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_SNYC / 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 static
 211 int
 212 hammer2_vop_fsync(struct vop_fsync_args *ap)
 213 {
 214         hammer2_inode_t *ip;
 215         struct vnode *vp;
 216
 217         vp = ap->a_vp;
 218         ip = VTOI(vp);
 219
 220 #if 0
 221         /* XXX can't do this yet */
 222         hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH);
 223         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 224 #endif
 225         hammer2_trans_init(ip->pmp, 0);
 226         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 227
 228         /*
 229          * Calling chain_flush here creates a lot of duplicative
 230          * COW operations due to non-optimal vnode ordering.
 231          *
 232          * Only do it for an actual fsync() syscall.  The other forms
 233          * which call this function will eventually call chain_flush
 234          * on the volume root as a catch-all, which is far more optimal.
 235          */
 236         hammer2_inode_lock(ip, 0);
 237         if (ip->flags & HAMMER2_INODE_MODIFIED)
 238                 hammer2_inode_chain_sync(ip);
 239         hammer2_inode_unlock(ip);
 240         hammer2_trans_done(ip->pmp);
 241
 242         return (0);
 243 }
 244
 245 static
 246 int
 247 hammer2_vop_access(struct vop_access_args *ap)
 248 {
 249         hammer2_inode_t *ip = VTOI(ap->a_vp);
 250         uid_t uid;
 251         gid_t gid;
 252         int error;
 253
 254         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 255         uid = hammer2_to_unix_xid(&ip->meta.uid);
 256         gid = hammer2_to_unix_xid(&ip->meta.gid);
 257         error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
 258         hammer2_inode_unlock(ip);
 259
 260         return (error);
 261 }
 262
 263 static
 264 int
 265 hammer2_vop_getattr(struct vop_getattr_args *ap)
 266 {
 267         hammer2_pfs_t *pmp;
 268         hammer2_inode_t *ip;
 269         struct vnode *vp;
 270         struct vattr *vap;
 271         hammer2_chain_t *chain;
 272         int i;
 273
 274         vp = ap->a_vp;
 275         vap = ap->a_vap;
 276
 277         ip = VTOI(vp);
 278         pmp = ip->pmp;
 279
 280         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 281
 282         vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
 283         vap->va_fileid = ip->meta.inum;
 284         vap->va_mode = ip->meta.mode;
 285         vap->va_nlink = ip->meta.nlinks;
 286         vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
 287         vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
 288         vap->va_rmajor = 0;
 289         vap->va_rminor = 0;
 290         vap->va_size = ip->meta.size;   /* protected by shared lock */
 291         vap->va_blocksize = HAMMER2_PBUFSIZE;
 292         vap->va_flags = ip->meta.uflags;
 293         hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
 294         hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
 295         hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
 296         vap->va_gen = 1;
 297         vap->va_bytes = 0;
 298         if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) {
 299                 /*
 300                  * Can't really calculate directory use sans the files under
 301                  * it, just assume one block for now.
 302                  */
 303                 vap->va_bytes += HAMMER2_INODE_BYTES;
 304         } else {
 305                 for (i = 0; i < ip->cluster.nchains; ++i) {
 306                         if ((chain = ip->cluster.array[i].chain) != NULL) {
 307                                 if (vap->va_bytes <
 308                                     chain->bref.embed.stats.data_count) {
 309                                         vap->va_bytes =
 310                                             chain->bref.embed.stats.data_count;
 311                                 }
 312                         }
 313                 }
 314         }
 315         vap->va_type = hammer2_get_vtype(ip->meta.type);
 316         vap->va_filerev = 0;
 317         vap->va_uid_uuid = ip->meta.uid;
 318         vap->va_gid_uuid = ip->meta.gid;
 319         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 320                           VA_FSID_UUID_VALID;
 321
 322         hammer2_inode_unlock(ip);
 323
 324         return (0);
 325 }
 326
 327 static
 328 int
 329 hammer2_vop_setattr(struct vop_setattr_args *ap)
 330 {
 331         hammer2_inode_t *ip;
 332         struct vnode *vp;
 333         struct vattr *vap;
 334         int error;
 335         int kflags = 0;
 336         uint64_t ctime;
 337
 338         vp = ap->a_vp;
 339         vap = ap->a_vap;
 340         hammer2_update_time(&ctime);
 341
 342         ip = VTOI(vp);
 343
 344         if (ip->pmp->ronly)
 345                 return (EROFS);
 346         if (hammer2_vfs_enospace(ip, 0, ap->a_cred) > 1)
 347                 return (ENOSPC);
 348
 349         hammer2_pfs_memory_wait(ip->pmp);
 350         hammer2_trans_init(ip->pmp, 0);
 351         hammer2_inode_lock(ip, 0);
 352         error = 0;
 353
 354         if (vap->va_flags != VNOVAL) {
 355                 uint32_t flags;
 356
 357                 flags = ip->meta.uflags;
 358                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
 359                                      hammer2_to_unix_xid(&ip->meta.uid),
 360                                      ap->a_cred);
 361                 if (error == 0) {
 362                         if (ip->meta.uflags != flags) {
 363                                 hammer2_inode_modify(ip);
 364                                 ip->meta.uflags = flags;
 365                                 ip->meta.ctime = ctime;
 366                                 kflags |= NOTE_ATTRIB;
 367                         }
 368                         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 369                                 error = 0;
 370                                 goto done;
 371                         }
 372                 }
 373                 goto done;
 374         }
 375         if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
 376                 error = EPERM;
 377                 goto done;
 378         }
 379         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
 380                 mode_t cur_mode = ip->meta.mode;
 381                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 382                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 383                 uuid_t uuid_uid;
 384                 uuid_t uuid_gid;
 385
 386                 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
 387                                          ap->a_cred,
 388                                          &cur_uid, &cur_gid, &cur_mode);
 389                 if (error == 0) {
 390                         hammer2_guid_to_uuid(&uuid_uid, cur_uid);
 391                         hammer2_guid_to_uuid(&uuid_gid, cur_gid);
 392                         if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
 393                             bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
 394                             ip->meta.mode != cur_mode
 395                         ) {
 396                                 hammer2_inode_modify(ip);
 397                                 ip->meta.uid = uuid_uid;
 398                                 ip->meta.gid = uuid_gid;
 399                                 ip->meta.mode = cur_mode;
 400                                 ip->meta.ctime = ctime;
 401                         }
 402                         kflags |= NOTE_ATTRIB;
 403                 }
 404         }
 405
 406         /*
 407          * Resize the file
 408          */
 409         if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
 410                 switch(vp->v_type) {
 411                 case VREG:
 412                         if (vap->va_size == ip->meta.size)
 413                                 break;
 414                         if (vap->va_size < ip->meta.size) {
 415                                 hammer2_mtx_ex(&ip->truncate_lock);
 416                                 hammer2_truncate_file(ip, vap->va_size);
 417                                 hammer2_mtx_unlock(&ip->truncate_lock);
 418                                 kflags |= NOTE_WRITE;
 419                         } else {
 420                                 hammer2_extend_file(ip, vap->va_size);
 421                                 kflags |= NOTE_WRITE | NOTE_EXTEND;
 422                         }
 423                         hammer2_inode_modify(ip);
 424                         ip->meta.mtime = ctime;
 425                         break;
 426                 default:
 427                         error = EINVAL;
 428                         goto done;
 429                 }
 430         }
 431 #if 0
 432         /* atime not supported */
 433         if (vap->va_atime.tv_sec != VNOVAL) {
 434                 hammer2_inode_modify(ip);
 435                 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
 436                 kflags |= NOTE_ATTRIB;
 437         }
 438 #endif
 439         if (vap->va_mode != (mode_t)VNOVAL) {
 440                 mode_t cur_mode = ip->meta.mode;
 441                 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
 442                 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
 443
 444                 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
 445                                          cur_uid, cur_gid, &cur_mode);
 446                 if (error == 0 && ip->meta.mode != cur_mode) {
 447                         hammer2_inode_modify(ip);
 448                         ip->meta.mode = cur_mode;
 449                         ip->meta.ctime = ctime;
 450                         kflags |= NOTE_ATTRIB;
 451                 }
 452         }
 453
 454         if (vap->va_mtime.tv_sec != VNOVAL) {
 455                 hammer2_inode_modify(ip);
 456                 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
 457                 kflags |= NOTE_ATTRIB;
 458         }
 459
 460 done:
 461         /*
 462          * If a truncation occurred we must call inode_fsync() now in order
 463          * to trim the related data chains, otherwise a later expansion can
 464          * cause havoc.
 465          *
 466          * If an extend occured that changed the DIRECTDATA state, we must
 467          * call inode_fsync now in order to prepare the inode's indirect
 468          * block table.
 469          */
 470         if (ip->flags & HAMMER2_INODE_RESIZED)
 471                 hammer2_inode_chain_sync(ip);
 472
 473         /*
 474          * Cleanup.
