sys/vfs/hammer2/hammer2_vnops.c

   1 /*
   2  * Copyright (c) 2011-2015 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@dragonflybsd.org>
   6  * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
   7  * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  *
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in
  17  *    the documentation and/or other materials provided with the
  18  *    distribution.
  19  * 3. Neither the name of The DragonFly Project nor the names of its
  20  *    contributors may be used to endorse or promote products derived
  21  *    from this software without specific, prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  26  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  27  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  28  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  29  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  30  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  31  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  32  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  33  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  */
  36 /*
  37  * Kernel Filesystem interface
  38  *
  39  * NOTE! local ipdata pointers must be reloaded on any modifying operation
  40  *       to the inode as its underlying chain may have changed.
  41  */
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/kernel.h>
  46 #include <sys/fcntl.h>
  47 #include <sys/buf.h>
  48 #include <sys/proc.h>
  49 #include <sys/namei.h>
  50 #include <sys/mount.h>
  51 #include <sys/vnode.h>
  52 #include <sys/mountctl.h>
  53 #include <sys/dirent.h>
  54 #include <sys/uio.h>
  55 #include <sys/objcache.h>
  56 #include <sys/event.h>
  57 #include <sys/file.h>
  58 #include <vfs/fifofs/fifo.h>
  59
  60 #include "hammer2.h"
  61
  62 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
  63                                 int seqcount);
  64 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
  65                                 int ioflag, int seqcount);
  66 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
  67 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);
  68
  69 struct objcache *cache_xops;
  70
  71 static __inline
  72 void
  73 hammer2_knote(struct vnode *vp, int flags)
  74 {
  75         if (flags)
  76                 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
  77 }
  78
  79 /*
  80  * Last reference to a vnode is going away but it is still cached.
  81  */
  82 static
  83 int
  84 hammer2_vop_inactive(struct vop_inactive_args *ap)
  85 {
  86         hammer2_inode_t *ip;
  87         struct vnode *vp;
  88
  89         LOCKSTART;
  90         vp = ap->a_vp;
  91         ip = VTOI(vp);
  92
  93         /*
  94          * Degenerate case
  95          */
  96         if (ip == NULL) {
  97                 vrecycle(vp);
  98                 LOCKSTOP;
  99                 return (0);
 100         }
 101
 102         /*
 103          * Check for deleted inodes and recycle immediately on the last
 104          * release.  Be sure to destroy any left-over buffer cache buffers
 105          * so we do not waste time trying to flush them.
 106          *
 107          * Note that deleting the file block chains under the inode chain
 108          * would just be a waste of energy, so don't do it.
 109          *
 110          * WARNING: nvtruncbuf() can only be safely called without the inode
 111          *          lock held due to the way our write thread works.
 112          */
 113         if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
 114                 hammer2_key_t lbase;
 115                 int nblksize;
 116
 117                 /*
 118                  * Detect updates to the embedded data which may be
 119                  * synchronized by the strategy code.  Simply mark the
 120                  * inode modified so it gets picked up by our normal flush.
 121                  */
 122                 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
 123                 nvtruncbuf(vp, 0, nblksize, 0, 0);
 124                 vrecycle(vp);
 125         }
 126         LOCKSTOP;
 127         return (0);
 128 }
 129
 130 /*
 131  * Reclaim a vnode so that it can be reused; after the inode is
 132  * disassociated, the filesystem must manage it alone.
 133  */
 134 static
 135 int
 136 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
 137 {
 138         hammer2_inode_t *ip;
 139         hammer2_pfs_t *pmp;
 140         struct vnode *vp;
 141
 142         LOCKSTART;
 143         vp = ap->a_vp;
 144         ip = VTOI(vp);
 145         if (ip == NULL) {
 146                 LOCKSTOP;
 147                 return(0);
 148         }
 149         pmp = ip->pmp;
 150
 151         /*
 152          * The final close of a deleted file or directory marks it for
 153          * destruction.  The DELETED flag allows the flusher to shortcut
 154          * any modified blocks still unflushed (that is, just ignore them).
 155          *
 156          * HAMMER2 usually does not try to optimize the freemap by returning
 157          * deleted blocks to it as it does not usually know how many snapshots
 158          * might be referencing portions of the file/dir.
 159          */
 160         vp->v_data = NULL;
 161         ip->vp = NULL;
 162
 163         /*
 164          * NOTE! We do not attempt to flush chains here, flushing is
 165          *       really fragile and could also deadlock.
 166          */
 167         vclrisdirty(vp);
 168
 169         /*
 170          * An unlinked inode may have been relinked to the ihidden directory.
 171          * This occurs if the inode was unlinked while open.  Reclamation of
 172          * these inodes requires processing we cannot safely do here so add
 173          * the inode to the sideq in that situation.
 174          *
 175          * A modified inode may require chain synchronization which will no
 176          * longer be driven by a sync or fsync without the vnode, also use
 177          * the sideq for that.
 178          *
 179          * A reclaim can occur at any time so we cannot safely start a
 180          * transaction to handle reclamation of unlinked files.  Instead,
 181          * the ip is left with a reference and placed on a linked list and
 182          * handled later on.
 183          */
 184
 185         if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
 186                           HAMMER2_INODE_MODIFIED |
 187                           HAMMER2_INODE_RESIZED)) &&
 188             (ip->flags & HAMMER2_INODE_ISDELETED) == 0) {
 189                 hammer2_inode_sideq_t *ipul;
 190
 191                 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
 192                 ipul->ip = ip;
 193
 194                 hammer2_spin_ex(&pmp->list_spin);
 195                 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
 196                         /* ref -> sideq */
 197                         atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ);
 198                         TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry);
 199                         hammer2_spin_unex(&pmp->list_spin);
 200                 } else {
 201                         hammer2_spin_unex(&pmp->list_spin);
 202                         kfree(ipul, pmp->minode);
 203                         hammer2_inode_drop(ip);         /* vp ref */
 204                 }
 205                 /* retain ref from vp for ipul */
 206         } else {
 207                 hammer2_inode_drop(ip);                 /* vp ref */
 208         }
 209
 210         /*
 211          * XXX handle background sync when ip dirty, kernel will no longer
 212          * notify us regarding this inode because there is no longer a
 213          * vnode attached to it.
 214          */
 215
 216         LOCKSTOP;
 217         return (0);
 218 }
 219
 220 static
 221 int
 222 hammer2_vop_fsync(struct vop_fsync_args *ap)
 223 {
 224         hammer2_inode_t *ip;
 225         struct vnode *vp;
 226
 227         LOCKSTART;
 228         vp = ap->a_vp;
 229         ip = VTOI(vp);
 230
 231 #if 0
 232         /* XXX can't do this yet */
 233         hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH);
 234         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 235 #endif
 236         hammer2_trans_init(ip->pmp, 0);
 237         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 238
 239         /*
 240          * Calling chain_flush here creates a lot of duplicative
 241          * COW operations due to non-optimal vnode ordering.
 242          *
 243          * Only do it for an actual fsync() syscall.  The other forms
 244          * which call this function will eventually call chain_flush
 245          * on the volume root as a catch-all, which is far more optimal.
 246          */
 247         hammer2_inode_lock(ip, 0);
 248         if (ip->flags & HAMMER2_INODE_MODIFIED)
 249                 hammer2_inode_chain_sync(ip);
 250         hammer2_inode_unlock(ip);
 251         hammer2_trans_done(ip->pmp);
 252
 253         LOCKSTOP;
 254         return (0);
 255 }
 256
 257 static
 258 int
 259 hammer2_vop_access(struct vop_access_args *ap)
 260 {
 261         hammer2_inode_t *ip = VTOI(ap->a_vp);
 262         uid_t uid;
 263         gid_t gid;
 264         int error;
 265
 266         LOCKSTART;
 267         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 268         uid = hammer2_to_unix_xid(&ip->meta.uid);
 269         gid = hammer2_to_unix_xid(&ip->meta.gid);
 270         error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
 271         hammer2_inode_unlock(ip);
 272
 273         LOCKSTOP;
 274         return (error);
 275 }
 276
 277 static
 278 int
 279 hammer2_vop_getattr(struct vop_getattr_args *ap)
 280 {
 281         hammer2_pfs_t *pmp;
 282         hammer2_inode_t *ip;
 283         struct vnode *vp;
 284         struct vattr *vap;
 285         hammer2_chain_t *chain;
 286         int i;
 287
 288         LOCKSTART;
 289         vp = ap->a_vp;
 290         vap = ap->a_vap;
 291
 292         ip = VTOI(vp);
 293         pmp = ip->pmp;
 294
 295         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 296
 297         vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
 298         vap->va_fileid = ip->meta.inum;
 299         vap->va_mode = ip->meta.mode;
 300         vap->va_nlink = ip->meta.nlinks;
 301         vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
 302         vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
 303         vap->va_rmajor = 0;
 304         vap->va_rminor = 0;
 305         vap->va_size = ip->meta.size;   /* protected by shared lock */
 306         vap->va_blocksize = HAMMER2_PBUFSIZE;
 307         vap->va_flags = ip->meta.uflags;
 308         hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
 309         hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
 310         hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
 311         vap->va_gen = 1;
 312         vap->va_bytes = 0;
 313         if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) {
 314                 /*
 315                  * Can't really calculate directory use sans the files under
 316                  * it, just assume one block for now.
