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         vp = ap->a_vp;
  90         ip = VTOI(vp);
  91
  92         /*
  93          * Degenerate case
  94          */
  95         if (ip == NULL) {
  96                 vrecycle(vp);
  97                 return (0);
  98         }
  99
 100         /*
 101          * Check for deleted inodes and recycle immediately on the last
 102          * release.  Be sure to destroy any left-over buffer cache buffers
 103          * so we do not waste time trying to flush them.
 104          *
 105          * Note that deleting the file block chains under the inode chain
 106          * would just be a waste of energy, so don't do it.
 107          *
 108          * WARNING: nvtruncbuf() can only be safely called without the inode
 109          *          lock held due to the way our write thread works.
 110          */
 111         if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
 112                 hammer2_key_t lbase;
 113                 int nblksize;
 114
 115                 /*
 116                  * Detect updates to the embedded data which may be
 117                  * synchronized by the strategy code.  Simply mark the
 118                  * inode modified so it gets picked up by our normal flush.
 119                  */
 120                 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
 121                 nvtruncbuf(vp, 0, nblksize, 0, 0);
 122                 vrecycle(vp);
 123         }
 124         return (0);
 125 }
 126
 127 /*
 128  * Reclaim a vnode so that it can be reused; after the inode is
 129  * disassociated, the filesystem must manage it alone.
 130  */
 131 static
 132 int
 133 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
 134 {
 135         hammer2_inode_t *ip;
 136         hammer2_pfs_t *pmp;
 137         struct vnode *vp;
 138
 139         vp = ap->a_vp;
 140         ip = VTOI(vp);
 141         if (ip == NULL) {
 142                 return(0);
 143         }
 144         pmp = ip->pmp;
 145
 146         /*
 147          * The final close of a deleted file or directory marks it for
 148          * destruction.  The DELETED flag allows the flusher to shortcut
 149          * any modified blocks still unflushed (that is, just ignore them).
 150          *
 151          * HAMMER2 usually does not try to optimize the freemap by returning
 152          * deleted blocks to it as it does not usually know how many snapshots
 153          * might be referencing portions of the file/dir.
 154          */
 155         vp->v_data = NULL;
 156         ip->vp = NULL;
 157
 158         /*
 159          * NOTE! We do not attempt to flush chains here, flushing is
 160          *       really fragile and could also deadlock.
 161          */
 162         vclrisdirty(vp);
 163
 164         /*
 165          * A modified inode may require chain synchronization.  This
 166          * synchronization is usually handled by VOP_SYNC / VOP_FSYNC
 167          * when vfsync() is called.  However, that requires a vnode.
 168          *
 169          * When the vnode is disassociated we must keep track of any modified
 170          * inode via the sideq so that it is properly flushed.  We cannot
 171          * safely synchronize the inode from inside the reclaim due to
 172          * potentially deep locks held as-of when the reclaim occurs.
 173          * Interactions and potential deadlocks abound.
 174          */
 175         if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
 176                           HAMMER2_INODE_MODIFIED |
 177                           HAMMER2_INODE_RESIZED)) &&
 178             (ip->flags & HAMMER2_INODE_ISDELETED) == 0) {
 179                 hammer2_inode_sideq_t *ipul;
 180
 181                 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
 182                 ipul->ip = ip;
 183
 184                 hammer2_spin_ex(&pmp->list_spin);
 185                 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
 186                         /* ref -> sideq */
 187                         atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ);
 188                         TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry);
 189                         ++pmp->sideq_count;
 190                         hammer2_spin_unex(&pmp->list_spin);
 191                 } else {
 192                         hammer2_spin_unex(&pmp->list_spin);
 193                         kfree(ipul, pmp->minode);
 194                         hammer2_inode_drop(ip);         /* vp ref */
 195                 }
 196                 /* retain ref from vp for ipul */
 197         } else {
 198                 hammer2_inode_drop(ip);                 /* vp ref */
 199         }
 200
 201         /*
 202          * XXX handle background sync when ip dirty, kernel will no longer
 203          * notify us regarding this inode because there is no longer a
 204          * vnode attached to it.
 205          */
 206
 207         return (0);
 208 }
 209
 210 /*
 211  * Currently this function synchronizes the front-end inode state to the
 212  * backend chain topology, then flushes the inode's chain and sub-topology
 213  * to backend media.  This function does not flush the root topology down to
 214  * the inode.
 215  */
 216 static
 217 int
 218 hammer2_vop_fsync(struct vop_fsync_args *ap)
 219 {
 220         hammer2_inode_t *ip;
 221         struct vnode *vp;
 222         int error1;
 223         int error2;
 224
 225         vp = ap->a_vp;
 226         ip = VTOI(vp);
 227         error1 = 0;
 228
 229         hammer2_trans_init(ip->pmp, 0);
 230
 231         /*
 232          * Clean out buffer cache, wait for I/O's to complete.
 233          */
 234         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 235         bio_track_wait(&vp->v_track_write, 0, 0);
 236
 237         /*
 238          * Flush any inode changes
 239          */
 240         hammer2_inode_lock(ip, 0);
 241         if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MODIFIED))
 242                 error1 = hammer2_inode_chain_sync(ip);
 243
 244         /*
 245          * Flush dirty chains related to the inode.
 246          *
 247          * NOTE! XXX We do not currently flush to the volume root, ultimately
 248          *       we will want to have a shortcut for the flushed inode stored
 249          *       in the volume root for recovery purposes.
 250          */
 251         error2 = hammer2_inode_chain_flush(ip);
 252         if (error2)
 253                 error1 = error2;
 254         hammer2_inode_unlock(ip);
 255         hammer2_trans_done(ip->pmp);
 256
 257         return (error1);
 258 }
 259
 260 static
 261 int
 262 hammer2_vop_access(struct vop_access_args *ap)
 263 {
 264         hammer2_inode_t *ip = VTOI(ap->a_vp);
 265         uid_t uid;
 266         gid_t gid;
 267         int error;
 268
 269         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 270         uid = hammer2_to_unix_xid(&ip->meta.uid);
 271         gid = hammer2_to_unix_xid(&ip->meta.gid);
 272         error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
 273         hammer2_inode_unlock(ip);
 274
 275         return (error);
 276 }
 277
 278 static
 279 int
 280 hammer2_vop_getattr(struct vop_getattr_args *ap)
 281 {
 282         hammer2_pfs_t *pmp;
 283         hammer2_inode_t *ip;
 284         struct vnode *vp;
 285         struct vattr *vap;
 286         hammer2_chain_t *chain;
 287         int i;
 288
 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 <
 323                                     chain->bref.embed.stats.data_count) {
 324                                         vap->va_bytes =
 325                                             chain->bref.embed.stats.data_count;
 326                                 }
 327                         }
 328                 }
 329         }
 330         vap->va_type = hammer2_get_vtype(ip->meta.type);
 331         vap->va_filerev = 0;
 332         vap->va_uid_uuid = ip->meta.uid;
 333         vap->va_gid_uuid = ip->meta.gid;
 334         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 335                           VA_FSID_UUID_VALID;
 336
 337         hammer2_inode_unlock(ip);
 338
 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         vp = ap->a_vp;
 354         vap = ap->a_vap;
 355         hammer2_update_time(&ctime);
 356
 357         ip = VTOI(vp);
 358
 359         if (ip->pmp->ronly)
 360                 return (EROFS);
 361         if (hammer2_vfs_enospace(ip, 0, ap->a_cred) > 1)
 362                 return (ENOSPC);
 363
 364         hammer2_pfs_memory_wait(ip->pmp);
 365         hammer2_trans_init(ip->pmp, 0);
 366         hammer2_inode_lock(ip, 0);
 367         error = 0;
 368
 369         if (vap->va_flags != VNOVAL) {
 370                 uint32_t flags;
 371
 372                 flags = ip->meta.uflags;
 373                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
 374                                      hammer2_to_unix_xid(&ip->meta.uid),
 375                                      ap->a_cred);
 376                 if (error == 0) {
 377                         if (ip->meta.uflags != flags) {
 378                                 hammer2_inode_modify(ip);
 379                                 ip->meta.uflags = flags;
 380                                 ip->meta.ctime = ctime;
 381                                 kflags |= NOTE_ATTRIB;
 382                         }
 383                         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 384                                 error = 0;
 385                                 goto done;
 386                         }
 387                 }
 388                 goto done;
 389         }
 390         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 391                 error = EPERM;
 392                 goto done;
 393         }
 394         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
 395                 mode_t cur_mode = ip->meta.mode;
 396                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 397                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 398                 uuid_t uuid_uid;
 399                 uuid_t uuid_gid;
 400
 401                 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
 402                                          ap->a_cred,
 403                                          &cur_uid, &cur_gid, &cur_mode);
 404                 if (error == 0) {
 405                         hammer2_guid_to_uuid(&uuid_uid, cur_uid);
 406                         hammer2_guid_to_uuid(&uuid_gid, cur_gid);
 407                         if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
 408                             bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
 409                             ip->meta.mode != cur_mode
 410                         ) {
 411                                 hammer2_inode_modify(ip);
 412                                 ip->meta.uid = uuid_uid;
 413                                 ip->meta.gid = uuid_gid;
 414                                 ip->meta.mode = cur_mode;
 415                                 ip->meta.ctime = ctime;
 416                         }
 417                         kflags |= NOTE_ATTRIB;
 418                 }
 419         }
 420
 421         /*
 422          * Resize the file
 423          */
 424         if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
 425                 switch(vp->v_type) {
 426                 case VREG:
 427                         if (vap->va_size == ip->meta.size)
 428                                 break;
 429                         if (vap->va_size < ip->meta.size) {
 430                                 hammer2_mtx_ex(&ip->truncate_lock);
 431                                 hammer2_truncate_file(ip, vap->va_size);
 432                                 hammer2_mtx_unlock(&ip->truncate_lock);
 433                                 kflags |= NOTE_WRITE;
 434                         } else {
 435                                 hammer2_extend_file(ip, vap->va_size);
 436                                 kflags |= NOTE_WRITE | NOTE_EXTEND;
 437                         }
 438                         hammer2_inode_modify(ip);
 439                         ip->meta.mtime = ctime;
 440                         vclrflags(vp, VLASTWRITETS);
 441                         break;
 442                 default:
 443                         error = EINVAL;
 444                         goto done;
 445                 }
 446         }
 447 #if 0
 448         /* atime not supported */
 449         if (vap->va_atime.tv_sec != VNOVAL) {
 450                 hammer2_inode_modify(ip);
 451                 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
 452                 kflags |= NOTE_ATTRIB;
 453         }
 454 #endif
 455         if (vap->va_mode != (mode_t)VNOVAL) {
 456                 mode_t cur_mode = ip->meta.mode;
 457                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 458                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 459
 460                 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
 461                                          cur_uid, cur_gid, &cur_mode);
 462                 if (error == 0 && ip->meta.mode != cur_mode) {
 463                         hammer2_inode_modify(ip);
 464                         ip->meta.mode = cur_mode;
 465                         ip->meta.ctime = ctime;
 466                         kflags |= NOTE_ATTRIB;
 467                 }
 468         }
 469
 470         if (vap->va_mtime.tv_sec != VNOVAL) {
 471                 hammer2_inode_modify(ip);
 472                 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
 473                 kflags |= NOTE_ATTRIB;
 474                 vclrflags(vp, VLASTWRITETS);
 475         }
 476
 477 done:
 478         /*
 479          * If a truncation occurred we must call chain_sync() now in order
 480          * to trim the related data chains, otherwise a later expansion can
 481          * cause havoc.