 475          */
 476         hammer2_inode_unlock(ip);
 477         hammer2_trans_done(ip->pmp);
 478         hammer2_knote(ip->vp, kflags);
 479
 480         return (error);
 481 }
 482
 483 static
 484 int
 485 hammer2_vop_readdir(struct vop_readdir_args *ap)
 486 {
 487         hammer2_xop_readdir_t *xop;
 488         hammer2_blockref_t bref;
 489         hammer2_inode_t *ip;
 490         hammer2_tid_t inum;
 491         hammer2_key_t lkey;
 492         struct uio *uio;
 493         off_t *cookies;
 494         off_t saveoff;
 495         int cookie_index;
 496         int ncookies;
 497         int error;
 498         int eofflag;
 499         int r;
 500
 501         ip = VTOI(ap->a_vp);
 502         uio = ap->a_uio;
 503         saveoff = uio->uio_offset;
 504         eofflag = 0;
 505         error = 0;
 506
 507         /*
 508          * Setup cookies directory entry cookies if requested
 509          */
 510         if (ap->a_ncookies) {
 511                 ncookies = uio->uio_resid / 16 + 1;
 512                 if (ncookies > 1024)
 513                         ncookies = 1024;
 514                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 515         } else {
 516                 ncookies = -1;
 517                 cookies = NULL;
 518         }
 519         cookie_index = 0;
 520
 521         hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
 522
 523         /*
 524          * Handle artificial entries.  To ensure that only positive 64 bit
 525          * quantities are returned to userland we always strip off bit 63.
 526          * The hash code is designed such that codes 0x0000-0x7FFF are not
 527          * used, allowing us to use these codes for articial entries.
 528          *
 529          * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
 530          * allow '..' to cross the mount point into (e.g.) the super-root.
 531          */
 532         if (saveoff == 0) {
 533                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 534                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
 535                 if (r)
 536                         goto done;
 537                 if (cookies)
 538                         cookies[cookie_index] = saveoff;
 539                 ++saveoff;
 540                 ++cookie_index;
 541                 if (cookie_index == ncookies)
 542                         goto done;
 543         }
 544
 545         if (saveoff == 1) {
 546                 /*
 547                  * Be careful with lockorder when accessing ".."
 548                  *
 549                  * (ip is the current dir. xip is the parent dir).
 550                  */
 551                 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
 552                 if (ip != ip->pmp->iroot)
 553                         inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK;
 554                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
 555                 if (r)
 556                         goto done;
 557                 if (cookies)
 558                         cookies[cookie_index] = saveoff;
 559                 ++saveoff;
 560                 ++cookie_index;
 561                 if (cookie_index == ncookies)
 562                         goto done;
 563         }
 564
 565         lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
 566         if (hammer2_debug & 0x0020)
 567                 kprintf("readdir: lkey %016jx\n", lkey);
 568         if (error)
 569                 goto done;
 570
 571         /*
 572          * Use XOP for cluster scan.
 573          *
 574          * parent is the inode cluster, already locked for us.  Don't
 575          * double lock shared locks as this will screw up upgrades.
 576          */
 577         xop = hammer2_xop_alloc(ip, 0);
 578         xop->lkey = lkey;
 579         hammer2_xop_start(&xop->head, hammer2_xop_readdir);
 580
 581         for (;;) {
 582                 const hammer2_inode_data_t *ripdata;
 583                 const char *dname;
 584                 int dtype;
 585
 586                 error = hammer2_xop_collect(&xop->head, 0);
 587                 error = hammer2_error_to_errno(error);
 588                 if (error) {
 589                         break;
 590                 }
 591                 if (cookie_index == ncookies)
 592                         break;
 593                 if (hammer2_debug & 0x0020)
 594                 kprintf("cluster chain %p %p\n",
 595                         xop->head.cluster.focus,
 596                         (xop->head.cluster.focus ?
 597                          xop->head.cluster.focus->data : (void *)-1));
 598                 hammer2_cluster_bref(&xop->head.cluster, &bref);
 599
 600                 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
 601                         ripdata =
 602                             &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
 603                         dtype = hammer2_get_dtype(ripdata->meta.type);
 604                         saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 605                         r = vop_write_dirent(&error, uio,
 606                                              ripdata->meta.inum &
 607                                               HAMMER2_DIRHASH_USERMSK,
 608                                              dtype,
 609                                              ripdata->meta.name_len,
 610                                              ripdata->filename);
 611                         if (r)
 612                                 break;
 613                         if (cookies)
 614                                 cookies[cookie_index] = saveoff;
 615                         ++cookie_index;
 616                 } else if (bref.type == HAMMER2_BREF_TYPE_DIRENT) {
 617                         dtype = hammer2_get_dtype(bref.embed.dirent.type);
 618                         saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 619                         if (bref.embed.dirent.namlen <=
 620                             sizeof(bref.check.buf)) {
 621                                 dname = bref.check.buf;
 622                         } else {
 623                                 dname =
 624                                  hammer2_cluster_rdata(&xop->head.cluster)->buf;
 625                         }
 626                         r = vop_write_dirent(&error, uio,
 627                                              bref.embed.dirent.inum,
 628                                              dtype,
 629                                              bref.embed.dirent.namlen,
 630                                              dname);
 631                         if (r)
 632                                 break;
 633                         if (cookies)
 634                                 cookies[cookie_index] = saveoff;
 635                         ++cookie_index;
 636                 } else {
 637                         /* XXX chain error */
 638                         kprintf("bad chain type readdir %d\n", bref.type);
 639                 }
 640         }
 641         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
 642         if (error == ENOENT) {
 643                 error = 0;
 644                 eofflag = 1;
 645                 saveoff = (hammer2_key_t)-1;
 646         } else {
 647                 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
 648         }
 649 done:
 650         hammer2_inode_unlock(ip);
 651         if (ap->a_eofflag)
 652                 *ap->a_eofflag = eofflag;
 653         if (hammer2_debug & 0x0020)
 654                 kprintf("readdir: done at %016jx\n", saveoff);
 655         uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
 656         if (error && cookie_index == 0) {
 657                 if (cookies) {
 658                         kfree(cookies, M_TEMP);
 659                         *ap->a_ncookies = 0;
 660                         *ap->a_cookies = NULL;
 661                 }
 662         } else {
 663                 if (cookies) {
 664                         *ap->a_ncookies = cookie_index;
 665                         *ap->a_cookies = cookies;
 666                 }
 667         }
 668         return (error);
 669 }
 670
 671 /*
 672  * hammer2_vop_readlink { vp, uio, cred }
 673  */
 674 static
 675 int
 676 hammer2_vop_readlink(struct vop_readlink_args *ap)
 677 {
 678         struct vnode *vp;
 679         hammer2_inode_t *ip;
 680         int error;
 681
 682         vp = ap->a_vp;
 683         if (vp->v_type != VLNK)
 684                 return (EINVAL);
 685         ip = VTOI(vp);
 686
 687         error = hammer2_read_file(ip, ap->a_uio, 0);
 688         return (error);
 689 }
 690
 691 static
 692 int
 693 hammer2_vop_read(struct vop_read_args *ap)
 694 {
 695         struct vnode *vp;
 696         hammer2_inode_t *ip;
 697         struct uio *uio;
 698         int error;
 699         int seqcount;
 700         int bigread;
 701
 702         /*
 703          * Read operations supported on this vnode?
 704          */
 705         vp = ap->a_vp;
 706         if (vp->v_type != VREG)
 707                 return (EINVAL);
 708
 709         /*
 710          * Misc
 711          */
 712         ip = VTOI(vp);
 713         uio = ap->a_uio;
 714         error = 0;
 715
 716         seqcount = ap->a_ioflag >> 16;
 717         bigread = (uio->uio_resid > 100 * 1024 * 1024);
 718
 719         error = hammer2_read_file(ip, uio, seqcount);
 720         return (error);
 721 }
 722
 723 static
 724 int
 725 hammer2_vop_write(struct vop_write_args *ap)
 726 {
 727         hammer2_inode_t *ip;
 728         thread_t td;
 729         struct vnode *vp;
 730         struct uio *uio;
 731         int error;
 732         int seqcount;
 733         int ioflag;
 734
 735         /*
 736          * Read operations supported on this vnode?
 737          */
 738         vp = ap->a_vp;
 739         if (vp->v_type != VREG)
 740                 return (EINVAL);
 741
 742         /*
 743          * Misc
 744          */
 745         ip = VTOI(vp);
 746         ioflag = ap->a_ioflag;
 747         uio = ap->a_uio;
 748         error = 0;
 749         if (ip->pmp->ronly)
 750                 return (EROFS);
 751         switch (hammer2_vfs_enospace(ip, uio->uio_resid, ap->a_cred)) {
 752         case 2:
 753                 return (ENOSPC);
 754         case 1:
 755                 ioflag |= IO_DIRECT;    /* semi-synchronous */
 756                 /* fall through */
 757         default:
 758                 break;
 759         }
 760
 761         seqcount = ioflag >> 16;
 762
 763         /*
 764          * Check resource limit
 765          */
 766         if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
 767             uio->uio_offset + uio->uio_resid >
 768              td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 769                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 770                 return (EFBIG);
 771         }
 772
 773         /*
 774          * The transaction interlocks against flush initiations
 775          * (note: but will run concurrently with the actual flush).
 776          *
 777          * To avoid deadlocking against the VM system, we must flag any
 778          * transaction related to the buffer cache or other direct
 779          * VM page manipulation.
