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          * This occurs if the inode was unlinked while open.  Reclamation of
 171          * these inodes requires processing we cannot safely do here so add
 172          * the inode to the sideq in that situation.
 173          *
 174          * A modified inode may require chain synchronization which will no
 175          * longer be driven by a sync or fsync without the vnode, also use
 176          * the sideq for that.
 177          *
 178          * A reclaim can occur at any time so we cannot safely start a
 179          * transaction to handle reclamation of unlinked files.  Instead,
 180          * the ip is left with a reference and placed on a linked list and
 181          * handled later on.
 182          */
 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 <
 322                                     chain->bref.embed.stats.data_count) {
 323                                         vap->va_bytes =
 324                                             chain->bref.embed.stats.data_count;
 325                                 }
 326                         }
 327                 }
 328         }
 329         vap->va_type = hammer2_get_vtype(ip->meta.type);
 330         vap->va_filerev = 0;
 331         vap->va_uid_uuid = ip->meta.uid;
 332         vap->va_gid_uuid = ip->meta.gid;
 333         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 334                           VA_FSID_UUID_VALID;
 335
 336         hammer2_inode_unlock(ip);
 337
 338         LOCKSTOP;
 339         return (0);
 340 }
 341
 342 static
 343 int
 344 hammer2_vop_setattr(struct vop_setattr_args *ap)
 345 {
 346         hammer2_inode_t *ip;
 347         struct vnode *vp;
 348         struct vattr *vap;
 349         int error;
 350         int kflags = 0;
 351         uint64_t ctime;
 352
 353         LOCKSTART;
 354         vp = ap->a_vp;
 355         vap = ap->a_vap;
 356         hammer2_update_time(&ctime);
 357
 358         ip = VTOI(vp);
 359
 360         if (ip->pmp->ronly) {
 361                 LOCKSTOP;
 362                 return(EROFS);
 363         }
 364
 365         hammer2_pfs_memory_wait(ip->pmp);
 366         hammer2_trans_init(ip->pmp, 0);
 367         hammer2_inode_lock(ip, 0);
 368         error = 0;
 369
 370         if (vap->va_flags != VNOVAL) {
 371                 uint32_t flags;
 372
 373                 flags = ip->meta.uflags;
 374                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
 375                                      hammer2_to_unix_xid(&ip->meta.uid),
 376                                      ap->a_cred);
 377                 if (error == 0) {
 378                         if (ip->meta.uflags != flags) {
 379                                 hammer2_inode_modify(ip);
 380                                 ip->meta.uflags = flags;
 381                                 ip->meta.ctime = ctime;
 382                                 kflags |= NOTE_ATTRIB;
 383                         }
 384                         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 385                                 error = 0;
 386                                 goto done;
 387                         }
 388                 }
 389                 goto done;
 390         }
 391         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 392                 error = EPERM;
 393                 goto done;
 394         }
 395         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
 396                 mode_t cur_mode = ip->meta.mode;
 397                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 398                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 399                 uuid_t uuid_uid;
 400                 uuid_t uuid_gid;
 401
 402                 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
 403                                          ap->a_cred,
 404                                          &cur_uid, &cur_gid, &cur_mode);
 405                 if (error == 0) {
 406                         hammer2_guid_to_uuid(&uuid_uid, cur_uid);
 407                         hammer2_guid_to_uuid(&uuid_gid, cur_gid);
 408                         if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
 409                             bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
 410                             ip->meta.mode != cur_mode
 411                         ) {
 412                                 hammer2_inode_modify(ip);
 413                                 ip->meta.uid = uuid_uid;
 414                                 ip->meta.gid = uuid_gid;
 415                                 ip->meta.mode = cur_mode;
 416                                 ip->meta.ctime = ctime;
 417                         }
 418                         kflags |= NOTE_ATTRIB;
 419                 }
 420         }
 421
 422         /*
 423          * Resize the file
 424          */
 425         if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
 426                 switch(vp->v_type) {
 427                 case VREG:
 428                         if (vap->va_size == ip->meta.size)
 429                                 break;
 430                         if (vap->va_size < ip->meta.size) {
 431                                 hammer2_mtx_ex(&ip->truncate_lock);
 432                                 hammer2_truncate_file(ip, vap->va_size);
 433                                 hammer2_mtx_unlock(&ip->truncate_lock);
 434                         } else {
 435                                 hammer2_extend_file(ip, vap->va_size);
 436                         }
 437                         hammer2_inode_modify(ip);
 438                         ip->meta.mtime = ctime;
 439                         break;
 440                 default:
 441                         error = EINVAL;
 442                         goto done;
 443                 }
 444         }
 445 #if 0
 446         /* atime not supported */
 447         if (vap->va_atime.tv_sec != VNOVAL) {
 448                 hammer2_inode_modify(ip);
 449                 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
 450                 kflags |= NOTE_ATTRIB;
 451         }
 452 #endif
 453         if (vap->va_mode != (mode_t)VNOVAL) {
 454                 mode_t cur_mode = ip->meta.mode;
 455                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 456                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 457
 458                 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
 459                                          cur_uid, cur_gid, &cur_mode);
 460                 if (error == 0 && ip->meta.mode != cur_mode) {
 461                         hammer2_inode_modify(ip);
 462                         ip->meta.mode = cur_mode;
 463                         ip->meta.ctime = ctime;
 464                         kflags |= NOTE_ATTRIB;
 465                 }
 466         }
 467
 468         if (vap->va_mtime.tv_sec != VNOVAL) {
 469                 hammer2_inode_modify(ip);
 470                 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
 471                 kflags |= NOTE_ATTRIB;
 472         }
 473
 474 done:
 475         /*
 476          * If a truncation occurred we must call inode_fsync() now in order
 477          * to trim the related data chains, otherwise a later expansion can
 478          * cause havoc.
 479          *
 480          * If an extend occured that changed the DIRECTDATA state, we must
 481          * call inode_fsync now in order to prepare the inode's indirect
 482          * block table.
 483          */
 484         if (ip->flags & HAMMER2_INODE_RESIZED)
 485                 hammer2_inode_chain_sync(ip);
 486
 487         /*
 488          * Cleanup.
 489          */
 490         hammer2_inode_unlock(ip);
 491         hammer2_trans_done(ip->pmp);
 492         hammer2_knote(ip->vp, kflags);
 493
 494         LOCKSTOP;
 495         return (error);
 496 }
 497
 498 static
 499 int
 500 hammer2_vop_readdir(struct vop_readdir_args *ap)
 501 {
 502         hammer2_xop_readdir_t *xop;
 503         hammer2_blockref_t bref;
 504         hammer2_inode_t *ip;
 505         hammer2_tid_t inum;
 506         hammer2_key_t lkey;
 507         struct uio *uio;
 508         off_t *cookies;
 509         off_t saveoff;
 510         int cookie_index;
 511         int ncookies;
 512         int error;
 513         int eofflag;
 514         int dtype;
 515         int r;
 516
 517         LOCKSTART;
 518         ip = VTOI(ap->a_vp);
 519         uio = ap->a_uio;
 520         saveoff = uio->uio_offset;
 521         eofflag = 0;
 522         error = 0;
 523
 524         /*
 525          * Setup cookies directory entry cookies if requested
 526          */
 527         if (ap->a_ncookies) {
 528                 ncookies = uio->uio_resid / 16 + 1;
 529                 if (ncookies > 1024)
 530                         ncookies = 1024;
 531                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 532         } else {
 533                 ncookies = -1;
 534                 cookies = NULL;
 535         }
 536         cookie_index = 0;
 537
 538         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 539
 540         /*
 541          * Handle artificial entries.  To ensure that only positive 64 bit
 542          * quantities are returned to userland we always strip off bit 63.
 543          * The hash code is designed such that codes 0x0000-0x7FFF are not
 544          * used, allowing us to use these codes for articial entries.
 545          *
 546          * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
 547          * allow '..' to cross the mount point into (e.g.) the super-root.
 548          */
 549         if (saveoff == 0) {
 550                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 551                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
 552                 if (r)
 553                         goto done;
 554                 if (cookies)
 555                         cookies[cookie_index] = saveoff;
 556                 ++saveoff;
 557                 ++cookie_index;
 558                 if (cookie_index == ncookies)
 559                         goto done;
 560         }
 561
 562         if (saveoff == 1) {
 563                 /*
 564                  * Be careful with lockorder when accessing ".."
