sys/vfs/hammer2/hammer2_vnops.c

   1 /*
   2  * Copyright (c) 2011-2015 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@dragonflybsd.org>
   6  * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
   7  * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  *
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in
  17  *    the documentation and/or other materials provided with the
  18  *    distribution.
  19  * 3. Neither the name of The DragonFly Project nor the names of its
  20  *    contributors may be used to endorse or promote products derived
  21  *    from this software without specific, prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  26  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  27  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  28  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  29  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  30  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  31  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  32  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  33  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  */
  36 /*
  37  * Kernel Filesystem interface
  38  *
  39  * NOTE! local ipdata pointers must be reloaded on any modifying operation
  40  *       to the inode as its underlying chain may have changed.
  41  */
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/kernel.h>
  46 #include <sys/fcntl.h>
  47 #include <sys/buf.h>
  48 #include <sys/proc.h>
  49 #include <sys/namei.h>
  50 #include <sys/mount.h>
  51 #include <sys/vnode.h>
  52 #include <sys/mountctl.h>
  53 #include <sys/dirent.h>
  54 #include <sys/uio.h>
  55 #include <sys/objcache.h>
  56 #include <sys/event.h>
  57 #include <sys/file.h>
  58 #include <vfs/fifofs/fifo.h>
  59
  60 #include "hammer2.h"
  61
  62 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
  63                                 int seqcount);
  64 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
  65                                 int ioflag, int seqcount);
  66 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
  67 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);
  68
  69 struct objcache *cache_xops;
  70
  71 static __inline
  72 void
  73 hammer2_knote(struct vnode *vp, int flags)
  74 {
  75         if (flags)
  76                 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
  77 }
  78
  79 /*
  80  * Last reference to a vnode is going away but it is still cached.
  81  */
  82 static
  83 int
  84 hammer2_vop_inactive(struct vop_inactive_args *ap)
  85 {
  86         hammer2_inode_t *ip;
  87         struct vnode *vp;
  88
  89         LOCKSTART;
  90         vp = ap->a_vp;
  91         ip = VTOI(vp);
  92
  93         /*
  94          * Degenerate case
  95          */
  96         if (ip == NULL) {
  97                 vrecycle(vp);
  98                 LOCKSTOP;
  99                 return (0);
 100         }
 101
 102         /*
 103          * Check for deleted inodes and recycle immediately on the last
 104          * release.  Be sure to destroy any left-over buffer cache buffers
 105          * so we do not waste time trying to flush them.
 106          *
 107          * WARNING: nvtruncbuf() can only be safely called without the inode
 108          *          lock held due to the way our write thread works.
 109          */
 110         if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
 111                 hammer2_key_t lbase;
 112                 int nblksize;
 113
 114                 /*
 115                  * Detect updates to the embedded data which may be
 116                  * synchronized by the strategy code.  Simply mark the
 117                  * inode modified so it gets picked up by our normal flush.
 118                  */
 119                 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
 120                 nvtruncbuf(vp, 0, nblksize, 0, 0);
 121                 vrecycle(vp);
 122         }
 123         LOCKSTOP;
 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         LOCKSTART;
 140         vp = ap->a_vp;
 141         ip = VTOI(vp);
 142         if (ip == NULL) {
 143                 LOCKSTOP;
 144                 return(0);
 145         }
 146         pmp = ip->pmp;
 147
 148         /*
 149          * The final close of a deleted file or directory marks it for
 150          * destruction.  The DELETED flag allows the flusher to shortcut
 151          * any modified blocks still unflushed (that is, just ignore them).
 152          *
 153          * HAMMER2 usually does not try to optimize the freemap by returning
 154          * deleted blocks to it as it does not usually know how many snapshots
 155          * might be referencing portions of the file/dir.
 156          */
 157         vp->v_data = NULL;
 158         ip->vp = NULL;
 159
 160         /*
 161          * NOTE! We do not attempt to flush chains here, flushing is
 162          *       really fragile and could also deadlock.
 163          */
 164         vclrisdirty(vp);
 165
 166         /*
 167          * Once reclaimed the inode is disconnected from the normal flush
 168          * mechanism and must be tracked
 169          *
 170          * A reclaim can occur at any time so we cannot safely start a
 171          * transaction to handle reclamation of unlinked files.  Instead,
 172          * the ip is left with a reference and placed on a linked list and
 173          * handled later on.
 174          */
 175         if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
 176                 hammer2_inode_unlink_t *ipul;
 177
 178                 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
 179                 ipul->ip = ip;
 180
 181                 hammer2_spin_ex(&pmp->list_spin);
 182                 TAILQ_INSERT_TAIL(&pmp->unlinkq, ipul, entry);
 183                 hammer2_spin_unex(&pmp->list_spin);
 184                 /* retain ref from vp for ipul */
 185         } else {
 186                 hammer2_inode_drop(ip);                 /* vp ref */
 187         }
 188
 189         /*
 190          * XXX handle background sync when ip dirty, kernel will no longer
 191          * notify us regarding this inode because there is no longer a
 192          * vnode attached to it.
 193          */
 194
 195         LOCKSTOP;
 196         return (0);
 197 }
 198
 199 static
 200 int
 201 hammer2_vop_fsync(struct vop_fsync_args *ap)
 202 {
 203         hammer2_inode_t *ip;
 204         struct vnode *vp;
 205
 206         LOCKSTART;
 207         vp = ap->a_vp;
 208         ip = VTOI(vp);
 209
 210 #if 0
 211         /* XXX can't do this yet */
 212         hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH);
 213         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 214 #endif
 215         hammer2_trans_init(ip->pmp, 0);
 216         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 217
 218         /*
 219          * Calling chain_flush here creates a lot of duplicative
 220          * COW operations due to non-optimal vnode ordering.
 221          *
 222          * Only do it for an actual fsync() syscall.  The other forms
 223          * which call this function will eventually call chain_flush
 224          * on the volume root as a catch-all, which is far more optimal.
 225          */
 226         hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
 227         if (ip->flags & HAMMER2_INODE_MODIFIED)
 228                 hammer2_inode_fsync(ip, NULL);
 229         hammer2_inode_unlock(ip, NULL);
 230         hammer2_trans_done(ip->pmp);
 231
 232         LOCKSTOP;
 233         return (0);
 234 }
 235
 236 static
 237 int
 238 hammer2_vop_access(struct vop_access_args *ap)
 239 {
 240         hammer2_inode_t *ip = VTOI(ap->a_vp);
 241         uid_t uid;
 242         gid_t gid;
 243         int error;
 244
 245         LOCKSTART;
 246         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 247         uid = hammer2_to_unix_xid(&ip->meta.uid);
 248         gid = hammer2_to_unix_xid(&ip->meta.gid);
 249         error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
 250         hammer2_inode_unlock(ip, NULL);
 251
 252         LOCKSTOP;
 253         return (error);
 254 }
 255
 256 static
 257 int
 258 hammer2_vop_getattr(struct vop_getattr_args *ap)
 259 {
 260         hammer2_pfs_t *pmp;
 261         hammer2_inode_t *ip;
 262         struct vnode *vp;
 263         struct vattr *vap;
 264
 265         LOCKSTART;
 266         vp = ap->a_vp;
 267         vap = ap->a_vap;
 268
 269         ip = VTOI(vp);
 270         pmp = ip->pmp;
 271
 272         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 273
 274         vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
 275         vap->va_fileid = ip->meta.inum;
 276         vap->va_mode = ip->meta.mode;
 277         vap->va_nlink = ip->meta.nlinks;
 278         vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
 279         vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
 280         vap->va_rmajor = 0;
 281         vap->va_rminor = 0;
 282         vap->va_size = ip->meta.size;   /* protected by shared lock */
 283         vap->va_blocksize = HAMMER2_PBUFSIZE;
 284         vap->va_flags = ip->meta.uflags;
 285         hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
 286         hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
 287         hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
 288         vap->va_gen = 1;
 289         vap->va_bytes = ip->bref.data_count;
 290         vap->va_type = hammer2_get_vtype(ip->meta.type);
 291         vap->va_filerev = 0;
 292         vap->va_uid_uuid = ip->meta.uid;
 293         vap->va_gid_uuid = ip->meta.gid;
 294         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 295                           VA_FSID_UUID_VALID;
 296
 297         hammer2_inode_unlock(ip, NULL);
 298
 299         LOCKSTOP;
 300         return (0);
 301 }
 302
 303 static
 304 int
 305 hammer2_vop_setattr(struct vop_setattr_args *ap)
 306 {
 307         hammer2_inode_t *ip;
 308         struct vnode *vp;
 309         struct vattr *vap;
 310         int error;
 311         int kflags = 0;
 312         uint64_t ctime;
 313
 314         LOCKSTART;