 780          */
 781         if (uio->uio_segflg == UIO_NOCOPY)
 782                 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_BUFCACHE);
 783         else
 784                 hammer2_trans_init(ip->pmp, 0);
 785         error = hammer2_write_file(ip, uio, ioflag, seqcount);
 786         hammer2_trans_done(ip->pmp);
 787
 788         return (error);
 789 }
 790
 791 /*
 792  * Perform read operations on a file or symlink given an UNLOCKED
 793  * inode and uio.
 794  *
 795  * The passed ip is not locked.
 796  */
 797 static
 798 int
 799 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
 800 {
 801         hammer2_off_t size;
 802         struct buf *bp;
 803         int error;
 804
 805         error = 0;
 806
 807         /*
 808          * UIO read loop.
 809          *
 810          * WARNING! Assumes that the kernel interlocks size changes at the
 811          *          vnode level.
 812          */
 813         hammer2_mtx_sh(&ip->lock);
 814         hammer2_mtx_sh(&ip->truncate_lock);
 815         size = ip->meta.size;
 816         hammer2_mtx_unlock(&ip->lock);
 817
 818         while (uio->uio_resid > 0 && uio->uio_offset < size) {
 819                 hammer2_key_t lbase;
 820                 hammer2_key_t leof;
 821                 int lblksize;
 822                 int loff;
 823                 int n;
 824
 825                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 826                                                 &lbase, &leof);
 827
 828 #if 1
 829                 error = cluster_read(ip->vp, leof, lbase, lblksize,
 830                                      uio->uio_resid, seqcount * MAXBSIZE,
 831                                      &bp);
 832 #else
 833                 if (uio->uio_segflg == UIO_NOCOPY) {
 834                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 835                         if (bp->b_flags & B_CACHE) {
 836                                 int i;
 837                                 int j = 0;
 838                                 if (bp->b_xio.xio_npages != 16)
 839                                         kprintf("NPAGES BAD\n");
 840                                 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
 841                                         vm_page_t m;
 842                                         m = bp->b_xio.xio_pages[i];
 843                                         if (m == NULL || m->valid == 0) {
 844                                                 kprintf("bp %016jx %016jx pg %d inv",
 845                                                         lbase, leof, i);
 846                                                 if (m)
 847                                                         kprintf("m->object %p/%p", m->object, ip->vp->v_object);
 848                                                 kprintf("\n");
 849                                                 j = 1;
 850                                         }
 851                                 }
 852                                 if (j)
 853                                         kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error);
 854                         }
 855                         bqrelse(bp);
 856                 }
 857                 error = bread(ip->vp, lbase, lblksize, &bp);
 858 #endif
 859                 if (error) {
 860                         brelse(bp);
 861                         break;
 862                 }
 863                 loff = (int)(uio->uio_offset - lbase);
 864                 n = lblksize - loff;
 865                 if (n > uio->uio_resid)
 866                         n = uio->uio_resid;
 867                 if (n > size - uio->uio_offset)
 868                         n = (int)(size - uio->uio_offset);
 869                 bp->b_flags |= B_AGE;
 870                 uiomovebp(bp, (char *)bp->b_data + loff, n, uio);
 871                 bqrelse(bp);
 872         }
 873         hammer2_mtx_unlock(&ip->truncate_lock);
 874
 875         return (error);
 876 }
 877
 878 /*
 879  * Write to the file represented by the inode via the logical buffer cache.
 880  * The inode may represent a regular file or a symlink.
 881  *
 882  * The inode must not be locked.
 883  */
 884 static
 885 int
 886 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
 887                    int ioflag, int seqcount)
 888 {
 889         hammer2_key_t old_eof;
 890         hammer2_key_t new_eof;
 891         struct buf *bp;
 892         int kflags;
 893         int error;
 894         int modified;
 895
 896         /*
 897          * Setup if append
 898          *
 899          * WARNING! Assumes that the kernel interlocks size changes at the
 900          *          vnode level.
 901          */
 902         hammer2_mtx_ex(&ip->lock);
 903         hammer2_mtx_sh(&ip->truncate_lock);
 904         if (ioflag & IO_APPEND)
 905                 uio->uio_offset = ip->meta.size;
 906         old_eof = ip->meta.size;
 907
 908         /*
 909          * Extend the file if necessary.  If the write fails at some point
 910          * we will truncate it back down to cover as much as we were able
 911          * to write.
 912          *
 913          * Doing this now makes it easier to calculate buffer sizes in
 914          * the loop.
 915          */
 916         kflags = 0;
 917         error = 0;
 918         modified = 0;
 919
 920         if (uio->uio_offset + uio->uio_resid > old_eof) {
 921                 new_eof = uio->uio_offset + uio->uio_resid;
 922                 modified = 1;
 923                 hammer2_extend_file(ip, new_eof);
 924                 kflags |= NOTE_EXTEND;
 925         } else {
 926                 new_eof = old_eof;
 927         }
 928         hammer2_mtx_unlock(&ip->lock);
 929
 930         /*
 931          * UIO write loop
 932          */
 933         while (uio->uio_resid > 0) {
 934                 hammer2_key_t lbase;
 935                 int trivial;
 936                 int endofblk;
 937                 int lblksize;
 938                 int loff;
 939                 int n;
 940
 941                 /*
 942                  * Don't allow the buffer build to blow out the buffer
 943                  * cache.
 944                  */
 945                 if ((ioflag & IO_RECURSE) == 0)
 946                         bwillwrite(HAMMER2_PBUFSIZE);
 947
 948                 /*
 949                  * This nominally tells us how much we can cluster and
 950                  * what the logical buffer size needs to be.  Currently
 951                  * we don't try to cluster the write and just handle one
 952                  * block at a time.
 953                  */
 954                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 955                                                 &lbase, NULL);
 956                 loff = (int)(uio->uio_offset - lbase);
 957
 958                 KKASSERT(lblksize <= 65536);
 959
 960                 /*
 961                  * Calculate bytes to copy this transfer and whether the
 962                  * copy completely covers the buffer or not.
 963                  */
 964                 trivial = 0;
 965                 n = lblksize - loff;
 966                 if (n > uio->uio_resid) {
 967                         n = uio->uio_resid;
 968                         if (loff == lbase && uio->uio_offset + n == new_eof)
 969                                 trivial = 1;
 970                         endofblk = 0;
 971                 } else {
 972                         if (loff == 0)
 973                                 trivial = 1;
 974                         endofblk = 1;
 975                 }
 976                 if (lbase >= new_eof)
 977                         trivial = 1;
 978
 979                 /*
 980                  * Get the buffer
 981                  */
 982                 if (uio->uio_segflg == UIO_NOCOPY) {
 983                         /*
 984                          * Issuing a write with the same data backing the
 985                          * buffer.  Instantiate the buffer to collect the
 986                          * backing vm pages, then read-in any missing bits.
 987                          *
 988                          * This case is used by vop_stdputpages().
 989                          */
 990                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 991                         if ((bp->b_flags & B_CACHE) == 0) {
 992                                 bqrelse(bp);
 993                                 error = bread(ip->vp, lbase, lblksize, &bp);
 994                         }
 995                 } else if (trivial) {
 996                         /*
 997                          * Even though we are entirely overwriting the buffer
 998                          * we may still have to zero it out to avoid a
 999                          * mmap/write visibility issue.
1000                          */
1001                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1002                         if ((bp->b_flags & B_CACHE) == 0)
1003                                 vfs_bio_clrbuf(bp);
1004                 } else {
1005                         /*
1006                          * Partial overwrite, read in any missing bits then
1007                          * replace the portion being written.
1008                          *
1009                          * (The strategy code will detect zero-fill physical
1010                          * blocks for this case).
1011                          */
1012                         error = bread(ip->vp, lbase, lblksize, &bp);
1013                         if (error == 0)
1014                                 bheavy(bp);
1015                 }
1016
1017                 if (error) {
1018                         brelse(bp);
1019                         break;
1020                 }
1021
1022                 /*
1023                  * Ok, copy the data in
1024                  */
1025                 error = uiomovebp(bp, bp->b_data + loff, n, uio);
1026                 kflags |= NOTE_WRITE;
1027                 modified = 1;
1028                 if (error) {
1029                         brelse(bp);
1030                         break;
1031                 }
1032
1033                 /*
1034                  * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1035                  *          with IO_SYNC or IO_ASYNC set.  These writes
1036                  *          must be handled as the pageout daemon expects.
1037                  *
1038                  * NOTE!    H2 relies on cluster_write() here because it
1039                  *          cannot preallocate disk blocks at the logical
1040                  *          level due to not knowing what the compression
1041                  *          size will be at this time.
1042                  *
1043                  *          We must use cluster_write() here and we depend
1044                  *          on the write-behind feature to flush buffers
1045                  *          appropriately.  If we let the buffer daemons do
1046                  *          it the block allocations will be all over the
1047                  *          map.
1048                  */
1049                 if (ioflag & IO_SYNC) {
1050                         bwrite(bp);
1051                 } else if ((ioflag & IO_DIRECT) && endofblk) {
1052                         bawrite(bp);
1053                 } else if (ioflag & IO_ASYNC) {
1054                         bawrite(bp);
1055                 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) {
1056                         bdwrite(bp);
1057                 } else {
1058 #if 1
1059                         bp->b_flags |= B_CLUSTEROK;
1060                         cluster_write(bp, new_eof, lblksize, seqcount);
1061 #else
1062                         bp->b_flags |= B_CLUSTEROK;
1063                         bdwrite(bp);
1064 #endif
1065                 }
1066         }
1067
1068         /*
1069          * Cleanup.  If we extended the file EOF but failed to write through
1070          * the entire write is a failure and we have to back-up.