 482          *
 483          * If an extend occured that changed the DIRECTDATA state, we must
 484          * call inode_fsync now in order to prepare the inode's indirect
 485          * block table.
 486          *
 487          * WARNING! This means we are making an adjustment to the inode's
 488          * chain outside of sync/fsync, and not just to inode->meta, which
 489          * may result in some consistency issues if a crash were to occur
 490          * at just the wrong time.
 491          */
 492         if (ip->flags & HAMMER2_INODE_RESIZED)
 493                 hammer2_inode_chain_sync(ip);
 494
 495         /*
 496          * Cleanup.
 497          */
 498         hammer2_inode_unlock(ip);
 499         hammer2_trans_done(ip->pmp);
 500         hammer2_knote(ip->vp, kflags);
 501
 502         return (error);
 503 }
 504
 505 static
 506 int
 507 hammer2_vop_readdir(struct vop_readdir_args *ap)
 508 {
 509         hammer2_xop_readdir_t *xop;
 510         hammer2_blockref_t bref;
 511         hammer2_inode_t *ip;
 512         hammer2_tid_t inum;
 513         hammer2_key_t lkey;
 514         struct uio *uio;
 515         off_t *cookies;
 516         off_t saveoff;
 517         int cookie_index;
 518         int ncookies;
 519         int error;
 520         int eofflag;
 521         int r;
 522
 523         ip = VTOI(ap->a_vp);
 524         uio = ap->a_uio;
 525         saveoff = uio->uio_offset;
 526         eofflag = 0;
 527         error = 0;
 528
 529         /*
 530          * Setup cookies directory entry cookies if requested
 531          */
 532         if (ap->a_ncookies) {
 533                 ncookies = uio->uio_resid / 16 + 1;
 534                 if (ncookies > 1024)
 535                         ncookies = 1024;
 536                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 537         } else {
 538                 ncookies = -1;
 539                 cookies = NULL;
 540         }
 541         cookie_index = 0;
 542
 543         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 544
 545         /*
 546          * Handle artificial entries.  To ensure that only positive 64 bit
 547          * quantities are returned to userland we always strip off bit 63.
 548          * The hash code is designed such that codes 0x0000-0x7FFF are not
 549          * used, allowing us to use these codes for articial entries.
 550          *
 551          * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
 552          * allow '..' to cross the mount point into (e.g.) the super-root.
 553          */
 554         if (saveoff == 0) {
 555                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 556                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
 557                 if (r)
 558                         goto done;
 559                 if (cookies)
 560                         cookies[cookie_index] = saveoff;
 561                 ++saveoff;
 562                 ++cookie_index;
 563                 if (cookie_index == ncookies)
 564                         goto done;
 565         }
 566
 567         if (saveoff == 1) {
 568                 /*
 569                  * Be careful with lockorder when accessing ".."
 570                  *
 571                  * (ip is the current dir. xip is the parent dir).
 572                  */
 573                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 574                 if (ip != ip->pmp->iroot)
 575                         inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK;
 576                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
 577                 if (r)
 578                         goto done;
 579                 if (cookies)
 580                         cookies[cookie_index] = saveoff;
 581                 ++saveoff;
 582                 ++cookie_index;
 583                 if (cookie_index == ncookies)
 584                         goto done;
 585         }
 586
 587         lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
 588         if (hammer2_debug & 0x0020)
 589                 kprintf("readdir: lkey %016jx\n", lkey);
 590         if (error)
 591                 goto done;
 592
 593         /*
 594          * Use XOP for cluster scan.
 595          *
 596          * parent is the inode cluster, already locked for us.  Don't
 597          * double lock shared locks as this will screw up upgrades.
 598          */
 599         xop = hammer2_xop_alloc(ip, 0);
 600         xop->lkey = lkey;
 601         hammer2_xop_start(&xop->head, hammer2_xop_readdir);
 602
 603         for (;;) {
 604                 const hammer2_inode_data_t *ripdata;
 605                 const char *dname;
 606                 int dtype;
 607
 608                 error = hammer2_xop_collect(&xop->head, 0);
 609                 error = hammer2_error_to_errno(error);
 610                 if (error) {
 611                         break;
 612                 }
 613                 if (cookie_index == ncookies)
 614                         break;
 615                 if (hammer2_debug & 0x0020)
 616                 kprintf("cluster chain %p %p\n",
 617                         xop->head.cluster.focus,
 618                         (xop->head.cluster.focus ?
 619                          xop->head.cluster.focus->data : (void *)-1));
 620                 hammer2_cluster_bref(&xop->head.cluster, &bref);
 621
 622                 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
 623                         ripdata =
 624                             &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
 625                         dtype = hammer2_get_dtype(ripdata->meta.type);
 626                         saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 627                         r = vop_write_dirent(&error, uio,
 628                                              ripdata->meta.inum &
 629                                               HAMMER2_DIRHASH_USERMSK,
 630                                              dtype,
 631                                              ripdata->meta.name_len,
 632                                              ripdata->filename);
 633                         if (r)
 634                                 break;
 635                         if (cookies)
 636                                 cookies[cookie_index] = saveoff;
 637                         ++cookie_index;
 638                 } else if (bref.type == HAMMER2_BREF_TYPE_DIRENT) {
 639                         dtype = hammer2_get_dtype(bref.embed.dirent.type);
 640                         saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 641                         if (bref.embed.dirent.namlen <=
 642                             sizeof(bref.check.buf)) {
 643                                 dname = bref.check.buf;
 644                         } else {
 645                                 dname =
 646                                  hammer2_cluster_rdata(&xop->head.cluster)->buf;
 647                         }
 648                         r = vop_write_dirent(&error, uio,
 649                                              bref.embed.dirent.inum,
 650                                              dtype,
 651                                              bref.embed.dirent.namlen,
 652                                              dname);
 653                         if (r)
 654                                 break;
 655                         if (cookies)
 656                                 cookies[cookie_index] = saveoff;
 657                         ++cookie_index;
 658                 } else {
 659                         /* XXX chain error */
 660                         kprintf("bad chain type readdir %d\n", bref.type);
 661                 }
 662         }
 663         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
 664         if (error == ENOENT) {
 665                 error = 0;
 666                 eofflag = 1;
 667                 saveoff = (hammer2_key_t)-1;
 668         } else {
 669                 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 670         }
 671 done:
 672         hammer2_inode_unlock(ip);
 673         if (ap->a_eofflag)
 674                 *ap->a_eofflag = eofflag;
 675         if (hammer2_debug & 0x0020)
 676                 kprintf("readdir: done at %016jx\n", saveoff);
 677         uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
 678         if (error && cookie_index == 0) {
 679                 if (cookies) {
 680                         kfree(cookies, M_TEMP);
 681                         *ap->a_ncookies = 0;
 682                         *ap->a_cookies = NULL;
 683                 }
 684         } else {
 685                 if (cookies) {
 686                         *ap->a_ncookies = cookie_index;
 687                         *ap->a_cookies = cookies;
 688                 }
 689         }
 690         return (error);
 691 }
 692
 693 /*
 694  * hammer2_vop_readlink { vp, uio, cred }
 695  */
 696 static
 697 int
 698 hammer2_vop_readlink(struct vop_readlink_args *ap)
 699 {
 700         struct vnode *vp;
 701         hammer2_inode_t *ip;
 702         int error;
 703
 704         vp = ap->a_vp;
 705         if (vp->v_type != VLNK)
 706                 return (EINVAL);
 707         ip = VTOI(vp);
 708
 709         error = hammer2_read_file(ip, ap->a_uio, 0);
 710         return (error);
 711 }
 712
 713 static
 714 int
 715 hammer2_vop_read(struct vop_read_args *ap)
 716 {
 717         struct vnode *vp;
 718         hammer2_inode_t *ip;
 719         struct uio *uio;
 720         int error;
 721         int seqcount;
 722         int bigread;
 723
 724         /*
 725          * Read operations supported on this vnode?
