fsync,
2185         .vop_read =             vop_stdnoread,
2186         .vop_write =            vop_stdnowrite,
2187         .vop_access =           hammer2_vop_access,
2188         .vop_close =            hammer2_vop_close,
2189         .vop_markatime =        hammer2_vop_markatime,
2190         .vop_getattr =          hammer2_vop_getattr,
2191         .vop_inactive =         hammer2_vop_inactive,
2192         .vop_reclaim =          hammer2_vop_reclaim,
2193         .vop_setattr =          hammer2_vop_setattr
2194 };
2195
2196 struct vop_ops hammer2_fifo_vops = {
2197         .vop_default =          fifo_vnoperate,
2198         .vop_fsync =            hammer2_vop_fsync,
2199 #if 0
2200         .vop_read =             hammer2_vop_fiforead,
2201         .vop_write =            hammer2_vop_fifowrite,
2202 #endif
2203         .vop_access =           hammer2_vop_access,
2204 #if 0
2205         .vop_close =            hammer2_vop_fifoclose,
2206 #endif
2207         .vop_markatime =        hammer2_vop_markatime,
2208         .vop_getattr =          hammer2_vop_getattr,
2209         .vop_inactive =         hammer2_vop_inactive,
2210         .vop_reclaim =          hammer2_vop_reclaim,
2211         .vop_setattr =          hammer2_vop_setattr,
2212         .vop_kqfilter =         hammer2_vop_fifokqfilter
2213 };
2214