1071          */
1072         if (error && new_eof != old_eof) {
1073                 hammer2_mtx_unlock(&ip->truncate_lock);
1074                 hammer2_mtx_ex(&ip->lock);
1075                 hammer2_mtx_ex(&ip->truncate_lock);
1076                 hammer2_truncate_file(ip, old_eof);
1077                 if (ip->flags & HAMMER2_INODE_MODIFIED)
1078                         hammer2_inode_chain_sync(ip);
1079                 hammer2_mtx_unlock(&ip->lock);
1080         } else if (modified) {
1081                 hammer2_mtx_ex(&ip->lock);
1082                 hammer2_inode_modify(ip);
1083                 hammer2_update_time(&ip->meta.mtime);
1084                 if (ip->flags & HAMMER2_INODE_MODIFIED)
1085                         hammer2_inode_chain_sync(ip);
1086                 hammer2_mtx_unlock(&ip->lock);
1087                 hammer2_knote(ip->vp, kflags);
1088         }
1089         hammer2_trans_assert_strategy(ip->pmp);
1090         hammer2_mtx_unlock(&ip->truncate_lock);
1091
1092         return error;
1093 }
1094
1095 /*
1096  * Truncate the size of a file.  The inode must not be locked.
1097  *
1098  * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1099  * ensure that any on-media data beyond the new file EOF has been destroyed.
1100  *
1101  * WARNING: nvtruncbuf() can only be safely called without the inode lock
1102  *          held due to the way our write thread works.  If the truncation
1103  *          occurs in the middle of a buffer, nvtruncbuf() is responsible
1104  *          for dirtying that buffer and zeroing out trailing bytes.
1105  *
1106  * WARNING! Assumes that the kernel interlocks size changes at the
1107  *          vnode level.
1108  *
1109  * WARNING! Caller assumes responsibility for removing dead blocks
1110  *          if INODE_RESIZED is set.
1111  */
1112 static
1113 void
1114 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1115 {
1116         hammer2_key_t lbase;
1117         int nblksize;
1118
1119         hammer2_mtx_unlock(&ip->lock);
1120         if (ip->vp) {
1121                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1122                 nvtruncbuf(ip->vp, nsize,
1123                            nblksize, (int)nsize & (nblksize - 1),
1124                            0);
1125         }
1126         hammer2_mtx_ex(&ip->lock);
1127         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1128         ip->osize = ip->meta.size;
1129         ip->meta.size = nsize;
1130         atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1131         hammer2_inode_modify(ip);
1132 }
1133
1134 /*
1135  * Extend the size of a file.  The inode must not be locked.
1136  *
1137  * Even though the file size is changing, we do not have to set the
1138  * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1139  * boundary.  When this occurs a hammer2_inode_chain_sync() is required
1140  * to prepare the inode cluster's indirect block table, otherwise
1141  * async execution of the strategy code will implode on us.
1142  *
1143  * WARNING! Assumes that the kernel interlocks size changes at the
1144  *          vnode level.
1145  *
1146  * WARNING! Caller assumes responsibility for transitioning out
1147  *          of the inode DIRECTDATA mode if INODE_RESIZED is set.
1148  */
1149 static
1150 void
1151 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1152 {
1153         hammer2_key_t lbase;
1154         hammer2_key_t osize;
1155         int oblksize;
1156         int nblksize;
1157
1158         KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1159         hammer2_inode_modify(ip);
1160         osize = ip->meta.size;
1161         ip->osize = osize;
1162         ip->meta.size = nsize;
1163
1164         if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1165                 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1166                 hammer2_inode_chain_sync(ip);
1167         }
1168
1169         hammer2_mtx_unlock(&ip->lock);
1170         if (ip->vp) {
1171                 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1172                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1173                 nvextendbuf(ip->vp,
1174                             osize, nsize,
1175                             oblksize, nblksize,
1176                             -1, -1, 0);
1177         }
1178         hammer2_mtx_ex(&ip->lock);
1179 }
1180
1181 static
1182 int
1183 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1184 {
1185         hammer2_xop_nresolve_t *xop;
1186         hammer2_inode_t *ip;
1187         hammer2_inode_t *dip;
1188         struct namecache *ncp;
1189         struct vnode *vp;
1190         int error;
1191
1192         dip = VTOI(ap->a_dvp);
1193         xop = hammer2_xop_alloc(dip, 0);
1194
1195         ncp = ap->a_nch->ncp;
1196         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1197
1198         /*
1199          * Note: In DragonFly the kernel handles '.' and '..'.
1200          */
1201         hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1202         hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1203
1204         error = hammer2_xop_collect(&xop->head, 0);
1205         error = hammer2_error_to_errno(error);
1206         if (error) {
1207                 ip = NULL;
1208         } else {
1209                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1210         }
1211         hammer2_inode_unlock(dip);
1212
1213         /*
1214          * Acquire the related vnode
1215          *
1216          * NOTE: For error processing, only ENOENT resolves the namecache
1217          *       entry to NULL, otherwise we just return the error and
1218          *       leave the namecache unresolved.
1219          *
1220          * NOTE: multiple hammer2_inode structures can be aliased to the
1221          *       same chain element, for example for hardlinks.  This
1222          *       use case does not 'reattach' inode associations that
1223          *       might already exist, but always allocates a new one.
1224          *
1225          * WARNING: inode structure is locked exclusively via inode_get
1226          *          but chain was locked shared.  inode_unlock()
1227          *          will handle it properly.
1228          */
1229         if (ip) {
1230                 vp = hammer2_igetv(ip, &error); /* error set to UNIX error */
1231                 if (error == 0) {
1232                         vn_unlock(vp);
1233                         cache_setvp(ap->a_nch, vp);
1234                 } else if (error == ENOENT) {
1235                         cache_setvp(ap->a_nch, NULL);
1236                 }
1237                 hammer2_inode_unlock(ip);
1238
1239                 /*
1240                  * The vp should not be released until after we've disposed
1241                  * of our locks, because it might cause vop_inactive() to
1242                  * be called.
1243                  */
1244                 if (vp)
1245                         vrele(vp);
1246         } else {
1247                 error = ENOENT;
1248                 cache_setvp(ap->a_nch, NULL);
1249         }
1250         hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1251         KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1252                 ("resolve error %d/%p ap %p\n",
1253                  error, ap->a_nch->ncp->nc_vp, ap));
1254
1255         return error;
1256 }
1257
1258 static
1259 int
1260 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1261 {
1262         hammer2_inode_t *dip;
1263         hammer2_tid_t inum;
1264         int error;
1265
1266         dip = VTOI(ap->a_dvp);
1267         inum = dip->meta.iparent;
1268         *ap->a_vpp = NULL;
1269
1270         if (inum) {
1271                 error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL,
1272                                          inum, ap->a_vpp);
1273         } else {
1274                 error = ENOENT;
1275         }
1276         return error;
1277 }
1278
1279 static
1280 int
1281 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1282 {
1283         hammer2_inode_t *dip;
1284         hammer2_inode_t *nip;
1285         struct namecache *ncp;
1286         const uint8_t *name;
1287         size_t name_len;
1288         hammer2_tid_t inum;
1289         int error;
1290
1291         dip = VTOI(ap->a_dvp);
1292         if (dip->pmp->ronly)
1293                 return (EROFS);
1294         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1295                 return (ENOSPC);
1296
1297         ncp = ap->a_nch->ncp;
1298         name = ncp->nc_name;
1299         name_len = ncp->nc_nlen;
1300
1301         hammer2_pfs_memory_wait(dip->pmp);
1302         hammer2_trans_init(dip->pmp, 0);
1303
1304         inum = hammer2_trans_newinum(dip->pmp);
1305
1306         /*
1307          * Create the actual inode as a hidden file in the iroot, then
1308          * create the directory entry.  The creation of the actual inode
1309          * sets its nlinks to 1 which is the value we desire.
1310          */
1311         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1312                                    NULL, 0, inum,
1313                                    inum, 0, 0,
1314                                    0, &error);
1315         if (error) {
1316                 error = hammer2_error_to_errno(error);
1317         } else {
1318                 error = hammer2_dirent_create(dip, name, name_len,
1319                                               nip->meta.inum, nip->meta.type);
1320                 /* returns UNIX error code */
1321         }
1322         if (error) {
1323                 if (nip) {
1324                         hammer2_inode_unlink_finisher(nip, 0);
1325                         hammer2_inode_unlock(nip);
1326                         nip = NULL;
1327                 }
1328                 *ap->a_vpp = NULL;
1329         } else {
1330                 *ap->a_vpp = hammer2_igetv(nip, &error);
1331                 hammer2_inode_unlock(nip);
1332         }
1333
1334         /*
1335          * Update dip's mtime
1336          *
1337          * We can use a shared inode lock and allow the meta.mtime update
1338          * SMP race.  hammer2_inode_modify() is MPSAFE w/a shared lock.