 565                  *
 566                  * (ip is the current dir. xip is the parent dir).
 567                  */
 568                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 569                 if (ip != ip->pmp->iroot)
 570                         inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK;
 571                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
 572                 if (r)
 573                         goto done;
 574                 if (cookies)
 575                         cookies[cookie_index] = saveoff;
 576                 ++saveoff;
 577                 ++cookie_index;
 578                 if (cookie_index == ncookies)
 579                         goto done;
 580         }
 581
 582         lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
 583         if (hammer2_debug & 0x0020)
 584                 kprintf("readdir: lkey %016jx\n", lkey);
 585         if (error)
 586                 goto done;
 587
 588         /*
 589          * Use XOP for cluster scan.
 590          *
 591          * parent is the inode cluster, already locked for us.  Don't
 592          * double lock shared locks as this will screw up upgrades.
 593          */
 594         xop = hammer2_xop_alloc(ip, 0);
 595         xop->lkey = lkey;
 596         hammer2_xop_start(&xop->head, hammer2_xop_readdir);
 597
 598         for (;;) {
 599                 const hammer2_inode_data_t *ripdata;
 600
 601                 error = hammer2_xop_collect(&xop->head, 0);
 602                 if (error)
 603                         break;
 604                 if (cookie_index == ncookies)
 605                         break;
 606                 if (hammer2_debug & 0x0020)
 607                 kprintf("cluster chain %p %p\n",
 608                         xop->head.cluster.focus,
 609                         (xop->head.cluster.focus ?
 610                          xop->head.cluster.focus->data : (void *)-1));
 611                 ripdata = &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
 612                 hammer2_cluster_bref(&xop->head.cluster, &bref);
 613                 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
 614                         dtype = hammer2_get_dtype(ripdata);
 615                         saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 616                         r = vop_write_dirent(&error, uio,
 617                                              ripdata->meta.inum &
 618                                               HAMMER2_DIRHASH_USERMSK,
 619                                              dtype,
 620                                              ripdata->meta.name_len,
 621                                              ripdata->filename);
 622                         if (r)
 623                                 break;
 624                         if (cookies)
 625                                 cookies[cookie_index] = saveoff;
 626                         ++cookie_index;
 627                 } else {
 628                         /* XXX chain error */
 629                         kprintf("bad chain type readdir %d\n", bref.type);
 630                 }
 631         }
 632         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
 633         if (error == ENOENT) {
 634                 error = 0;
 635                 eofflag = 1;
 636                 saveoff = (hammer2_key_t)-1;
 637         } else {
 638                 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 639         }
 640 done:
 641         hammer2_inode_unlock(ip);
 642         if (ap->a_eofflag)
 643                 *ap->a_eofflag = eofflag;
 644         if (hammer2_debug & 0x0020)
 645                 kprintf("readdir: done at %016jx\n", saveoff);
 646         uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
 647         if (error && cookie_index == 0) {
 648                 if (cookies) {
 649                         kfree(cookies, M_TEMP);
 650                         *ap->a_ncookies = 0;
 651                         *ap->a_cookies = NULL;
 652                 }
 653         } else {
 654                 if (cookies) {
 655                         *ap->a_ncookies = cookie_index;
 656                         *ap->a_cookies = cookies;
 657                 }
 658         }
 659         LOCKSTOP;
 660         return (error);
 661 }
 662
 663 /*
 664  * hammer2_vop_readlink { vp, uio, cred }
 665  */
 666 static
 667 int
 668 hammer2_vop_readlink(struct vop_readlink_args *ap)
 669 {
 670         struct vnode *vp;
 671         hammer2_inode_t *ip;
 672         int error;
 673
 674         vp = ap->a_vp;
 675         if (vp->v_type != VLNK)
 676                 return (EINVAL);
 677         ip = VTOI(vp);
 678
 679         error = hammer2_read_file(ip, ap->a_uio, 0);
 680         return (error);
 681 }
 682
 683 static
 684 int
 685 hammer2_vop_read(struct vop_read_args *ap)
 686 {
 687         struct vnode *vp;
 688         hammer2_inode_t *ip;
 689         struct uio *uio;
 690         int error;
 691         int seqcount;
 692         int bigread;
 693
 694         /*
 695          * Read operations supported on this vnode?
 696          */
 697         vp = ap->a_vp;
 698         if (vp->v_type != VREG)
 699                 return (EINVAL);
 700
 701         /*
 702          * Misc
 703          */
 704         ip = VTOI(vp);
 705         uio = ap->a_uio;
 706         error = 0;
 707
 708         seqcount = ap->a_ioflag >> 16;
 709         bigread = (uio->uio_resid > 100 * 1024 * 1024);
 710
 711         error = hammer2_read_file(ip, uio, seqcount);
 712         return (error);
 713 }
 714
 715 static
 716 int
 717 hammer2_vop_write(struct vop_write_args *ap)
 718 {
 719         hammer2_inode_t *ip;
 720         thread_t td;
 721         struct vnode *vp;
 722         struct uio *uio;
 723         int error;
 724         int seqcount;
 725
 726         /*
 727          * Read operations supported on this vnode?
 728          */
 729         vp = ap->a_vp;
 730         if (vp->v_type != VREG)
 731                 return (EINVAL);
 732
 733         /*
 734          * Misc
 735          */
 736         ip = VTOI(vp);
 737         uio = ap->a_uio;
 738         error = 0;
 739         if (ip->pmp->ronly) {
 740                 return (EROFS);
 741         }
 742
 743         seqcount = ap->a_ioflag >> 16;
 744
 745         /*
 746          * Check resource limit
 747          */
 748         if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
 749             uio->uio_offset + uio->uio_resid >
 750              td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 751                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 752                 return (EFBIG);
 753         }
 754
 755         /*
 756          * The transaction interlocks against flush initiations
 757          * (note: but will run concurrently with the actual flush).
 758          *
 759          * To avoid deadlocking against the VM system, we must flag any
 760          * transaction related to the buffer cache or other direct
 761          * VM page manipulation.
 762          */
 763         if (uio->uio_segflg == UIO_NOCOPY)
 764                 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_BUFCACHE);
 765         else
 766                 hammer2_trans_init(ip->pmp, 0);
 767         error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
 768         hammer2_trans_done(ip->pmp);
 769
 770         return (error);
 771 }
 772
 773 /*
 774  * Perform read operations on a file or symlink given an UNLOCKED
 775  * inode and uio.
 776  *
 777  * The passed ip is not locked.
 778  */
 779 static
 780 int
 781 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
 782 {
 783         hammer2_off_t size;
 784         struct buf *bp;
 785         int error;
 786
 787         error = 0;
 788
 789         /*
 790          * UIO read loop.
 791          *
 792          * WARNING! Assumes that the kernel interlocks size changes at the
 793          *          vnode level.
 794          */
 795         hammer2_mtx_sh(&ip->lock);
 796         hammer2_mtx_sh(&ip->truncate_lock);
 797         size = ip->meta.size;
 798         hammer2_mtx_unlock(&ip->lock);
 799
 800         while (uio->uio_resid > 0 && uio->uio_offset < size) {
 801                 hammer2_key_t lbase;
 802                 hammer2_key_t leof;
 803                 int lblksize;
 804                 int loff;
 805                 int n;
 806
 807                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 808                                                 &lbase, &leof);
 809
 810 #if 1
 811                 error = cluster_read(ip->vp, leof, lbase, lblksize,
 812                                      uio->uio_resid, seqcount * MAXBSIZE,
 813                                      &bp);
 814 #else
 815                 if (uio->uio_segflg == UIO_NOCOPY) {
 816                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 817                         if (bp->b_flags & B_CACHE) {
 818                                 int i;
 819                                 int j = 0;
 820                                 if (bp->b_xio.xio_npages != 16)
 821                                         kprintf("NPAGES BAD\n");
 822                                 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
 823                                         vm_page_t m;
 824                                         m = bp->b_xio.xio_pages[i];
 825                                         if (m == NULL || m->valid == 0) {
 826                                                 kprintf("bp %016jx %016jx pg %d inv",
 827                                                         lbase, leof, i);
 828                                                 if (m)
 829                                                         kprintf("m->object %p/%p", m->object, ip->vp->v_object);
 830                                                 kprintf("\n");
 831                                                 j = 1;
 832                                         }
 833                                 }
 834                                 if (j)
 835                                         kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error);
 836                         }
 837                         bqrelse(bp);
 838                 }
 839                 error = bread(ip->vp, lbase, lblksize, &bp);
 840 #endif
 841                 if (error) {
 842                         brelse(bp);
 843                         break;
 844                 }
 845                 loff = (int)(uio->uio_offset - lbase);
 846                 n = lblksize - loff;
 847                 if (n > uio->uio_resid)
 848                         n = uio->uio_resid;
 849                 if (n > size - uio->uio_offset)
 850                         n = (int)(size - uio->uio_offset);
 851                 bp->b_flags |= B_AGE;
 852                 uiomovebp(bp, (char *)bp->b_data + loff, n, uio);
 853                 bqrelse(bp);
 854         }
 855         hammer2_mtx_unlock(&ip->truncate_lock);
 856
 857         return (error);
 858 }
 859
 860 /*
 861  * Write to the file represented by the inode via the logical buffer cache.