 317                  */
 318                 vap->va_bytes += HAMMER2_INODE_BYTES;
 319         } else {
 320                 for (i = 0; i < ip->cluster.nchains; ++i) {
 321                         if ((chain = ip->cluster.array[i].chain) != NULL) {
 322                                 if (vap->va_bytes < chain->bref.data_count)
 323                                         vap->va_bytes = chain->bref.data_count;
 324                         }
 325                 }
 326         }
 327         vap->va_type = hammer2_get_vtype(ip->meta.type);
 328         vap->va_filerev = 0;
 329         vap->va_uid_uuid = ip->meta.uid;
 330         vap->va_gid_uuid = ip->meta.gid;
 331         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 332                           VA_FSID_UUID_VALID;
 333
 334         hammer2_inode_unlock(ip);
 335
 336         LOCKSTOP;
 337         return (0);
 338 }
 339
 340 static
 341 int
 342 hammer2_vop_setattr(struct vop_setattr_args *ap)
 343 {
 344         hammer2_inode_t *ip;
 345         struct vnode *vp;
 346         struct vattr *vap;
 347         int error;
 348         int kflags = 0;
 349         uint64_t ctime;
 350
 351         LOCKSTART;
 352         vp = ap->a_vp;
 353         vap = ap->a_vap;
 354         hammer2_update_time(&ctime);
 355
 356         ip = VTOI(vp);
 357
 358         if (ip->pmp->ronly) {
 359                 LOCKSTOP;
 360                 return(EROFS);
 361         }
 362
 363         hammer2_pfs_memory_wait(ip->pmp);
 364         hammer2_trans_init(ip->pmp, 0);
 365         hammer2_inode_lock(ip, 0);
 366         error = 0;
 367
 368         if (vap->va_flags != VNOVAL) {
 369                 u_int32_t flags;
 370
 371                 flags = ip->meta.uflags;
 372                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
 373                                      hammer2_to_unix_xid(&ip->meta.uid),
 374                                      ap->a_cred);
 375                 if (error == 0) {
 376                         if (ip->meta.uflags != flags) {
 377                                 hammer2_inode_modify(ip);
 378                                 ip->meta.uflags = flags;
 379                                 ip->meta.ctime = ctime;
 380                                 kflags |= NOTE_ATTRIB;
 381                         }
 382                         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 383                                 error = 0;
 384                                 goto done;
 385                         }
 386                 }
 387                 goto done;
 388         }
 389         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 390                 error = EPERM;
 391                 goto done;
 392         }
 393         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
 394                 mode_t cur_mode = ip->meta.mode;
 395                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 396                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 397                 uuid_t uuid_uid;
 398                 uuid_t uuid_gid;
 399
 400                 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
 401                                          ap->a_cred,
 402                                          &cur_uid, &cur_gid, &cur_mode);
 403                 if (error == 0) {
 404                         hammer2_guid_to_uuid(&uuid_uid, cur_uid);
 405                         hammer2_guid_to_uuid(&uuid_gid, cur_gid);
 406                         if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
 407                             bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
 408                             ip->meta.mode != cur_mode
 409                         ) {
 410                                 hammer2_inode_modify(ip);
 411                                 ip->meta.uid = uuid_uid;
 412                                 ip->meta.gid = uuid_gid;
 413                                 ip->meta.mode = cur_mode;
 414                                 ip->meta.ctime = ctime;
 415                         }
 416                         kflags |= NOTE_ATTRIB;
 417                 }
 418         }
 419
 420         /*
 421          * Resize the file
 422          */
 423         if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
 424                 switch(vp->v_type) {
 425                 case VREG:
 426                         if (vap->va_size == ip->meta.size)
 427                                 break;
 428                         if (vap->va_size < ip->meta.size) {
 429                                 hammer2_truncate_file(ip, vap->va_size);
 430                         } else {
 431                                 hammer2_extend_file(ip, vap->va_size);
 432                         }
 433                         hammer2_inode_modify(ip);
 434                         ip->meta.mtime = ctime;
 435                         break;
 436                 default:
 437                         error = EINVAL;
 438                         goto done;
 439                 }
 440         }
 441 #if 0
 442         /* atime not supported */
 443         if (vap->va_atime.tv_sec != VNOVAL) {
 444                 hammer2_inode_modify(ip);
 445                 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
 446                 kflags |= NOTE_ATTRIB;
 447         }
 448 #endif
 449         if (vap->va_mode != (mode_t)VNOVAL) {
 450                 mode_t cur_mode = ip->meta.mode;
 451                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 452                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 453
 454                 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
 455                                          cur_uid, cur_gid, &cur_mode);
 456                 if (error == 0 && ip->meta.mode != cur_mode) {
 457                         hammer2_inode_modify(ip);
 458                         ip->meta.mode = cur_mode;
 459                         ip->meta.ctime = ctime;
 460                         kflags |= NOTE_ATTRIB;
 461                 }
 462         }
 463
 464         if (vap->va_mtime.tv_sec != VNOVAL) {
 465                 hammer2_inode_modify(ip);
 466                 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
 467                 kflags |= NOTE_ATTRIB;
 468         }
 469
 470 done:
 471         /*
 472          * If a truncation occurred we must call inode_fsync() now in order
 473          * to trim the related data chains, otherwise a later expansion can
 474          * cause havoc.
 475          *
 476          * If an extend occured that changed the DIRECTDATA state, we must
 477          * call inode_fsync now in order to prepare the inode's indirect
 478          * block table.
 479          */
 480         if (ip->flags & HAMMER2_INODE_RESIZED)
 481                 hammer2_inode_chain_sync(ip);
 482
 483         /*
 484          * Cleanup.
 485          */
 486         hammer2_inode_unlock(ip);
 487         hammer2_trans_done(ip->pmp);
 488         hammer2_knote(ip->vp, kflags);
 489
 490         LOCKSTOP;
 491         return (error);
 492 }
 493
 494 static
 495 int
 496 hammer2_vop_readdir(struct vop_readdir_args *ap)
 497 {
 498         hammer2_xop_readdir_t *xop;
 499         hammer2_blockref_t bref;
 500         hammer2_inode_t *ip;
 501         hammer2_tid_t inum;
 502         hammer2_key_t lkey;
 503         struct uio *uio;
 504         off_t *cookies;
 505         off_t saveoff;
 506         int cookie_index;
 507         int ncookies;
 508         int error;
 509         int eofflag;
 510         int dtype;
 511         int r;
 512
 513         LOCKSTART;
 514         ip = VTOI(ap->a_vp);
 515         uio = ap->a_uio;
 516         saveoff = uio->uio_offset;
 517         eofflag = 0;
 518         error = 0;
 519
 520         /*
 521          * Setup cookies directory entry cookies if requested
 522          */
 523         if (ap->a_ncookies) {
 524                 ncookies = uio->uio_resid / 16 + 1;
 525                 if (ncookies > 1024)
 526                         ncookies = 1024;
 527                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 528         } else {
 529                 ncookies = -1;
 530                 cookies = NULL;
 531         }
 532         cookie_index = 0;
 533
 534         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 535
 536         /*
 537          * Handle artificial entries.  To ensure that only positive 64 bit
 538          * quantities are returned to userland we always strip off bit 63.
 539          * The hash code is designed such that codes 0x0000-0x7FFF are not
 540          * used, allowing us to use these codes for articial entries.
 541          *
 542          * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
 543          * allow '..' to cross the mount point into (e.g.) the super-root.
 544          */
 545         if (saveoff == 0) {
 546                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 547                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
 548                 if (r)
 549                         goto done;
 550                 if (cookies)
 551                         cookies[cookie_index] = saveoff;
 552                 ++saveoff;
 553                 ++cookie_index;
 554                 if (cookie_index == ncookies)
 555                         goto done;
 556         }
 557
 558         if (saveoff == 1) {
 559                 /*
 560                  * Be careful with lockorder when accessing ".."
 561                  *
 562                  * (ip is the current dir. xip is the parent dir).
 563                  */
 564                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 565                 if (ip->pip && ip != ip->pmp->iroot)
 566                         inum = ip->pip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 567                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
 568                 if (r)
 569                         goto done;
 570                 if (cookies)
 571                         cookies[cookie_index] = saveoff;
 572                 ++saveoff;
 573                 ++cookie_index;
 574                 if (cookie_index == ncookies)
 575                         goto done;
 576         }
 577
 578         lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
 579         if (hammer2_debug & 0x0020)
 580                 kprintf("readdir: lkey %016jx\n", lkey);
 581         if (error)
 582                 goto done;
 583
 584         /*
 585          * Use XOP for cluster scan.