 315         vp = ap->a_vp;
 316         vap = ap->a_vap;
 317         hammer2_update_time(&ctime);
 318
 319         ip = VTOI(vp);
 320
 321         if (ip->pmp->ronly) {
 322                 LOCKSTOP;
 323                 return(EROFS);
 324         }
 325
 326         hammer2_pfs_memory_wait(ip->pmp);
 327         hammer2_trans_init(ip->pmp, 0);
 328         hammer2_inode_lock(ip, 0);
 329         error = 0;
 330
 331         if (vap->va_flags != VNOVAL) {
 332                 u_int32_t flags;
 333
 334                 flags = ip->meta.uflags;
 335                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
 336                                      hammer2_to_unix_xid(&ip->meta.uid),
 337                                      ap->a_cred);
 338                 if (error == 0) {
 339                         if (ip->meta.uflags != flags) {
 340                                 hammer2_inode_modify(ip);
 341                                 ip->meta.uflags = flags;
 342                                 ip->meta.ctime = ctime;
 343                                 kflags |= NOTE_ATTRIB;
 344                         }
 345                         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 346                                 error = 0;
 347                                 goto done;
 348                         }
 349                 }
 350                 goto done;
 351         }
 352         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 353                 error = EPERM;
 354                 goto done;
 355         }
 356         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
 357                 mode_t cur_mode = ip->meta.mode;
 358                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 359                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 360                 uuid_t uuid_uid;
 361                 uuid_t uuid_gid;
 362
 363                 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
 364                                          ap->a_cred,
 365                                          &cur_uid, &cur_gid, &cur_mode);
 366                 if (error == 0) {
 367                         hammer2_guid_to_uuid(&uuid_uid, cur_uid);
 368                         hammer2_guid_to_uuid(&uuid_gid, cur_gid);
 369                         if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
 370                             bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
 371                             ip->meta.mode != cur_mode
 372                         ) {
 373                                 hammer2_inode_modify(ip);
 374                                 ip->meta.uid = uuid_uid;
 375                                 ip->meta.gid = uuid_gid;
 376                                 ip->meta.mode = cur_mode;
 377                                 ip->meta.ctime = ctime;
 378                         }
 379                         kflags |= NOTE_ATTRIB;
 380                 }
 381         }
 382
 383         /*
 384          * Resize the file
 385          */
 386         if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
 387                 switch(vp->v_type) {
 388                 case VREG:
 389                         if (vap->va_size == ip->meta.size)
 390                                 break;
 391                         if (vap->va_size < ip->meta.size) {
 392                                 hammer2_truncate_file(ip, vap->va_size);
 393                         } else {
 394                                 hammer2_extend_file(ip, vap->va_size);
 395                         }
 396                         hammer2_inode_modify(ip);
 397                         ip->meta.mtime = ctime;
 398                         break;
 399                 default:
 400                         error = EINVAL;
 401                         goto done;
 402                 }
 403         }
 404 #if 0
 405         /* atime not supported */
 406         if (vap->va_atime.tv_sec != VNOVAL) {
 407                 hammer2_inode_modify(ip);
 408                 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
 409                 kflags |= NOTE_ATTRIB;
 410         }
 411 #endif
 412         if (vap->va_mode != (mode_t)VNOVAL) {
 413                 mode_t cur_mode = ip->meta.mode;
 414                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 415                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 416
 417                 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
 418                                          cur_uid, cur_gid, &cur_mode);
 419                 if (error == 0 && ip->meta.mode != cur_mode) {
 420                         hammer2_inode_modify(ip);
 421                         ip->meta.mode = cur_mode;
 422                         ip->meta.ctime = ctime;
 423                         kflags |= NOTE_ATTRIB;
 424                 }
 425         }
 426
 427         if (vap->va_mtime.tv_sec != VNOVAL) {
 428                 hammer2_inode_modify(ip);
 429                 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
 430                 kflags |= NOTE_ATTRIB;
 431         }
 432
 433 done:
 434         /*
 435          * If a truncation occurred we must call inode_fsync() now in order
 436          * to trim the related data chains, otherwise a later expansion can
 437          * cause havoc.
 438          *
 439          * If an extend occured that changed the DIRECTDATA state, we must
 440          * call inode_fsync now in order to prepare the inode's indirect
 441          * block table.
 442          */
 443         if (ip->flags & HAMMER2_INODE_RESIZED)
 444                 hammer2_inode_fsync(ip, NULL);
 445
 446         /*
 447          * Cleanup.
 448          */
 449         hammer2_inode_unlock(ip, NULL);
 450         hammer2_trans_done(ip->pmp);
 451         hammer2_knote(ip->vp, kflags);
 452
 453         LOCKSTOP;
 454         return (error);
 455 }
 456
 457 static
 458 int
 459 hammer2_vop_readdir(struct vop_readdir_args *ap)
 460 {
 461         hammer2_xop_readdir_t *xop;
 462         hammer2_blockref_t bref;
 463         hammer2_inode_t *ip;
 464         hammer2_tid_t inum;
 465         hammer2_key_t lkey;
 466         struct uio *uio;
 467         off_t *cookies;
 468         off_t saveoff;
 469         int cookie_index;
 470         int ncookies;
 471         int error;
 472         int eofflag;
 473         int dtype;
 474         int r;
 475
 476         LOCKSTART;
 477         ip = VTOI(ap->a_vp);
 478         uio = ap->a_uio;
 479         saveoff = uio->uio_offset;
 480         eofflag = 0;
 481         error = 0;
 482
 483         /*
 484          * Setup cookies directory entry cookies if requested
 485          */
 486         if (ap->a_ncookies) {
 487                 ncookies = uio->uio_resid / 16 + 1;
 488                 if (ncookies > 1024)
 489                         ncookies = 1024;
 490                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 491         } else {
 492                 ncookies = -1;
 493                 cookies = NULL;
 494         }
 495         cookie_index = 0;
 496
 497         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 498
 499         /*
 500          * Handle artificial entries.  To ensure that only positive 64 bit
 501          * quantities are returned to userland we always strip off bit 63.
 502          * The hash code is designed such that codes 0x0000-0x7FFF are not
 503          * used, allowing us to use these codes for articial entries.
 504          *
 505          * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
 506          * allow '..' to cross the mount point into (e.g.) the super-root.
 507          */
 508         if (saveoff == 0) {
 509                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 510                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
 511                 if (r)
 512                         goto done;
 513                 if (cookies)
 514                         cookies[cookie_index] = saveoff;
 515                 ++saveoff;
 516                 ++cookie_index;
 517                 if (cookie_index == ncookies)
 518                         goto done;
 519         }
 520
 521         if (saveoff == 1) {
 522                 /*
 523                  * Be careful with lockorder when accessing ".."
 524                  *
 525                  * (ip is the current dir. xip is the parent dir).
 526                  */
 527                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 528                 if (ip->pip && ip != ip->pmp->iroot)
 529                         inum = ip->pip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 530                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
 531                 if (r)
 532                         goto done;
 533                 if (cookies)
 534                         cookies[cookie_index] = saveoff;
 535                 ++saveoff;
 536                 ++cookie_index;
 537                 if (cookie_index == ncookies)
 538                         goto done;
 539         }
 540
 541         lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
 542         if (hammer2_debug & 0x0020)
 543                 kprintf("readdir: lkey %016jx\n", lkey);
 544         if (error)
 545                 goto done;
 546
 547         /*
 548          * Use XOP for cluster scan.
 549          *
 550          * parent is the inode cluster, already locked for us.  Don't
 551          * double lock shared locks as this will screw up upgrades.
 552          */
 553         xop = &hammer2_xop_alloc(ip)->xop_readdir;
 554         xop->head.lkey = lkey;
 555         hammer2_xop_start(&xop->head, hammer2_xop_readdir);
 556
 557         for (;;) {
 558                 const hammer2_inode_data_t *ripdata;
 559
 560                 error = hammer2_xop_collect(&xop->head, 0);
 561                 if (error)
 562                         break;
 563                 if (cookie_index == ncookies)
 564                         break;
 565                 if (hammer2_debug & 0x0020)
 566                 kprintf("cluster chain %p %p\n",
 567                         xop->head.cluster.focus,
 568                         (xop->head.cluster.focus ?