 862  * The inode may represent a regular file or a symlink.
 863  *
 864  * The inode must not be locked.
 865  */
 866 static
 867 int
 868 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
 869                    int ioflag, int seqcount)
 870 {
 871         hammer2_key_t old_eof;
 872         hammer2_key_t new_eof;
 873         struct buf *bp;
 874         int kflags;
 875         int error;
 876         int modified;
 877
 878         /*
 879          * Setup if append
 880          *
 881          * WARNING! Assumes that the kernel interlocks size changes at the
 882          *          vnode level.
 883          */
 884         hammer2_mtx_ex(&ip->lock);
 885         hammer2_mtx_sh(&ip->truncate_lock);
 886         if (ioflag & IO_APPEND)
 887                 uio->uio_offset = ip->meta.size;
 888         old_eof = ip->meta.size;
 889
 890         /*
 891          * Extend the file if necessary.  If the write fails at some point
 892          * we will truncate it back down to cover as much as we were able
 893          * to write.
 894          *
 895          * Doing this now makes it easier to calculate buffer sizes in
 896          * the loop.
 897          */
 898         kflags = 0;
 899         error = 0;
 900         modified = 0;
 901
 902         if (uio->uio_offset + uio->uio_resid > old_eof) {
 903                 new_eof = uio->uio_offset + uio->uio_resid;
 904                 modified = 1;
 905                 hammer2_extend_file(ip, new_eof);
 906                 kflags |= NOTE_EXTEND;
 907         } else {
 908                 new_eof = old_eof;
 909         }
 910         hammer2_mtx_unlock(&ip->lock);
 911
 912         /*
 913          * UIO write loop
 914          */
 915         while (uio->uio_resid > 0) {
 916                 hammer2_key_t lbase;
 917                 int trivial;
 918                 int endofblk;
 919                 int lblksize;
 920                 int loff;
 921                 int n;
 922
 923                 /*
 924                  * Don't allow the buffer build to blow out the buffer
 925                  * cache.
 926                  */
 927                 if ((ioflag & IO_RECURSE) == 0)
 928                         bwillwrite(HAMMER2_PBUFSIZE);
 929
 930                 /*
 931                  * This nominally tells us how much we can cluster and
 932                  * what the logical buffer size needs to be.  Currently
 933                  * we don't try to cluster the write and just handle one
 934                  * block at a time.
 935                  */
 936                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 937                                                 &lbase, NULL);
 938                 loff = (int)(uio->uio_offset - lbase);
 939
 940                 KKASSERT(lblksize <= 65536);
 941
 942                 /*
 943                  * Calculate bytes to copy this transfer and whether the
 944                  * copy completely covers the buffer or not.
 945                  */
 946                 trivial = 0;
 947                 n = lblksize - loff;
 948                 if (n > uio->uio_resid) {
 949                         n = uio->uio_resid;
 950                         if (loff == lbase && uio->uio_offset + n == new_eof)
 951                                 trivial = 1;
 952                         endofblk = 0;
 953                 } else {
 954                         if (loff == 0)
 955                                 trivial = 1;
 956                         endofblk = 1;
 957                 }
 958                 if (lbase >= new_eof)
 959                         trivial = 1;
 960
 961                 /*
 962                  * Get the buffer
 963                  */
 964                 if (uio->uio_segflg == UIO_NOCOPY) {
 965                         /*
 966                          * Issuing a write with the same data backing the
 967                          * buffer.  Instantiate the buffer to collect the
 968                          * backing vm pages, then read-in any missing bits.
 969                          *
 970                          * This case is used by vop_stdputpages().
 971                          */
 972                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 973                         if ((bp->b_flags & B_CACHE) == 0) {
 974                                 bqrelse(bp);
 975                                 error = bread(ip->vp, lbase, lblksize, &bp);
 976                         }
 977                 } else if (trivial) {
 978                         /*
 979                          * Even though we are entirely overwriting the buffer
 980                          * we may still have to zero it out to avoid a
 981                          * mmap/write visibility issue.
 982                          */
 983                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 984                         if ((bp->b_flags & B_CACHE) == 0)
 985                                 vfs_bio_clrbuf(bp);
 986                 } else {
 987                         /*
 988                          * Partial overwrite, read in any missing bits then
 989                          * replace the portion being written.
 990                          *
 991                          * (The strategy code will detect zero-fill physical
 992                          * blocks for this case).
 993                          */
 994                         error = bread(ip->vp, lbase, lblksize, &bp);
 995                         if (error == 0)
 996                                 bheavy(bp);
 997                 }
 998
 999                 if (error) {
1000                         brelse(bp);
1001                         break;
1002                 }
1003
1004                 /*
1005                  * Ok, copy the data in
1006                  */
1007                 error = uiomovebp(bp, bp->b_data + loff, n, uio);
1008                 kflags |= NOTE_WRITE;
1009                 modified = 1;
1010                 if (error) {
1011                         brelse(bp);
1012                         break;
1013                 }
1014
1015                 /*
1016                  * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1017                  *          with IO_SYNC or IO_ASYNC set.  These writes
1018                  *          must be handled as the pageout daemon expects.
1019                  *
1020                  * NOTE!    H2 relies on cluster_write() here because it
1021                  *          cannot preallocate disk blocks at the logical
1022                  *          level due to not knowing what the compression
1023                  *          size will be at this time.
1024                  *
1025                  *          We must use cluster_write() here and we depend
1026                  *          on the write-behind feature to flush buffers
1027                  *          appropriately.  If we let the buffer daemons do
1028                  *          it the block allocations will be all over the
1029                  *          map.
1030                  */
1031                 if (ioflag & IO_SYNC) {
1032                         bwrite(bp);
1033                 } else if ((ioflag & IO_DIRECT) && endofblk) {
1034                         bawrite(bp);
1035                 } else if (ioflag & IO_ASYNC) {
1036                         bawrite(bp);
1037                 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) {
1038                         bdwrite(bp);
1039                 } else {
1040 #if 1
1041                         bp->b_flags |= B_CLUSTEROK;
1042                         cluster_write(bp, new_eof, lblksize, seqcount);
1043 #else
1044                         bp->b_flags |= B_CLUSTEROK;
1045                         bdwrite(bp);
1046 #endif
1047                 }
1048         }
1049
1050         /*
1051          * Cleanup.  If we extended the file EOF but failed to write through
1052          * the entire write is a failure and we have to back-up.
1053          */
1054         if (error && new_eof != old_eof) {
1055                 hammer2_mtx_unlock(&ip->truncate_lock);
1056                 hammer2_mtx_ex(&ip->lock);
1057                 hammer2_mtx_ex(&ip->truncate_lock);
1058                 hammer2_truncate_file(ip, old_eof);
1059                 if (ip->flags & HAMMER2_INODE_MODIFIED)
1060                         hammer2_inode_chain_sync(ip);
1061                 hammer2_mtx_unlock(&ip->lock);
1062         } else if (modified) {
1063                 hammer2_mtx_ex(&ip->lock);
1064                 hammer2_inode_modify(ip);
1065                 hammer2_update_time(&ip->meta.mtime);
1066                 if (ip->flags & HAMMER2_INODE_MODIFIED)
1067                         hammer2_inode_chain_sync(ip);
1068                 hammer2_mtx_unlock(&ip->lock);
1069                 hammer2_knote(ip->vp, kflags);
1070         }
1071         hammer2_trans_assert_strategy(ip->pmp);
1072         hammer2_mtx_unlock(&ip->truncate_lock);
1073
1074         return error;
1075 }
1076
1077 /*
1078  * Truncate the size of a file.  The inode must not be locked.