1339          */
1340         if (error == 0) {
1341                 uint64_t mtime;
1342
1343                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1344                 hammer2_update_time(&mtime);
1345                 hammer2_inode_modify(dip);
1346                 dip->meta.mtime = mtime;
1347                 hammer2_inode_unlock(dip);
1348         }
1349
1350         hammer2_trans_done(dip->pmp);
1351
1352         if (error == 0) {
1353                 cache_setunresolved(ap->a_nch);
1354                 cache_setvp(ap->a_nch, *ap->a_vpp);
1355                 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
1356         }
1357         return error;
1358 }
1359
1360 static
1361 int
1362 hammer2_vop_open(struct vop_open_args *ap)
1363 {
1364         return vop_stdopen(ap);
1365 }
1366
1367 /*
1368  * hammer2_vop_advlock { vp, id, op, fl, flags }
1369  */
1370 static
1371 int
1372 hammer2_vop_advlock(struct vop_advlock_args *ap)
1373 {
1374         hammer2_inode_t *ip = VTOI(ap->a_vp);
1375         hammer2_off_t size;
1376
1377         size = ip->meta.size;
1378         return (lf_advlock(ap, &ip->advlock, size));
1379 }
1380
1381 static
1382 int
1383 hammer2_vop_close(struct vop_close_args *ap)
1384 {
1385         return vop_stdclose(ap);
1386 }
1387
1388 /*
1389  * hammer2_vop_nlink { nch, dvp, vp, cred }
1390  *
1391  * Create a hardlink from (vp) to {dvp, nch}.
1392  */
1393 static
1394 int
1395 hammer2_vop_nlink(struct vop_nlink_args *ap)
1396 {
1397         hammer2_inode_t *tdip;  /* target directory to create link in */
1398         hammer2_inode_t *ip;    /* inode we are hardlinking to */
1399         struct namecache *ncp;
1400         const uint8_t *name;
1401         size_t name_len;
1402         int error;
1403
1404         if (ap->a_dvp->v_mount != ap->a_vp->v_mount)
1405                 return(EXDEV);
1406
1407         tdip = VTOI(ap->a_dvp);
1408         if (tdip->pmp->ronly)
1409                 return (EROFS);
1410         if (hammer2_vfs_enospace(tdip, 0, ap->a_cred) > 1)
1411                 return (ENOSPC);
1412
1413         ncp = ap->a_nch->ncp;
1414         name = ncp->nc_name;
1415         name_len = ncp->nc_nlen;
1416
1417         /*
1418          * ip represents the file being hardlinked.  The file could be a
1419          * normal file or a hardlink target if it has already been hardlinked.
1420          * (with the new semantics, it will almost always be a hardlink
1421          * target).
1422          *
1423          * Bump nlinks and potentially also create or move the hardlink
1424          * target in the parent directory common to (ip) and (tdip).  The
1425          * consolidation code can modify ip->cluster.  The returned cluster
1426          * is locked.
1427          */
1428         ip = VTOI(ap->a_vp);
1429         KASSERT(ip->pmp, ("ip->pmp is NULL %p %p", ip, ip->pmp));
1430         hammer2_pfs_memory_wait(ip->pmp);
1431         hammer2_trans_init(ip->pmp, 0);
1432
1433         /*
1434          * Target should be an indexed inode or there's no way we will ever
1435          * be able to find it!
1436          */
1437         KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1438
1439         error = 0;
1440
1441         /*
1442          * Can return NULL and error == EXDEV if the common parent
1443          * crosses a directory with the xlink flag set.
1444          */
1445         hammer2_inode_lock(tdip, 0);
1446         hammer2_inode_lock(ip, 0);
1447
1448         /*
1449          * Create the directory entry and bump nlinks.
1450          */
1451         if (error == 0) {
1452                 error = hammer2_dirent_create(tdip, name, name_len,
1453                                               ip->meta.inum, ip->meta.type);
1454                 hammer2_inode_modify(ip);
1455                 ++ip->meta.nlinks;
1456         }
1457         if (error == 0) {
1458                 /*
1459                  * Update dip's mtime
1460                  */
1461                 uint64_t mtime;
1462
1463                 hammer2_update_time(&mtime);
1464                 hammer2_inode_modify(tdip);
1465                 tdip->meta.mtime = mtime;
1466
1467                 cache_setunresolved(ap->a_nch);
1468                 cache_setvp(ap->a_nch, ap->a_vp);
1469         }
1470         hammer2_inode_unlock(ip);
1471         hammer2_inode_unlock(tdip);
1472
1473         hammer2_trans_done(ip->pmp);
1474         hammer2_knote(ap->a_vp, NOTE_LINK);
1475         hammer2_knote(ap->a_dvp, NOTE_WRITE);
1476
1477         return error;
1478 }
1479
1480 /*
1481  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1482  *
1483  * The operating system has already ensured that the directory entry
1484  * does not exist and done all appropriate namespace locking.
1485  */
1486 static
1487 int
1488 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1489 {
1490         hammer2_inode_t *dip;
1491         hammer2_inode_t *nip;
1492         struct namecache *ncp;
1493         const uint8_t *name;
1494         size_t name_len;
1495         hammer2_tid_t inum;
1496         int error;
1497
1498         dip = VTOI(ap->a_dvp);
1499         if (dip->pmp->ronly)
1500                 return (EROFS);
1501         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1502                 return (ENOSPC);
1503
1504         ncp = ap->a_nch->ncp;
1505         name = ncp->nc_name;
1506         name_len = ncp->nc_nlen;
1507         hammer2_pfs_memory_wait(dip->pmp);
1508         hammer2_trans_init(dip->pmp, 0);
1509
1510         inum = hammer2_trans_newinum(dip->pmp);
1511
1512         /*
1513          * Create the actual inode as a hidden file in the iroot, then
1514          * create the directory entry.  The creation of the actual inode
1515          * sets its nlinks to 1 which is the value we desire.
1516          */
1517         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1518                                    NULL, 0, inum,
1519                                    inum, 0, 0,
1520                                    0, &error);
1521
1522         if (error) {
1523                 error = hammer2_error_to_errno(error);
1524         } else {
1525                 error = hammer2_dirent_create(dip, name, name_len,
1526                                               nip->meta.inum, nip->meta.type);
1527         }
1528         if (error) {
1529                 if (nip) {
1530                         hammer2_inode_unlink_finisher(nip, 0);
1531                         hammer2_inode_unlock(nip);
1532                         nip = NULL;
1533                 }
1534                 *ap->a_vpp = NULL;
1535         } else {
1536                 *ap->a_vpp = hammer2_igetv(nip, &error);
1537                 hammer2_inode_unlock(nip);
1538         }
1539
1540         /*
1541          * Update dip's mtime
1542          */
1543         if (error == 0) {
1544                 uint64_t mtime;
1545
1546                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1547                 hammer2_update_time(&mtime);
1548                 hammer2_inode_modify(dip);
1549                 dip->meta.mtime = mtime;
1550                 hammer2_inode_unlock(dip);
1551         }
1552
1553         hammer2_trans_done(dip->pmp);
1554
1555         if (error == 0) {
1556                 cache_setunresolved(ap->a_nch);
1557                 cache_setvp(ap->a_nch, *ap->a_vpp);
1558                 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1559         }
1560         return error;
1561 }
1562
1563 /*
1564  * Make a device node (typically a fifo)
1565  */
1566 static
1567 int
1568 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1569 {
1570         hammer2_inode_t *dip;
1571         hammer2_inode_t *nip;
1572         struct namecache *ncp;
1573         const uint8_t *name;
1574         size_t name_len;
1575         hammer2_tid_t inum;
1576         int error;
1577
1578         dip = VTOI(ap->a_dvp);
1579         if (dip->pmp->ronly)
1580                 return (EROFS);
1581         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1582                 return (ENOSPC);
1583
1584         ncp = ap->a_nch->ncp;
1585         name = ncp->nc_name;
1586         name_len = ncp->nc_nlen;
1587         hammer2_pfs_memory_wait(dip->pmp);
1588         hammer2_trans_init(dip->pmp, 0);
1589
1590         /*
1591          * Create the device inode and then create the directory entry.
1592          */
1593         inum = hammer2_trans_newinum(dip->pmp);
1594         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1595                                    NULL, 0, inum,
1596                                    inum, 0, 0,
1597                                    0, &error);
1598         if (error == 0) {
1599                 error = hammer2_dirent_create(dip, name, name_len,
1600                                               nip->meta.inum, nip->meta.type);
1601         }
1602         if (error) {
1603                 if (nip) {
1604                         hammer2_inode_unlink_finisher(nip, 0);
1605                         hammer2_inode_unlock(nip);
1606                         nip = NULL;
1607                 }
1608                 *ap->a_vpp = NULL;
1609         } else {
1610                 *ap->a_vpp = hammer2_igetv(nip, &error);
1611                 hammer2_inode_unlock(nip);
1612         }
1613
1614         /*
1615          * Update dip's mtime
1616          */
1617         if (error == 0) {
1618                 uint64_t mtime;
1619
1620                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1621                 hammer2_update_time(&mtime);
1622                 hammer2_inode_modify(dip);
1623                 dip->meta.mtime = mtime;
1624                 hammer2_inode_unlock(dip);
1625         }
1626
1627         hammer2_trans_done(dip->pmp);
1628
1629         if (error == 0) {
1630                 cache_setunresolved(ap->a_nch);
1631                 cache_setvp(ap->a_nch, *ap->a_vpp);
1632                 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1633         }
1634         return error;
1635 }
1636
1637 /*
1638  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1639  */
1640 static
1641 int
1642 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1643 {
1644         hammer2_inode_t *dip;
1645         hammer2_inode_t *nip;
1646         struct namecache *ncp;
1647         const uint8_t *name;
1648         size_t name_len;
1649         hammer2_tid_t inum;
1650         int error;
1651
1652         dip = VTOI(ap->a_dvp);
1653         if (dip->pmp->ronly)
1654                 return (EROFS);
1655         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1656                 return (ENOSPC);
1657
1658         ncp = ap->a_nch->ncp;
1659         name = ncp->nc_name;
1660         name_len = ncp->nc_nlen;
1661         hammer2_pfs_memory_wait(dip->pmp);
1662         hammer2_trans_init(dip->pmp, 0);
1663
1664         ap->a_vap->va_type = VLNK;      /* enforce type */
1665
1666         /*
1667          * Create the softlink as an inode and then create the directory
1668          * entry.