 569                          xop->head.cluster.focus->data : (void *)-1));
 570                 ripdata = &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
 571                 hammer2_cluster_bref(&xop->head.cluster, &bref);
 572                 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
 573                         dtype = hammer2_get_dtype(ripdata);
 574                         saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 575                         r = vop_write_dirent(&error, uio,
 576                                              ripdata->meta.inum &
 577                                               HAMMER2_DIRHASH_USERMSK,
 578                                              dtype,
 579                                              ripdata->meta.name_len,
 580                                              ripdata->filename);
 581                         if (r)
 582                                 break;
 583                         if (cookies)
 584                                 cookies[cookie_index] = saveoff;
 585                         ++cookie_index;
 586                 } else {
 587                         /* XXX chain error */
 588                         kprintf("bad chain type readdir %d\n", bref.type);
 589                 }
 590         }
 591         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
 592         if (error == ENOENT) {
 593                 error = 0;
 594                 eofflag = 1;
 595                 saveoff = (hammer2_key_t)-1;
 596         } else {
 597                 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 598         }
 599 done:
 600         hammer2_inode_unlock(ip, NULL);
 601         if (ap->a_eofflag)
 602                 *ap->a_eofflag = eofflag;
 603         if (hammer2_debug & 0x0020)
 604                 kprintf("readdir: done at %016jx\n", saveoff);
 605         uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
 606         if (error && cookie_index == 0) {
 607                 if (cookies) {
 608                         kfree(cookies, M_TEMP);
 609                         *ap->a_ncookies = 0;
 610                         *ap->a_cookies = NULL;
 611                 }
 612         } else {
 613                 if (cookies) {
 614                         *ap->a_ncookies = cookie_index;
 615                         *ap->a_cookies = cookies;
 616                 }
 617         }
 618         LOCKSTOP;
 619         return (error);
 620 }
 621
 622 /*
 623  * hammer2_vop_readlink { vp, uio, cred }
 624  */
 625 static
 626 int
 627 hammer2_vop_readlink(struct vop_readlink_args *ap)
 628 {
 629         struct vnode *vp;
 630         hammer2_inode_t *ip;
 631         int error;
 632
 633         vp = ap->a_vp;
 634         if (vp->v_type != VLNK)
 635                 return (EINVAL);
 636         ip = VTOI(vp);
 637
 638         error = hammer2_read_file(ip, ap->a_uio, 0);
 639         return (error);
 640 }
 641
 642 static
 643 int
 644 hammer2_vop_read(struct vop_read_args *ap)
 645 {
 646         struct vnode *vp;
 647         hammer2_inode_t *ip;
 648         struct uio *uio;
 649         int error;
 650         int seqcount;
 651         int bigread;
 652
 653         /*
 654          * Read operations supported on this vnode?
 655          */
 656         vp = ap->a_vp;
 657         if (vp->v_type != VREG)
 658                 return (EINVAL);
 659
 660         /*
 661          * Misc
 662          */
 663         ip = VTOI(vp);
 664         uio = ap->a_uio;
 665         error = 0;
 666
 667         seqcount = ap->a_ioflag >> 16;
 668         bigread = (uio->uio_resid > 100 * 1024 * 1024);
 669
 670         error = hammer2_read_file(ip, uio, seqcount);
 671         return (error);
 672 }
 673
 674 static
 675 int
 676 hammer2_vop_write(struct vop_write_args *ap)
 677 {
 678         hammer2_inode_t *ip;
 679         thread_t td;
 680         struct vnode *vp;
 681         struct uio *uio;
 682         int error;
 683         int seqcount;
 684
 685         /*
 686          * Read operations supported on this vnode?
 687          */
 688         vp = ap->a_vp;
 689         if (vp->v_type != VREG)
 690                 return (EINVAL);
 691
 692         /*
 693          * Misc
 694          */
 695         ip = VTOI(vp);
 696         uio = ap->a_uio;
 697         error = 0;
 698         if (ip->pmp->ronly) {
 699                 return (EROFS);
 700         }
 701
 702         seqcount = ap->a_ioflag >> 16;
 703
 704         /*
 705          * Check resource limit
 706          */
 707         if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
 708             uio->uio_offset + uio->uio_resid >
 709              td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 710                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 711                 return (EFBIG);
 712         }
 713
 714         /*
 715          * The transaction interlocks against flushes initiations
 716          * (note: but will run concurrently with the actual flush).
 717          */
 718         hammer2_trans_init(ip->pmp, 0);
 719         error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
 720         hammer2_trans_done(ip->pmp);
 721
 722         return (error);
 723 }
 724
 725 /*
 726  * Perform read operations on a file or symlink given an UNLOCKED
 727  * inode and uio.
 728  *
 729  * The passed ip is not locked.
 730  */
 731 static
 732 int
 733 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
 734 {
 735         hammer2_off_t size;
 736         struct buf *bp;
 737         int error;
 738
 739         error = 0;
 740
 741         /*
 742          * UIO read loop.
 743          *
 744          * WARNING! Assumes that the kernel interlocks size changes at the
 745          *          vnode level.
 746          */
 747         hammer2_mtx_sh(&ip->lock);
 748         size = ip->meta.size;
 749         hammer2_mtx_unlock(&ip->lock);
 750
 751         while (uio->uio_resid > 0 && uio->uio_offset < size) {
 752                 hammer2_key_t lbase;
 753                 hammer2_key_t leof;
 754                 int lblksize;
 755                 int loff;
 756                 int n;
 757
 758                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 759                                                 &lbase, &leof);
 760
 761                 error = cluster_read(ip->vp, leof, lbase, lblksize,
 762                                      uio->uio_resid, seqcount * BKVASIZE,
 763                                      &bp);
 764
 765                 if (error)
 766                         break;
 767                 loff = (int)(uio->uio_offset - lbase);
 768                 n = lblksize - loff;
 769                 if (n > uio->uio_resid)
 770                         n = uio->uio_resid;
 771                 if (n > size - uio->uio_offset)
 772                         n = (int)(size - uio->uio_offset);
 773                 bp->b_flags |= B_AGE;
 774                 uiomove((char *)bp->b_data + loff, n, uio);
 775                 bqrelse(bp);
 776         }
 777         return (error);
 778 }
 779
 780 /*
 781  * Write to the file represented by the inode via the logical buffer cache.
 782  * The inode may represent a regular file or a symlink.
 783  *
 784  * The inode must not be locked.
 785  */
 786 static
 787 int
 788 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
 789                    int ioflag, int seqcount)
 790 {
 791         hammer2_key_t old_eof;
 792         hammer2_key_t new_eof;
 793         struct buf *bp;
 794         int kflags;
 795         int error;
 796         int modified;
 797
 798         /*
 799          * Setup if append
 800          *
 801          * WARNING! Assumes that the kernel interlocks size changes at the
 802          *          vnode level.
 803          */
 804         hammer2_mtx_ex(&ip->lock);
 805         if (ioflag & IO_APPEND)
 806                 uio->uio_offset = ip->meta.size;
 807         old_eof = ip->meta.size;
 808
 809         /*
 810          * Extend the file if necessary.  If the write fails at some point
 811          * we will truncate it back down to cover as much as we were able
 812          * to write.
 813          *
 814          * Doing this now makes it easier to calculate buffer sizes in
 815          * the loop.
 816          */
 817         kflags = 0;
 818         error = 0;
 819         modified = 0;
 820
 821         if (uio->uio_offset + uio->uio_resid > old_eof) {
 822                 new_eof = uio->uio_offset + uio->uio_resid;
 823                 modified = 1;
 824                 hammer2_extend_file(ip, new_eof);
 825                 kflags |= NOTE_EXTEND;
 826         } else {
 827                 new_eof = old_eof;
 828         }
 829         hammer2_mtx_unlock(&ip->lock);
 830
 831         /*
 832          * UIO write loop
 833          */
 834         while (uio->uio_resid > 0) {
 835                 hammer2_key_t lbase;
 836                 int trivial;
 837                 int endofblk;
 838                 int lblksize;
 839                 int loff;
 840                 int n;
 841
 842                 /*
 843                  * Don't allow the buffer build to blow out the buffer
 844                  * cache.
 845                  */
 846                 if ((ioflag & IO_RECURSE) == 0)
 847                         bwillwrite(HAMMER2_PBUFSIZE);
 848
 849                 /*
 850                  * This nominally tells us how much we can cluster and
 851                  * what the logical buffer size needs to be.  Currently
 852                  * we don't try to cluster the write and just handle one
 853                  * block at a time.
 854                  */
 855                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 856                                                 &lbase, NULL);
 857                 loff = (int)(uio->uio_offset - lbase);
 858
 859                 KKASSERT(lblksize <= 65536);
 860
 861                 /*
 862                  * Calculate bytes to copy this transfer and whether the
 863                  * copy completely covers the buffer or not.
 864                  */
 865                 trivial = 0;
 866                 n = lblksize - loff;
 867                 if (n > uio->uio_resid) {
 868                         n = uio->uio_resid;
 869                         if (loff == lbase && uio->uio_offset + n == new_eof)
 870                                 trivial = 1;
 871                         endofblk = 0;
 872                 } else {
 873                         if (loff == 0)
 874                                 trivial = 1;
 875                         endofblk = 1;
 876                 }
 877
 878                 /*
 879                  * Get the buffer
 880                  */
 881                 if (uio->uio_segflg == UIO_NOCOPY) {
 882                         /*
 883                          * Issuing a write with the same data backing the
 884                          * buffer.  Instantiate the buffer to collect the
 885                          * backing vm pages, then read-in any missing bits.