1079  *
1080  * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1081  * ensure that any on-media data beyond the new file EOF has been destroyed.
1082  *
1083  * WARNING: nvtruncbuf() can only be safely called without the inode lock
1084  *          held due to the way our write thread works.  If the truncation
1085  *          occurs in the middle of a buffer, nvtruncbuf() is responsible
1086  *          for dirtying that buffer and zeroing out trailing bytes.
1087  *
1088  * WARNING! Assumes that the kernel interlocks size changes at the
1089  *          vnode level.
1090  *
1091  * WARNING! Caller assumes responsibility for removing dead blocks
1092  *          if INODE_RESIZED is set.
1093  */
1094 static
1095 void
1096 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1097 {
1098         hammer2_key_t lbase;
1099         int nblksize;
1100
1101         LOCKSTART;
1102         hammer2_mtx_unlock(&ip->lock);
1103         if (ip->vp) {
1104                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1105                 nvtruncbuf(ip->vp, nsize,
1106                            nblksize, (int)nsize & (nblksize - 1),
1107                            0);
1108         }
1109         hammer2_mtx_ex(&ip->lock);
1110         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1111         ip->osize = ip->meta.size;
1112         ip->meta.size = nsize;
1113         atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1114         hammer2_inode_modify(ip);
1115         LOCKSTOP;
1116 }
1117
1118 /*
1119  * Extend the size of a file.  The inode must not be locked.
1120  *
1121  * Even though the file size is changing, we do not have to set the
1122  * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1123  * boundary.  When this occurs a hammer2_inode_chain_sync() is required
1124  * to prepare the inode cluster's indirect block table, otherwise
1125  * async execution of the strategy code will implode on us.
1126  *
1127  * WARNING! Assumes that the kernel interlocks size changes at the
1128  *          vnode level.
1129  *
1130  * WARNING! Caller assumes responsibility for transitioning out
1131  *          of the inode DIRECTDATA mode if INODE_RESIZED is set.
1132  */
1133 static
1134 void
1135 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1136 {
1137         hammer2_key_t lbase;
1138         hammer2_key_t osize;
1139         int oblksize;
1140         int nblksize;
1141
1142         LOCKSTART;
1143
1144         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1145         hammer2_inode_modify(ip);
1146         osize = ip->meta.size;
1147         ip->osize = osize;
1148         ip->meta.size = nsize;
1149
1150         if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1151                 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1152                 hammer2_inode_chain_sync(ip);
1153         }
1154
1155         hammer2_mtx_unlock(&ip->lock);
1156         if (ip->vp) {
1157                 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1158                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1159                 nvextendbuf(ip->vp,
1160                             osize, nsize,
1161                             oblksize, nblksize,
1162                             -1, -1, 0);
1163         }
1164         hammer2_mtx_ex(&ip->lock);
1165
1166         LOCKSTOP;
1167 }
1168
1169 static
1170 int
1171 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1172 {
1173         hammer2_xop_nresolve_t *xop;
1174         hammer2_inode_t *ip;
1175         hammer2_inode_t *dip;
1176         struct namecache *ncp;
1177         struct vnode *vp;
1178         int error;
1179
1180         LOCKSTART;
1181         dip = VTOI(ap->a_dvp);
1182         xop = hammer2_xop_alloc(dip, 0);
1183
1184         ncp = ap->a_nch->ncp;
1185         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1186
1187         /*
1188          * Note: In DragonFly the kernel handles '.' and '..'.
1189          */
1190         hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1191         hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1192
1193         error = hammer2_xop_collect(&xop->head, 0);
1194         if (error) {
1195                 ip = NULL;
1196         } else {
1197                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1198         }
1199         hammer2_inode_unlock(dip);
1200
1201         /*
1202          * Acquire the related vnode
1203          *
1204          * NOTE: For error processing, only ENOENT resolves the namecache
1205          *       entry to NULL, otherwise we just return the error and
1206          *       leave the namecache unresolved.
1207          *
1208          * NOTE: multiple hammer2_inode structures can be aliased to the
1209          *       same chain element, for example for hardlinks.  This
1210          *       use case does not 'reattach' inode associations that
1211          *       might already exist, but always allocates a new one.
1212          *
1213          * WARNING: inode structure is locked exclusively via inode_get
1214          *          but chain was locked shared.  inode_unlock()
1215          *          will handle it properly.
1216          */
1217         if (ip) {
1218                 vp = hammer2_igetv(ip, &error);
1219                 if (error == 0) {
1220                         vn_unlock(vp);
1221                         cache_setvp(ap->a_nch, vp);
1222                 } else if (error == ENOENT) {
1223                         cache_setvp(ap->a_nch, NULL);
1224                 }
1225                 hammer2_inode_unlock(ip);
1226
1227                 /*
1228                  * The vp should not be released until after we've disposed
1229                  * of our locks, because it might cause vop_inactive() to
1230                  * be called.
1231                  */
1232                 if (vp)
1233                         vrele(vp);
1234         } else {
1235                 error = ENOENT;
1236                 cache_setvp(ap->a_nch, NULL);
1237         }
1238         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1239         KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1240                 ("resolve error %d/%p ap %p\n",
1241                  error, ap->a_nch->ncp->nc_vp, ap));
1242         LOCKSTOP;
1243
1244         return error;
1245 }
1246
1247 static
1248 int
1249 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1250 {
1251         hammer2_inode_t *dip;
1252         hammer2_tid_t inum;
1253         int error;
1254
1255         LOCKSTART;
1256         dip = VTOI(ap->a_dvp);
1257         inum = dip->meta.iparent;
1258         *ap->a_vpp = NULL;
1259
1260         if (inum) {
1261                 error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL,
1262                                          inum, ap->a_vpp);
1263         } else {
1264                 error = ENOENT;
1265         }
1266         LOCKSTOP;
1267         return error;
1268 }
1269
1270 static
1271 int
1272 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1273 {
1274         hammer2_inode_t *dip;
1275         hammer2_inode_t *nip;
1276         struct namecache *ncp;
1277         const uint8_t *name;
1278         size_t name_len;
1279         hammer2_tid_t inum;
1280         int error;
1281
1282         LOCKSTART;
1283         dip = VTOI(ap->a_dvp);
1284         if (dip->pmp->ronly) {
1285                 LOCKSTOP;
1286                 return (EROFS);
1287         }
1288
1289         ncp = ap->a_nch->ncp;
1290         name = ncp->nc_name;
1291         name_len = ncp->nc_nlen;
1292
1293         hammer2_pfs_memory_wait(dip->pmp);
1294         hammer2_trans_init(dip->pmp, 0);
1295
1296         inum = hammer2_trans_newinum(dip->pmp);
1297
1298         /*
1299          * Create the actual inode as a hidden file in the iroot, then
1300          * create the directory entry as a hardlink to it.  The creation
1301          * of the actual inode sets its nlinks to 1 which is the value
1302          * we desire.
1303          */
1304         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1305                                    NULL, 0, inum,
1306                                    inum, 0, 0,
1307                                    0, &error);
1308         if (error == 0) {
1309                 hammer2_inode_create(dip, dip, NULL, NULL,
1310                                      name, name_len, 0,
1311                                      nip->meta.inum,
1312                                      HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1313                                      0, &error);
1314         }
1315
1316         if (error) {
1317                 KKASSERT(nip == NULL);
1318                 *ap->a_vpp = NULL;
1319         } else {
1320                 *ap->a_vpp = hammer2_igetv(nip, &error);
1321                 hammer2_inode_unlock(nip);
1322         }
1323
1324         /*
1325          * Update dip's mtime
1326          */
1327         if (error == 0) {
1328                 uint64_t mtime;
1329
1330                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1331                 hammer2_update_time(&mtime);
1332                 hammer2_inode_modify(dip);
1333                 dip->meta.mtime = mtime;
1334                 hammer2_inode_unlock(dip);
1335         }
1336
1337         hammer2_trans_done(dip->pmp);
1338
1339         if (error == 0) {
1340                 cache_setunresolved(ap->a_nch);
1341                 cache_setvp(ap->a_nch, *ap->a_vpp);
1342         }
1343         LOCKSTOP;
1344         return error;
1345 }
1346
1347 static
1348 int
1349 hammer2_vop_open(struct vop_open_args *ap)
1350 {
1351         return vop_stdopen(ap);
1352 }
1353
1354 /*
1355  * hammer2_vop_advlock { vp, id, op, fl, flags }
1356  */
1357 static
1358 int
1359 hammer2_vop_advlock(struct vop_advlock_args *ap)
1360 {
1361         hammer2_inode_t *ip = VTOI(ap->a_vp);
1362         hammer2_off_t size;
1363
1364         size = ip->meta.size;
1365         return (lf_advlock(ap, &ip->advlock, size));
1366 }
1367
1368 static
1369 int
1370 hammer2_vop_close(struct vop_close_args *ap)
1371 {
1372         return vop_stdclose(ap);
1373 }
1374
1375 /*
1376  * hammer2_vop_nlink { nch, dvp, vp, cred }
1377  *
1378  * Create a hardlink from (vp) to {dvp, nch}.