 586          *
 587          * parent is the inode cluster, already locked for us.  Don't
 588          * double lock shared locks as this will screw up upgrades.
 589          */
 590         xop = hammer2_xop_alloc(ip, 0);
 591         xop->lkey = lkey;
 592         hammer2_xop_start(&xop->head, hammer2_xop_readdir);
 593
 594         for (;;) {
 595                 const hammer2_inode_data_t *ripdata;
 596
 597                 error = hammer2_xop_collect(&xop->head, 0);
 598                 if (error)
 599                         break;
 600                 if (cookie_index == ncookies)
 601                         break;
 602                 if (hammer2_debug & 0x0020)
 603                 kprintf("cluster chain %p %p\n",
 604                         xop->head.cluster.focus,
 605                         (xop->head.cluster.focus ?
 606                          xop->head.cluster.focus->data : (void *)-1));
 607                 ripdata = &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
 608                 hammer2_cluster_bref(&xop->head.cluster, &bref);
 609                 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
 610                         dtype = hammer2_get_dtype(ripdata);
 611                         saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 612                         r = vop_write_dirent(&error, uio,
 613                                              ripdata->meta.inum &
 614                                               HAMMER2_DIRHASH_USERMSK,
 615                                              dtype,
 616                                              ripdata->meta.name_len,
 617                                              ripdata->filename);
 618                         if (r)
 619                                 break;
 620                         if (cookies)
 621                                 cookies[cookie_index] = saveoff;
 622                         ++cookie_index;
 623                 } else {
 624                         /* XXX chain error */
 625                         kprintf("bad chain type readdir %d\n", bref.type);
 626                 }
 627         }
 628         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
 629         if (error == ENOENT) {
 630                 error = 0;
 631                 eofflag = 1;
 632                 saveoff = (hammer2_key_t)-1;
 633         } else {
 634                 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 635         }
 636 done:
 637         hammer2_inode_unlock(ip);
 638         if (ap->a_eofflag)
 639                 *ap->a_eofflag = eofflag;
 640         if (hammer2_debug & 0x0020)
 641                 kprintf("readdir: done at %016jx\n", saveoff);
 642         uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
 643         if (error && cookie_index == 0) {
 644                 if (cookies) {
 645                         kfree(cookies, M_TEMP);
 646                         *ap->a_ncookies = 0;
 647                         *ap->a_cookies = NULL;
 648                 }
 649         } else {
 650                 if (cookies) {
 651                         *ap->a_ncookies = cookie_index;
 652                         *ap->a_cookies = cookies;
 653                 }
 654         }
 655         LOCKSTOP;
 656         return (error);
 657 }
 658
 659 /*
 660  * hammer2_vop_readlink { vp, uio, cred }
 661  */
 662 static
 663 int
 664 hammer2_vop_readlink(struct vop_readlink_args *ap)
 665 {
 666         struct vnode *vp;
 667         hammer2_inode_t *ip;
 668         int error;
 669
 670         vp = ap->a_vp;
 671         if (vp->v_type != VLNK)
 672                 return (EINVAL);
 673         ip = VTOI(vp);
 674
 675         error = hammer2_read_file(ip, ap->a_uio, 0);
 676         return (error);
 677 }
 678
 679 static
 680 int
 681 hammer2_vop_read(struct vop_read_args *ap)
 682 {
 683         struct vnode *vp;
 684         hammer2_inode_t *ip;
 685         struct uio *uio;
 686         int error;
 687         int seqcount;
 688         int bigread;
 689
 690         /*
 691          * Read operations supported on this vnode?
 692          */
 693         vp = ap->a_vp;
 694         if (vp->v_type != VREG)
 695                 return (EINVAL);
 696
 697         /*
 698          * Misc
 699          */
 700         ip = VTOI(vp);
 701         uio = ap->a_uio;
 702         error = 0;
 703
 704         seqcount = ap->a_ioflag >> 16;
 705         bigread = (uio->uio_resid > 100 * 1024 * 1024);
 706
 707         error = hammer2_read_file(ip, uio, seqcount);
 708         return (error);
 709 }
 710
 711 static
 712 int
 713 hammer2_vop_write(struct vop_write_args *ap)
 714 {
 715         hammer2_inode_t *ip;
 716         thread_t td;
 717         struct vnode *vp;
 718         struct uio *uio;
 719         int error;
 720         int seqcount;
 721
 722         /*
 723          * Read operations supported on this vnode?
 724          */
 725         vp = ap->a_vp;
 726         if (vp->v_type != VREG)
 727                 return (EINVAL);
 728
 729         /*
 730          * Misc
 731          */
 732         ip = VTOI(vp);
 733         uio = ap->a_uio;
 734         error = 0;
 735         if (ip->pmp->ronly) {
 736                 return (EROFS);
 737         }
 738
 739         seqcount = ap->a_ioflag >> 16;
 740
 741         /*
 742          * Check resource limit
 743          */
 744         if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
 745             uio->uio_offset + uio->uio_resid >
 746              td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 747                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 748                 return (EFBIG);
 749         }
 750
 751         /*
 752          * The transaction interlocks against flushes initiations
 753          * (note: but will run concurrently with the actual flush).
 754          */
 755         hammer2_trans_init(ip->pmp, 0);
 756         error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
 757         hammer2_trans_done(ip->pmp);
 758
 759         return (error);
 760 }
 761
 762 /*
 763  * Perform read operations on a file or symlink given an UNLOCKED
 764  * inode and uio.
 765  *
 766  * The passed ip is not locked.
 767  */
 768 static
 769 int
 770 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
 771 {
 772         hammer2_off_t size;
 773         struct buf *bp;
 774         int error;
 775
 776         error = 0;
 777
 778         /*
 779          * UIO read loop.
 780          *
 781          * WARNING! Assumes that the kernel interlocks size changes at the
 782          *          vnode level.
 783          */
 784         hammer2_mtx_sh(&ip->lock);
 785         size = ip->meta.size;
 786         hammer2_mtx_unlock(&ip->lock);
 787
 788         while (uio->uio_resid > 0 && uio->uio_offset < size) {
 789                 hammer2_key_t lbase;
 790                 hammer2_key_t leof;
 791                 int lblksize;
 792                 int loff;
 793                 int n;
 794
 795                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 796                                                 &lbase, &leof);
 797
 798                 error = cluster_read(ip->vp, leof, lbase, lblksize,
 799                                      uio->uio_resid, seqcount * BKVASIZE,
 800                                      &bp);
 801
 802                 if (error)
 803                         break;
 804                 loff = (int)(uio->uio_offset - lbase);
 805                 n = lblksize - loff;
 806                 if (n > uio->uio_resid)
 807                         n = uio->uio_resid;
 808                 if (n > size - uio->uio_offset)
 809                         n = (int)(size - uio->uio_offset);
 810                 bp->b_flags |= B_AGE;
 811                 uiomove((char *)bp->b_data + loff, n, uio);
 812                 bqrelse(bp);
 813         }
 814         return (error);
 815 }
 816
 817 /*
 818  * Write to the file represented by the inode via the logical buffer cache.
 819  * The inode may represent a regular file or a symlink.
 820  *
 821  * The inode must not be locked.
 822  */
 823 static
 824 int
 825 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
 826                    int ioflag, int seqcount)
 827 {
 828         hammer2_key_t old_eof;
 829         hammer2_key_t new_eof;
 830         struct buf *bp;
 831         int kflags;
 832         int error;
 833         int modified;
 834
 835         /*
 836          * Setup if append
 837          *
 838          * WARNING! Assumes that the kernel interlocks size changes at the
 839          *          vnode level.
 840          */
 841         hammer2_mtx_ex(&ip->lock);
 842         if (ioflag & IO_APPEND)
 843                 uio->uio_offset = ip->meta.size;
 844         old_eof = ip->meta.size;
 845
 846         /*
 847          * Extend the file if necessary.  If the write fails at some point
 848          * we will truncate it back down to cover as much as we were able
 849          * to write.
 850          *
 851          * Doing this now makes it easier to calculate buffer sizes in
 852          * the loop.
 853          */
 854         kflags = 0;
 855         error = 0;
 856         modified = 0;
 857
 858         if (uio->uio_offset + uio->uio_resid > old_eof) {
 859                 new_eof = uio->uio_offset + uio->uio_resid;
 860                 modified = 1;
 861                 hammer2_extend_file(ip, new_eof);
 862                 kflags |= NOTE_EXTEND;
 863         } else {
 864                 new_eof = old_eof;
 865         }
 866         hammer2_mtx_unlock(&ip->lock);
 867
 868         /*
 869          * UIO write loop
 870          */
 871         while (uio->uio_resid > 0) {
 872                 hammer2_key_t lbase;
 873                 int trivial;
 874                 int endofblk;
 875                 int lblksize;
 876                 int loff;
 877                 int n;
 878
 879                 /*
 880                  * Don't allow the buffer build to blow out the buffer
 881                  * cache.