1669          */
1670         inum = hammer2_trans_newinum(dip->pmp);
1671
1672         nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1673                                    NULL, 0, inum,
1674                                    inum, 0, 0,
1675                                    0, &error);
1676         if (error == 0) {
1677                 error = hammer2_dirent_create(dip, name, name_len,
1678                                               nip->meta.inum, nip->meta.type);
1679         }
1680         if (error) {
1681                 if (nip) {
1682                         hammer2_inode_unlink_finisher(nip, 0);
1683                         hammer2_inode_unlock(nip);
1684                         nip = NULL;
1685                 }
1686                 *ap->a_vpp = NULL;
1687                 hammer2_trans_done(dip->pmp);
1688                 return error;
1689         }
1690         *ap->a_vpp = hammer2_igetv(nip, &error);
1691
1692         /*
1693          * Build the softlink (~like file data) and finalize the namecache.
1694          */
1695         if (error == 0) {
1696                 size_t bytes;
1697                 struct uio auio;
1698                 struct iovec aiov;
1699
1700                 bytes = strlen(ap->a_target);
1701
1702                 hammer2_inode_unlock(nip);
1703                 bzero(&auio, sizeof(auio));
1704                 bzero(&aiov, sizeof(aiov));
1705                 auio.uio_iov = &aiov;
1706                 auio.uio_segflg = UIO_SYSSPACE;
1707                 auio.uio_rw = UIO_WRITE;
1708                 auio.uio_resid = bytes;
1709                 auio.uio_iovcnt = 1;
1710                 auio.uio_td = curthread;
1711                 aiov.iov_base = ap->a_target;
1712                 aiov.iov_len = bytes;
1713                 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1714                 /* XXX handle error */
1715                 error = 0;
1716         } else {
1717                 hammer2_inode_unlock(nip);
1718         }
1719
1720         /*
1721          * Update dip's mtime
1722          */
1723         if (error == 0) {
1724                 uint64_t mtime;
1725
1726                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1727                 hammer2_update_time(&mtime);
1728                 hammer2_inode_modify(dip);
1729                 dip->meta.mtime = mtime;
1730                 hammer2_inode_unlock(dip);
1731         }
1732
1733         hammer2_trans_done(dip->pmp);
1734
1735         /*
1736          * Finalize namecache
1737          */
1738         if (error == 0) {
1739                 cache_setunresolved(ap->a_nch);
1740                 cache_setvp(ap->a_nch, *ap->a_vpp);
1741                 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1742         }
1743         return error;
1744 }
1745
1746 /*
1747  * hammer2_vop_nremove { nch, dvp, cred }
1748  */
1749 static
1750 int
1751 hammer2_vop_nremove(struct vop_nremove_args *ap)
1752 {
1753         hammer2_xop_unlink_t *xop;
1754         hammer2_inode_t *dip;
1755         hammer2_inode_t *ip;
1756         struct namecache *ncp;
1757         int error;
1758         int isopen;
1759
1760         dip = VTOI(ap->a_dvp);
1761         if (dip->pmp->ronly)
1762                 return (EROFS);
1763 #if 0
1764         /* allow removals, except user to also bulkfree */
1765         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1766                 return (ENOSPC);
1767 #endif
1768
1769         ncp = ap->a_nch->ncp;
1770
1771         hammer2_pfs_memory_wait(dip->pmp);
1772         hammer2_trans_init(dip->pmp, 0);
1773         hammer2_inode_lock(dip, 0);
1774
1775         /*
1776          * The unlink XOP unlinks the path from the directory and
1777          * locates and returns the cluster associated with the real inode.
1778          * We have to handle nlinks here on the frontend.
1779          */
1780         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1781         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1782
1783         /*
1784          * The namecache entry is locked so nobody can use this namespace.
1785          * Calculate isopen to determine if this namespace has an open vp
1786          * associated with it and resolve the vp only if it does.
1787          *
1788          * We try to avoid resolving the vnode if nobody has it open, but
1789          * note that the test is via this namespace only.
1790          */
1791         isopen = cache_isopen(ap->a_nch);
1792         xop->isdir = 0;
1793         xop->dopermanent = 0;
1794         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1795
1796         /*
1797          * Collect the real inode and adjust nlinks, destroy the real
1798          * inode if nlinks transitions to 0 and it was the real inode
1799          * (else it has already been removed).
1800          */
1801         error = hammer2_xop_collect(&xop->head, 0);
1802         error = hammer2_error_to_errno(error);
1803         hammer2_inode_unlock(dip);
1804
1805         if (error == 0) {
1806                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1807                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1808                 if (ip) {
1809                         hammer2_inode_unlink_finisher(ip, isopen);
1810                         hammer2_inode_unlock(ip);
1811                 }
1812         } else {
1813                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1814         }
1815
1816         /*
1817          * Update dip's mtime
1818          */
1819         if (error == 0) {
1820                 uint64_t mtime;
1821
1822                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1823                 hammer2_update_time(&mtime);
1824                 hammer2_inode_modify(dip);
1825                 dip->meta.mtime = mtime;
1826                 hammer2_inode_unlock(dip);
1827         }
1828
1829         hammer2_inode_run_sideq(dip->pmp, 0);
1830         hammer2_trans_done(dip->pmp);
1831         if (error == 0) {
1832                 cache_unlink(ap->a_nch);
1833                 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1834         }
1835         return (error);
1836 }
1837
1838 /*
1839  * hammer2_vop_nrmdir { nch, dvp, cred }
1840  */
1841 static
1842 int
1843 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1844 {
1845         hammer2_xop_unlink_t *xop;
1846         hammer2_inode_t *dip;
1847         hammer2_inode_t *ip;
1848         struct namecache *ncp;
1849         int isopen;
1850         int error;
1851
1852         dip = VTOI(ap->a_dvp);
1853         if (dip->pmp->ronly)
1854                 return (EROFS);
1855 #if 0
1856         /* allow removals, except user to also bulkfree */
1857         if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1858                 return (ENOSPC);
1859 #endif
1860
1861         hammer2_pfs_memory_wait(dip->pmp);
1862         hammer2_trans_init(dip->pmp, 0);
1863         hammer2_inode_lock(dip, 0);
1864
1865         xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1866
1867         ncp = ap->a_nch->ncp;
1868         hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1869         isopen = cache_isopen(ap->a_nch);
1870         xop->isdir = 1;
1871         xop->dopermanent = 0;
1872         hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1873
1874         /*
1875          * Collect the real inode and adjust nlinks, destroy the real
1876          * inode if nlinks transitions to 0 and it was the real inode
1877          * (else it has already been removed).
1878          */
1879         error = hammer2_xop_collect(&xop->head, 0);
1880         error = hammer2_error_to_errno(error);
1881         hammer2_inode_unlock(dip);
1882
1883         if (error == 0) {
1884                 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1885                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1886                 if (ip) {
1887                         hammer2_inode_unlink_finisher(ip, isopen);
1888                         hammer2_inode_unlock(ip);
1889                 }
1890         } else {
1891                 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1892         }
1893
1894         /*
1895          * Update dip's mtime
1896          */
1897         if (error == 0) {
1898                 uint64_t mtime;
1899
1900                 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1901                 hammer2_update_time(&mtime);
1902                 hammer2_inode_modify(dip);
1903                 dip->meta.mtime = mtime;
1904                 hammer2_inode_unlock(dip);
1905         }
1906
1907         hammer2_inode_run_sideq(dip->pmp, 0);
1908         hammer2_trans_done(dip->pmp);
1909         if (error == 0) {
1910                 cache_unlink(ap->a_nch);
1911                 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
1912         }
1913         return (error);
1914 }
1915
1916 /*
1917  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1918  */
1919 static
1920 int
1921 hammer2_vop_nrename(struct vop_nrename_args *ap)
1922 {
1923         struct namecache *fncp;
1924         struct namecache *tncp;
1925         hammer2_inode_t *fdip;  /* source directory */
1926         hammer2_inode_t *tdip;  /* target directory */
1927         hammer2_inode_t *ip;    /* file being renamed */
1928         hammer2_inode_t *tip;   /* replaced target during rename or NULL */
1929         const uint8_t *fname;
1930         size_t fname_len;
1931         const uint8_t *tname;
1932         size_t tname_len;
1933         int error;
1934         int update_tdip;
1935         int update_fdip;
1936         hammer2_key_t tlhc;
1937
1938         if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1939                 return(EXDEV);
1940         if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1941                 return(EXDEV);
1942
1943         fdip = VTOI(ap->a_fdvp);        /* source directory */
1944         tdip = VTOI(ap->a_tdvp);        /* target directory */
1945
1946         if (fdip->pmp->ronly)
1947                 return (EROFS);
1948         if (hammer2_vfs_enospace(fdip, 0, ap->a_cred) > 1)
1949                 return (ENOSPC);
1950
1951         fncp = ap->a_fnch->ncp;         /* entry name in source */
1952         fname = fncp->nc_name;
1953         fname_len = fncp->nc_nlen;
1954
1955         tncp = ap->a_tnch->ncp;         /* entry name in target */
1956         tname = tncp->nc_name;
1957         tname_len = tncp->nc_nlen;
1958
1959         hammer2_pfs_memory_wait(tdip->pmp);
1960         hammer2_trans_init(tdip->pmp, 0);
1961
1962         update_tdip = 0;
1963         update_fdip = 0;
1964
1965         ip = VTOI(fncp->nc_vp);
1966         hammer2_inode_ref(ip);          /* extra ref */
1967
1968         /*
1969          * Lookup the target name to determine if a directory entry
1970          * is being overwritten.  We only hold related inode locks
1971          * temporarily, the operating system is expected to protect
1972          * against rename races.