 726          */
 727         vp = ap->a_vp;
 728         if (vp->v_type != VREG)
 729                 return (EINVAL);
 730
 731         /*
 732          * Misc
 733          */
 734         ip = VTOI(vp);
 735         uio = ap->a_uio;
 736         error = 0;
 737
 738         seqcount = ap->a_ioflag >> 16;
 739         bigread = (uio->uio_resid > 100 * 1024 * 1024);
 740
 741         error = hammer2_read_file(ip, uio, seqcount);
 742         return (error);
 743 }
 744
 745 static
 746 int
 747 hammer2_vop_write(struct vop_write_args *ap)
 748 {
 749         hammer2_inode_t *ip;
 750         thread_t td;
 751         struct vnode *vp;
 752         struct uio *uio;
 753         int error;
 754         int seqcount;
 755         int ioflag;
 756
 757         /*
 758          * Read operations supported on this vnode?
 759          */
 760         vp = ap->a_vp;
 761         if (vp->v_type != VREG)
 762                 return (EINVAL);
 763
 764         /*
 765          * Misc
 766          */
 767         ip = VTOI(vp);
 768         ioflag = ap->a_ioflag;
 769         uio = ap->a_uio;
 770         error = 0;
 771         if (ip->pmp->ronly)
 772                 return (EROFS);
 773         switch (hammer2_vfs_enospace(ip, uio->uio_resid, ap->a_cred)) {
 774         case 2:
 775                 return (ENOSPC);
 776         case 1:
 777                 ioflag |= IO_DIRECT;    /* semi-synchronous */
 778                 /* fall through */
 779         default:
 780                 break;
 781         }
 782
 783         seqcount = ioflag >> 16;
 784
 785         /*
 786          * Check resource limit
 787          */
 788         if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
 789             uio->uio_offset + uio->uio_resid >
 790              td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 791                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 792                 return (EFBIG);
 793         }
 794
 795         /*
 796          * The transaction interlocks against flush initiations
 797          * (note: but will run concurrently with the actual flush).
 798          *
 799          * To avoid deadlocking against the VM system, we must flag any
 800          * transaction related to the buffer cache or other direct
 801          * VM page manipulation.
 802          */
 803         if (uio->uio_segflg == UIO_NOCOPY)
 804                 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_BUFCACHE);
 805         else
 806                 hammer2_trans_init(ip->pmp, 0);
 807         error = hammer2_write_file(ip, uio, ioflag, seqcount);
 808         hammer2_trans_done(ip->pmp);
 809
 810         return (error);
 811 }
 812
 813 /*
 814  * Perform read operations on a file or symlink given an UNLOCKED
 815  * inode and uio.
 816  *
 817  * The passed ip is not locked.
 818  */
 819 static
 820 int
 821 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
 822 {
 823         hammer2_off_t size;
 824         struct buf *bp;
 825         int error;
 826
 827         error = 0;
 828
 829         /*
 830          * UIO read loop.
 831          *
 832          * WARNING! Assumes that the kernel interlocks size changes at the
 833          *          vnode level.
 834          */
 835         hammer2_mtx_sh(&ip->lock);
 836         hammer2_mtx_sh(&ip->truncate_lock);
 837         size = ip->meta.size;
 838         hammer2_mtx_unlock(&ip->lock);
 839
 840         while (uio->uio_resid > 0 && uio->uio_offset < size) {
 841                 hammer2_key_t lbase;
 842                 hammer2_key_t leof;
 843                 int lblksize;
 844                 int loff;
 845                 int n;
 846
 847                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 848                                                 &lbase, &leof);
 849
 850 #if 1
 851                 bp = NULL;
 852                 error = cluster_readx(ip->vp, leof, lbase, lblksize,
 853                                       B_NOTMETA | B_KVABIO,
 854                                       uio->uio_resid,
 855                                       seqcount * MAXBSIZE,
 856                                       &bp);
 857 #else
 858                 if (uio->uio_segflg == UIO_NOCOPY) {
 859                         bp = getblk(ip->vp, lbase, lblksize,
 860                                     GETBLK_BHEAVY | GETBLK_KVABIO, 0);
 861                         if (bp->b_flags & B_CACHE) {
 862                                 int i;
 863                                 int j = 0;
 864                                 if (bp->b_xio.xio_npages != 16)
 865                                         kprintf("NPAGES BAD\n");
 866                                 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
 867                                         vm_page_t m;
 868                                         m = bp->b_xio.xio_pages[i];
 869                                         if (m == NULL || m->valid == 0) {
 870                                                 kprintf("bp %016jx %016jx pg %d inv",
 871                                                         lbase, leof, i);
 872                                                 if (m)
 873                                                         kprintf("m->object %p/%p", m->object, ip->vp->v_object);
 874                                                 kprintf("\n");
 875                                                 j = 1;
 876                                         }
 877                                 }
 878                                 if (j)
 879                                         kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error);
 880                         }
 881                         bqrelse(bp);
 882                 }
 883                 error = bread_kvabio(ip->vp, lbase, lblksize, &bp);
 884 #endif
 885                 if (error) {
 886                         brelse(bp);
 887                         break;
 888                 }
 889                 bkvasync(bp);
 890                 loff = (int)(uio->uio_offset - lbase);
 891                 n = lblksize - loff;
 892                 if (n > uio->uio_resid)
 893                         n = uio->uio_resid;
 894                 if (n > size - uio->uio_offset)
 895                         n = (int)(size - uio->uio_offset);
 896                 bp->b_flags |= B_AGE;
 897                 uiomovebp(bp, (char *)bp->b_data + loff, n, uio);
 898                 bqrelse(bp);
 899         }
 900         hammer2_mtx_unlock(&ip->truncate_lock);
 901
 902         return (error);
 903 }
 904
 905 /*
 906  * Write to the file represented by the inode via the logical buffer cache.
 907  * The inode may represent a regular file or a symlink.
 908  *
 909  * The inode must not be locked.
 910  */
 911 static
 912 int
 913 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
 914                    int ioflag, int seqcount)
 915 {
 916         hammer2_key_t old_eof;
 917         hammer2_key_t new_eof;
 918         struct buf *bp;
 919         int kflags;
 920         int error;
 921         int modified;
 922
 923         /*
 924          * Setup if append
 925          *
 926          * WARNING! Assumes that the kernel interlocks size changes at the
 927          *          vnode level.
 928          */
 929         hammer2_mtx_ex(&ip->lock);
 930         hammer2_mtx_sh(&ip->truncate_lock);
 931         if (ioflag & IO_APPEND)
 932                 uio->uio_offset = ip->meta.size;
 933         old_eof = ip->meta.size;
 934
 935         /*
 936          * Extend the file if necessary.  If the write fails at some point
 937          * we will truncate it back down to cover as much as we were able
 938          * to write.
 939          *
 940          * Doing this now makes it easier to calculate buffer sizes in
 941          * the loop.
 942          */
 943         kflags = 0;
 944         error = 0;
 945         modified = 0;
 946
 947         if (uio->uio_offset + uio->uio_resid > old_eof) {
 948                 new_eof = uio->uio_offset + uio->uio_resid;
 949                 modified = 1;
 950                 hammer2_extend_file(ip, new_eof);
 951                 kflags |= NOTE_EXTEND;
 952         } else {
 953                 new_eof = old_eof;
 954         }
 955         hammer2_mtx_unlock(&ip->lock);
 956
 957         /*
 958          * UIO write loop
 959          */
 960         while (uio->uio_resid > 0) {
 961                 hammer2_key_t lbase;
 962                 int trivial;
 963                 int endofblk;
 964                 int lblksize;
 965                 int loff;
 966                 int n;
 967
 968                 /*
 969                  * Don't allow the buffer build to blow out the buffer
 970                  * cache.
 971                  */
 972                 if ((ioflag & IO_RECURSE) == 0)
 973                         bwillwrite(HAMMER2_PBUFSIZE);
 974
 975                 /*
 976                  * This nominally tells us how much we can cluster and
 977                  * what the logical buffer size needs to be.  Currently
 978                  * we don't try to cluster the write and just handle one
 979                  * block at a time.
 980                  */
 981                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 982                                                 &lbase, NULL);
 983                 loff = (int)(uio->uio_offset - lbase);
 984
 985                 KKASSERT(lblksize <= 65536);
 986
 987                 /*
 988                  * Calculate bytes to copy this transfer and whether the
 989                  * copy completely covers the buffer or not.
 990                  */
 991                 trivial = 0;
 992                 n = lblksize - loff;
 993                 if (n > uio->uio_resid) {
 994                         n = uio->uio_resid;
 995                         if (loff == lbase && uio->uio_offset + n == new_eof)
 996                                 trivial = 1;
 997                         endofblk = 0;
 998                 } else {
 999                         if (loff == 0)
1000                                 trivial = 1;
1001                         endofblk = 1;
1002                 }
1003                 if (lbase >= new_eof)
1004                         trivial = 1;
1005
1006                 /*
1007                  * Get the buffer
1008                  */
1009                 if (uio->uio_segflg == UIO_NOCOPY) {
1010                         /*
1011                          * Issuing a write with the same data backing the
1012                          * buffer.  Instantiate the buffer to collect the
1013                          * backing vm pages, then read-in any missing bits.
1014                          *
1015                          * This case is used by vop_stdputpages().
1016                          */
1017                         bp = getblk(ip->vp, lbase, lblksize,
1018                                     GETBLK_BHEAVY | GETBLK_KVABIO, 0);
1019                         if ((bp->b_flags & B_CACHE) == 0) {
1020                                 bqrelse(bp);
1021                                 error = bread_kvabio(ip->vp, lbase,
1022                                                      lblksize, &bp);
1023                         }
1024                 } else if (trivial) {
1025                         /*
1026                          * Even though we are entirely overwriting the buffer
1027                          * we may still have to zero it out to avoid a
1028                          * mmap/write visibility issue.
1029                          */
1030                         bp = getblk(ip->vp, lbase, lblksize,
1031                                     GETBLK_BHEAVY | GETBLK_KVABIO, 0);
1032                         if ((bp->b_flags & B_CACHE) == 0)
1033                                 vfs_bio_clrbuf(bp);
1034                 } else {
1035                         /*
1036                          * Partial overwrite, read in any missing bits then
1037                          * replace the portion being written.