 882                  */
 883                 if ((ioflag & IO_RECURSE) == 0)
 884                         bwillwrite(HAMMER2_PBUFSIZE);
 885
 886                 /*
 887                  * This nominally tells us how much we can cluster and
 888                  * what the logical buffer size needs to be.  Currently
 889                  * we don't try to cluster the write and just handle one
 890                  * block at a time.
 891                  */
 892                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 893                                                 &lbase, NULL);
 894                 loff = (int)(uio->uio_offset - lbase);
 895
 896                 KKASSERT(lblksize <= 65536);
 897
 898                 /*
 899                  * Calculate bytes to copy this transfer and whether the
 900                  * copy completely covers the buffer or not.
 901                  */
 902                 trivial = 0;
 903                 n = lblksize - loff;
 904                 if (n > uio->uio_resid) {
 905                         n = uio->uio_resid;
 906                         if (loff == lbase && uio->uio_offset + n == new_eof)
 907                                 trivial = 1;
 908                         endofblk = 0;
 909                 } else {
 910                         if (loff == 0)
 911                                 trivial = 1;
 912                         endofblk = 1;
 913                 }
 914
 915                 /*
 916                  * Get the buffer
 917                  */
 918                 if (uio->uio_segflg == UIO_NOCOPY) {
 919                         /*
 920                          * Issuing a write with the same data backing the
 921                          * buffer.  Instantiate the buffer to collect the
 922                          * backing vm pages, then read-in any missing bits.
 923                          *
 924                          * This case is used by vop_stdputpages().
 925                          */
 926                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 927                         if ((bp->b_flags & B_CACHE) == 0) {
 928                                 bqrelse(bp);
 929                                 error = bread(ip->vp, lbase, lblksize, &bp);
 930                         }
 931                 } else if (trivial) {
 932                         /*
 933                          * Even though we are entirely overwriting the buffer
 934                          * we may still have to zero it out to avoid a
 935                          * mmap/write visibility issue.
 936                          */
 937                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 938                         if ((bp->b_flags & B_CACHE) == 0)
 939                                 vfs_bio_clrbuf(bp);
 940                 } else {
 941                         /*
 942                          * Partial overwrite, read in any missing bits then
 943                          * replace the portion being written.
 944                          *
 945                          * (The strategy code will detect zero-fill physical
 946                          * blocks for this case).
 947                          */
 948                         error = bread(ip->vp, lbase, lblksize, &bp);
 949                         if (error == 0)
 950                                 bheavy(bp);
 951                 }
 952
 953                 if (error) {
 954                         brelse(bp);
 955                         break;
 956                 }
 957
 958                 /*
 959                  * Ok, copy the data in
 960                  */
 961                 error = uiomove(bp->b_data + loff, n, uio);
 962                 kflags |= NOTE_WRITE;
 963                 modified = 1;
 964                 if (error) {
 965                         brelse(bp);
 966                         break;
 967                 }
 968
 969                 /*
 970                  * WARNING: Pageout daemon will issue UIO_NOCOPY writes
 971                  *          with IO_SYNC or IO_ASYNC set.  These writes
 972                  *          must be handled as the pageout daemon expects.
 973                  */
 974                 if (ioflag & IO_SYNC) {
 975                         bwrite(bp);
 976                 } else if ((ioflag & IO_DIRECT) && endofblk) {
 977                         bawrite(bp);
 978                 } else if (ioflag & IO_ASYNC) {
 979                         bawrite(bp);
 980                 } else {
 981                         bdwrite(bp);
 982                 }
 983         }
 984
 985         /*
 986          * Cleanup.  If we extended the file EOF but failed to write through
 987          * the entire write is a failure and we have to back-up.
 988          */
 989         if (error && new_eof != old_eof) {
 990                 hammer2_mtx_ex(&ip->lock);
 991                 hammer2_truncate_file(ip, old_eof);
 992                 if (ip->flags & HAMMER2_INODE_MODIFIED)
 993                         hammer2_inode_chain_sync(ip);
 994                 hammer2_mtx_unlock(&ip->lock);
 995         } else if (modified) {
 996                 hammer2_mtx_ex(&ip->lock);
 997                 hammer2_inode_modify(ip);
 998                 hammer2_update_time(&ip->meta.mtime);
 999                 if (ip->flags & HAMMER2_INODE_MODIFIED)
1000                         hammer2_inode_chain_sync(ip);
1001                 hammer2_mtx_unlock(&ip->lock);
1002                 hammer2_knote(ip->vp, kflags);
1003         }
1004         hammer2_trans_assert_strategy(ip->pmp);
1005
1006         return error;
1007 }
1008
1009 /*
1010  * Truncate the size of a file.  The inode must not be locked.
1011  *
1012  * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1013  * ensure that any on-media data beyond the new file EOF has been destroyed.
1014  *
1015  * WARNING: nvtruncbuf() can only be safely called without the inode lock
1016  *          held due to the way our write thread works.  If the truncation
1017  *          occurs in the middle of a buffer, nvtruncbuf() is responsible
1018  *          for dirtying that buffer and zeroing out trailing bytes.
1019  *
1020  * WARNING! Assumes that the kernel interlocks size changes at the
1021  *          vnode level.
1022  *
1023  * WARNING! Caller assumes responsibility for removing dead blocks
1024  *          if INODE_RESIZED is set.
1025  */
1026 static
1027 void
1028 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1029 {
1030         hammer2_key_t lbase;
1031         int nblksize;
1032
1033         LOCKSTART;
1034         hammer2_mtx_unlock(&ip->lock);
1035         if (ip->vp) {
1036                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1037                 nvtruncbuf(ip->vp, nsize,
1038                            nblksize, (int)nsize & (nblksize - 1),
1039                            0);
1040         }
1041         hammer2_mtx_ex(&ip->lock);
1042         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1043         ip->osize = ip->meta.size;
1044         ip->meta.size = nsize;
1045         atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1046         hammer2_inode_modify(ip);
1047         LOCKSTOP;
1048 }
1049
1050 /*
1051  * Extend the size of a file.  The inode must not be locked.
1052  *
1053  * Even though the file size is changing, we do not have to set the
1054  * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1055  * boundary.  When this occurs a hammer2_inode_chain_sync() is required
1056  * to prepare the inode cluster's indirect block table, otherwise
1057  * async execution of the strategy code will implode on us.
1058  *
1059  * WARNING! Assumes that the kernel interlocks size changes at the
1060  *          vnode level.
1061  *
1062  * WARNING! Caller assumes responsibility for transitioning out
1063  *          of the inode DIRECTDATA mode if INODE_RESIZED is set.
1064  */
1065 static
1066 void
1067 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1068 {
1069         hammer2_key_t lbase;
1070         hammer2_key_t osize;
1071         int oblksize;
1072         int nblksize;
1073
1074         LOCKSTART;
1075
1076         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1077         hammer2_inode_modify(ip);
1078         osize = ip->meta.size;
1079         ip->osize = osize;
1080         ip->meta.size = nsize;
1081
1082         if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1083                 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1084                 hammer2_inode_chain_sync(ip);
1085         }
1086
1087         hammer2_mtx_unlock(&ip->lock);
1088         if (ip->vp) {
1089                 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1090                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1091                 nvextendbuf(ip->vp,
1092                             osize, nsize,
1093                             oblksize, nblksize,
1094                             -1, -1, 0);
1095         }
1096         hammer2_mtx_ex(&ip->lock);
1097
1098         LOCKSTOP;
1099 }
1100
1101 static
1102 int
1103 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1104 {
1105         hammer2_xop_nresolve_t *xop;
1106         hammer2_inode_t *ip;
1107         hammer2_inode_t *dip;
1108         struct namecache *ncp;
1109         struct vnode *vp;
1110         int error;
1111
1112         LOCKSTART;
1113         dip = VTOI(ap->a_dvp);
1114         xop = hammer2_xop_alloc(dip, 0);
1115
1116         ncp = ap->a_nch->ncp;
1117         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1118
1119         /*
1120          * Note: In DragonFly the kernel handles '.' and '..'.
1121          */
1122         hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1123         hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1124
1125         error = hammer2_xop_collect(&xop->head, 0);
1126         if (error) {
1127                 ip = NULL;
1128         } else {
1129                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1130         }
1131         hammer2_inode_unlock(dip);
1132
1133         /*
1134          * Acquire the related vnode
1135          *
1136          * NOTE: For error processing, only ENOENT resolves the namecache
1137          *       entry to NULL, otherwise we just return the error and
1138          *       leave the namecache unresolved.
1139          *
1140          * NOTE: multiple hammer2_inode structures can be aliased to the
1141          *       same chain element, for example for hardlinks.  This
1142          *       use case does not 'reattach' inode associations that
1143          *       might already exist, but always allocates a new one.
1144          *
1145          * WARNING: inode structure is locked exclusively via inode_get
1146          *          but chain was locked shared.  inode_unlock()
1147          *          will handle it properly.