1379  */
1380 static
1381 int
1382 hammer2_vop_nlink(struct vop_nlink_args *ap)
1383 {
1384         hammer2_inode_t *tdip;  /* target directory to create link in */
1385         hammer2_inode_t *ip;    /* inode we are hardlinking to */
1386         struct namecache *ncp;
1387         const uint8_t *name;
1388         size_t name_len;
1389         int error;
1390
1391         if (ap->a_dvp->v_mount != ap->a_vp->v_mount)
1392                 return(EXDEV);
1393
1394         LOCKSTART;
1395         tdip = VTOI(ap->a_dvp);
1396         if (tdip->pmp->ronly) {
1397                 LOCKSTOP;
1398                 return (EROFS);
1399         }
1400
1401         ncp = ap->a_nch->ncp;
1402         name = ncp->nc_name;
1403         name_len = ncp->nc_nlen;
1404
1405         /*
1406          * ip represents the file being hardlinked.  The file could be a
1407          * normal file or a hardlink target if it has already been hardlinked.
1408          * (with the new semantics, it will almost always be a hardlink
1409          * target).
1410          *
1411          * Bump nlinks and potentially also create or move the hardlink
1412          * target in the parent directory common to (ip) and (tdip).  The
1413          * consolidation code can modify ip->cluster.  The returned cluster
1414          * is locked.
1415          */
1416         ip = VTOI(ap->a_vp);
1417         KASSERT(ip->pmp, ("ip->pmp is NULL %p %p", ip, ip->pmp));
1418         hammer2_pfs_memory_wait(ip->pmp);
1419         hammer2_trans_init(ip->pmp, 0);
1420
1421         /*
1422          * Target should be an indexed inode or there's no way we will ever
1423          * be able to find it!
1424          */
1425         KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1426
1427         error = 0;
1428
1429         /*
1430          * Can return NULL and error == EXDEV if the common parent
1431          * crosses a directory with the xlink flag set.
1432          */
1433         hammer2_inode_lock(tdip, 0);
1434         hammer2_inode_lock(ip, 0);
1435
1436         /*
1437          * Create the hardlink target and bump nlinks.
1438          */
1439         if (error == 0) {
1440                 hammer2_inode_create(tdip, tdip, NULL, NULL,
1441                                      name, name_len, 0,
1442                                      ip->meta.inum,
1443                                      HAMMER2_OBJTYPE_HARDLINK, ip->meta.type,
1444                                      0, &error);
1445                 hammer2_inode_modify(ip);
1446                 ++ip->meta.nlinks;
1447         }
1448         if (error == 0) {
1449                 /*
1450                  * Update dip's mtime
1451                  */
1452                 uint64_t mtime;
1453
1454                 hammer2_update_time(&mtime);
1455                 hammer2_inode_modify(tdip);
1456                 tdip->meta.mtime = mtime;
1457
1458                 cache_setunresolved(ap->a_nch);
1459                 cache_setvp(ap->a_nch, ap->a_vp);
1460         }
1461         hammer2_inode_unlock(ip);
1462         hammer2_inode_unlock(tdip);
1463
1464         hammer2_trans_done(ip->pmp);
1465
1466         LOCKSTOP;
1467         return error;
1468 }
1469
1470 /*
1471  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1472  *
1473  * The operating system has already ensured that the directory entry
1474  * does not exist and done all appropriate namespace locking.
1475  */
1476 static
1477 int
1478 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1479 {
1480         hammer2_inode_t *dip;
1481         hammer2_inode_t *nip;
1482         struct namecache *ncp;
1483         const uint8_t *name;
1484         size_t name_len;
1485         hammer2_tid_t inum;
1486         int error;
1487
1488         LOCKSTART;
1489         dip = VTOI(ap->a_dvp);
1490         if (dip->pmp->ronly) {
1491                 LOCKSTOP;
1492                 return (EROFS);
1493         }
1494
1495         ncp = ap->a_nch->ncp;
1496         name = ncp->nc_name;
1497         name_len = ncp->nc_nlen;
1498         hammer2_pfs_memory_wait(dip->pmp);
1499         hammer2_trans_init(dip->pmp, 0);
1500
1501         inum = hammer2_trans_newinum(dip->pmp);
1502
1503         /*
1504          * Create the actual inode as a hidden file in the iroot, then
1505          * create the directory entry as a hardlink to it.  The creation
1506          * of the actual inode sets its nlinks to 1 which is the value
1507          * we desire.
1508          */
1509         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1510                                    NULL, 0, inum,
1511                                    inum, 0, 0,
1512                                    0, &error);
1513
1514         if (error == 0) {
1515                 hammer2_inode_create(dip, dip, NULL, NULL,
1516                                      name, name_len, 0,
1517                                      nip->meta.inum,
1518                                      HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1519                                      0, &error);
1520         }
1521         if (error) {
1522                 KKASSERT(nip == NULL);
1523                 *ap->a_vpp = NULL;
1524         } else {
1525                 *ap->a_vpp = hammer2_igetv(nip, &error);
1526                 hammer2_inode_unlock(nip);
1527         }
1528
1529         /*
1530          * Update dip's mtime
1531          */
1532         if (error == 0) {
1533                 uint64_t mtime;
1534
1535                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1536                 hammer2_update_time(&mtime);
1537                 hammer2_inode_modify(dip);
1538                 dip->meta.mtime = mtime;
1539                 hammer2_inode_unlock(dip);
1540         }
1541
1542         hammer2_trans_done(dip->pmp);
1543
1544         if (error == 0) {
1545                 cache_setunresolved(ap->a_nch);
1546                 cache_setvp(ap->a_nch, *ap->a_vpp);
1547         }
1548         LOCKSTOP;
1549         return error;
1550 }
1551
1552 /*
1553  * Make a device node (typically a fifo)
1554  */
1555 static
1556 int
1557 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1558 {
1559         hammer2_inode_t *dip;
1560         hammer2_inode_t *nip;
1561         struct namecache *ncp;
1562         const uint8_t *name;
1563         size_t name_len;
1564         hammer2_tid_t inum;
1565         int error;
1566
1567         LOCKSTART;
1568         dip = VTOI(ap->a_dvp);
1569         if (dip->pmp->ronly) {
1570                 LOCKSTOP;
1571                 return (EROFS);
1572         }
1573
1574         ncp = ap->a_nch->ncp;
1575         name = ncp->nc_name;
1576         name_len = ncp->nc_nlen;
1577         hammer2_pfs_memory_wait(dip->pmp);
1578         hammer2_trans_init(dip->pmp, 0);
1579
1580         /*
1581          * The device node is entered as the directory entry itself and not
1582          * as a hardlink to an inode.  Since one cannot obtain a
1583          * file handle on the filesystem entry representing the device, we
1584          * do not have to worry about indexing its inode.