 886                          *
 887                          * This case is used by vop_stdputpages().
 888                          */
 889                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 890                         if ((bp->b_flags & B_CACHE) == 0) {
 891                                 bqrelse(bp);
 892                                 error = bread(ip->vp, lbase, lblksize, &bp);
 893                         }
 894                 } else if (trivial) {
 895                         /*
 896                          * Even though we are entirely overwriting the buffer
 897                          * we may still have to zero it out to avoid a
 898                          * mmap/write visibility issue.
 899                          */
 900                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 901                         if ((bp->b_flags & B_CACHE) == 0)
 902                                 vfs_bio_clrbuf(bp);
 903                 } else {
 904                         /*
 905                          * Partial overwrite, read in any missing bits then
 906                          * replace the portion being written.
 907                          *
 908                          * (The strategy code will detect zero-fill physical
 909                          * blocks for this case).
 910                          */
 911                         error = bread(ip->vp, lbase, lblksize, &bp);
 912                         if (error == 0)
 913                                 bheavy(bp);
 914                 }
 915
 916                 if (error) {
 917                         brelse(bp);
 918                         break;
 919                 }
 920
 921                 /*
 922                  * Ok, copy the data in
 923                  */
 924                 error = uiomove(bp->b_data + loff, n, uio);
 925                 kflags |= NOTE_WRITE;
 926                 modified = 1;
 927                 if (error) {
 928                         brelse(bp);
 929                         break;
 930                 }
 931
 932                 /*
 933                  * WARNING: Pageout daemon will issue UIO_NOCOPY writes
 934                  *          with IO_SYNC or IO_ASYNC set.  These writes
 935                  *          must be handled as the pageout daemon expects.
 936                  */
 937                 if (ioflag & IO_SYNC) {
 938                         bwrite(bp);
 939                 } else if ((ioflag & IO_DIRECT) && endofblk) {
 940                         bawrite(bp);
 941                 } else if (ioflag & IO_ASYNC) {
 942                         bawrite(bp);
 943                 } else {
 944                         bdwrite(bp);
 945                 }
 946         }
 947
 948         /*
 949          * Cleanup.  If we extended the file EOF but failed to write through
 950          * the entire write is a failure and we have to back-up.
 951          */
 952         if (error && new_eof != old_eof) {
 953                 hammer2_mtx_ex(&ip->lock);
 954                 hammer2_truncate_file(ip, old_eof);
 955                 if (ip->flags & HAMMER2_INODE_MODIFIED)
 956                         hammer2_inode_fsync(ip, NULL);
 957                 hammer2_mtx_unlock(&ip->lock);
 958         } else if (modified) {
 959                 hammer2_mtx_ex(&ip->lock);
 960                 hammer2_inode_modify(ip);
 961                 hammer2_update_time(&ip->meta.mtime);
 962                 if (ip->flags & HAMMER2_INODE_MODIFIED)
 963                         hammer2_inode_fsync(ip, NULL);
 964                 hammer2_mtx_unlock(&ip->lock);
 965                 hammer2_knote(ip->vp, kflags);
 966         }
 967         hammer2_trans_assert_strategy(ip->pmp);
 968
 969         return error;
 970 }
 971
 972 /*
 973  * Truncate the size of a file.  The inode must not be locked.
 974  *
 975  * We must unconditionally set HAMMER2_INODE_RESIZED to properly
 976  * ensure that any on-media data beyond the new file EOF has been destroyed.
 977  *
 978  * WARNING: nvtruncbuf() can only be safely called without the inode lock
 979  *          held due to the way our write thread works.  If the truncation
 980  *          occurs in the middle of a buffer, nvtruncbuf() is responsible
 981  *          for dirtying that buffer and zeroing out trailing bytes.
 982  *
 983  * WARNING! Assumes that the kernel interlocks size changes at the
 984  *          vnode level.
 985  *
 986  * WARNING! Caller assumes responsibility for removing dead blocks
 987  *          if INODE_RESIZED is set.
 988  */
 989 static
 990 void
 991 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
 992 {
 993         hammer2_key_t lbase;
 994         int nblksize;
 995
 996         LOCKSTART;
 997         hammer2_mtx_unlock(&ip->lock);
 998         if (ip->vp) {
 999                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1000                 nvtruncbuf(ip->vp, nsize,
1001                            nblksize, (int)nsize & (nblksize - 1),
1002                            0);
1003         }
1004         hammer2_mtx_ex(&ip->lock);
1005         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1006         ip->osize = ip->meta.size;
1007         ip->meta.size = nsize;
1008         atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED |
1009                                    HAMMER2_INODE_RESIZED);
1010         LOCKSTOP;
1011 }
1012
1013 /*
1014  * Extend the size of a file.  The inode must not be locked.
1015  *
1016  * Even though the file size is changing, we do not have to set the
1017  * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1018  * boundary.  When this occurs a hammer2_inode_fsync() is required
1019  * to prepare the inode cluster's indirect block table.
1020  *
1021  * WARNING! Assumes that the kernel interlocks size changes at the
1022  *          vnode level.
1023  *
1024  * WARNING! Caller assumes responsibility for transitioning out
1025  *          of the inode DIRECTDATA mode if INODE_RESIZED is set.
1026  */
1027 static
1028 void
1029 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1030 {
1031         hammer2_key_t lbase;
1032         hammer2_key_t osize;
1033         int oblksize;
1034         int nblksize;
1035
1036         LOCKSTART;
1037
1038         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1039         osize = ip->meta.size;
1040         ip->osize = osize;
1041         ip->meta.size = nsize;
1042         atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1043
1044         if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES)
1045                 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1046
1047         hammer2_mtx_unlock(&ip->lock);
1048         if (ip->vp) {
1049                 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1050                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1051                 nvextendbuf(ip->vp,
1052                             osize, nsize,
1053                             oblksize, nblksize,
1054                             -1, -1, 0);
1055         }
1056         hammer2_mtx_ex(&ip->lock);
1057
1058         LOCKSTOP;
1059 }
1060
1061 static
1062 int
1063 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1064 {
1065         hammer2_xop_nresolve_t *xop;
1066         hammer2_inode_t *ip;
1067         hammer2_inode_t *dip;
1068         struct namecache *ncp;
1069         struct vnode *vp;
1070         int error;
1071
1072         LOCKSTART;
1073         dip = VTOI(ap->a_dvp);
1074         xop = &hammer2_xop_alloc(dip)->xop_nresolve;
1075
1076         ncp = ap->a_nch->ncp;
1077         xop->name = ncp->nc_name;
1078         xop->name_len = ncp->nc_nlen;
1079
1080         /*
1081          * Note: In DragonFly the kernel handles '.' and '..'.
1082          */
1083         hammer2_inode_lock(dip, HAMMER2_RESOLVE_ALWAYS |
1084                                 HAMMER2_RESOLVE_SHARED);
1085         hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1086
1087         error = hammer2_xop_collect(&xop->head, 0);
1088         if (error) {
1089                 ip = NULL;
1090         } else {
1091                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster);
1092         }
1093         hammer2_inode_unlock(dip, NULL);
1094
1095         /*
1096          * Acquire the related vnode
1097          *
1098          * NOTE: For error processing, only ENOENT resolves the namecache
1099          *       entry to NULL, otherwise we just return the error and
1100          *       leave the namecache unresolved.
1101          *
1102          * NOTE: multiple hammer2_inode structures can be aliased to the
1103          *       same chain element, for example for hardlinks.  This
1104          *       use case does not 'reattach' inode associations that
1105          *       might already exist, but always allocates a new one.
1106          *
1107          * WARNING: inode structure is locked exclusively via inode_get
1108          *          but chain was locked shared.  inode_unlock()
1109          *          will handle it properly.
1110          */
1111         if (ip) {
1112                 vp = hammer2_igetv(ip, &error);
1113                 if (error == 0) {
1114                         vn_unlock(vp);
1115                         cache_setvp(ap->a_nch, vp);
1116                 } else if (error == ENOENT) {
1117                         cache_setvp(ap->a_nch, NULL);
1118                 }
1119                 hammer2_inode_unlock(ip, NULL);
1120
1121                 /*
1122                  * The vp should not be released until after we've disposed
1123                  * of our locks, because it might cause vop_inactive() to
1124                  * be called.