1973          */
1974         tip = tncp->nc_vp ? VTOI(tncp->nc_vp) : NULL;
1975         if (tip)
1976                 hammer2_inode_ref(tip); /* extra ref */
1977
1978         /*
1979          * Can return NULL and error == EXDEV if the common parent
1980          * crosses a directory with the xlink flag set.
1981          *
1982          * For now try to avoid deadlocks with a simple pointer address
1983          * test.  (tip) can be NULL.
1984          */
1985         error = 0;
1986         if (fdip <= tdip) {
1987                 hammer2_inode_lock(fdip, 0);
1988                 hammer2_inode_lock(tdip, 0);
1989         } else {
1990                 hammer2_inode_lock(tdip, 0);
1991                 hammer2_inode_lock(fdip, 0);
1992         }
1993         if (tip) {
1994                 if (ip <= tip) {
1995                         hammer2_inode_lock(ip, 0);
1996                         hammer2_inode_lock(tip, 0);
1997                 } else {
1998                         hammer2_inode_lock(tip, 0);
1999                         hammer2_inode_lock(ip, 0);
2000                 }
2001         } else {
2002                 hammer2_inode_lock(ip, 0);
2003         }
2004
2005 #if 0
2006         /*
2007          * Delete the target namespace.
2008          *
2009          * REMOVED - NOW FOLDED INTO XOP_NRENAME OPERATION
2010          */
2011         {
2012                 hammer2_xop_unlink_t *xop2;
2013                 hammer2_inode_t *tip;
2014                 int isopen;
2015
2016                 /*
2017                  * The unlink XOP unlinks the path from the directory and
2018                  * locates and returns the cluster associated with the real
2019                  * inode.  We have to handle nlinks here on the frontend.
2020                  */
2021                 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2022                 hammer2_xop_setname(&xop2->head, tname, tname_len);
2023                 isopen = cache_isopen(ap->a_tnch);
2024                 xop2->isdir = -1;
2025                 xop2->dopermanent = 0;
2026                 hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
2027
2028                 /*
2029                  * Collect the real inode and adjust nlinks, destroy the real
2030                  * inode if nlinks transitions to 0 and it was the real inode
2031                  * (else it has already been removed).
2032                  */
2033                 tnch_error = hammer2_xop_collect(&xop2->head, 0);
2034                 tnch_error = hammer2_error_to_errno(tnch_error);
2035                 /* hammer2_inode_unlock(tdip); */
2036
2037                 if (tnch_error == 0) {
2038                         tip = hammer2_inode_get(tdip->pmp, NULL,
2039                                                 &xop2->head.cluster, -1);
2040                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
2041                         if (tip) {
2042                                 hammer2_inode_unlink_finisher(tip, isopen);
2043                                 hammer2_inode_unlock(tip);
2044                         }
2045                 } else {
2046                         hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
2047                 }
2048                 /* hammer2_inode_lock(tdip, 0); */
2049
2050                 if (tnch_error && tnch_error != ENOENT) {
2051                         error = tnch_error;
2052                         goto done2;
2053                 }
2054                 update_tdip = 1;
2055         }
2056 #endif
2057
2058         /*
2059          * Resolve the collision space for (tdip, tname, tname_len)
2060          *
2061          * tdip must be held exclusively locked to prevent races since
2062          * multiple filenames can end up in the same collision space.
2063          */
2064         {
2065                 hammer2_xop_scanlhc_t *sxop;
2066                 hammer2_tid_t lhcbase;
2067
2068                 tlhc = hammer2_dirhash(tname, tname_len);
2069                 lhcbase = tlhc;
2070                 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2071                 sxop->lhc = tlhc;
2072                 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
2073                 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
2074                         if (tlhc != sxop->head.cluster.focus->bref.key)
2075                                 break;
2076                         ++tlhc;
2077                 }
2078                 error = hammer2_error_to_errno(error);
2079                 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
2080
2081                 if (error) {
2082                         if (error != ENOENT)
2083                                 goto done2;
2084                         ++tlhc;
2085                         error = 0;
2086                 }
2087                 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
2088                         error = ENOSPC;
2089                         goto done2;
2090                 }
2091         }
2092
2093         /*
2094          * Ready to go, issue the rename to the backend.  Note that meta-data
2095          * updates to the related inodes occur separately from the rename
2096          * operation.
2097          *
2098          * NOTE: While it is not necessary to update ip->meta.name*, doing
2099          *       so aids catastrophic recovery and debugging.
2100          */
2101         if (error == 0) {
2102                 hammer2_xop_nrename_t *xop4;
2103
2104                 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
2105                 xop4->lhc = tlhc;
2106                 xop4->ip_key = ip->meta.name_key;
2107                 hammer2_xop_setip2(&xop4->head, ip);
2108                 hammer2_xop_setip3(&xop4->head, tdip);
2109                 hammer2_xop_setname(&xop4->head, fname, fname_len);
2110                 hammer2_xop_setname2(&xop4->head, tname, tname_len);
2111                 hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
2112
2113                 error = hammer2_xop_collect(&xop4->head, 0);
2114                 error = hammer2_error_to_errno(error);
2115                 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
2116
2117                 if (error == ENOENT)
2118                         error = 0;
2119
2120                 /*
2121                  * Update inode meta-data.
2122                  *
2123                  * WARNING!  The in-memory inode (ip) structure does not
2124                  *           maintain a copy of the inode's filename buffer.
2125                  */
2126                 if (error == 0 &&
2127                     (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
2128                         hammer2_inode_modify(ip);
2129                         ip->meta.name_len = tname_len;
2130                         ip->meta.name_key = tlhc;
2131                 }
2132                 if (error == 0) {
2133                         hammer2_inode_modify(ip);
2134                         ip->meta.iparent = tdip->meta.inum;
2135                 }
2136                 update_fdip = 1;
2137                 update_tdip = 1;
2138         }
2139
2140 done2:
2141         /*
2142          * If no error, the backend has replaced the target directory entry.
2143          * We must adjust nlinks on the original replace target if it exists.
2144          */
2145         if (error == 0 && tip) {
2146                 int isopen;
2147
2148                 isopen = cache_isopen(ap->a_tnch);
2149                 hammer2_inode_unlink_finisher(tip, isopen);
2150         }
2151
2152         /*
2153          * Update directory mtimes to represent the something changed.
2154          */
2155         if (update_fdip || update_tdip) {
2156                 uint64_t mtime;
2157
2158                 hammer2_update_time(&mtime);
2159                 if (update_fdip) {
2160                         hammer2_inode_modify(fdip);
2161                         fdip->meta.mtime = mtime;
2162                 }
2163                 if (update_tdip) {
2164                         hammer2_inode_modify(tdip);
2165                         tdip->meta.mtime = mtime;
2166                 }
2167         }
2168         if (tip) {
2169                 hammer2_inode_unlock(tip);
2170                 hammer2_inode_drop(tip);
2171         }
2172         hammer2_inode_unlock(ip);
2173         hammer2_inode_unlock(tdip);
2174         hammer2_inode_unlock(fdip);
2175         hammer2_inode_drop(ip);
2176         hammer2_inode_run_sideq(fdip->pmp, 0);
2177
2178         hammer2_trans_done(tdip->pmp);
2179
2180         /*
2181          * Issue the namecache update after unlocking all the internal
2182          * hammer2 structures, otherwise we might deadlock.
2183          *
2184          * WARNING! The target namespace must be updated atomically,
2185          *          and we depend on cache_rename() to handle that for
2186          *          us.  Do not do a separate cache_unlink() because
2187          *          that leaves a small window of opportunity for other
2188          *          threads to allocate the target namespace before we
2189          *          manage to complete our rename.