1585          */
1586         inum = hammer2_trans_newinum(dip->pmp);
1587         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1588                                    NULL, 0, inum,
1589                                    inum, 0, 0,
1590                                    0, &error);
1591         if (error == 0) {
1592                 hammer2_inode_create(dip, dip, NULL, NULL,
1593                                      name, name_len, 0,
1594                                      nip->meta.inum,
1595                                      HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1596                                      0, &error);
1597         }
1598
1599
1600         if (error) {
1601                 KKASSERT(nip == NULL);
1602                 *ap->a_vpp = NULL;
1603         } else {
1604                 *ap->a_vpp = hammer2_igetv(nip, &error);
1605                 hammer2_inode_unlock(nip);
1606         }
1607
1608         /*
1609          * Update dip's mtime
1610          */
1611         if (error == 0) {
1612                 uint64_t mtime;
1613
1614                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1615                 hammer2_update_time(&mtime);
1616                 hammer2_inode_modify(dip);
1617                 dip->meta.mtime = mtime;
1618                 hammer2_inode_unlock(dip);
1619         }
1620
1621         hammer2_trans_done(dip->pmp);
1622
1623         if (error == 0) {
1624                 cache_setunresolved(ap->a_nch);
1625                 cache_setvp(ap->a_nch, *ap->a_vpp);
1626         }
1627         LOCKSTOP;
1628         return error;
1629 }
1630
1631 /*
1632  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1633  */
1634 static
1635 int
1636 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1637 {
1638         hammer2_inode_t *dip;
1639         hammer2_inode_t *nip;
1640         struct namecache *ncp;
1641         const uint8_t *name;
1642         size_t name_len;
1643         hammer2_tid_t inum;
1644         int error;
1645
1646         dip = VTOI(ap->a_dvp);
1647         if (dip->pmp->ronly)
1648                 return (EROFS);
1649
1650         ncp = ap->a_nch->ncp;
1651         name = ncp->nc_name;
1652         name_len = ncp->nc_nlen;
1653         hammer2_pfs_memory_wait(dip->pmp);
1654         hammer2_trans_init(dip->pmp, 0);
1655
1656         ap->a_vap->va_type = VLNK;      /* enforce type */
1657
1658         /*
1659          * The softlink is entered into the directory itself and not
1660          * as a hardlink to an inode.  Since one cannot obtain a
1661          * file handle on the softlink itself we do not have to worry
1662          * about indexing its inode.
1663          */
1664         inum = hammer2_trans_newinum(dip->pmp);
1665
1666         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1667                                    NULL, 0, inum,
1668                                    inum, 0, 0,
1669                                    0, &error);
1670         if (error == 0) {
1671                 hammer2_inode_create(dip, dip, NULL, NULL,
1672                                      name, name_len, 0,
1673                                      nip->meta.inum,
1674                                      HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1675                                      0, &error);
1676         }
1677
1678
1679         if (error) {
1680                 KKASSERT(nip == NULL);
1681                 *ap->a_vpp = NULL;
1682                 hammer2_trans_done(dip->pmp);
1683                 return error;
1684         }
1685         *ap->a_vpp = hammer2_igetv(nip, &error);
1686
1687         /*
1688          * Build the softlink (~like file data) and finalize the namecache.
1689          */
1690         if (error == 0) {
1691                 size_t bytes;
1692                 struct uio auio;
1693                 struct iovec aiov;
1694
1695                 bytes = strlen(ap->a_target);
1696
1697                 hammer2_inode_unlock(nip);
1698                 bzero(&auio, sizeof(auio));
1699                 bzero(&aiov, sizeof(aiov));
1700                 auio.uio_iov = &aiov;
1701                 auio.uio_segflg = UIO_SYSSPACE;
1702                 auio.uio_rw = UIO_WRITE;
1703                 auio.uio_resid = bytes;
1704                 auio.uio_iovcnt = 1;
1705                 auio.uio_td = curthread;
1706                 aiov.iov_base = ap->a_target;
1707                 aiov.iov_len = bytes;
1708                 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1709                 /* XXX handle error */
1710                 error = 0;
1711         } else {
1712                 hammer2_inode_unlock(nip);
1713         }
1714
1715         /*
1716          * Update dip's mtime
1717          */
1718         if (error == 0) {
1719                 uint64_t mtime;
1720
1721                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1722                 hammer2_update_time(&mtime);
1723                 hammer2_inode_modify(dip);
1724                 dip->meta.mtime = mtime;
1725                 hammer2_inode_unlock(dip);
1726         }
1727
1728         hammer2_trans_done(dip->pmp);
1729
1730         /*
1731          * Finalize namecache
1732          */
1733         if (error == 0) {
1734                 cache_setunresolved(ap->a_nch);
1735                 cache_setvp(ap->a_nch, *ap->a_vpp);
1736                 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1737         }
1738         return error;
1739 }
1740
1741 /*
1742  * hammer2_vop_nremove { nch, dvp, cred }
1743  */
1744 static
1745 int
1746 hammer2_vop_nremove(struct vop_nremove_args *ap)
1747 {
1748         hammer2_xop_unlink_t *xop;
1749         hammer2_inode_t *dip;
1750         hammer2_inode_t *ip;
1751         struct namecache *ncp;
1752         int error;
1753         int isopen;
1754
1755         LOCKSTART;
1756         dip = VTOI(ap->a_dvp);
1757         if (dip->pmp->ronly) {
1758                 LOCKSTOP;
1759                 return(EROFS);
1760         }
1761
1762         ncp = ap->a_nch->ncp;
1763
1764         hammer2_pfs_memory_wait(dip->pmp);
1765         hammer2_trans_init(dip->pmp, 0);
1766         hammer2_inode_lock(dip, 0);
1767
1768         /*
1769          * The unlink XOP unlinks the path from the directory and
1770          * locates and returns the cluster associated with the real inode.
1771          * We have to handle nlinks here on the frontend.
1772          */
1773         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1774         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1775         isopen = cache_isopen(ap->a_nch);
1776         xop->isdir = 0;
1777         xop->dopermanent = 0;
1778         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1779
1780         /*
1781          * Collect the real inode and adjust nlinks, destroy the real
1782          * inode if nlinks transitions to 0 and it was the real inode
1783          * (else it has already been removed).
1784          */
1785         error = hammer2_xop_collect(&xop->head, 0);
1786         hammer2_inode_unlock(dip);
1787
1788         if (error == 0) {
1789                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1790                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1791                 if (ip) {
1792                         hammer2_inode_unlink_finisher(ip, isopen);
1793                         hammer2_inode_unlock(ip);
1794                 }
1795         } else {
1796                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1797         }
1798
1799         /*
1800          * Update dip's mtime
1801          */
1802         if (error == 0) {
1803                 uint64_t mtime;
1804
1805                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1806                 hammer2_update_time(&mtime);
1807                 hammer2_inode_modify(dip);
1808                 dip->meta.mtime = mtime;
1809                 hammer2_inode_unlock(dip);
1810         }
1811
1812         hammer2_inode_run_sideq(dip->pmp);
1813         hammer2_trans_done(dip->pmp);
1814         if (error == 0)
1815                 cache_unlink(ap->a_nch);
1816         LOCKSTOP;
1817         return (error);
1818 }
1819
1820 /*
1821  * hammer2_vop_nrmdir { nch, dvp, cred }
1822  */
1823 static
1824 int
1825 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1826 {
1827         hammer2_xop_unlink_t *xop;
1828         hammer2_inode_t *dip;
1829         hammer2_inode_t *ip;
1830         struct namecache *ncp;
1831         int isopen;
1832         int error;
1833
1834         LOCKSTART;
1835         dip = VTOI(ap->a_dvp);
1836         if (dip->pmp->ronly) {
1837                 LOCKSTOP;
1838                 return(EROFS);
1839         }
1840
1841         hammer2_pfs_memory_wait(dip->pmp);
1842         hammer2_trans_init(dip->pmp, 0);
1843         hammer2_inode_lock(dip, 0);
1844
1845         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1846
1847         ncp = ap->a_nch->ncp;
1848         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1849         isopen = cache_isopen(ap->a_nch);
1850         xop->isdir = 1;
1851         xop->dopermanent = 0;
1852         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1853
1854         /*
1855          * Collect the real inode and adjust nlinks, destroy the real
1856          * inode if nlinks transitions to 0 and it was the real inode
1857          * (else it has already been removed).