1148          */
1149         if (ip) {
1150                 vp = hammer2_igetv(ip, &error);
1151                 if (error == 0) {
1152                         vn_unlock(vp);
1153                         cache_setvp(ap->a_nch, vp);
1154                 } else if (error == ENOENT) {
1155                         cache_setvp(ap->a_nch, NULL);
1156                 }
1157                 hammer2_inode_unlock(ip);
1158
1159                 /*
1160                  * The vp should not be released until after we've disposed
1161                  * of our locks, because it might cause vop_inactive() to
1162                  * be called.
1163                  */
1164                 if (vp)
1165                         vrele(vp);
1166         } else {
1167                 error = ENOENT;
1168                 cache_setvp(ap->a_nch, NULL);
1169         }
1170         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1171         KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1172                 ("resolve error %d/%p ap %p\n",
1173                  error, ap->a_nch->ncp->nc_vp, ap));
1174         LOCKSTOP;
1175
1176         return error;
1177 }
1178
1179 static
1180 int
1181 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1182 {
1183         hammer2_inode_t *dip;
1184         hammer2_inode_t *ip;
1185         int error;
1186
1187         LOCKSTART;
1188         dip = VTOI(ap->a_dvp);
1189
1190         if ((ip = dip->pip) == NULL) {
1191                 *ap->a_vpp = NULL;
1192                 LOCKSTOP;
1193                 return ENOENT;
1194         }
1195         hammer2_inode_lock(ip, 0);
1196         *ap->a_vpp = hammer2_igetv(ip, &error);
1197         hammer2_inode_unlock(ip);
1198
1199         LOCKSTOP;
1200         return error;
1201 }
1202
1203 static
1204 int
1205 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1206 {
1207         hammer2_inode_t *dip;
1208         hammer2_inode_t *nip;
1209         struct namecache *ncp;
1210         const uint8_t *name;
1211         size_t name_len;
1212         int error;
1213
1214         LOCKSTART;
1215         dip = VTOI(ap->a_dvp);
1216         if (dip->pmp->ronly) {
1217                 LOCKSTOP;
1218                 return (EROFS);
1219         }
1220
1221         ncp = ap->a_nch->ncp;
1222         name = ncp->nc_name;
1223         name_len = ncp->nc_nlen;
1224
1225         hammer2_pfs_memory_wait(dip->pmp);
1226         hammer2_trans_init(dip->pmp, 0);
1227         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1228                                    name, name_len, 0,
1229                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1230                                    0, &error);
1231         if (error) {
1232                 KKASSERT(nip == NULL);
1233                 *ap->a_vpp = NULL;
1234         } else {
1235                 *ap->a_vpp = hammer2_igetv(nip, &error);
1236                 hammer2_inode_unlock(nip);
1237         }
1238         hammer2_trans_done(dip->pmp);
1239
1240         if (error == 0) {
1241                 cache_setunresolved(ap->a_nch);
1242                 cache_setvp(ap->a_nch, *ap->a_vpp);
1243         }
1244         LOCKSTOP;
1245         return error;
1246 }
1247
1248 static
1249 int
1250 hammer2_vop_open(struct vop_open_args *ap)
1251 {
1252         return vop_stdopen(ap);
1253 }
1254
1255 /*
1256  * hammer2_vop_advlock { vp, id, op, fl, flags }
1257  */
1258 static
1259 int
1260 hammer2_vop_advlock(struct vop_advlock_args *ap)
1261 {
1262         hammer2_inode_t *ip = VTOI(ap->a_vp);
1263         hammer2_off_t size;
1264
1265         size = ip->meta.size;
1266         return (lf_advlock(ap, &ip->advlock, size));
1267 }
1268
1269 static
1270 int
1271 hammer2_vop_close(struct vop_close_args *ap)
1272 {
1273         return vop_stdclose(ap);
1274 }
1275
1276 /*
1277  * hammer2_vop_nlink { nch, dvp, vp, cred }
1278  *
1279  * Create a hardlink from (vp) to {dvp, nch}.
1280  */
1281 static
1282 int
1283 hammer2_vop_nlink(struct vop_nlink_args *ap)
1284 {
1285         hammer2_xop_nlink_t *xop1;
1286         hammer2_inode_t *fdip;  /* target directory to create link in */
1287         hammer2_inode_t *tdip;  /* target directory to create link in */
1288         hammer2_inode_t *cdip;  /* common parent directory */
1289         hammer2_inode_t *ip;    /* inode we are hardlinking to */
1290         struct namecache *ncp;
1291         const uint8_t *name;
1292         size_t name_len;
1293         int nlink_locked;
1294         int error;
1295
1296         LOCKSTART;
1297         tdip = VTOI(ap->a_dvp);
1298         if (tdip->pmp->ronly) {
1299                 LOCKSTOP;
1300                 return (EROFS);
1301         }
1302
1303         ncp = ap->a_nch->ncp;
1304         name = ncp->nc_name;
1305         name_len = ncp->nc_nlen;
1306
1307         /*
1308          * ip represents the file being hardlinked.  The file could be a
1309          * normal file or a hardlink target if it has already been hardlinked.
1310          * If ip is a hardlinked target then ip->pip represents the location
1311          * of the hardlinked target, NOT the location of the hardlink pointer.
1312          *
1313          * Bump nlinks and potentially also create or move the hardlink
1314          * target in the parent directory common to (ip) and (tdip).  The
1315          * consolidation code can modify ip->cluster and ip->pip.  The
1316          * returned cluster is locked.
1317          */
1318         ip = VTOI(ap->a_vp);
1319         hammer2_pfs_memory_wait(ip->pmp);
1320         hammer2_trans_init(ip->pmp, 0);
1321
1322         /*
1323          * The common parent directory must be locked first to avoid deadlocks.
1324          * Also note that fdip and/or tdip might match cdip.
1325          *
1326          * WARNING!  The kernel's namecache locks are insufficient for
1327          *           protecting us from hardlink shifts, since unrelated
1328          *           rename() or link() calls on parent directories might
1329          *           cause a shift.  A PFS-wide lock is required for this
1330          *           situation.
1331          */
1332         if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY ||
1333             (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
1334                 lockmgr(&ip->pmp->lock_nlink, LK_EXCLUSIVE);
1335                 nlink_locked = 1;
1336         } else {
1337                 nlink_locked = 0;
1338         }
1339         fdip = ip->pip;
1340         cdip = hammer2_inode_common_parent(fdip, tdip);
1341         hammer2_inode_lock(cdip, 0);
1342         hammer2_inode_lock(fdip, 0);
1343         hammer2_inode_lock(tdip, 0);
1344         hammer2_inode_lock(ip, 0);
1345         error = 0;
1346
1347         /*
1348          * Dispatch xop_nlink unconditionally since we have to update nlinks.
1349          *
1350          * Otherwise we'd be able to avoid the XOP if the ip does not have
1351          * to be converted or moved.
1352          * If ip is not a hardlink target we must convert it to a hardlink.
1353          * If fdip != cdip we must shift the inode to cdip.
1354          *
1355          * XXX this and other nlink update usage should be passed top-down
1356          *     and not updated with a delta bottom-up.
1357          */
1358 #if 0
1359         if (fdip != cdip || (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE))
1360 #endif
1361         {
1362                 xop1 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1363                 hammer2_xop_setip2(&xop1->head, ip);
1364                 hammer2_xop_setip3(&xop1->head, cdip);
1365                 xop1->nlinks_delta = 1;
1366
1367                 hammer2_xop_start(&xop1->head, hammer2_xop_nlink);
1368                 error = hammer2_xop_collect(&xop1->head, 0);
1369                 hammer2_xop_retire(&xop1->head, HAMMER2_XOPMASK_VOP);
1370                 if (error == ENOENT)
1371                         error = 0;
1372         }
1373
1374         /*
1375          * Must synchronize original inode whos chains are now a hardlink
1376          * target.  We must match what the backend XOP did to the
1377          * chains.
1378          */
1379         if (error == 0 && (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1380                 hammer2_inode_modify(ip);
1381                 ip->meta.name_key = ip->meta.inum;
1382                 ip->meta.name_len = 18; /* "0x%016jx" */
1383         }
1384
1385         /*
1386          * Create the hardlink target and bump nlinks.
1387          */
1388         if (error == 0) {
1389                 hammer2_inode_create(tdip, NULL, NULL,
1390                                      name, name_len, 0,
1391                                      ip->meta.inum,
1392                                      HAMMER2_OBJTYPE_HARDLINK, ip->meta.type,
1393                                      0, &error);
1394                 hammer2_inode_modify(ip);
1395                 ++ip->meta.nlinks;
1396         }
1397         if (error == 0) {
1398                 cache_setunresolved(ap->a_nch);
1399                 cache_setvp(ap->a_nch, ap->a_vp);
1400         }
1401         hammer2_inode_unlock(ip);
1402         hammer2_inode_unlock(tdip);
1403         hammer2_inode_unlock(fdip);
1404         hammer2_inode_unlock(cdip);
1405         hammer2_inode_drop(cdip);
1406
1407         if (nlink_locked)
1408                 lockmgr(&ip->pmp->lock_nlink, LK_RELEASE);
1409         hammer2_trans_done(ip->pmp);
1410
1411         LOCKSTOP;
1412         return error;
1413 }
1414
1415 /*
1416  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1417  *
1418  * The operating system has already ensured that the directory entry
1419  * does not exist and done all appropriate namespace locking.