1125                  */
1126                 if (vp)
1127                         vrele(vp);
1128         } else {
1129                 error = ENOENT;
1130                 cache_setvp(ap->a_nch, NULL);
1131         }
1132         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1133         KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1134                 ("resolve error %d/%p ap %p\n",
1135                  error, ap->a_nch->ncp->nc_vp, ap));
1136         LOCKSTOP;
1137
1138         return error;
1139 }
1140
1141 static
1142 int
1143 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1144 {
1145         hammer2_inode_t *dip;
1146         hammer2_inode_t *ip;
1147         int error;
1148
1149         LOCKSTART;
1150         dip = VTOI(ap->a_dvp);
1151
1152         if ((ip = dip->pip) == NULL) {
1153                 *ap->a_vpp = NULL;
1154                 LOCKSTOP;
1155                 return ENOENT;
1156         }
1157         hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
1158         *ap->a_vpp = hammer2_igetv(ip, &error);
1159         hammer2_inode_unlock(ip, NULL);
1160
1161         LOCKSTOP;
1162         return error;
1163 }
1164
1165 static
1166 int
1167 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1168 {
1169         hammer2_inode_t *dip;
1170         hammer2_inode_t *nip;
1171         struct namecache *ncp;
1172         const uint8_t *name;
1173         size_t name_len;
1174         int error;
1175
1176         LOCKSTART;
1177         dip = VTOI(ap->a_dvp);
1178         if (dip->pmp->ronly) {
1179                 LOCKSTOP;
1180                 return (EROFS);
1181         }
1182
1183         ncp = ap->a_nch->ncp;
1184         name = ncp->nc_name;
1185         name_len = ncp->nc_nlen;
1186
1187         hammer2_pfs_memory_wait(dip->pmp);
1188         hammer2_trans_init(dip->pmp, 0);
1189         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1190                                    name, name_len, 0, &error);
1191         if (error) {
1192                 KKASSERT(nip == NULL);
1193                 *ap->a_vpp = NULL;
1194         } else {
1195                 *ap->a_vpp = hammer2_igetv(nip, &error);
1196                 hammer2_inode_unlock(nip, NULL);
1197         }
1198         hammer2_trans_done(dip->pmp);
1199
1200         if (error == 0) {
1201                 cache_setunresolved(ap->a_nch);
1202                 cache_setvp(ap->a_nch, *ap->a_vpp);
1203         }
1204         LOCKSTOP;
1205         return error;
1206 }
1207
1208 static
1209 int
1210 hammer2_vop_open(struct vop_open_args *ap)
1211 {
1212         return vop_stdopen(ap);
1213 }
1214
1215 /*
1216  * hammer2_vop_advlock { vp, id, op, fl, flags }
1217  */
1218 static
1219 int
1220 hammer2_vop_advlock(struct vop_advlock_args *ap)
1221 {
1222         hammer2_inode_t *ip = VTOI(ap->a_vp);
1223         const hammer2_inode_data_t *ripdata;
1224         hammer2_cluster_t *cparent;
1225         hammer2_off_t size;
1226
1227         hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
1228                                HAMMER2_RESOLVE_SHARED);
1229         cparent = hammer2_inode_cluster(ip, HAMMER2_RESOLVE_ALWAYS |
1230                                             HAMMER2_RESOLVE_SHARED);
1231         ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
1232         size = ripdata->meta.size;
1233         hammer2_inode_unlock(ip, cparent);
1234         return (lf_advlock(ap, &ip->advlock, size));
1235 }
1236
1237
1238 static
1239 int
1240 hammer2_vop_close(struct vop_close_args *ap)
1241 {
1242         return vop_stdclose(ap);
1243 }
1244
1245 /*
1246  * hammer2_vop_nlink { nch, dvp, vp, cred }
1247  *
1248  * Create a hardlink from (vp) to {dvp, nch}.
1249  */
1250 static
1251 int
1252 hammer2_vop_nlink(struct vop_nlink_args *ap)
1253 {
1254         hammer2_inode_t *fdip;  /* target directory to create link in */
1255         hammer2_inode_t *tdip;  /* target directory to create link in */
1256         hammer2_inode_t *cdip;  /* common parent directory */
1257         hammer2_inode_t *ip;    /* inode we are hardlinking to */
1258         hammer2_cluster_t *cluster;
1259         hammer2_cluster_t *fdcluster;
1260         hammer2_cluster_t *tdcluster;
1261         hammer2_cluster_t *cdcluster;
1262         struct namecache *ncp;
1263         const uint8_t *name;
1264         size_t name_len;
1265         int error;
1266
1267         LOCKSTART;
1268         tdip = VTOI(ap->a_dvp);
1269         if (tdip->pmp->ronly) {
1270                 LOCKSTOP;
1271                 return (EROFS);
1272         }
1273
1274         ncp = ap->a_nch->ncp;
1275         name = ncp->nc_name;
1276         name_len = ncp->nc_nlen;
1277
1278         /*
1279          * ip represents the file being hardlinked.  The file could be a
1280          * normal file or a hardlink target if it has already been hardlinked.
1281          * If ip is a hardlinked target then ip->pip represents the location
1282          * of the hardlinked target, NOT the location of the hardlink pointer.
1283          *
1284          * Bump nlinks and potentially also create or move the hardlink
1285          * target in the parent directory common to (ip) and (tdip).  The
1286          * consolidation code can modify ip->cluster and ip->pip.  The
1287          * returned cluster is locked.
1288          */
1289         ip = VTOI(ap->a_vp);
1290         hammer2_pfs_memory_wait(ip->pmp);
1291         hammer2_trans_init(ip->pmp, 0);
1292
1293         /*
1294          * The common parent directory must be locked first to avoid deadlocks.
1295          * Also note that fdip and/or tdip might match cdip.
1296          */
1297         fdip = ip->pip;
1298         cdip = hammer2_inode_common_parent(fdip, tdip);
1299         hammer2_inode_lock(cdip, HAMMER2_RESOLVE_ALWAYS);
1300         hammer2_inode_lock(fdip, HAMMER2_RESOLVE_ALWAYS);
1301         hammer2_inode_lock(tdip, HAMMER2_RESOLVE_ALWAYS);
1302         cdcluster = hammer2_inode_cluster(cdip, HAMMER2_RESOLVE_ALWAYS);
1303         fdcluster = hammer2_inode_cluster(fdip, HAMMER2_RESOLVE_ALWAYS);
1304         tdcluster = hammer2_inode_cluster(tdip, HAMMER2_RESOLVE_ALWAYS);
1305
1306         hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
1307         cluster = hammer2_inode_cluster(ip, HAMMER2_RESOLVE_ALWAYS);
1308
1309         error = hammer2_cluster_hardlink_consolidate(ip, &cluster,
1310                                                      cdip, cdcluster, 1);
1311         if (error)
1312                 goto done;
1313
1314         /*
1315          * Create a directory entry connected to the specified cluster.
1316          *
1317          * WARNING! chain can get moved by the connect (indirectly due to
1318          *          potential indirect block creation).
1319          */
1320         error = hammer2_inode_connect(ip, &cluster, 1,
1321                                       tdip, tdcluster,
1322                                       name, name_len, 0);
1323         if (error == 0) {
1324                 cache_setunresolved(ap->a_nch);
1325                 cache_setvp(ap->a_nch, ap->a_vp);
1326         }
1327 done:
1328         hammer2_inode_unlock(ip, cluster);
1329         hammer2_inode_unlock(tdip, tdcluster);
1330         hammer2_inode_unlock(fdip, fdcluster);
1331         hammer2_inode_unlock(cdip, cdcluster);
1332         hammer2_inode_drop(cdip);
1333         hammer2_trans_done(ip->pmp);
1334
1335         LOCKSTOP;
1336         return error;
1337 }
1338
1339 /*
1340  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1341  *
1342  * The operating system has already ensured that the directory entry
1343  * does not exist and done all appropriate namespace locking.