1038                          *
1039                          * (The strategy code will detect zero-fill physical
1040                          * blocks for this case).
1041                          */
1042                         error = bread_kvabio(ip->vp, lbase, lblksize, &bp);
1043                         if (error == 0)
1044                                 bheavy(bp);
1045                 }
1046
1047                 if (error) {
1048                         brelse(bp);
1049                         break;
1050                 }
1051
1052                 /*
1053                  * Ok, copy the data in
1054                  */
1055                 bkvasync(bp);
1056                 error = uiomovebp(bp, bp->b_data + loff, n, uio);
1057                 kflags |= NOTE_WRITE;
1058                 modified = 1;
1059                 if (error) {
1060                         brelse(bp);
1061                         break;
1062                 }
1063
1064                 /*
1065                  * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1066                  *          with IO_SYNC or IO_ASYNC set.  These writes
1067                  *          must be handled as the pageout daemon expects.
1068                  *
1069                  * NOTE!    H2 relies on cluster_write() here because it
1070                  *          cannot preallocate disk blocks at the logical
1071                  *          level due to not knowing what the compression
1072                  *          size will be at this time.
1073                  *
1074                  *          We must use cluster_write() here and we depend
1075                  *          on the write-behind feature to flush buffers
1076                  *          appropriately.  If we let the buffer daemons do
1077                  *          it the block allocations will be all over the
1078                  *          map.
1079                  */
1080                 if (ioflag & IO_SYNC) {
1081                         bwrite(bp);
1082                 } else if ((ioflag & IO_DIRECT) && endofblk) {
1083                         bawrite(bp);
1084                 } else if (ioflag & IO_ASYNC) {
1085                         bawrite(bp);
1086                 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) {
1087                         bdwrite(bp);
1088                 } else {
1089 #if 1
1090                         bp->b_flags |= B_CLUSTEROK;
1091                         cluster_write(bp, new_eof, lblksize, seqcount);
1092 #else
1093                         bp->b_flags |= B_CLUSTEROK;
1094                         bdwrite(bp);
1095 #endif
1096                 }
1097         }
1098
1099         /*
1100          * Cleanup.  If we extended the file EOF but failed to write through
1101          * the entire write is a failure and we have to back-up.
1102          */
1103         if (error && new_eof != old_eof) {
1104                 hammer2_mtx_unlock(&ip->truncate_lock);
1105                 hammer2_mtx_ex(&ip->lock);
1106                 hammer2_mtx_ex(&ip->truncate_lock);
1107                 hammer2_truncate_file(ip, old_eof);
1108                 if (ip->flags & HAMMER2_INODE_MODIFIED)
1109                         hammer2_inode_chain_sync(ip);
1110                 hammer2_mtx_unlock(&ip->lock);
1111         } else if (modified) {
1112                 struct vnode *vp = ip->vp;
1113
1114                 hammer2_mtx_ex(&ip->lock);
1115                 hammer2_inode_modify(ip);
1116                 if (uio->uio_segflg == UIO_NOCOPY) {
1117                         if (vp->v_flag & VLASTWRITETS) {
1118                                 ip->meta.mtime =
1119                                     (unsigned long)vp->v_lastwrite_ts.tv_sec *
1120                                     1000000 +
1121                                     vp->v_lastwrite_ts.tv_nsec / 1000;
1122                         }
1123                 } else {
1124                         hammer2_update_time(&ip->meta.mtime);
1125                         vclrflags(vp, VLASTWRITETS);
1126                 }
1127
1128 #if 0
1129                 /*
1130                  * REMOVED - handled by hammer2_extend_file().  Do not issue
1131                  * a chain_sync() outside of a sync/fsync except for DIRECTDATA
1132                  * state changes.
1133                  *
1134                  * Under normal conditions we only issue a chain_sync if
1135                  * the inode's DIRECTDATA state changed.
1136                  */
1137                 if (ip->flags & HAMMER2_INODE_RESIZED)
1138                         hammer2_inode_chain_sync(ip);
1139 #endif
1140                 hammer2_mtx_unlock(&ip->lock);
1141                 hammer2_knote(ip->vp, kflags);
1142         }
1143         hammer2_trans_assert_strategy(ip->pmp);
1144         hammer2_mtx_unlock(&ip->truncate_lock);
1145
1146         return error;
1147 }
1148
1149 /*
1150  * Truncate the size of a file.  The inode must not be locked.
1151  *
1152  * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1153  * ensure that any on-media data beyond the new file EOF has been destroyed.
1154  *
1155  * WARNING: nvtruncbuf() can only be safely called without the inode lock
1156  *          held due to the way our write thread works.  If the truncation
1157  *          occurs in the middle of a buffer, nvtruncbuf() is responsible
1158  *          for dirtying that buffer and zeroing out trailing bytes.
1159  *
1160  * WARNING! Assumes that the kernel interlocks size changes at the
1161  *          vnode level.
1162  *
1163  * WARNING! Caller assumes responsibility for removing dead blocks
1164  *          if INODE_RESIZED is set.
1165  */
1166 static
1167 void
1168 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1169 {
1170         hammer2_key_t lbase;
1171         int nblksize;
1172
1173         hammer2_mtx_unlock(&ip->lock);
1174         if (ip->vp) {
1175                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1176                 nvtruncbuf(ip->vp, nsize,
1177                            nblksize, (int)nsize & (nblksize - 1),
1178                            0);
1179         }
1180         hammer2_mtx_ex(&ip->lock);
1181         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1182         ip->osize = ip->meta.size;
1183         ip->meta.size = nsize;
1184         atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1185         hammer2_inode_modify(ip);
1186 }
1187
1188 /*
1189  * Extend the size of a file.  The inode must not be locked.
1190  *
1191  * Even though the file size is changing, we do not have to set the
1192  * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1193  * boundary.  When this occurs a hammer2_inode_chain_sync() is required
1194  * to prepare the inode cluster's indirect block table, otherwise
1195  * async execution of the strategy code will implode on us.
1196  *
1197  * WARNING! Assumes that the kernel interlocks size changes at the
1198  *          vnode level.
1199  *
1200  * WARNING! Caller assumes responsibility for transitioning out
1201  *          of the inode DIRECTDATA mode if INODE_RESIZED is set.
1202  */
1203 static
1204 void
1205 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1206 {
1207         hammer2_key_t lbase;
1208         hammer2_key_t osize;
1209         int oblksize;
1210         int nblksize;
1211
1212         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1213         hammer2_inode_modify(ip);
1214         osize = ip->meta.size;
1215         ip->osize = osize;
1216         ip->meta.size = nsize;
1217
1218         /*
1219          * We must issue a chain_sync() when the DIRECTDATA state changes
1220          * to prevent confusion between the flush code and the in-memory
1221          * state.  This is not perfect because we are doing it outside of
1222          * a sync/fsync operation, so it might not be fully synchronized
1223          * with the meta-data topology flush.
1224          */
1225         if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1226                 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1227                 hammer2_inode_chain_sync(ip);
1228         }
1229
1230         hammer2_mtx_unlock(&ip->lock);
1231         if (ip->vp) {
1232                 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1233                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1234                 nvextendbuf(ip->vp,
1235                             osize, nsize,
1236                             oblksize, nblksize,
1237                             -1, -1, 0);
1238         }
1239         hammer2_mtx_ex(&ip->lock);
1240 }
1241
1242 static
1243 int
1244 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1245 {
1246         hammer2_xop_nresolve_t *xop;
1247         hammer2_inode_t *ip;
1248         hammer2_inode_t *dip;
1249         struct namecache *ncp;
1250         struct vnode *vp;
1251         int error;
1252
1253         dip = VTOI(ap->a_dvp);
1254         xop = hammer2_xop_alloc(dip, 0);
1255
1256         ncp = ap->a_nch->ncp;
1257         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1258
1259         /*
1260          * Note: In DragonFly the kernel handles '.' and '..'.
1261          */
1262         hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1263         hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1264
1265         error = hammer2_xop_collect(&xop->head, 0);
1266         error = hammer2_error_to_errno(error);
1267         if (error) {
1268                 ip = NULL;
1269         } else {
1270                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1271         }
1272         hammer2_inode_unlock(dip);
1273
1274         /*
1275          * Acquire the related vnode
1276          *
1277          * NOTE: For error processing, only ENOENT resolves the namecache
1278          *       entry to NULL, otherwise we just return the error and
1279          *       leave the namecache unresolved.
1280          *
1281          * NOTE: multiple hammer2_inode structures can be aliased to the
1282          *       same chain element, for example for hardlinks.  This
1283          *       use case does not 'reattach' inode associations that
1284          *       might already exist, but always allocates a new one.
1285          *
1286          * WARNING: inode structure is locked exclusively via inode_get
1287          *          but chain was locked shared.  inode_unlock()
1288          *          will handle it properly.
1289          */
1290         if (ip) {
1291                 vp = hammer2_igetv(ip, &error); /* error set to UNIX error */
1292                 if (error == 0) {
1293                         vn_unlock(vp);
1294                         cache_setvp(ap->a_nch, vp);
1295                 } else if (error == ENOENT) {
1296                         cache_setvp(ap->a_nch, NULL);
1297                 }
1298                 hammer2_inode_unlock(ip);
1299
1300                 /*
1301                  * The vp should not be released until after we've disposed
1302                  * of our locks, because it might cause vop_inactive() to
1303                  * be called.