1858          */
1859         error = hammer2_xop_collect(&xop->head, 0);
1860         hammer2_inode_unlock(dip);
1861
1862         if (error == 0) {
1863                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1864                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1865                 if (ip) {
1866                         hammer2_inode_unlink_finisher(ip, isopen);
1867                         hammer2_inode_unlock(ip);
1868                 }
1869         } else {
1870                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1871         }
1872
1873         /*
1874          * Update dip's mtime
1875          */
1876         if (error == 0) {
1877                 uint64_t mtime;
1878
1879                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1880                 hammer2_update_time(&mtime);
1881                 hammer2_inode_modify(dip);
1882                 dip->meta.mtime = mtime;
1883                 hammer2_inode_unlock(dip);
1884         }
1885
1886         hammer2_inode_run_sideq(dip->pmp);
1887         hammer2_trans_done(dip->pmp);
1888         if (error == 0)
1889                 cache_unlink(ap->a_nch);
1890         LOCKSTOP;
1891         return (error);
1892 }
1893
1894 /*
1895  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1896  */
1897 static
1898 int
1899 hammer2_vop_nrename(struct vop_nrename_args *ap)
1900 {
1901         struct namecache *fncp;
1902         struct namecache *tncp;
1903         hammer2_inode_t *fdip;
1904         hammer2_inode_t *tdip;
1905         hammer2_inode_t *ip;
1906         const uint8_t *fname;
1907         size_t fname_len;
1908         const uint8_t *tname;
1909         size_t tname_len;
1910         int error;
1911         int tnch_error;
1912         int update_tdip;
1913         int update_fdip;
1914         hammer2_key_t tlhc;
1915
1916         if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1917                 return(EXDEV);
1918         if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1919                 return(EXDEV);
1920
1921         fdip = VTOI(ap->a_fdvp);        /* source directory */
1922         tdip = VTOI(ap->a_tdvp);        /* target directory */
1923
1924         if (fdip->pmp->ronly)
1925                 return(EROFS);
1926
1927         LOCKSTART;
1928         fncp = ap->a_fnch->ncp;         /* entry name in source */
1929         fname = fncp->nc_name;
1930         fname_len = fncp->nc_nlen;
1931
1932         tncp = ap->a_tnch->ncp;         /* entry name in target */
1933         tname = tncp->nc_name;
1934         tname_len = tncp->nc_nlen;
1935
1936         hammer2_pfs_memory_wait(tdip->pmp);
1937         hammer2_trans_init(tdip->pmp, 0);
1938
1939         update_tdip = 0;
1940         update_fdip = 0;
1941
1942         /*
1943          * ip is the inode being renamed.  If this is a hardlink then
1944          * ip represents the actual file and not the hardlink marker.
1945          */
1946         ip = VTOI(fncp->nc_vp);
1947
1948         KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1949
1950         /*
1951          * Can return NULL and error == EXDEV if the common parent
1952          * crosses a directory with the xlink flag set.
1953          */
1954         error = 0;
1955         hammer2_inode_lock(fdip, 0);
1956         hammer2_inode_lock(tdip, 0);
1957         hammer2_inode_ref(ip);          /* extra ref */
1958
1959         hammer2_inode_lock(ip, 0);
1960
1961         /*
1962          * Delete the target namespace.
1963          */
1964         {
1965                 hammer2_xop_unlink_t *xop2;
1966                 hammer2_inode_t *tip;
1967                 int isopen;
1968
1969                 /*
1970                  * The unlink XOP unlinks the path from the directory and
1971                  * locates and returns the cluster associated with the real
1972                  * inode.  We have to handle nlinks here on the frontend.
1973                  */
1974                 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1975                 hammer2_xop_setname(&xop2->head, tname, tname_len);
1976                 isopen = cache_isopen(ap->a_tnch);
1977                 xop2->isdir = -1;
1978                 xop2->dopermanent = 0;
1979                 hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
1980
1981                 /*
1982                  * Collect the real inode and adjust nlinks, destroy the real
1983                  * inode if nlinks transitions to 0 and it was the real inode
1984                  * (else it has already been removed).
1985                  */
1986                 tnch_error = hammer2_xop_collect(&xop2->head, 0);
1987                 /* hammer2_inode_unlock(tdip); */
1988
1989                 if (tnch_error == 0) {
1990                         tip = hammer2_inode_get(tdip->pmp, NULL,
1991                                                 &xop2->head.cluster, -1);
1992                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1993                         if (tip) {
1994                                 hammer2_inode_unlink_finisher(tip, isopen);
1995                                 hammer2_inode_unlock(tip);
1996                         }
1997                 } else {
1998                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1999                 }
2000                 /* hammer2_inode_lock(tdip, 0); */
2001
2002                 if (tnch_error && tnch_error != ENOENT) {
2003                         error = tnch_error;
2004                         goto done2;
2005                 }
2006                 update_tdip = 1;
2007         }
2008
2009         /*
2010          * Resolve the collision space for (tdip, tname, tname_len)
2011          *
2012          * tdip must be held exclusively locked to prevent races.
2013          */
2014         {
2015                 hammer2_xop_scanlhc_t *sxop;
2016                 hammer2_tid_t lhcbase;
2017
2018                 tlhc = hammer2_dirhash(tname, tname_len);
2019                 lhcbase = tlhc;
2020                 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2021                 sxop->lhc = tlhc;
2022                 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
2023                 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
2024                         if (tlhc != sxop->head.cluster.focus->bref.key)
2025                                 break;
2026                         ++tlhc;
2027                 }
2028                 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
2029
2030                 if (error) {
2031                         if (error != ENOENT)
2032                                 goto done2;
2033                         ++tlhc;
2034                         error = 0;
2035                 }
2036                 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
2037                         error = ENOSPC;
2038                         goto done2;
2039                 }
2040         }
2041
2042         /*
2043          * Everything is setup, do the rename.
2044          *
2045          * We have to synchronize ip->meta to the underlying operation.
2046          *
2047          * NOTE: To avoid deadlocks we cannot lock (ip) while we are
2048          *       unlinking elements from their directories.  Locking
2049          *       the nlinks field does not lock the whole inode.
2050          */
2051         /* hammer2_inode_lock(ip, 0); */
2052         if (error == 0) {
2053                 hammer2_xop_nrename_t *xop4;
2054
2055                 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
2056                 xop4->lhc = tlhc;
2057                 xop4->ip_key = ip->meta.name_key;
2058                 hammer2_xop_setip2(&xop4->head, ip);
2059                 hammer2_xop_setip3(&xop4->head, tdip);
2060                 hammer2_xop_setname(&xop4->head, fname, fname_len);
2061                 hammer2_xop_setname2(&xop4->head, tname, tname_len);
2062                 hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
2063
2064                 error = hammer2_xop_collect(&xop4->head, 0);
2065                 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
2066
2067                 if (error == ENOENT)
2068                         error = 0;
2069                 if (error == 0 &&
2070                     (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
2071                         hammer2_inode_modify(ip);
2072                         ip->meta.name_len = tname_len;
2073                         ip->meta.name_key = tlhc;
2074
2075                 }
2076                 update_fdip = 1;
2077                 update_fdip = 1;
2078         }
2079
2080 done2:
2081         /*
2082          * Update directory mtimes to represent the something changed.
2083          */
2084         if (update_fdip || update_tdip) {
2085                 uint64_t mtime;
2086
2087                 hammer2_update_time(&mtime);
2088                 if (update_fdip) {
2089                         hammer2_inode_modify(fdip);
2090                         fdip->meta.mtime = mtime;
2091                 }
2092                 if (update_tdip) {
2093                         hammer2_inode_modify(tdip);
2094                         tdip->meta.mtime = mtime;
2095                 }
2096         }
2097         hammer2_inode_unlock(ip);
2098         hammer2_inode_unlock(tdip);
2099         hammer2_inode_unlock(fdip);
2100         hammer2_inode_drop(ip);
2101         hammer2_inode_run_sideq(fdip->pmp);
2102
2103         hammer2_trans_done(tdip->pmp);
2104
2105         /*
2106          * Issue the namecache update after unlocking all the internal
2107          * hammer structures, otherwise we might deadlock.