1344  */
1345 static
1346 int
1347 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1348 {
1349         hammer2_inode_t *dip;
1350         hammer2_inode_t *nip;
1351         struct namecache *ncp;
1352         const uint8_t *name;
1353         size_t name_len;
1354         int error;
1355
1356         LOCKSTART;
1357         dip = VTOI(ap->a_dvp);
1358         if (dip->pmp->ronly) {
1359                 LOCKSTOP;
1360                 return (EROFS);
1361         }
1362
1363         ncp = ap->a_nch->ncp;
1364         name = ncp->nc_name;
1365         name_len = ncp->nc_nlen;
1366         hammer2_pfs_memory_wait(dip->pmp);
1367         hammer2_trans_init(dip->pmp, 0);
1368
1369         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1370                                    name, name_len, 0, &error);
1371         if (error) {
1372                 KKASSERT(nip == NULL);
1373                 *ap->a_vpp = NULL;
1374         } else {
1375                 *ap->a_vpp = hammer2_igetv(nip, &error);
1376                 hammer2_inode_unlock(nip, NULL);
1377         }
1378         hammer2_trans_done(dip->pmp);
1379
1380         if (error == 0) {
1381                 cache_setunresolved(ap->a_nch);
1382                 cache_setvp(ap->a_nch, *ap->a_vpp);
1383         }
1384         LOCKSTOP;
1385         return error;
1386 }
1387
1388 /*
1389  * Make a device node (typically a fifo)
1390  */
1391 static
1392 int
1393 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1394 {
1395         hammer2_inode_t *dip;
1396         hammer2_inode_t *nip;
1397         struct namecache *ncp;
1398         const uint8_t *name;
1399         size_t name_len;
1400         int error;
1401
1402         LOCKSTART;
1403         dip = VTOI(ap->a_dvp);
1404         if (dip->pmp->ronly) {
1405                 LOCKSTOP;
1406                 return (EROFS);
1407         }
1408
1409         ncp = ap->a_nch->ncp;
1410         name = ncp->nc_name;
1411         name_len = ncp->nc_nlen;
1412         hammer2_pfs_memory_wait(dip->pmp);
1413         hammer2_trans_init(dip->pmp, 0);
1414
1415         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1416                                    name, name_len, 0, &error);
1417         if (error) {
1418                 KKASSERT(nip == NULL);
1419                 *ap->a_vpp = NULL;
1420         } else {
1421                 *ap->a_vpp = hammer2_igetv(nip, &error);
1422                 hammer2_inode_unlock(nip, NULL);
1423         }
1424         hammer2_trans_done(dip->pmp);
1425
1426         if (error == 0) {
1427                 cache_setunresolved(ap->a_nch);
1428                 cache_setvp(ap->a_nch, *ap->a_vpp);
1429         }
1430         LOCKSTOP;
1431         return error;
1432 }
1433
1434 /*
1435  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1436  */
1437 static
1438 int
1439 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1440 {
1441         hammer2_inode_t *dip;
1442         hammer2_inode_t *nip;
1443         struct namecache *ncp;
1444         const uint8_t *name;
1445         size_t name_len;
1446         int error;
1447
1448         dip = VTOI(ap->a_dvp);
1449         if (dip->pmp->ronly)
1450                 return (EROFS);
1451
1452         ncp = ap->a_nch->ncp;
1453         name = ncp->nc_name;
1454         name_len = ncp->nc_nlen;
1455         hammer2_pfs_memory_wait(dip->pmp);
1456         hammer2_trans_init(dip->pmp, 0);
1457
1458         ap->a_vap->va_type = VLNK;      /* enforce type */
1459
1460         nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1461                                    name, name_len, 0, &error);
1462         if (error) {
1463                 KKASSERT(nip == NULL);
1464                 *ap->a_vpp = NULL;
1465                 hammer2_trans_done(dip->pmp);
1466                 return error;
1467         }
1468         *ap->a_vpp = hammer2_igetv(nip, &error);
1469
1470         /*
1471          * Build the softlink (~like file data) and finalize the namecache.
1472          */
1473         if (error == 0) {
1474                 size_t bytes;
1475                 struct uio auio;
1476                 struct iovec aiov;
1477
1478                 bytes = strlen(ap->a_target);
1479
1480 #if 0
1481                 if (bytes <= HAMMER2_EMBEDDED_BYTES) {
1482                         KKASSERT(nipdata->meta.op_flags &
1483                                  HAMMER2_OPFLAG_DIRECTDATA);
1484                         bcopy(ap->a_target, nipdata->u.data, bytes);
1485                         nipdata->meta.size = bytes;
1486                         nip->meta.size = bytes;
1487                         hammer2_cluster_modsync(ncparent);
1488                         hammer2_inode_unlock(nip, ncparent);
1489                         /* nipdata = NULL; not needed */
1490                 } else
1491 #endif
1492                 {
1493                         hammer2_inode_unlock(nip, NULL);
1494                         bzero(&auio, sizeof(auio));
1495                         bzero(&aiov, sizeof(aiov));
1496                         auio.uio_iov = &aiov;
1497                         auio.uio_segflg = UIO_SYSSPACE;
1498                         auio.uio_rw = UIO_WRITE;
1499                         auio.uio_resid = bytes;
1500                         auio.uio_iovcnt = 1;
1501                         auio.uio_td = curthread;
1502                         aiov.iov_base = ap->a_target;
1503                         aiov.iov_len = bytes;
1504                         error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1505                         /* XXX handle error */
1506                         error = 0;
1507                 }
1508         } else {
1509                 hammer2_inode_unlock(nip, NULL);
1510         }
1511         hammer2_trans_done(dip->pmp);
1512
1513         /*
1514          * Finalize namecache
1515          */
1516         if (error == 0) {
1517                 cache_setunresolved(ap->a_nch);
1518                 cache_setvp(ap->a_nch, *ap->a_vpp);
1519                 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1520         }
1521         return error;
1522 }
1523
1524 /*
1525  * hammer2_vop_nremove { nch, dvp, cred }
1526  */
1527 static
1528 int
1529 hammer2_vop_nremove(struct vop_nremove_args *ap)
1530 {
1531         hammer2_inode_t *dip;
1532         struct namecache *ncp;
1533         const uint8_t *name;
1534         size_t name_len;
1535         int error;
1536
1537         LOCKSTART;
1538         dip = VTOI(ap->a_dvp);
1539         if (dip->pmp->ronly) {
1540                 LOCKSTOP;
1541                 return(EROFS);
1542         }
1543
1544         ncp = ap->a_nch->ncp;
1545         name = ncp->nc_name;
1546         name_len = ncp->nc_nlen;
1547
1548         hammer2_pfs_memory_wait(dip->pmp);
1549         hammer2_trans_init(dip->pmp, 0);
1550         error = hammer2_unlink_file(dip, NULL, name, name_len,
1551                                     0, NULL, ap->a_nch, -1);
1552         hammer2_run_unlinkq(dip->pmp);
1553         hammer2_trans_done(dip->pmp);
1554         if (error == 0)
1555                 cache_unlink(ap->a_nch);
1556         LOCKSTOP;
1557         return (error);
1558 }
1559
1560 /*
1561  * hammer2_vop_nrmdir { nch, dvp, cred }
1562  */
1563 static
1564 int
1565 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1566 {
1567         hammer2_inode_t *dip;
1568         struct namecache *ncp;
1569         const uint8_t *name;
1570         size_t name_len;
1571         int error;
1572
1573         LOCKSTART;
1574         dip = VTOI(ap->a_dvp);
1575         if (dip->pmp->ronly) {
1576                 LOCKSTOP;
1577                 return(EROFS);
1578         }
1579
1580         ncp = ap->a_nch->ncp;
1581         name = ncp->nc_name;
1582         name_len = ncp->nc_nlen;
1583
1584         hammer2_pfs_memory_wait(dip->pmp);
1585         hammer2_trans_init(dip->pmp, 0);
1586         hammer2_run_unlinkq(dip->pmp);
1587         error = hammer2_unlink_file(dip, NULL, name, name_len,
1588                                     1, NULL, ap->a_nch, -1);
1589         hammer2_trans_done(dip->pmp);
1590         if (error == 0)
1591                 cache_unlink(ap->a_nch);
1592         LOCKSTOP;
1593         return (error);
1594 }
1595
1596 /*
1597  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1598  */
1599 static
1600 int
1601 hammer2_vop_nrename(struct vop_nrename_args *ap)
1602 {
1603         struct namecache *fncp;
1604         struct namecache *tncp;
1605         hammer2_inode_t *cdip;
1606         hammer2_inode_t *fdip;
1607         hammer2_inode_t *tdip;
1608         hammer2_inode_t *ip;
1609         hammer2_cluster_t *cluster;
1610         hammer2_cluster_t *fdcluster;
1611         hammer2_cluster_t *tdcluster;
1612         hammer2_cluster_t *cdcluster;
1613         const uint8_t *fname;
1614         size_t fname_len;
1615         const uint8_t *tname;
1616         size_t tname_len;
1617         int error;
1618         int tnch_error;
1619         int hlink;
1620
1621         if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1622                 return(EXDEV);
1623         if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1624                 return(EXDEV);
1625
1626         fdip = VTOI(ap->a_fdvp);        /* source directory */
1627         tdip = VTOI(ap->a_tdvp);        /* target directory */
1628
1629         if (fdip->pmp->ronly)
1630                 return(EROFS);
1631
1632         LOCKSTART;
1633         fncp = ap->a_fnch->ncp;         /* entry name in source */
1634         fname = fncp->nc_name;
1635         fname_len = fncp->nc_nlen;
1636
1637         tncp = ap->a_tnch->ncp;         /* entry name in target */
1638         tname = tncp->nc_name;
1639         tname_len = tncp->nc_nlen;
1640
1641         hammer2_pfs_memory_wait(tdip->pmp);
1642         hammer2_trans_init(tdip->pmp, 0);
1643
1644         /*
1645          * ip is the inode being renamed.  If this is a hardlink then
1646          * ip represents the actual file and not the hardlink marker.