1304                  */
1305                 if (vp)
1306                         vrele(vp);
1307         } else {
1308                 error = ENOENT;
1309                 cache_setvp(ap->a_nch, NULL);
1310         }
1311         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1312         KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1313                 ("resolve error %d/%p ap %p\n",
1314                  error, ap->a_nch->ncp->nc_vp, ap));
1315
1316         return error;
1317 }
1318
1319 static
1320 int
1321 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1322 {
1323         hammer2_inode_t *dip;
1324         hammer2_tid_t inum;
1325         int error;
1326
1327         dip = VTOI(ap->a_dvp);
1328         inum = dip->meta.iparent;
1329         *ap->a_vpp = NULL;
1330
1331         if (inum) {
1332                 error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL,
1333                                          inum, ap->a_vpp);
1334         } else {
1335                 error = ENOENT;
1336         }
1337         return error;
1338 }
1339
1340 static
1341 int
1342 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1343 {
1344         hammer2_inode_t *dip;
1345         hammer2_inode_t *nip;
1346         struct namecache *ncp;
1347         const uint8_t *name;
1348         size_t name_len;
1349         hammer2_tid_t inum;
1350         int error;
1351
1352         dip = VTOI(ap->a_dvp);
1353         if (dip->pmp->ronly)
1354                 return (EROFS);
1355         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1356                 return (ENOSPC);
1357
1358         ncp = ap->a_nch->ncp;
1359         name = ncp->nc_name;
1360         name_len = ncp->nc_nlen;
1361
1362         hammer2_pfs_memory_wait(dip->pmp);
1363         hammer2_trans_init(dip->pmp, 0);
1364
1365         inum = hammer2_trans_newinum(dip->pmp);
1366
1367         /*
1368          * Create the actual inode as a hidden file in the iroot, then
1369          * create the directory entry.  The creation of the actual inode
1370          * sets its nlinks to 1 which is the value we desire.
1371          */
1372         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1373                                    NULL, 0, inum,
1374                                    inum, 0, 0,
1375                                    0, &error);
1376         if (error) {
1377                 error = hammer2_error_to_errno(error);
1378         } else {
1379                 error = hammer2_dirent_create(dip, name, name_len,
1380                                               nip->meta.inum, nip->meta.type);
1381                 /* returns UNIX error code */
1382         }
1383         if (error) {
1384                 if (nip) {
1385                         hammer2_inode_unlink_finisher(nip, 0);
1386                         hammer2_inode_unlock(nip);
1387                         nip = NULL;
1388                 }
1389                 *ap->a_vpp = NULL;
1390         } else {
1391                 *ap->a_vpp = hammer2_igetv(nip, &error);
1392                 hammer2_inode_unlock(nip);
1393         }
1394
1395         /*
1396          * Update dip's mtime
1397          *
1398          * We can use a shared inode lock and allow the meta.mtime update
1399          * SMP race.  hammer2_inode_modify() is MPSAFE w/a shared lock.
1400          */
1401         if (error == 0) {
1402                 uint64_t mtime;
1403
1404                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1405                 hammer2_update_time(&mtime);
1406                 hammer2_inode_modify(dip);
1407                 dip->meta.mtime = mtime;
1408                 hammer2_inode_unlock(dip);
1409         }
1410
1411         hammer2_trans_done(dip->pmp);
1412
1413         if (error == 0) {
1414                 cache_setunresolved(ap->a_nch);
1415                 cache_setvp(ap->a_nch, *ap->a_vpp);
1416                 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
1417         }
1418         return error;
1419 }
1420
1421 static
1422 int
1423 hammer2_vop_open(struct vop_open_args *ap)
1424 {
1425         return vop_stdopen(ap);
1426 }
1427
1428 /*
1429  * hammer2_vop_advlock { vp, id, op, fl, flags }
1430  */
1431 static
1432 int
1433 hammer2_vop_advlock(struct vop_advlock_args *ap)
1434 {
1435         hammer2_inode_t *ip = VTOI(ap->a_vp);
1436         hammer2_off_t size;
1437
1438         size = ip->meta.size;
1439         return (lf_advlock(ap, &ip->advlock, size));
1440 }
1441
1442 static
1443 int
1444 hammer2_vop_close(struct vop_close_args *ap)
1445 {
1446         return vop_stdclose(ap);
1447 }
1448
1449 /*
1450  * hammer2_vop_nlink { nch, dvp, vp, cred }
1451  *
1452  * Create a hardlink from (vp) to {dvp, nch}.
1453  */
1454 static
1455 int
1456 hammer2_vop_nlink(struct vop_nlink_args *ap)
1457 {
1458         hammer2_inode_t *tdip;  /* target directory to create link in */
1459         hammer2_inode_t *ip;    /* inode we are hardlinking to */
1460         struct namecache *ncp;
1461         const uint8_t *name;
1462         size_t name_len;
1463         int error;
1464
1465         if (ap->a_dvp->v_mount != ap->a_vp->v_mount)
1466                 return(EXDEV);
1467
1468         tdip = VTOI(ap->a_dvp);
1469         if (tdip->pmp->ronly)
1470                 return (EROFS);
1471         if (hammer2_vfs_enospace(tdip, 0, ap->a_cred) > 1)
1472                 return (ENOSPC);
1473
1474         ncp = ap->a_nch->ncp;
1475         name = ncp->nc_name;
1476         name_len = ncp->nc_nlen;
1477
1478         /*
1479          * ip represents the file being hardlinked.  The file could be a
1480          * normal file or a hardlink target if it has already been hardlinked.
1481          * (with the new semantics, it will almost always be a hardlink
1482          * target).
1483          *
1484          * Bump nlinks and potentially also create or move the hardlink
1485          * target in the parent directory common to (ip) and (tdip).  The
1486          * consolidation code can modify ip->cluster.  The returned cluster
1487          * is locked.
1488          */
1489         ip = VTOI(ap->a_vp);
1490         KASSERT(ip->pmp, ("ip->pmp is NULL %p %p", ip, ip->pmp));
1491         hammer2_pfs_memory_wait(ip->pmp);
1492         hammer2_trans_init(ip->pmp, 0);
1493
1494         /*
1495          * Target should be an indexed inode or there's no way we will ever
1496          * be able to find it!
1497          */
1498         KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1499
1500         error = 0;
1501
1502         /*
1503          * Can return NULL and error == EXDEV if the common parent
1504          * crosses a directory with the xlink flag set.
1505          */
1506         hammer2_inode_lock(tdip, 0);
1507         hammer2_inode_lock(ip, 0);
1508
1509         /*
1510          * Create the directory entry and bump nlinks.
1511          */
1512         if (error == 0) {
1513                 error = hammer2_dirent_create(tdip, name, name_len,
1514                                               ip->meta.inum, ip->meta.type);
1515                 hammer2_inode_modify(ip);
1516                 ++ip->meta.nlinks;
1517         }
1518         if (error == 0) {
1519                 /*
1520                  * Update dip's mtime
1521                  */
1522                 uint64_t mtime;
1523
1524                 hammer2_update_time(&mtime);
1525                 hammer2_inode_modify(tdip);
1526                 tdip->meta.mtime = mtime;
1527
1528                 cache_setunresolved(ap->a_nch);
1529                 cache_setvp(ap->a_nch, ap->a_vp);
1530         }
1531         hammer2_inode_unlock(ip);
1532         hammer2_inode_unlock(tdip);
1533
1534         hammer2_trans_done(ip->pmp);
1535         hammer2_knote(ap->a_vp, NOTE_LINK);
1536         hammer2_knote(ap->a_dvp, NOTE_WRITE);
1537
1538         return error;
1539 }
1540
1541 /*
1542  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1543  *
1544  * The operating system has already ensured that the directory entry
1545  * does not exist and done all appropriate namespace locking.
1546  */
1547 static
1548 int
1549 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1550 {
1551         hammer2_inode_t *dip;
1552         hammer2_inode_t *nip;
1553         struct namecache *ncp;
1554         const uint8_t *name;
1555         size_t name_len;
1556         hammer2_tid_t inum;
1557         int error;
1558
1559         dip = VTOI(ap->a_dvp);
1560         if (dip->pmp->ronly)
1561                 return (EROFS);
1562         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1563                 return (ENOSPC);
1564
1565         ncp = ap->a_nch->ncp;
1566         name = ncp->nc_name;
1567         name_len = ncp->nc_nlen;
1568         hammer2_pfs_memory_wait(dip->pmp);
1569         hammer2_trans_init(dip->pmp, 0);
1570
1571         inum = hammer2_trans_newinum(dip->pmp);
1572
1573         /*
1574          * Create the actual inode as a hidden file in the iroot, then
1575          * create the directory entry.  The creation of the actual inode
1576          * sets its nlinks to 1 which is the value we desire.
1577          */
1578         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1579                                    NULL, 0, inum,
1580                                    inum, 0, 0,
1581                                    0, &error);
1582
1583         if (error) {
1584                 error = hammer2_error_to_errno(error);
1585         } else {
1586                 error = hammer2_dirent_create(dip, name, name_len,
1587                                               nip->meta.inum, nip->meta.type);
1588         }
1589         if (error) {
1590                 if (nip) {
1591                         hammer2_inode_unlink_finisher(nip, 0);
1592                         hammer2_inode_unlock(nip);
1593                         nip = NULL;
1594                 }
1595                 *ap->a_vpp = NULL;
1596         } else {
1597                 *ap->a_vpp = hammer2_igetv(nip, &error);
1598                 hammer2_inode_unlock(nip);
1599         }
1600
1601         /*
1602          * Update dip's mtime
1603          */
1604         if (error == 0) {
1605                 uint64_t mtime;
1606
1607                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1608                 hammer2_update_time(&mtime);
1609                 hammer2_inode_modify(dip);
1610                 dip->meta.mtime = mtime;
1611                 hammer2_inode_unlock(dip);
1612         }
1613
1614         hammer2_trans_done(dip->pmp);
1615
1616         if (error == 0) {
1617                 cache_setunresolved(ap->a_nch);
1618                 cache_setvp(ap->a_nch, *ap->a_vpp);
1619                 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1620         }
1621         return error;
1622 }
1623
1624 /*
1625  * Make a device node (typically a fifo)
1626  */
1627 static
1628 int
1629 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1630 {
1631         hammer2_inode_t *dip;
1632         hammer2_inode_t *nip;
1633         struct namecache *ncp;
1634         const uint8_t *name;
1635         size_t name_len;
1636         hammer2_tid_t inum;
1637         int error;
1638
1639         dip = VTOI(ap->a_dvp);
1640         if (dip->pmp->ronly)
1641                 return (EROFS);
1642         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1643                 return (ENOSPC);
1644
1645         ncp = ap->a_nch->ncp;
1646         name = ncp->nc_name;
1647         name_len = ncp->nc_nlen;
1648         hammer2_pfs_memory_wait(dip->pmp);
1649         hammer2_trans_init(dip->pmp, 0);
1650
1651         /*
1652          * Create the device inode and then create the directory entry.