1420  */
1421 static
1422 int
1423 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1424 {
1425         hammer2_inode_t *dip;
1426         hammer2_inode_t *nip;
1427         struct namecache *ncp;
1428         const uint8_t *name;
1429         size_t name_len;
1430         int error;
1431
1432         LOCKSTART;
1433         dip = VTOI(ap->a_dvp);
1434         if (dip->pmp->ronly) {
1435                 LOCKSTOP;
1436                 return (EROFS);
1437         }
1438
1439         ncp = ap->a_nch->ncp;
1440         name = ncp->nc_name;
1441         name_len = ncp->nc_nlen;
1442         hammer2_pfs_memory_wait(dip->pmp);
1443         hammer2_trans_init(dip->pmp, 0);
1444
1445         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1446                                    name, name_len, 0,
1447                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1448                                    0, &error);
1449         if (error) {
1450                 KKASSERT(nip == NULL);
1451                 *ap->a_vpp = NULL;
1452         } else {
1453                 *ap->a_vpp = hammer2_igetv(nip, &error);
1454                 hammer2_inode_unlock(nip);
1455         }
1456         hammer2_trans_done(dip->pmp);
1457
1458         if (error == 0) {
1459                 cache_setunresolved(ap->a_nch);
1460                 cache_setvp(ap->a_nch, *ap->a_vpp);
1461         }
1462         LOCKSTOP;
1463         return error;
1464 }
1465
1466 /*
1467  * Make a device node (typically a fifo)
1468  */
1469 static
1470 int
1471 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1472 {
1473         hammer2_inode_t *dip;
1474         hammer2_inode_t *nip;
1475         struct namecache *ncp;
1476         const uint8_t *name;
1477         size_t name_len;
1478         int error;
1479
1480         LOCKSTART;
1481         dip = VTOI(ap->a_dvp);
1482         if (dip->pmp->ronly) {
1483                 LOCKSTOP;
1484                 return (EROFS);
1485         }
1486
1487         ncp = ap->a_nch->ncp;
1488         name = ncp->nc_name;
1489         name_len = ncp->nc_nlen;
1490         hammer2_pfs_memory_wait(dip->pmp);
1491         hammer2_trans_init(dip->pmp, 0);
1492
1493         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1494                                    name, name_len, 0,
1495                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1496                                    0, &error);
1497         if (error) {
1498                 KKASSERT(nip == NULL);
1499                 *ap->a_vpp = NULL;
1500         } else {
1501                 *ap->a_vpp = hammer2_igetv(nip, &error);
1502                 hammer2_inode_unlock(nip);
1503         }
1504         hammer2_trans_done(dip->pmp);
1505
1506         if (error == 0) {
1507                 cache_setunresolved(ap->a_nch);
1508                 cache_setvp(ap->a_nch, *ap->a_vpp);
1509         }
1510         LOCKSTOP;
1511         return error;
1512 }
1513
1514 /*
1515  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1516  */
1517 static
1518 int
1519 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1520 {
1521         hammer2_inode_t *dip;
1522         hammer2_inode_t *nip;
1523         struct namecache *ncp;
1524         const uint8_t *name;
1525         size_t name_len;
1526         int error;
1527
1528         dip = VTOI(ap->a_dvp);
1529         if (dip->pmp->ronly)
1530                 return (EROFS);
1531
1532         ncp = ap->a_nch->ncp;
1533         name = ncp->nc_name;
1534         name_len = ncp->nc_nlen;
1535         hammer2_pfs_memory_wait(dip->pmp);
1536         hammer2_trans_init(dip->pmp, 0);
1537
1538         ap->a_vap->va_type = VLNK;      /* enforce type */
1539
1540         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1541                                    name, name_len, 0,
1542                                    hammer2_trans_newinum(dip->pmp), 0, 0,
1543                                    0, &error);
1544         if (error) {
1545                 KKASSERT(nip == NULL);
1546                 *ap->a_vpp = NULL;
1547                 hammer2_trans_done(dip->pmp);
1548                 return error;
1549         }
1550         *ap->a_vpp = hammer2_igetv(nip, &error);
1551
1552         /*
1553          * Build the softlink (~like file data) and finalize the namecache.
1554          */
1555         if (error == 0) {
1556                 size_t bytes;
1557                 struct uio auio;
1558                 struct iovec aiov;
1559
1560                 bytes = strlen(ap->a_target);
1561
1562                 hammer2_inode_unlock(nip);
1563                 bzero(&auio, sizeof(auio));
1564                 bzero(&aiov, sizeof(aiov));
1565                 auio.uio_iov = &aiov;
1566                 auio.uio_segflg = UIO_SYSSPACE;
1567                 auio.uio_rw = UIO_WRITE;
1568                 auio.uio_resid = bytes;
1569                 auio.uio_iovcnt = 1;
1570                 auio.uio_td = curthread;
1571                 aiov.iov_base = ap->a_target;
1572                 aiov.iov_len = bytes;
1573                 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1574                 /* XXX handle error */
1575                 error = 0;
1576         } else {
1577                 hammer2_inode_unlock(nip);
1578         }
1579         hammer2_trans_done(dip->pmp);
1580
1581         /*
1582          * Finalize namecache
1583          */
1584         if (error == 0) {
1585                 cache_setunresolved(ap->a_nch);
1586                 cache_setvp(ap->a_nch, *ap->a_vpp);
1587                 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1588         }
1589         return error;
1590 }
1591
1592 /*
1593  * hammer2_vop_nremove { nch, dvp, cred }
1594  */
1595 static
1596 int
1597 hammer2_vop_nremove(struct vop_nremove_args *ap)
1598 {
1599         hammer2_xop_unlink_t *xop;
1600         hammer2_inode_t *dip;
1601         hammer2_inode_t *ip;
1602         struct namecache *ncp;
1603         int error;
1604         int isopen;
1605
1606         LOCKSTART;
1607         dip = VTOI(ap->a_dvp);
1608         if (dip->pmp->ronly) {
1609                 LOCKSTOP;
1610                 return(EROFS);
1611         }
1612
1613         ncp = ap->a_nch->ncp;
1614
1615         hammer2_pfs_memory_wait(dip->pmp);
1616         hammer2_trans_init(dip->pmp, 0);
1617         hammer2_inode_lock(dip, 0);
1618
1619         /*
1620          * The unlink XOP unlinks the path from the directory and
1621          * locates and returns the cluster associated with the real inode.
1622          * We have to handle nlinks here on the frontend.
1623          */
1624         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1625         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1626         isopen = cache_isopen(ap->a_nch);
1627         xop->isdir = 0;
1628         xop->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1629         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1630
1631         /*
1632          * Collect the real inode and adjust nlinks, destroy the real
1633          * inode if nlinks transitions to 0 and it was the real inode
1634          * (else it has already been removed).
1635          */
1636         error = hammer2_xop_collect(&xop->head, 0);
1637         hammer2_inode_unlock(dip);
1638
1639         if (error == 0) {
1640                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1641                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1642                 if (ip) {
1643                         hammer2_inode_unlink_finisher(ip, isopen);
1644                         hammer2_inode_unlock(ip);
1645                 }
1646         } else {
1647                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1648         }
1649
1650         hammer2_inode_run_sideq(dip->pmp);
1651         hammer2_trans_done(dip->pmp);
1652         if (error == 0)
1653                 cache_unlink(ap->a_nch);
1654         LOCKSTOP;
1655         return (error);
1656 }
1657
1658 /*
1659  * hammer2_vop_nrmdir { nch, dvp, cred }
1660  */
1661 static
1662 int
1663 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1664 {
1665         hammer2_xop_unlink_t *xop;
1666         hammer2_inode_t *dip;
1667         hammer2_inode_t *ip;
1668         struct namecache *ncp;
1669         int isopen;
1670         int error;
1671
1672         LOCKSTART;
1673         dip = VTOI(ap->a_dvp);
1674         if (dip->pmp->ronly) {
1675                 LOCKSTOP;
1676                 return(EROFS);
1677         }
1678
1679         hammer2_pfs_memory_wait(dip->pmp);
1680         hammer2_trans_init(dip->pmp, 0);
1681         hammer2_inode_lock(dip, 0);
1682
1683         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1684
1685         ncp = ap->a_nch->ncp;
1686         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1687         isopen = cache_isopen(ap->a_nch);
1688         xop->isdir = 1;
1689         xop->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1690         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1691
1692         /*
1693          * Collect the real inode and adjust nlinks, destroy the real
1694          * inode if nlinks transitions to 0 and it was the real inode
1695          * (else it has already been removed).