2190          *
2191          * WARNING! cache_rename() (and cache_unlink()) will properly
2192          *          set VREF_FINALIZE on any attached vnode.  Do not
2193          *          call cache_setunresolved() manually before-hand as
2194          *          this will prevent the flag from being set later via
2195          *          cache_rename().  If VREF_FINALIZE is not properly set
2196          *          and the inode is no longer in the topology, related
2197          *          chains can remain dirty indefinitely.
2198          */
2199         if (error == 0 && tip) {
2200                 /*cache_unlink(ap->a_tnch); see above */
2201                 /*cache_setunresolved(ap->a_tnch); see above */
2202         }
2203         if (error == 0) {
2204                 cache_rename(ap->a_fnch, ap->a_tnch);
2205                 hammer2_knote(ap->a_fdvp, NOTE_WRITE);
2206                 hammer2_knote(ap->a_tdvp, NOTE_WRITE);
2207                 hammer2_knote(fncp->nc_vp, NOTE_RENAME);
2208         }
2209
2210         return (error);
2211 }
2212
2213 /*
2214  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2215  */
2216 static
2217 int
2218 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2219 {
2220         hammer2_inode_t *ip;
2221         int error;
2222
2223         ip = VTOI(ap->a_vp);
2224
2225         error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2226                               ap->a_fflag, ap->a_cred);
2227         return (error);
2228 }
2229
2230 static
2231 int
2232 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2233 {
2234         struct mount *mp;
2235         hammer2_pfs_t *pmp;
2236         int rc;
2237
2238         switch (ap->a_op) {
2239         case (MOUNTCTL_SET_EXPORT):
2240                 mp = ap->a_head.a_ops->head.vv_mount;
2241                 pmp = MPTOPMP(mp);
2242
2243                 if (ap->a_ctllen != sizeof(struct export_args))
2244                         rc = (EINVAL);
2245                 else
2246                         rc = vfs_export(mp, &pmp->export,
2247                                         (const struct export_args *)ap->a_ctl);
2248                 break;
2249         default:
2250                 rc = vop_stdmountctl(ap);
2251                 break;
2252         }
2253         return (rc);
2254 }
2255
2256 /*
2257  * KQFILTER
2258  */
2259 static void filt_hammer2detach(struct knote *kn);
2260 static int filt_hammer2read(struct knote *kn, long hint);
2261 static int filt_hammer2write(struct knote *kn, long hint);
2262 static int filt_hammer2vnode(struct knote *kn, long hint);
2263
2264 static struct filterops hammer2read_filtops =
2265         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2266           NULL, filt_hammer2detach, filt_hammer2read };
2267 static struct filterops hammer2write_filtops =
2268         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2269           NULL, filt_hammer2detach, filt_hammer2write };
2270 static struct filterops hammer2vnode_filtops =
2271         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2272           NULL, filt_hammer2detach, filt_hammer2vnode };
2273
2274 static
2275 int
2276 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2277 {
2278         struct vnode *vp = ap->a_vp;
2279         struct knote *kn = ap->a_kn;
2280
2281         switch (kn->kn_filter) {
2282         case EVFILT_READ:
2283                 kn->kn_fop = &hammer2read_filtops;
2284                 break;
2285         case EVFILT_WRITE:
2286                 kn->kn_fop = &hammer2write_filtops;
2287                 break;
2288         case EVFILT_VNODE:
2289                 kn->kn_fop = &hammer2vnode_filtops;
2290                 break;
2291         default:
2292                 return (EOPNOTSUPP);
2293         }
2294
2295         kn->kn_hook = (caddr_t)vp;
2296
2297         knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2298
2299         return(0);
2300 }
2301
2302 static void
2303 filt_hammer2detach(struct knote *kn)
2304 {
2305         struct vnode *vp = (void *)kn->kn_hook;
2306
2307         knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2308 }
2309
2310 static int
2311 filt_hammer2read(struct knote *kn, long hint)
2312 {
2313         struct vnode *vp = (void *)kn->kn_hook;
2314         hammer2_inode_t *ip = VTOI(vp);
2315         off_t off;
2316
2317         if (hint == NOTE_REVOKE) {
2318                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2319                 return(1);
2320         }
2321         off = ip->meta.size - kn->kn_fp->f_offset;
2322         kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2323         if (kn->kn_sfflags & NOTE_OLDAPI)
2324                 return(1);
2325         return (kn->kn_data != 0);
2326 }
2327
2328
2329 static int
2330 filt_hammer2write(struct knote *kn, long hint)
2331 {
2332         if (hint == NOTE_REVOKE)
2333                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2334         kn->kn_data = 0;
2335         return (1);
2336 }
2337
2338 static int
2339 filt_hammer2vnode(struct knote *kn, long hint)
2340 {
2341         if (kn->kn_sfflags & hint)
2342                 kn->kn_fflags |= hint;
2343         if (hint == NOTE_REVOKE) {
2344                 kn->kn_flags |= (EV_EOF | EV_NODATA);
2345                 return (1);
2346         }
2347         return (kn->kn_fflags != 0);
2348 }
2349
2350 /*
2351  * FIFO VOPS
2352  */
2353 static
2354 int
2355 hammer2_vop_markatime(struct vop_markatime_args *ap)
2356 {
2357         hammer2_inode_t *ip;
2358         struct vnode *vp;
2359
2360         vp = ap->a_vp;
2361         ip = VTOI(vp);
2362
2363         if (ip->pmp->ronly)
2364                 return (EROFS);
2365         return(0);
2366 }
2367
2368 static
2369 int
2370 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2371 {
2372         int error;
2373
2374         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2375         if (error)
2376                 error = hammer2_vop_kqfilter(ap);
2377         return(error);
2378 }
2379
2380 /*
2381  * VOPS vector
2382  */
2383 struct vop_ops hammer2_vnode_vops = {
2384         .vop_default    = vop_defaultop,
2385         .vop_fsync      = hammer2_vop_fsync,
2386         .vop_getpages   = vop_stdgetpages,
2387         .vop_putpages   = vop_stdputpages,
2388         .vop_access     = hammer2_vop_access,
2389         .vop_advlock    = hammer2_vop_advlock,
2390         .vop_close      = hammer2_vop_close,
2391         .vop_nlink      = hammer2_vop_nlink,
2392         .vop_ncreate    = hammer2_vop_ncreate,
2393         .vop_nsymlink   = hammer2_vop_nsymlink,
2394         .vop_nremove    = hammer2_vop_nremove,
2395         .vop_nrmdir     = hammer2_vop_nrmdir,
2396         .vop_nrename    = hammer2_vop_nrename,
2397         .vop_getattr    = hammer2_vop_getattr,
2398         .vop_setattr    = hammer2_vop_setattr,
2399         .vop_readdir    = hammer2_vop_readdir,
2400         .vop_readlink   = hammer2_vop_readlink,
2401         .vop_getpages   = vop_stdgetpages,
2402         .vop_putpages   = vop_stdputpages,
2403         .vop_read       = hammer2_vop_read,
2404         .vop_write      = hammer2_vop_write,
2405         .vop_open       = hammer2_vop_open,
2406         .vop_inactive   = hammer2_vop_inactive,
2407         .vop_reclaim    = hammer2_vop_reclaim,
2408         .vop_nresolve   = hammer2_vop_nresolve,
2409         .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2410         .vop_nmkdir     = hammer2_vop_nmkdir,
2411         .vop_nmknod     = hammer2_vop_nmknod,
2412         .vop_ioctl      = hammer2_vop_ioctl,
2413         .vop_mountctl   = hammer2_vop_mountctl,
2414         .vop_bmap       = hammer2_vop_bmap,
2415         .vop_strategy   = hammer2_vop_strategy,
2416         .vop_kqfilter   = hammer2_vop_kqfilter
2417 };
2418
2419 struct vop_ops hammer2_spec_vops = {
2420         .vop_default =          vop_defaultop,
2421         .vop_fsync =            hammer2_vop_fsync,
2422         .vop_read =             vop_stdnoread,
2423         .vop_write =            vop_stdnowrite,
2424         .vop_access =           hammer2_vop_access,
2425         .vop_close =            hammer2_vop_close,
2426         .vop_markatime =        hammer2_vop_markatime,
2427         .vop_getattr =          hammer2_vop_getattr,
2428         .vop_inactive =         hammer2_vop_inactive,
2429         .vop_reclaim =          hammer2_vop_reclaim,
2430         .vop_setattr =          hammer2_vop_setattr
2431 };
2432
2433 struct vop_ops hammer2_fifo_vops = {
2434         .vop_default =          fifo_vnoperate,
2435         .vop_fsync =            hammer2_vop_fsync,
2436 #if 0
2437         .vop_read =             hammer2_vop_fiforead,
2438         .vop_write =            hammer2_vop_fifowrite,
2439 #endif
2440         .vop_access =           hammer2_vop_access,
2441 #if 0
2442         .vop_close =            hammer2_vop_fifoclose,
2443 #endif
2444         .vop_markatime =        hammer2_vop_markatime,
2445         .vop_getattr =          hammer2_vop_getattr,
2446         .vop_inactive =         hammer2_vop_inactive,
2447         .vop_reclaim =          hammer2_vop_reclaim,
2448         .vop_setattr =          hammer2_vop_setattr,
2449         .vop_kqfilter =         hammer2_vop_fifokqfilter
2450 };
2451