1647          */
1648         ip = VTOI(fncp->nc_vp);
1649         cluster = NULL;
1650
1651
1652         /*
1653          * The common parent directory must be locked first to avoid deadlocks.
1654          * Also note that fdip and/or tdip might match cdip.
1655          *
1656          * WARNING! fdip may not match ip->pip.  That is, if the source file
1657          *          is already a hardlink then what we are renaming is the
1658          *          hardlink pointer, not the hardlink itself.  The hardlink
1659          *          directory (ip->pip) will already be at a common parent
1660          *          of fdrip.
1661          *
1662          *          Be sure to use ip->pip when finding the common parent
1663          *          against tdip or we might accidently move the hardlink
1664          *          target into a subdirectory that makes it inaccessible to
1665          *          other pointers.
1666          */
1667         cdip = hammer2_inode_common_parent(ip->pip, tdip);
1668         hammer2_inode_lock(cdip, HAMMER2_RESOLVE_ALWAYS);
1669         hammer2_inode_lock(fdip, HAMMER2_RESOLVE_ALWAYS);
1670         hammer2_inode_lock(tdip, HAMMER2_RESOLVE_ALWAYS);
1671         cdcluster = hammer2_inode_cluster(cdip, HAMMER2_RESOLVE_ALWAYS);
1672         fdcluster = hammer2_inode_cluster(fdip, HAMMER2_RESOLVE_ALWAYS);
1673         tdcluster = hammer2_inode_cluster(tdip, HAMMER2_RESOLVE_ALWAYS);
1674
1675         /*
1676          * Keep a tight grip on the inode so the temporary unlinking from
1677          * the source location prior to linking to the target location
1678          * does not cause the cluster to be destroyed.
1679          *
1680          * NOTE: To avoid deadlocks we cannot lock (ip) while we are
1681          *       unlinking elements from their directories.  Locking
1682          *       the nlinks field does not lock the whole inode.
1683          */
1684         hammer2_inode_ref(ip);
1685
1686         /*
1687          * Remove target if it exists.
1688          */
1689         error = hammer2_unlink_file(tdip, NULL, tname, tname_len,
1690                                     -1, NULL, ap->a_tnch, -1);
1691         tnch_error = error;
1692         if (error && error != ENOENT)
1693                 goto done2;
1694
1695         /*
1696          * When renaming a hardlinked file we may have to re-consolidate
1697          * the location of the hardlink target.
1698          *
1699          * If ip represents a regular file the consolidation code essentially
1700          * does nothing other than return the same locked cluster that was
1701          * passed in.
1702          *
1703          * The returned cluster will be locked.
1704          *
1705          * WARNING!  We do not currently have a local copy of ipdata but
1706          *           we do use one later remember that it must be reloaded
1707          *           on any modification to the inode, including connects.
1708          */
1709         hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
1710         cluster = hammer2_inode_cluster(ip, HAMMER2_RESOLVE_ALWAYS);
1711         error = hammer2_cluster_hardlink_consolidate(ip, &cluster,
1712                                                      cdip, cdcluster, 0);
1713         if (error)
1714                 goto done1;
1715
1716         /*
1717          * Disconnect (fdip, fname) from the source directory.  This will
1718          * disconnect (ip) if it represents a direct file.  If (ip) represents
1719          * a hardlink the HARDLINK pointer object will be removed but the
1720          * hardlink will stay intact.
1721          *
1722          * Always pass nch as NULL because we intend to reconnect the inode,
1723          * so we don't want hammer2_unlink_file() to rename it to the hidden
1724          * open-but-unlinked directory.
1725          *
1726          * The target cluster may be marked DELETED but will not be destroyed
1727          * since we retain our hold on ip and cluster.
1728          *
1729          * NOTE: We pass nlinks as 0 (not -1) in order to retain the file's
1730          *       link count.
1731          */
1732         error = hammer2_unlink_file(fdip, ip, fname, fname_len,
1733                                     -1, &hlink, NULL, 0);
1734         KKASSERT(error != EAGAIN);
1735         if (error)
1736                 goto done1;
1737
1738         /*
1739          * Reconnect ip to target directory using cluster.  Chains cannot
1740          * actually be moved, so this will duplicate the cluster in the new
1741          * spot and assign it to the ip, replacing the old cluster.
1742          *
1743          * WARNING: Because recursive locks are allowed and we unlinked the
1744          *          file that we have a cluster-in-hand for just above, the
1745          *          cluster might have been delete-duplicated.  We must
1746          *          refactor the cluster.
1747          *
1748          * WARNING: Chain locks can lock buffer cache buffers, to avoid
1749          *          deadlocks we want to unlock before issuing a cache_*()
1750          *          op (that might have to lock a vnode).
1751          *
1752          * NOTE:    Pass nlinks as 0 because we retained the link count from
1753          *          the unlink, so we do not have to modify it.
1754          */
1755         error = hammer2_inode_connect(ip, &cluster, hlink,
1756                                       tdip, tdcluster,
1757                                       tname, tname_len, 0);
1758         if (error == 0) {
1759                 KKASSERT(cluster != NULL);
1760                 hammer2_inode_repoint(ip, (hlink ? ip->pip : tdip), cluster);
1761         }
1762 done1:
1763         hammer2_inode_unlock(ip, cluster);
1764 done2:
1765         hammer2_inode_unlock(tdip, tdcluster);
1766         hammer2_inode_unlock(fdip, fdcluster);
1767         hammer2_inode_unlock(cdip, cdcluster);
1768         hammer2_inode_drop(ip);
1769         hammer2_inode_drop(cdip);
1770         hammer2_run_unlinkq(fdip->pmp);
1771         hammer2_trans_done(tdip->pmp);
1772
1773         /*
1774          * Issue the namecache update after unlocking all the internal
1775          * hammer structures, otherwise we might deadlock.
1776          */
1777         if (tnch_error == 0) {
1778                 cache_unlink(ap->a_tnch);
1779                 cache_setunresolved(ap->a_tnch);
1780         }
1781         if (error == 0)
1782                 cache_rename(ap->a_fnch, ap->a_tnch);
1783
1784         LOCKSTOP;
1785         return (error);
1786 }
1787
1788 /*
1789  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
1790  */
1791 static
1792 int
1793 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
1794 {
1795         hammer2_inode_t *ip;
1796         int error;
1797
1798         LOCKSTART;
1799         ip = VTOI(ap->a_vp);
1800
1801         error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
1802                               ap->a_fflag, ap->a_cred);
1803         LOCKSTOP;
1804         return (error);
1805 }
1806
1807 static
1808 int
1809 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
1810 {
1811         struct mount *mp;
1812         hammer2_pfs_t *pmp;
1813         int rc;
1814
1815         LOCKSTART;
1816         switch (ap->a_op) {
1817         case (MOUNTCTL_SET_EXPORT):
1818                 mp = ap->a_head.a_ops->head.vv_mount;
1819                 pmp = MPTOPMP(mp);
1820
1821                 if (ap->a_ctllen != sizeof(struct export_args))
1822                         rc = (EINVAL);
1823                 else
1824                         rc = vfs_export(mp, &pmp->export,
1825                                         (const struct export_args *)ap->a_ctl);
1826                 break;
1827         default:
1828                 rc = vop_stdmountctl(ap);
1829                 break;
1830         }
1831         LOCKSTOP;
1832         return (rc);
1833 }
1834
1835 /*
1836  * This handles unlinked open files after the vnode is finally dereferenced.
1837  * To avoid deadlocks it cannot be called from the normal vnode recycling
1838  * path, so we call it (1) after a unlink, rmdir, or rename, (2) on every
1839  * flush, and (3) on umount.
1840  */
1841 void
1842 hammer2_run_unlinkq(hammer2_pfs_t *pmp)
1843 {
1844         const hammer2_inode_data_t *ripdata;
1845         hammer2_inode_unlink_t *ipul;
1846         hammer2_inode_t *ip;
1847         hammer2_cluster_t *cluster;
1848         hammer2_cluster_t *cparent;
1849
1850         if (TAILQ_EMPTY(&pmp->unlinkq))
1851                 return;
1852
1853         LOCKSTART;
1854         hammer2_spin_ex(&pmp->list_spin);
1855         while ((ipul = TAILQ_FIRST(&pmp->unlinkq)) != NULL) {
1856                 TAILQ_REMOVE(&pmp->unlinkq, ipul, entry);
1857                 hammer2_spin_unex(&pmp->list_spin);
1858                 ip = ipul->ip;
1859                 kfree(ipul, pmp->minode);
1860
1861                 hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
1862                 cluster = hammer2_inode_cluster(ip, HAMMER2_RESOLVE_ALWAYS);
1863                 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1864                 if (hammer2_debug & 0x400) {
1865                         kprintf("hammer2: unlink on reclaim: %s refs=%d\n",
1866                                 ripdata->filename, ip->refs);
1867                 }
1868
1869                 /*
1870                  * NOTE: Due to optimizations to avoid I/O on the inode for
1871                  *       the last unlink, ripdata->nlinks is not necessarily
1872                  *       0 here.