1653          */
1654         inum = hammer2_trans_newinum(dip->pmp);
1655         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1656                                    NULL, 0, inum,
1657                                    inum, 0, 0,
1658                                    0, &error);
1659         if (error == 0) {
1660                 error = hammer2_dirent_create(dip, name, name_len,
1661                                               nip->meta.inum, nip->meta.type);
1662         }
1663         if (error) {
1664                 if (nip) {
1665                         hammer2_inode_unlink_finisher(nip, 0);
1666                         hammer2_inode_unlock(nip);
1667                         nip = NULL;
1668                 }
1669                 *ap->a_vpp = NULL;
1670         } else {
1671                 *ap->a_vpp = hammer2_igetv(nip, &error);
1672                 hammer2_inode_unlock(nip);
1673         }
1674
1675         /*
1676          * Update dip's mtime
1677          */
1678         if (error == 0) {
1679                 uint64_t mtime;
1680
1681                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1682                 hammer2_update_time(&mtime);
1683                 hammer2_inode_modify(dip);
1684                 dip->meta.mtime = mtime;
1685                 hammer2_inode_unlock(dip);
1686         }
1687
1688         hammer2_trans_done(dip->pmp);
1689
1690         if (error == 0) {
1691                 cache_setunresolved(ap->a_nch);
1692                 cache_setvp(ap->a_nch, *ap->a_vpp);
1693                 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1694         }
1695         return error;
1696 }
1697
1698 /*
1699  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1700  */
1701 static
1702 int
1703 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1704 {
1705         hammer2_inode_t *dip;
1706         hammer2_inode_t *nip;
1707         struct namecache *ncp;
1708         const uint8_t *name;
1709         size_t name_len;
1710         hammer2_tid_t inum;
1711         int error;
1712
1713         dip = VTOI(ap->a_dvp);
1714         if (dip->pmp->ronly)
1715                 return (EROFS);
1716         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1717                 return (ENOSPC);
1718
1719         ncp = ap->a_nch->ncp;
1720         name = ncp->nc_name;
1721         name_len = ncp->nc_nlen;
1722         hammer2_pfs_memory_wait(dip->pmp);
1723         hammer2_trans_init(dip->pmp, 0);
1724
1725         ap->a_vap->va_type = VLNK;      /* enforce type */
1726
1727         /*
1728          * Create the softlink as an inode and then create the directory
1729          * entry.
1730          */
1731         inum = hammer2_trans_newinum(dip->pmp);
1732
1733         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1734                                    NULL, 0, inum,
1735                                    inum, 0, 0,
1736                                    0, &error);
1737         if (error == 0) {
1738                 error = hammer2_dirent_create(dip, name, name_len,
1739                                               nip->meta.inum, nip->meta.type);
1740         }
1741         if (error) {
1742                 if (nip) {
1743                         hammer2_inode_unlink_finisher(nip, 0);
1744                         hammer2_inode_unlock(nip);
1745                         nip = NULL;
1746                 }
1747                 *ap->a_vpp = NULL;
1748                 hammer2_trans_done(dip->pmp);
1749                 return error;
1750         }
1751         *ap->a_vpp = hammer2_igetv(nip, &error);
1752
1753         /*
1754          * Build the softlink (~like file data) and finalize the namecache.
1755          */
1756         if (error == 0) {
1757                 size_t bytes;
1758                 struct uio auio;
1759                 struct iovec aiov;
1760
1761                 bytes = strlen(ap->a_target);
1762
1763                 hammer2_inode_unlock(nip);
1764                 bzero(&auio, sizeof(auio));
1765                 bzero(&aiov, sizeof(aiov));
1766                 auio.uio_iov = &aiov;
1767                 auio.uio_segflg = UIO_SYSSPACE;
1768                 auio.uio_rw = UIO_WRITE;
1769                 auio.uio_resid = bytes;
1770                 auio.uio_iovcnt = 1;
1771                 auio.uio_td = curthread;
1772                 aiov.iov_base = ap->a_target;
1773                 aiov.iov_len = bytes;
1774                 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1775                 /* XXX handle error */
1776                 error = 0;
1777         } else {
1778                 hammer2_inode_unlock(nip);
1779         }
1780
1781         /*
1782          * Update dip's mtime
1783          */
1784         if (error == 0) {
1785                 uint64_t mtime;
1786
1787                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1788                 hammer2_update_time(&mtime);
1789                 hammer2_inode_modify(dip);
1790                 dip->meta.mtime = mtime;
1791                 hammer2_inode_unlock(dip);
1792         }
1793
1794         hammer2_trans_done(dip->pmp);
1795
1796         /*
1797          * Finalize namecache
1798          */
1799         if (error == 0) {
1800                 cache_setunresolved(ap->a_nch);
1801                 cache_setvp(ap->a_nch, *ap->a_vpp);
1802                 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1803         }
1804         return error;
1805 }
1806
1807 /*
1808  * hammer2_vop_nremove { nch, dvp, cred }
1809  */
1810 static
1811 int
1812 hammer2_vop_nremove(struct vop_nremove_args *ap)
1813 {
1814         hammer2_xop_unlink_t *xop;
1815         hammer2_inode_t *dip;
1816         hammer2_inode_t *ip;
1817         struct namecache *ncp;
1818         int error;
1819         int isopen;
1820
1821         dip = VTOI(ap->a_dvp);
1822         if (dip->pmp->ronly)
1823                 return (EROFS);
1824 #if 0
1825         /* allow removals, except user to also bulkfree */
1826         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1827                 return (ENOSPC);
1828 #endif
1829
1830         ncp = ap->a_nch->ncp;
1831
1832         hammer2_pfs_memory_wait(dip->pmp);
1833         hammer2_trans_init(dip->pmp, 0);
1834         hammer2_inode_lock(dip, 0);
1835
1836         /*
1837          * The unlink XOP unlinks the path from the directory and
1838          * locates and returns the cluster associated with the real inode.
1839          * We have to handle nlinks here on the frontend.
1840          */
1841         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1842         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1843
1844         /*
1845          * The namecache entry is locked so nobody can use this namespace.
1846          * Calculate isopen to determine if this namespace has an open vp
1847          * associated with it and resolve the vp only if it does.
1848          *
1849          * We try to avoid resolving the vnode if nobody has it open, but
1850          * note that the test is via this namespace only.
1851          */
1852         isopen = cache_isopen(ap->a_nch);
1853         xop->isdir = 0;
1854         xop->dopermanent = 0;
1855         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1856
1857         /*
1858          * Collect the real inode and adjust nlinks, destroy the real
1859          * inode if nlinks transitions to 0 and it was the real inode
1860          * (else it has already been removed).
1861          */
1862         error = hammer2_xop_collect(&xop->head, 0);
1863         error = hammer2_error_to_errno(error);
1864         hammer2_inode_unlock(dip);
1865
1866         if (error == 0) {
1867                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1868                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1869                 if (ip) {
1870                         hammer2_inode_unlink_finisher(ip, isopen);
1871                         hammer2_inode_unlock(ip);
1872                 }
1873         } else {
1874                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1875         }
1876
1877         /*
1878          * Update dip's mtime
1879          */
1880         if (error == 0) {
1881                 uint64_t mtime;
1882
1883                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1884                 hammer2_update_time(&mtime);
1885                 hammer2_inode_modify(dip);
1886                 dip->meta.mtime = mtime;
1887                 hammer2_inode_unlock(dip);
1888         }
1889
1890         hammer2_inode_run_sideq(dip->pmp, 0);
1891         hammer2_trans_done(dip->pmp);
1892         if (error == 0) {
1893                 cache_unlink(ap->a_nch);
1894                 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1895         }
1896         return (error);
1897 }
1898
1899 /*
1900  * hammer2_vop_nrmdir { nch, dvp, cred }
1901  */
1902 static
1903 int
1904 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1905 {
1906         hammer2_xop_unlink_t *xop;
1907         hammer2_inode_t *dip;
1908         hammer2_inode_t *ip;
1909         struct namecache *ncp;
1910         int isopen;
1911         int error;
1912
1913         dip = VTOI(ap->a_dvp);
1914         if (dip->pmp->ronly)
1915                 return (EROFS);
1916 #if 0
1917         /* allow removals, except user to also bulkfree */
1918         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1919                 return (ENOSPC);
1920 #endif
1921
1922         hammer2_pfs_memory_wait(dip->pmp);
1923         hammer2_trans_init(dip->pmp, 0);
1924         hammer2_inode_lock(dip, 0);
1925
1926         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1927
1928         ncp = ap->a_nch->ncp;
1929         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1930         isopen = cache_isopen(ap->a_nch);
1931         xop->isdir = 1;
1932         xop->dopermanent = 0;
1933         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1934
1935         /*
1936          * Collect the real inode and adjust nlinks, destroy the real
1937          * inode if nlinks transitions to 0 and it was the real inode
1938          * (else it has already been removed).