1696          */
1697         error = hammer2_xop_collect(&xop->head, 0);
1698         hammer2_inode_unlock(dip);
1699
1700         if (error == 0) {
1701                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1702                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1703                 if (ip) {
1704                         hammer2_inode_unlink_finisher(ip, isopen);
1705                         hammer2_inode_unlock(ip);
1706                 }
1707         } else {
1708                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1709         }
1710         hammer2_inode_run_sideq(dip->pmp);
1711         hammer2_trans_done(dip->pmp);
1712         if (error == 0)
1713                 cache_unlink(ap->a_nch);
1714         LOCKSTOP;
1715         return (error);
1716 }
1717
1718 /*
1719  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1720  */
1721 static
1722 int
1723 hammer2_vop_nrename(struct vop_nrename_args *ap)
1724 {
1725         struct namecache *fncp;
1726         struct namecache *tncp;
1727         hammer2_inode_t *cdip;
1728         hammer2_inode_t *fdip;
1729         hammer2_inode_t *tdip;
1730         hammer2_inode_t *ip;
1731         const uint8_t *fname;
1732         size_t fname_len;
1733         const uint8_t *tname;
1734         size_t tname_len;
1735         int error;
1736         int tnch_error;
1737         int nlink_locked;
1738         hammer2_key_t tlhc;
1739
1740         if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1741                 return(EXDEV);
1742         if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1743                 return(EXDEV);
1744
1745         fdip = VTOI(ap->a_fdvp);        /* source directory */
1746         tdip = VTOI(ap->a_tdvp);        /* target directory */
1747
1748         if (fdip->pmp->ronly)
1749                 return(EROFS);
1750
1751         LOCKSTART;
1752         fncp = ap->a_fnch->ncp;         /* entry name in source */
1753         fname = fncp->nc_name;
1754         fname_len = fncp->nc_nlen;
1755
1756         tncp = ap->a_tnch->ncp;         /* entry name in target */
1757         tname = tncp->nc_name;
1758         tname_len = tncp->nc_nlen;
1759
1760         hammer2_pfs_memory_wait(tdip->pmp);
1761         hammer2_trans_init(tdip->pmp, 0);
1762
1763         /*
1764          * ip is the inode being renamed.  If this is a hardlink then
1765          * ip represents the actual file and not the hardlink marker.
1766          */
1767         ip = VTOI(fncp->nc_vp);
1768
1769         /*
1770          * The common parent directory must be locked first to avoid deadlocks.
1771          * Also note that fdip and/or tdip might match cdip.
1772          *
1773          * WARNING!  The kernel's namecache locks are insufficient for
1774          *           protecting us from hardlink shifts, since unrelated
1775          *           rename() or link() calls on parent directories might
1776          *           cause a shift.  A PFS-wide lock is required for this
1777          *           situation.
1778          */
1779         if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY ||
1780             (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
1781                 lockmgr(&ip->pmp->lock_nlink, LK_EXCLUSIVE);
1782                 nlink_locked = 1;
1783         } else {
1784                 nlink_locked = 0;
1785         }
1786
1787         cdip = hammer2_inode_common_parent(ip->pip, tdip);
1788         hammer2_inode_lock(cdip, 0);
1789         hammer2_inode_lock(fdip, 0);
1790         hammer2_inode_lock(tdip, 0);
1791         hammer2_inode_ref(ip);          /* extra ref */
1792         error = 0;
1793
1794         /*
1795          * If ip is a hardlink target and fdip != cdip we must shift the
1796          * inode to cdip.
1797          */
1798         hammer2_inode_lock(ip, 0);
1799
1800         if (fdip != cdip &&
1801             (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
1802                 hammer2_xop_nlink_t *xop1;
1803
1804                 xop1 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1805                 hammer2_xop_setip2(&xop1->head, ip);
1806                 hammer2_xop_setip3(&xop1->head, cdip);
1807                 xop1->nlinks_delta = 0;
1808
1809                 hammer2_xop_start(&xop1->head, hammer2_xop_nlink);
1810                 error = hammer2_xop_collect(&xop1->head, 0);
1811                 hammer2_xop_retire(&xop1->head, HAMMER2_XOPMASK_VOP);
1812         }
1813         /* hammer2_inode_unlock(ip); */
1814
1815         /*
1816          * Delete the target namespace.
1817          */
1818         {
1819                 hammer2_xop_unlink_t *xop2;
1820                 hammer2_inode_t *tip;
1821                 int isopen;
1822
1823                 /*
1824                  * The unlink XOP unlinks the path from the directory and
1825                  * locates and returns the cluster associated with the real
1826                  * inode.  We have to handle nlinks here on the frontend.
1827                  */
1828                 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1829                 hammer2_xop_setname(&xop2->head, tname, tname_len);
1830                 isopen = cache_isopen(ap->a_tnch);
1831                 xop2->isdir = -1;
1832                 xop2->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1833                 hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
1834
1835                 /*
1836                  * Collect the real inode and adjust nlinks, destroy the real
1837                  * inode if nlinks transitions to 0 and it was the real inode
1838                  * (else it has already been removed).
1839                  */
1840                 tnch_error = hammer2_xop_collect(&xop2->head, 0);
1841                 /* hammer2_inode_unlock(tdip); */
1842
1843                 if (tnch_error == 0) {
1844                         tip = hammer2_inode_get(tdip->pmp, NULL,
1845                                                 &xop2->head.cluster, -1);
1846                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1847                         if (tip) {
1848                                 hammer2_inode_unlink_finisher(tip, isopen);
1849                                 hammer2_inode_unlock(tip);
1850                         }
1851                 } else {
1852                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1853                 }
1854                 /* hammer2_inode_lock(tdip, 0); */
1855
1856                 if (tnch_error && tnch_error != ENOENT) {
1857                         error = tnch_error;
1858                         goto done2;
1859                 }
1860         }
1861
1862         /*
1863          * Resolve the collision space for (tdip, tname, tname_len)
1864          *
1865          * tdip must be held exclusively locked to prevent races.
1866          */
1867         {
1868                 hammer2_xop_scanlhc_t *sxop;
1869                 hammer2_tid_t lhcbase;
1870
1871                 tlhc = hammer2_dirhash(tname, tname_len);
1872                 lhcbase = tlhc;
1873                 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1874                 sxop->lhc = tlhc;
1875                 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
1876                 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
1877                         if (tlhc != sxop->head.cluster.focus->bref.key)
1878                                 break;
1879                         ++tlhc;
1880                 }
1881                 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
1882
1883                 if (error) {
1884                         if (error != ENOENT)
1885                                 goto done2;
1886                         ++tlhc;
1887                         error = 0;
1888                 }
1889                 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
1890                         error = ENOSPC;
1891                         goto done2;
1892                 }
1893         }
1894
1895         /*
1896          * Everything is setup, do the rename.
1897          *
1898          * We have to synchronize ip->meta to the underlying operation.
1899          *
1900          * NOTE: To avoid deadlocks we cannot lock (ip) while we are
1901          *       unlinking elements from their directories.  Locking
1902          *       the nlinks field does not lock the whole inode.
1903          */
1904         /* hammer2_inode_lock(ip, 0); */
1905         if (error == 0) {
1906                 hammer2_xop_nrename_t *xop4;
1907
1908                 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1909                 xop4->lhc = tlhc;
1910                 xop4->ip_key = ip->meta.name_key;
1911                 hammer2_xop_setip2(&xop4->head, ip);
1912                 hammer2_xop_setip3(&xop4->head, tdip);
1913                 hammer2_xop_setname(&xop4->head, fname, fname_len);
1914                 hammer2_xop_setname2(&xop4->head, tname, tname_len);
1915                 hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
1916
1917                 error = hammer2_xop_collect(&xop4->head, 0);
1918                 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
1919
1920                 if (error == ENOENT)
1921                         error = 0;
1922                 if (error == 0 &&
1923                     (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1924                         hammer2_inode_modify(ip);
1925                         ip->meta.name_len = tname_len;
1926                         ip->meta.name_key = tlhc;
1927
1928                 }
1929         }
1930
1931         /*
1932          * Fixup ip->pip if we were renaming the actual file and not a
1933          * hardlink pointer.
1934          */
1935         if (error == 0 && (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1936                 hammer2_inode_t *opip;
1937
1938                 if (ip->pip != tdip) {
1939                         hammer2_inode_ref(tdip);
1940                         opip = ip->pip;
1941                         ip->pip = tdip;
1942                         if (opip)
1943                                 hammer2_inode_drop(opip);
1944                 }
1945         }
1946 done2:
1947         hammer2_inode_unlock(ip);
1948         hammer2_inode_unlock(tdip);
1949         hammer2_inode_unlock(fdip);
1950         hammer2_inode_unlock(cdip);
1951         hammer2_inode_drop(ip);
1952         hammer2_inode_drop(cdip);
1953         hammer2_inode_run_sideq(fdip->pmp);
1954
1955         if (nlink_locked)
1956                 lockmgr(&ip->pmp->lock_nlink, LK_RELEASE);
1957         hammer2_trans_done(tdip->pmp);
1958
1959         /*
1960          * Issue the namecache update after unlocking all the internal
1961          * hammer structures, otherwise we might deadlock.