2108          */
2109         if (tnch_error == 0) {
2110                 cache_unlink(ap->a_tnch);
2111                 cache_setunresolved(ap->a_tnch);
2112         }
2113         if (error == 0)
2114                 cache_rename(ap->a_fnch, ap->a_tnch);
2115
2116         LOCKSTOP;
2117         return (error);
2118 }
2119
2120 /*
2121  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2122  */
2123 static
2124 int
2125 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2126 {
2127         hammer2_inode_t *ip;
2128         int error;
2129
2130         LOCKSTART;
2131         ip = VTOI(ap->a_vp);
2132
2133         error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2134                               ap->a_fflag, ap->a_cred);
2135         LOCKSTOP;
2136         return (error);
2137 }
2138
2139 static
2140 int
2141 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2142 {
2143         struct mount *mp;
2144         hammer2_pfs_t *pmp;
2145         int rc;
2146
2147         LOCKSTART;
2148         switch (ap->a_op) {
2149         case (MOUNTCTL_SET_EXPORT):
2150                 mp = ap->a_head.a_ops->head.vv_mount;
2151                 pmp = MPTOPMP(mp);
2152
2153                 if (ap->a_ctllen != sizeof(struct export_args))
2154                         rc = (EINVAL);
2155                 else
2156                         rc = vfs_export(mp, &pmp->export,
2157                                         (const struct export_args *)ap->a_ctl);
2158                 break;
2159         default:
2160                 rc = vop_stdmountctl(ap);
2161                 break;
2162         }
2163         LOCKSTOP;
2164         return (rc);
2165 }
2166
2167 /*
2168  * KQFILTER
2169  */
2170 static void filt_hammer2detach(struct knote *kn);
2171 static int filt_hammer2read(struct knote *kn, long hint);
2172 static int filt_hammer2write(struct knote *kn, long hint);
2173 static int filt_hammer2vnode(struct knote *kn, long hint);
2174
2175 static struct filterops hammer2read_filtops =
2176         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2177           NULL, filt_hammer2detach, filt_hammer2read };
2178 static struct filterops hammer2write_filtops =
2179         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2180           NULL, filt_hammer2detach, filt_hammer2write };
2181 static struct filterops hammer2vnode_filtops =
2182         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2183           NULL, filt_hammer2detach, filt_hammer2vnode };
2184
2185 static
2186 int
2187 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2188 {
2189         struct vnode *vp = ap->a_vp;
2190         struct knote *kn = ap->a_kn;
2191
2192         switch (kn->kn_filter) {
2193         case EVFILT_READ:
2194                 kn->kn_fop = &hammer2read_filtops;
2195                 break;
2196         case EVFILT_WRITE:
2197                 kn->kn_fop = &hammer2write_filtops;
2198                 break;
2199         case EVFILT_VNODE:
2200                 kn->kn_fop = &hammer2vnode_filtops;
2201                 break;
2202         default:
2203                 return (EOPNOTSUPP);
2204         }
2205
2206         kn->kn_hook = (caddr_t)vp;
2207
2208         knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2209
2210         return(0);
2211 }
2212
2213 static void
2214 filt_hammer2detach(struct knote *kn)
2215 {
2216         struct vnode *vp = (void *)kn->kn_hook;
2217
2218         knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2219 }
2220
2221 static int
2222 filt_hammer2read(struct knote *kn, long hint)
2223 {
2224         struct vnode *vp = (void *)kn->kn_hook;
2225         hammer2_inode_t *ip = VTOI(vp);
2226         off_t off;
2227
2228         if (hint == NOTE_REVOKE) {
2229                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2230                 return(1);
2231         }
2232         off = ip->meta.size - kn->kn_fp->f_offset;
2233         kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2234         if (kn->kn_sfflags & NOTE_OLDAPI)
2235                 return(1);
2236         return (kn->kn_data != 0);
2237 }
2238
2239
2240 static int
2241 filt_hammer2write(struct knote *kn, long hint)
2242 {
2243         if (hint == NOTE_REVOKE)
2244                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2245         kn->kn_data = 0;
2246         return (1);
2247 }
2248
2249 static int
2250 filt_hammer2vnode(struct knote *kn, long hint)
2251 {
2252         if (kn->kn_sfflags & hint)
2253                 kn->kn_fflags |= hint;
2254         if (hint == NOTE_REVOKE) {
2255                 kn->kn_flags |= (EV_EOF | EV_NODATA);
2256                 return (1);
2257         }
2258         return (kn->kn_fflags != 0);
2259 }
2260
2261 /*
2262  * FIFO VOPS
2263  */
2264 static
2265 int
2266 hammer2_vop_markatime(struct vop_markatime_args *ap)
2267 {
2268         hammer2_inode_t *ip;
2269         struct vnode *vp;
2270
2271         vp = ap->a_vp;
2272         ip = VTOI(vp);
2273
2274         if (ip->pmp->ronly)
2275                 return(EROFS);
2276         return(0);
2277 }
2278
2279 static
2280 int
2281 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2282 {
2283         int error;
2284
2285         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2286         if (error)
2287                 error = hammer2_vop_kqfilter(ap);
2288         return(error);
2289 }
2290
2291 /*
2292  * VOPS vector
2293  */
2294 struct vop_ops hammer2_vnode_vops = {
2295         .vop_default    = vop_defaultop,
2296         .vop_fsync      = hammer2_vop_fsync,
2297         .vop_getpages   = vop_stdgetpages,
2298         .vop_putpages   = vop_stdputpages,
2299         .vop_access     = hammer2_vop_access,
2300         .vop_advlock    = hammer2_vop_advlock,
2301         .vop_close      = hammer2_vop_close,
2302         .vop_nlink      = hammer2_vop_nlink,
2303         .vop_ncreate    = hammer2_vop_ncreate,
2304         .vop_nsymlink   = hammer2_vop_nsymlink,
2305         .vop_nremove    = hammer2_vop_nremove,
2306         .vop_nrmdir     = hammer2_vop_nrmdir,
2307         .vop_nrename    = hammer2_vop_nrename,
2308         .vop_getattr    = hammer2_vop_getattr,
2309         .vop_setattr    = hammer2_vop_setattr,
2310         .vop_readdir    = hammer2_vop_readdir,
2311         .vop_readlink   = hammer2_vop_readlink,
2312         .vop_getpages   = vop_stdgetpages,
2313         .vop_putpages   = vop_stdputpages,
2314         .vop_read       = hammer2_vop_read,
2315         .vop_write      = hammer2_vop_write,
2316         .vop_open       = hammer2_vop_open,
2317         .vop_inactive   = hammer2_vop_inactive,
2318         .vop_reclaim    = hammer2_vop_reclaim,
2319         .vop_nresolve   = hammer2_vop_nresolve,
2320         .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2321         .vop_nmkdir     = hammer2_vop_nmkdir,
2322         .vop_nmknod     = hammer2_vop_nmknod,
2323         .vop_ioctl      = hammer2_vop_ioctl,
2324         .vop_mountctl   = hammer2_vop_mountctl,
2325         .vop_bmap       = hammer2_vop_bmap,
2326         .vop_strategy   = hammer2_vop_strategy,
2327         .vop_kqfilter   = hammer2_vop_kqfilter
2328 };
2329
2330 struct vop_ops hammer2_spec_vops = {
2331         .vop_default =          vop_defaultop,
2332         .vop_fsync =            hammer2_vop_fsync,
2333         .vop_read =             vop_stdnoread,
2334         .vop_write =            vop_stdnowrite,
2335         .vop_access =           hammer2_vop_access,
2336         .vop_close =            hammer2_vop_close,
2337         .vop_markatime =        hammer2_vop_markatime,
2338         .vop_getattr =          hammer2_vop_getattr,
2339         .vop_inactive =         hammer2_vop_inactive,
2340         .vop_reclaim =          hammer2_vop_reclaim,
2341         .vop_setattr =          hammer2_vop_setattr
2342 };
2343
2344 struct vop_ops hammer2_fifo_vops = {
2345         .vop_default =          fifo_vnoperate,
2346         .vop_fsync =            hammer2_vop_fsync,
2347 #if 0
2348         .vop_read =             hammer2_vop_fiforead,
2349         .vop_write =            hammer2_vop_fifowrite,
2350 #endif
2351         .vop_access =           hammer2_vop_access,
2352 #if 0
2353         .vop_close =            hammer2_vop_fifoclose,
2354 #endif
2355         .vop_markatime =        hammer2_vop_markatime,
2356         .vop_getattr =          hammer2_vop_getattr,
2357         .vop_inactive =         hammer2_vop_inactive,
2358         .vop_reclaim =          hammer2_vop_reclaim,
2359         .vop_setattr =          hammer2_vop_setattr,
2360         .vop_kqfilter =         hammer2_vop_fifokqfilter
2361 };
2362