1939          */
1940         error = hammer2_xop_collect(&xop->head, 0);
1941         error = hammer2_error_to_errno(error);
1942         hammer2_inode_unlock(dip);
1943
1944         if (error == 0) {
1945                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1946                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1947                 if (ip) {
1948                         hammer2_inode_unlink_finisher(ip, isopen);
1949                         hammer2_inode_unlock(ip);
1950                 }
1951         } else {
1952                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1953         }
1954
1955         /*
1956          * Update dip's mtime
1957          */
1958         if (error == 0) {
1959                 uint64_t mtime;
1960
1961                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1962                 hammer2_update_time(&mtime);
1963                 hammer2_inode_modify(dip);
1964                 dip->meta.mtime = mtime;
1965                 hammer2_inode_unlock(dip);
1966         }
1967
1968         hammer2_inode_run_sideq(dip->pmp, 0);
1969         hammer2_trans_done(dip->pmp);
1970         if (error == 0) {
1971                 cache_unlink(ap->a_nch);
1972                 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
1973         }
1974         return (error);
1975 }
1976
1977 /*
1978  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1979  */
1980 static
1981 int
1982 hammer2_vop_nrename(struct vop_nrename_args *ap)
1983 {
1984         struct namecache *fncp;
1985         struct namecache *tncp;
1986         hammer2_inode_t *fdip;  /* source directory */
1987         hammer2_inode_t *tdip;  /* target directory */
1988         hammer2_inode_t *ip;    /* file being renamed */
1989         hammer2_inode_t *tip;   /* replaced target during rename or NULL */
1990         const uint8_t *fname;
1991         size_t fname_len;
1992         const uint8_t *tname;
1993         size_t tname_len;
1994         int error;
1995         int update_tdip;
1996         int update_fdip;
1997         hammer2_key_t tlhc;
1998
1999         if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
2000                 return(EXDEV);
2001         if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
2002                 return(EXDEV);
2003
2004         fdip = VTOI(ap->a_fdvp);        /* source directory */
2005         tdip = VTOI(ap->a_tdvp);        /* target directory */
2006
2007         if (fdip->pmp->ronly)
2008                 return (EROFS);
2009         if (hammer2_vfs_enospace(fdip, 0, ap->a_cred) > 1)
2010                 return (ENOSPC);
2011
2012         fncp = ap->a_fnch->ncp;         /* entry name in source */
2013         fname = fncp->nc_name;
2014         fname_len = fncp->nc_nlen;
2015
2016         tncp = ap->a_tnch->ncp;         /* entry name in target */
2017         tname = tncp->nc_name;
2018         tname_len = tncp->nc_nlen;
2019
2020         hammer2_pfs_memory_wait(tdip->pmp);
2021         hammer2_trans_init(tdip->pmp, 0);
2022
2023         update_tdip = 0;
2024         update_fdip = 0;
2025
2026         ip = VTOI(fncp->nc_vp);
2027         hammer2_inode_ref(ip);          /* extra ref */
2028
2029         /*
2030          * Lookup the target name to determine if a directory entry
2031          * is being overwritten.  We only hold related inode locks
2032          * temporarily, the operating system is expected to protect
2033          * against rename races.
2034          */
2035         tip = tncp->nc_vp ? VTOI(tncp->nc_vp) : NULL;
2036         if (tip)
2037                 hammer2_inode_ref(tip); /* extra ref */
2038
2039         /*
2040          * Can return NULL and error == EXDEV if the common parent
2041          * crosses a directory with the xlink flag set.
2042          *
2043          * For now try to avoid deadlocks with a simple pointer address
2044          * test.  (tip) can be NULL.
2045          */
2046         error = 0;
2047         if (fdip <= tdip) {
2048                 hammer2_inode_lock(fdip, 0);
2049                 hammer2_inode_lock(tdip, 0);
2050         } else {
2051                 hammer2_inode_lock(tdip, 0);
2052                 hammer2_inode_lock(fdip, 0);
2053         }
2054         if (tip) {
2055                 if (ip <= tip) {
2056                         hammer2_inode_lock(ip, 0);
2057                         hammer2_inode_lock(tip, 0);
2058                 } else {
2059                         hammer2_inode_lock(tip, 0);
2060                         hammer2_inode_lock(ip, 0);
2061                 }
2062         } else {
2063                 hammer2_inode_lock(ip, 0);
2064         }
2065
2066 #if 0
2067         /*
2068          * Delete the target namespace.
2069          *
2070          * REMOVED - NOW FOLDED INTO XOP_NRENAME OPERATION
2071          */
2072         {
2073                 hammer2_xop_unlink_t *xop2;
2074                 hammer2_inode_t *tip;
2075                 int isopen;
2076
2077                 /*
2078                  * The unlink XOP unlinks the path from the directory and
2079                  * locates and returns the cluster associated with the real
2080                  * inode.  We have to handle nlinks here on the frontend.
2081                  */
2082                 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2083                 hammer2_xop_setname(&xop2->head, tname, tname_len);
2084                 isopen = cache_isopen(ap->a_tnch);
2085                 xop2->isdir = -1;
2086                 xop2->dopermanent = 0;
2087                 hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
2088
2089                 /*
2090                  * Collect the real inode and adjust nlinks, destroy the real
2091                  * inode if nlinks transitions to 0 and it was the real inode
2092                  * (else it has already been removed).
2093                  */
2094                 tnch_error = hammer2_xop_collect(&xop2->head, 0);
2095                 tnch_error = hammer2_error_to_errno(tnch_error);
2096                 /* hammer2_inode_unlock(tdip); */
2097
2098                 if (tnch_error == 0) {
2099                         tip = hammer2_inode_get(tdip->pmp, NULL,
2100                                                 &xop2->head.cluster, -1);
2101                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
2102                         if (tip) {
2103                                 hammer2_inode_unlink_finisher(tip, isopen);
2104                                 hammer2_inode_unlock(tip);
2105                         }
2106                 } else {
2107                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
2108                 }
2109                 /* hammer2_inode_lock(tdip, 0); */
2110
2111                 if (tnch_error && tnch_error != ENOENT) {
2112                         error = tnch_error;
2113                         goto done2;
2114                 }
2115                 update_tdip = 1;
2116         }
2117 #endif
2118
2119         /*
2120          * Resolve the collision space for (tdip, tname, tname_len)
2121          *
2122          * tdip must be held exclusively locked to prevent races since
2123          * multiple filenames can end up in the same collision space.
2124          */
2125         {
2126                 hammer2_xop_scanlhc_t *sxop;
2127                 hammer2_tid_t lhcbase;
2128
2129                 tlhc = hammer2_dirhash(tname, tname_len);
2130                 lhcbase = tlhc;
2131                 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2132                 sxop->lhc = tlhc;
2133                 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
2134                 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
2135                         if (tlhc != sxop->head.cluster.focus->bref.key)
2136                                 break;
2137                         ++tlhc;
2138                 }
2139                 error = hammer2_error_to_errno(error);
2140                 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
2141
2142                 if (error) {
2143                         if (error != ENOENT)
2144                                 goto done2;
2145                         ++tlhc;
2146                         error = 0;
2147                 }
2148                 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
2149                         error = ENOSPC;
2150                         goto done2;
2151                 }
2152         }
2153
2154         /*
2155          * Ready to go, issue the rename to the backend.  Note that meta-data
2156          * updates to the related inodes occur separately from the rename
2157          * operation.
2158          *
2159          * NOTE: While it is not necessary to update ip->meta.name*, doing
2160          *       so aids catastrophic recovery and debugging.
2161          */
2162         if (error == 0) {
2163                 hammer2_xop_nrename_t *xop4;
2164
2165                 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
2166                 xop4->lhc = tlhc;
2167                 xop4->ip_key = ip->meta.name_key;
2168                 hammer2_xop_setip2(&xop4->head, ip);
2169                 hammer2_xop_setip3(&xop4->head, tdip);
2170                 hammer2_xop_setname(&xop4->head, fname, fname_len);
2171                 hammer2_xop_setname2(&xop4->head, tname, tname_len);
2172                 hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
2173
2174                 error = hammer2_xop_collect(&xop4->head, 0);
2175                 error = hammer2_error_to_errno(error);
2176                 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
2177
2178                 if (error == ENOENT)
2179                         error = 0;
2180
2181                 /*
2182                  * Update inode meta-data.
2183                  *
2184                  * WARNING!  The in-memory inode (ip) structure does not
2185                  *           maintain a copy of the inode's filename buffer.
2186                  */
2187                 if (error == 0 &&
2188                     (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
2189                         hammer2_inode_modify(ip);
2190                         ip->meta.name_len = tname_len;
2191                         ip->meta.name_key = tlhc;
2192                 }
2193                 if (error == 0) {
2194                         hammer2_inode_modify(ip);
2195                         ip->meta.iparent = tdip->meta.inum;
2196                 }
2197                 update_fdip = 1;
2198                 update_tdip = 1;
2199         }
2200
2201 done2:
2202         /*
2203          * If no error, the backend has replaced the target directory entry.
2204          * We must adjust nlinks on the original replace target if it exists.
2205          */
2206         if (error == 0 && tip) {
2207                 int isopen;
2208
2209                 isopen = cache_isopen(ap->a_tnch);
2210                 hammer2_inode_unlink_finisher(tip, isopen);
2211         }
2212
2213         /*
2214          * Update directory mtimes to represent the something changed.
2215          */
2216         if (update_fdip || update_tdip) {
2217                 uint64_t mtime;
2218
2219                 hammer2_update_time(&mtime);
2220                 if (update_fdip) {
2221                         hammer2_inode_modify(fdip);
2222                         fdip->meta.mtime = mtime;
2223                 }
2224                 if (update_tdip) {
2225                         hammer2_inode_modify(tdip);
2226                         tdip->meta.mtime = mtime;
2227                 }
2228         }
2229         if (tip) {
2230                 hammer2_inode_unlock(tip);
2231                 hammer2_inode_drop(tip);
2232         }
2233         hammer2_inode_unlock(ip);
2234         hammer2_inode_unlock(tdip);
2235         hammer2_inode_unlock(fdip);
2236         hammer2_inode_drop(ip);
2237         hammer2_inode_run_sideq(fdip->pmp, 0);
2238
2239         hammer2_trans_done(tdip->pmp);
2240
2241         /*
2242          * Issue the namecache update after unlocking all the internal
2243          * hammer2 structures, otherwise we might deadlock.