1962          */
1963         if (tnch_error == 0) {
1964                 cache_unlink(ap->a_tnch);
1965                 cache_setunresolved(ap->a_tnch);
1966         }
1967         if (error == 0)
1968                 cache_rename(ap->a_fnch, ap->a_tnch);
1969
1970         LOCKSTOP;
1971         return (error);
1972 }
1973
1974 /*
1975  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
1976  */
1977 static
1978 int
1979 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
1980 {
1981         hammer2_inode_t *ip;
1982         int error;
1983
1984         LOCKSTART;
1985         ip = VTOI(ap->a_vp);
1986
1987         error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
1988                               ap->a_fflag, ap->a_cred);
1989         LOCKSTOP;
1990         return (error);
1991 }
1992
1993 static
1994 int
1995 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
1996 {
1997         struct mount *mp;
1998         hammer2_pfs_t *pmp;
1999         int rc;
2000
2001         LOCKSTART;
2002         switch (ap->a_op) {
2003         case (MOUNTCTL_SET_EXPORT):
2004                 mp = ap->a_head.a_ops->head.vv_mount;
2005                 pmp = MPTOPMP(mp);
2006
2007                 if (ap->a_ctllen != sizeof(struct export_args))
2008                         rc = (EINVAL);
2009                 else
2010                         rc = vfs_export(mp, &pmp->export,
2011                                         (const struct export_args *)ap->a_ctl);
2012                 break;
2013         default:
2014                 rc = vop_stdmountctl(ap);
2015                 break;
2016         }
2017         LOCKSTOP;
2018         return (rc);
2019 }
2020
2021 /*
2022  * KQFILTER
2023  */
2024 static void filt_hammer2detach(struct knote *kn);
2025 static int filt_hammer2read(struct knote *kn, long hint);
2026 static int filt_hammer2write(struct knote *kn, long hint);
2027 static int filt_hammer2vnode(struct knote *kn, long hint);
2028
2029 static struct filterops hammer2read_filtops =
2030         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2031           NULL, filt_hammer2detach, filt_hammer2read };
2032 static struct filterops hammer2write_filtops =
2033         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2034           NULL, filt_hammer2detach, filt_hammer2write };
2035 static struct filterops hammer2vnode_filtops =
2036         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2037           NULL, filt_hammer2detach, filt_hammer2vnode };
2038
2039 static
2040 int
2041 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2042 {
2043         struct vnode *vp = ap->a_vp;
2044         struct knote *kn = ap->a_kn;
2045
2046         switch (kn->kn_filter) {
2047         case EVFILT_READ:
2048                 kn->kn_fop = &hammer2read_filtops;
2049                 break;
2050         case EVFILT_WRITE:
2051                 kn->kn_fop = &hammer2write_filtops;
2052                 break;
2053         case EVFILT_VNODE:
2054                 kn->kn_fop = &hammer2vnode_filtops;
2055                 break;
2056         default:
2057                 return (EOPNOTSUPP);
2058         }
2059
2060         kn->kn_hook = (caddr_t)vp;
2061
2062         knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2063
2064         return(0);
2065 }
2066
2067 static void
2068 filt_hammer2detach(struct knote *kn)
2069 {
2070         struct vnode *vp = (void *)kn->kn_hook;
2071
2072         knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2073 }
2074
2075 static int
2076 filt_hammer2read(struct knote *kn, long hint)
2077 {
2078         struct vnode *vp = (void *)kn->kn_hook;
2079         hammer2_inode_t *ip = VTOI(vp);
2080         off_t off;
2081
2082         if (hint == NOTE_REVOKE) {
2083                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2084                 return(1);
2085         }
2086         off = ip->meta.size - kn->kn_fp->f_offset;
2087         kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2088         if (kn->kn_sfflags & NOTE_OLDAPI)
2089                 return(1);
2090         return (kn->kn_data != 0);
2091 }
2092
2093
2094 static int
2095 filt_hammer2write(struct knote *kn, long hint)
2096 {
2097         if (hint == NOTE_REVOKE)
2098                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2099         kn->kn_data = 0;
2100         return (1);
2101 }
2102
2103 static int
2104 filt_hammer2vnode(struct knote *kn, long hint)
2105 {
2106         if (kn->kn_sfflags & hint)
2107                 kn->kn_fflags |= hint;
2108         if (hint == NOTE_REVOKE) {
2109                 kn->kn_flags |= (EV_EOF | EV_NODATA);
2110                 return (1);
2111         }
2112         return (kn->kn_fflags != 0);
2113 }
2114
2115 /*
2116  * FIFO VOPS
2117  */
2118 static
2119 int
2120 hammer2_vop_markatime(struct vop_markatime_args *ap)
2121 {
2122         hammer2_inode_t *ip;
2123         struct vnode *vp;
2124
2125         vp = ap->a_vp;
2126         ip = VTOI(vp);
2127
2128         if (ip->pmp->ronly)
2129                 return(EROFS);
2130         return(0);
2131 }
2132
2133 static
2134 int
2135 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2136 {
2137         int error;
2138
2139         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2140         if (error)
2141                 error = hammer2_vop_kqfilter(ap);
2142         return(error);
2143 }
2144
2145 /*
2146  * VOPS vector
2147  */
2148 struct vop_ops hammer2_vnode_vops = {
2149         .vop_default    = vop_defaultop,
2150         .vop_fsync      = hammer2_vop_fsync,
2151         .vop_getpages   = vop_stdgetpages,
2152         .vop_putpages   = vop_stdputpages,
2153         .vop_access     = hammer2_vop_access,
2154         .vop_advlock    = hammer2_vop_advlock,
2155         .vop_close      = hammer2_vop_close,
2156         .vop_nlink      = hammer2_vop_nlink,
2157         .vop_ncreate    = hammer2_vop_ncreate,
2158         .vop_nsymlink   = hammer2_vop_nsymlink,
2159         .vop_nremove    = hammer2_vop_nremove,
2160         .vop_nrmdir     = hammer2_vop_nrmdir,
2161         .vop_nrename    = hammer2_vop_nrename,
2162         .vop_getattr    = hammer2_vop_getattr,
2163         .vop_setattr    = hammer2_vop_setattr,
2164         .vop_readdir    = hammer2_vop_readdir,
2165         .vop_readlink   = hammer2_vop_readlink,
2166         .vop_getpages   = vop_stdgetpages,
2167         .vop_putpages   = vop_stdputpages,
2168         .vop_read       = hammer2_vop_read,
2169         .vop_write      = hammer2_vop_write,
2170         .vop_open       = hammer2_vop_open,
2171         .vop_inactive   = hammer2_vop_inactive,
2172         .vop_reclaim    = hammer2_vop_reclaim,
2173         .vop_nresolve   = hammer2_vop_nresolve,
2174         .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2175         .vop_nmkdir     = hammer2_vop_nmkdir,
2176         .vop_nmknod     = hammer2_vop_nmknod,
2177         .vop_ioctl      = hammer2_vop_ioctl,
2178         .vop_mountctl   = hammer2_vop_mountctl,
2179         .vop_bmap       = hammer2_vop_bmap,
2180         .vop_strategy   = hammer2_vop_strategy,
2181         .vop_kqfilter   = hammer2_vop_kqfilter
2182 };
2183
2184 struct vop_ops hammer2_spec_vops = {
2185         .vop_default =          vop_defaultop,
2186         .vop_fsync =            hammer2_vop_fsync,
2187         .vop_read =             vop_stdnoread,
2188         .vop_write =            vop_stdnowrite,
2189         .vop_access =           hammer2_vop_access,
2190         .vop_close =            hammer2_vop_close,
2191         .vop_markatime =        hammer2_vop_markatime,
2192         .vop_getattr =          hammer2_vop_getattr,
2193         .vop_inactive =         hammer2_vop_inactive,
2194         .vop_reclaim =          hammer2_vop_reclaim,
2195         .vop_setattr =          hammer2_vop_setattr
2196 };
2197
2198 struct vop_ops hammer2_fifo_vops = {
2199         .vop_default =          fifo_vnoperate,
2200         .vop_fsync =            hammer2_vop_fsync,
2201 #if 0
2202         .vop_read =             hammer2_vop_fiforead,
2203         .vop_write =            hammer2_vop_fifowrite,
2204 #endif
2205         .vop_access =           hammer2_vop_access,
2206 #if 0
2207         .vop_close =            hammer2_vop_fifoclose,
2208 #endif
2209         .vop_markatime =        hammer2_vop_markatime,
2210         .vop_getattr =          hammer2_vop_getattr,
2211         .vop_inactive =         hammer2_vop_inactive,
2212         .vop_reclaim =          hammer2_vop_reclaim,
2213         .vop_setattr =          hammer2_vop_setattr,
2214         .vop_kqfilter =         hammer2_vop_fifokqfilter
2215 };
2216