1873                  */
1874                 /* KKASSERT(ripdata->nlinks == 0); (see NOTE) */
1875                 cparent = hammer2_cluster_parent(cluster);
1876                 hammer2_cluster_delete(cparent, cluster,
1877                                        HAMMER2_DELETE_PERMANENT);
1878                 hammer2_cluster_unlock(cparent);
1879                 hammer2_cluster_drop(cparent);
1880                 hammer2_inode_unlock(ip, cluster);      /* inode lock */
1881                 hammer2_inode_drop(ip);                 /* ipul ref */
1882
1883                 hammer2_spin_ex(&pmp->list_spin);
1884         }
1885         hammer2_spin_unex(&pmp->list_spin);
1886         LOCKSTOP;
1887 }
1888
1889
1890 /*
1891  * KQFILTER
1892  */
1893 static void filt_hammer2detach(struct knote *kn);
1894 static int filt_hammer2read(struct knote *kn, long hint);
1895 static int filt_hammer2write(struct knote *kn, long hint);
1896 static int filt_hammer2vnode(struct knote *kn, long hint);
1897
1898 static struct filterops hammer2read_filtops =
1899         { FILTEROP_ISFD | FILTEROP_MPSAFE,
1900           NULL, filt_hammer2detach, filt_hammer2read };
1901 static struct filterops hammer2write_filtops =
1902         { FILTEROP_ISFD | FILTEROP_MPSAFE,
1903           NULL, filt_hammer2detach, filt_hammer2write };
1904 static struct filterops hammer2vnode_filtops =
1905         { FILTEROP_ISFD | FILTEROP_MPSAFE,
1906           NULL, filt_hammer2detach, filt_hammer2vnode };
1907
1908 static
1909 int
1910 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
1911 {
1912         struct vnode *vp = ap->a_vp;
1913         struct knote *kn = ap->a_kn;
1914
1915         switch (kn->kn_filter) {
1916         case EVFILT_READ:
1917                 kn->kn_fop = &hammer2read_filtops;
1918                 break;
1919         case EVFILT_WRITE:
1920                 kn->kn_fop = &hammer2write_filtops;
1921                 break;
1922         case EVFILT_VNODE:
1923                 kn->kn_fop = &hammer2vnode_filtops;
1924                 break;
1925         default:
1926                 return (EOPNOTSUPP);
1927         }
1928
1929         kn->kn_hook = (caddr_t)vp;
1930
1931         knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1932
1933         return(0);
1934 }
1935
1936 static void
1937 filt_hammer2detach(struct knote *kn)
1938 {
1939         struct vnode *vp = (void *)kn->kn_hook;
1940
1941         knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1942 }
1943
1944 static int
1945 filt_hammer2read(struct knote *kn, long hint)
1946 {
1947         struct vnode *vp = (void *)kn->kn_hook;
1948         hammer2_inode_t *ip = VTOI(vp);
1949         off_t off;
1950
1951         if (hint == NOTE_REVOKE) {
1952                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1953                 return(1);
1954         }
1955         off = ip->meta.size - kn->kn_fp->f_offset;
1956         kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1957         if (kn->kn_sfflags & NOTE_OLDAPI)
1958                 return(1);
1959         return (kn->kn_data != 0);
1960 }
1961
1962
1963 static int
1964 filt_hammer2write(struct knote *kn, long hint)
1965 {
1966         if (hint == NOTE_REVOKE)
1967                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1968         kn->kn_data = 0;
1969         return (1);
1970 }
1971
1972 static int
1973 filt_hammer2vnode(struct knote *kn, long hint)
1974 {
1975         if (kn->kn_sfflags & hint)
1976                 kn->kn_fflags |= hint;
1977         if (hint == NOTE_REVOKE) {
1978                 kn->kn_flags |= (EV_EOF | EV_NODATA);
1979                 return (1);
1980         }
1981         return (kn->kn_fflags != 0);
1982 }
1983
1984 /*
1985  * FIFO VOPS
1986  */
1987 static
1988 int
1989 hammer2_vop_markatime(struct vop_markatime_args *ap)
1990 {
1991         hammer2_inode_t *ip;
1992         struct vnode *vp;
1993
1994         vp = ap->a_vp;
1995         ip = VTOI(vp);
1996
1997         if (ip->pmp->ronly)
1998                 return(EROFS);
1999         return(0);
2000 }
2001
2002 static
2003 int
2004 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2005 {
2006         int error;
2007
2008         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2009         if (error)
2010                 error = hammer2_vop_kqfilter(ap);
2011         return(error);
2012 }
2013
2014 /*
2015  * VOPS vector
2016  */
2017 struct vop_ops hammer2_vnode_vops = {
2018         .vop_default    = vop_defaultop,
2019         .vop_fsync      = hammer2_vop_fsync,
2020         .vop_getpages   = vop_stdgetpages,
2021         .vop_putpages   = vop_stdputpages,
2022         .vop_access     = hammer2_vop_access,
2023         .vop_advlock    = hammer2_vop_advlock,
2024         .vop_close      = hammer2_vop_close,
2025         .vop_nlink      = hammer2_vop_nlink,
2026         .vop_ncreate    = hammer2_vop_ncreate,
2027         .vop_nsymlink   = hammer2_vop_nsymlink,
2028         .vop_nremove    = hammer2_vop_nremove,
2029         .vop_nrmdir     = hammer2_vop_nrmdir,
2030         .vop_nrename    = hammer2_vop_nrename,
2031         .vop_getattr    = hammer2_vop_getattr,
2032         .vop_setattr    = hammer2_vop_setattr,
2033         .vop_readdir    = hammer2_vop_readdir,
2034         .vop_readlink   = hammer2_vop_readlink,
2035         .vop_getpages   = vop_stdgetpages,
2036         .vop_putpages   = vop_stdputpages,
2037         .vop_read       = hammer2_vop_read,
2038         .vop_write      = hammer2_vop_write,
2039         .vop_open       = hammer2_vop_open,
2040         .vop_inactive   = hammer2_vop_inactive,
2041         .vop_reclaim    = hammer2_vop_reclaim,
2042         .vop_nresolve   = hammer2_vop_nresolve,
2043         .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2044         .vop_nmkdir     = hammer2_vop_nmkdir,
2045         .vop_nmknod     = hammer2_vop_nmknod,
2046         .vop_ioctl      = hammer2_vop_ioctl,
2047         .vop_mountctl   = hammer2_vop_mountctl,
2048         .vop_bmap       = hammer2_vop_bmap,
2049         .vop_strategy   = hammer2_vop_strategy,
2050         .vop_kqfilter   = hammer2_vop_kqfilter
2051 };
2052
2053 struct vop_ops hammer2_spec_vops = {
2054         .vop_default =          vop_defaultop,
2055         .vop_fsync =            hammer2_vop_fsync,
2056         .vop_read =             vop_stdnoread,
2057         .vop_write =            vop_stdnowrite,
2058         .vop_access =           hammer2_vop_access,
2059         .vop_close =            hammer2_vop_close,
2060         .vop_markatime =        hammer2_vop_markatime,
2061         .vop_getattr =          hammer2_vop_getattr,
2062         .vop_inactive =         hammer2_vop_inactive,
2063         .vop_reclaim =          hammer2_vop_reclaim,
2064         .vop_setattr =          hammer2_vop_setattr
2065 };
2066
2067 struct vop_ops hammer2_fifo_vops = {
2068         .vop_default =          fifo_vnoperate,
2069         .vop_fsync =            hammer2_vop_fsync,
2070 #if 0
2071         .vop_read =             hammer2_vop_fiforead,
2072         .vop_write =            hammer2_vop_fifowrite,
2073 #endif
2074         .vop_access =           hammer2_vop_access,
2075 #if 0
2076         .vop_close =            hammer2_vop_fifoclose,
2077 #endif
2078         .vop_markatime =        hammer2_vop_markatime,
2079         .vop_getattr =          hammer2_vop_getattr,
2080         .vop_inactive =         hammer2_vop_inactive,
2081         .vop_reclaim =          hammer2_vop_reclaim,
2082         .vop_setattr =          hammer2_vop_setattr,
2083         .vop_kqfilter =         hammer2_vop_fifokqfilter
2084 };
2085