2244          *
2245          * WARNING! The target namespace must be updated atomically,
2246          *          and we depend on cache_rename() to handle that for
2247          *          us.  Do not do a separate cache_unlink() because
2248          *          that leaves a small window of opportunity for other
2249          *          threads to allocate the target namespace before we
2250          *          manage to complete our rename.
2251          *
2252          * WARNING! cache_rename() (and cache_unlink()) will properly
2253          *          set VREF_FINALIZE on any attached vnode.  Do not
2254          *          call cache_setunresolved() manually before-hand as
2255          *          this will prevent the flag from being set later via
2256          *          cache_rename().  If VREF_FINALIZE is not properly set
2257          *          and the inode is no longer in the topology, related
2258          *          chains can remain dirty indefinitely.
2259          */
2260         if (error == 0 && tip) {
2261                 /*cache_unlink(ap->a_tnch); see above */
2262                 /*cache_setunresolved(ap->a_tnch); see above */
2263         }
2264         if (error == 0) {
2265                 cache_rename(ap->a_fnch, ap->a_tnch);
2266                 hammer2_knote(ap->a_fdvp, NOTE_WRITE);
2267                 hammer2_knote(ap->a_tdvp, NOTE_WRITE);
2268                 hammer2_knote(fncp->nc_vp, NOTE_RENAME);
2269         }
2270
2271         return (error);
2272 }
2273
2274 /*
2275  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2276  */
2277 static
2278 int
2279 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2280 {
2281         hammer2_inode_t *ip;
2282         int error;
2283
2284         ip = VTOI(ap->a_vp);
2285
2286         error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2287                               ap->a_fflag, ap->a_cred);
2288         return (error);
2289 }
2290
2291 static
2292 int
2293 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2294 {
2295         struct mount *mp;
2296         hammer2_pfs_t *pmp;
2297         int rc;
2298
2299         switch (ap->a_op) {
2300         case (MOUNTCTL_SET_EXPORT):
2301                 mp = ap->a_head.a_ops->head.vv_mount;
2302                 pmp = MPTOPMP(mp);
2303
2304                 if (ap->a_ctllen != sizeof(struct export_args))
2305                         rc = (EINVAL);
2306                 else
2307                         rc = vfs_export(mp, &pmp->export,
2308                                         (const struct export_args *)ap->a_ctl);
2309                 break;
2310         default:
2311                 rc = vop_stdmountctl(ap);
2312                 break;
2313         }
2314         return (rc);
2315 }
2316
2317 /*
2318  * KQFILTER
2319  */
2320 static void filt_hammer2detach(struct knote *kn);
2321 static int filt_hammer2read(struct knote *kn, long hint);
2322 static int filt_hammer2write(struct knote *kn, long hint);
2323 static int filt_hammer2vnode(struct knote *kn, long hint);
2324
2325 static struct filterops hammer2read_filtops =
2326         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2327           NULL, filt_hammer2detach, filt_hammer2read };
2328 static struct filterops hammer2write_filtops =
2329         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2330           NULL, filt_hammer2detach, filt_hammer2write };
2331 static struct filterops hammer2vnode_filtops =
2332         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2333           NULL, filt_hammer2detach, filt_hammer2vnode };
2334
2335 static
2336 int
2337 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2338 {
2339         struct vnode *vp = ap->a_vp;
2340         struct knote *kn = ap->a_kn;
2341
2342         switch (kn->kn_filter) {
2343         case EVFILT_READ:
2344                 kn->kn_fop = &hammer2read_filtops;
2345                 break;
2346         case EVFILT_WRITE:
2347                 kn->kn_fop = &hammer2write_filtops;
2348                 break;
2349         case EVFILT_VNODE:
2350                 kn->kn_fop = &hammer2vnode_filtops;
2351                 break;
2352         default:
2353                 return (EOPNOTSUPP);
2354         }
2355
2356         kn->kn_hook = (caddr_t)vp;
2357
2358         knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2359
2360         return(0);
2361 }
2362
2363 static void
2364 filt_hammer2detach(struct knote *kn)
2365 {
2366         struct vnode *vp = (void *)kn->kn_hook;
2367
2368         knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2369 }
2370
2371 static int
2372 filt_hammer2read(struct knote *kn, long hint)
2373 {
2374         struct vnode *vp = (void *)kn->kn_hook;
2375         hammer2_inode_t *ip = VTOI(vp);
2376         off_t off;
2377
2378         if (hint == NOTE_REVOKE) {
2379                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2380                 return(1);
2381         }
2382         off = ip->meta.size - kn->kn_fp->f_offset;
2383         kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2384         if (kn->kn_sfflags & NOTE_OLDAPI)
2385                 return(1);
2386         return (kn->kn_data != 0);
2387 }
2388
2389
2390 static int
2391 filt_hammer2write(struct knote *kn, long hint)
2392 {
2393         if (hint == NOTE_REVOKE)
2394                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2395         kn->kn_data = 0;
2396         return (1);
2397 }
2398
2399 static int
2400 filt_hammer2vnode(struct knote *kn, long hint)
2401 {
2402         if (kn->kn_sfflags & hint)
2403                 kn->kn_fflags |= hint;
2404         if (hint == NOTE_REVOKE) {
2405                 kn->kn_flags |= (EV_EOF | EV_NODATA);
2406                 return (1);
2407         }
2408         return (kn->kn_fflags != 0);
2409 }
2410
2411 /*
2412  * FIFO VOPS
2413  */
2414 static
2415 int
2416 hammer2_vop_markatime(struct vop_markatime_args *ap)
2417 {
2418         hammer2_inode_t *ip;
2419         struct vnode *vp;
2420
2421         vp = ap->a_vp;
2422         ip = VTOI(vp);
2423
2424         if (ip->pmp->ronly)
2425                 return (EROFS);
2426         return(0);
2427 }
2428
2429 static
2430 int
2431 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2432 {
2433         int error;
2434
2435         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2436         if (error)
2437                 error = hammer2_vop_kqfilter(ap);
2438         return(error);
2439 }
2440
2441 /*
2442  * VOPS vector
2443  */
2444 struct vop_ops hammer2_vnode_vops = {
2445         .vop_default    = vop_defaultop,
2446         .vop_fsync      = hammer2_vop_fsync,
2447         .vop_getpages   = vop_stdgetpages,
2448         .vop_putpages   = vop_stdputpages,
2449         .vop_access     = hammer2_vop_access,
2450         .vop_advlock    = hammer2_vop_advlock,
2451         .vop_close      = hammer2_vop_close,
2452         .vop_nlink      = hammer2_vop_nlink,
2453         .vop_ncreate    = hammer2_vop_ncreate,
2454         .vop_nsymlink   = hammer2_vop_nsymlink,
2455         .vop_nremove    = hammer2_vop_nremove,
2456         .vop_nrmdir     = hammer2_vop_nrmdir,
2457         .vop_nrename    = hammer2_vop_nrename,
2458         .vop_getattr    = hammer2_vop_getattr,
2459         .vop_setattr    = hammer2_vop_setattr,
2460         .vop_readdir    = hammer2_vop_readdir,
2461         .vop_readlink   = hammer2_vop_readlink,
2462         .vop_read       = hammer2_vop_read,
2463         .vop_write      = hammer2_vop_write,
2464         .vop_open       = hammer2_vop_open,
2465         .vop_inactive   = hammer2_vop_inactive,
2466         .vop_reclaim    = hammer2_vop_reclaim,
2467         .vop_nresolve   = hammer2_vop_nresolve,
2468         .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2469         .vop_nmkdir     = hammer2_vop_nmkdir,
2470         .vop_nmknod     = hammer2_vop_nmknod,
2471         .vop_ioctl      = hammer2_vop_ioctl,
2472         .vop_mountctl   = hammer2_vop_mountctl,
2473         .vop_bmap       = hammer2_vop_bmap,
2474         .vop_strategy   = hammer2_vop_strategy,
2475         .vop_kqfilter   = hammer2_vop_kqfilter
2476 };
2477
2478 struct vop_ops hammer2_spec_vops = {
2479         .vop_default =          vop_defaultop,
2480         .vop_fsync =            hammer2_vop_fsync,
2481         .vop_read =             vop_stdnoread,
2482         .vop_write =            vop_stdnowrite,
2483         .vop_access =           hammer2_vop_access,
2484         .vop_close =            hammer2_vop_close,
2485         .vop_markatime =        hammer2_vop_markatime,
2486         .vop_getattr =          hammer2_vop_getattr,
2487         .vop_inactive =         hammer2_vop_inactive,
2488         .vop_reclaim =          hammer2_vop_reclaim,
2489         .vop_setattr =          hammer2_vop_setattr
2490 };
2491
2492 struct vop_ops hammer2_fifo_vops = {
2493         .vop_default =          fifo_vnoperate,
2494         .vop_fsync =            hammer2_vop_fsync,
2495 #if 0
2496         .vop_read =             hammer2_vop_fiforead,
2497         .vop_write =            hammer2_vop_fifowrite,
2498 #endif
2499         .vop_access =           hammer2_vop_access,
2500 #if 0
2501         .vop_close =            hammer2_vop_fifoclose,
2502 #endif
2503         .vop_markatime =        hammer2_vop_markatime,
2504         .vop_getattr =          hammer2_vop_getattr,
2505         .vop_inactive =         hammer2_vop_inactive,
2506         .vop_reclaim =          hammer2_vop_reclaim,
2507         .vop_setattr =          hammer2_vop_setattr,
2508         .vop_kqfilter =         hammer2_vop_fifokqfilter
2509 };
2510