sys/vfs/hammer2/hammer2_vnops.c

   1 /*
   2  * Copyright (c) 2011-2013 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 #include "hammer2_lz4.h"
  62
  63 #include "zlib/hammer2_zlib.h"
  64
  65 #define ZFOFFSET        (-2LL)
  66
  67 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
  68                                 int seqcount);
  69 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
  70                                 int ioflag, int seqcount);
  71 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
  72 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);
  73 static void hammer2_decompress_LZ4_callback(hammer2_io_t *dio,
  74                                 hammer2_chain_t *arg_c,
  75                                 void *arg_p, off_t arg_o);
  76 static void hammer2_decompress_ZLIB_callback(hammer2_io_t *dio,
  77                                 hammer2_chain_t *arg_c,
  78                                 void *arg_p, off_t arg_o);
  79
  80 struct objcache *cache_buffer_read;
  81 struct objcache *cache_buffer_write;
  82
  83 /*
  84  * Callback used in read path in case that a block is compressed with LZ4.
  85  */
  86 static
  87 void
  88 hammer2_decompress_LZ4_callback(hammer2_io_t *dio, hammer2_chain_t *arg_c,
  89                                 void *arg_p, off_t arg_o)
  90 {
  91         struct buf *obp;
  92         struct bio *obio = arg_p;
  93         char *bdata;
  94         int bytes = 1 << (int)(arg_o & HAMMER2_OFF_MASK_RADIX);
  95
  96         /*
  97          * If BIO_DONE is already set the device buffer was already
  98          * fully valid (B_CACHE).  If it is not set then I/O was issued
  99          * and we have to run I/O completion as the last bio.
 100          *
 101          * Nobody is waiting for our device I/O to complete, we are
 102          * responsible for bqrelse()ing it which means we also have to do
 103          * the equivalent of biowait() and clear BIO_DONE (which breadcb()
 104          * may have set).
 105          *
 106          * Any preexisting device buffer should match the requested size,
 107          * but due to bigblock recycling and other factors there is some
 108          * fragility there, so we assert that the device buffer covers
 109          * the request.
 110          */
 111         obp = obio->bio_buf;
 112
 113         if (dio->bp->b_flags & B_ERROR) {
 114                 obp->b_flags |= B_ERROR;
 115                 obp->b_error = dio->bp->b_error;
 116 #if 0
 117         } else if (obio->bio_caller_info2.index &&
 118                    obio->bio_caller_info1.uvalue32 !=
 119                     crc32(bp->b_data, bp->b_bufsize)) {
 120                 obp->b_flags |= B_ERROR;
 121                 obp->b_error = EIO;
 122 #endif
 123         } else {
 124                 char *compressed_buffer;
 125                 int *compressed_size;
 126                 int result;
 127
 128                 KKASSERT(obp->b_bufsize <= HAMMER2_PBUFSIZE);
 129                 bdata = hammer2_io_data(dio, arg_o);
 130                 compressed_size = (int *)bdata;
 131                 compressed_buffer = objcache_get(cache_buffer_read, M_INTWAIT);
 132                 KKASSERT((unsigned int)*compressed_size <= HAMMER2_PBUFSIZE);
 133                 result = LZ4_decompress_safe(&bdata[sizeof(int)],
 134                                              compressed_buffer,
 135                                              *compressed_size,
 136                                              obp->b_bufsize);
 137                 if (result < 0) {
 138                         kprintf("READ PATH: Error during decompression."
 139                                 "bio %016jx/%d log %016jx/%d\n",
 140                                 (intmax_t)dio->pbase, dio->psize,
 141                                 (intmax_t)arg_o, bytes);
 142                         /* make sure it isn't random garbage */
 143                         bzero(compressed_buffer, obp->b_bufsize);
 144                 }
 145                 KKASSERT(result <= obp->b_bufsize);
 146                 bcopy(compressed_buffer, obp->b_data, obp->b_bufsize);
 147                 if (result < obp->b_bufsize)
 148                         bzero(obp->b_data + result, obp->b_bufsize - result);
 149                 objcache_put(cache_buffer_read, compressed_buffer);
 150                 obp->b_resid = 0;
 151                 obp->b_flags |= B_AGE;
 152         }
 153         biodone(obio);
 154 }
 155
 156 /*
 157  * Callback used in read path in case that a block is compressed with ZLIB.
 158  * It is almost identical to LZ4 callback, so in theory they can be unified,
 159  * but we didn't want to make changes in bio structure for that.
 160  */
 161 static
 162 void
 163 hammer2_decompress_ZLIB_callback(hammer2_io_t *dio, hammer2_chain_t *arg_c,
 164                                  void *arg_p, off_t arg_o)
 165 {
 166         struct buf *obp;
 167         struct bio *obio = arg_p;
 168         char *bdata;
 169         int bytes = 1 << (int)(arg_o & HAMMER2_OFF_MASK_RADIX);
 170
 171         /*
 172          * If BIO_DONE is already set the device buffer was already
 173          * fully valid (B_CACHE).  If it is not set then I/O was issued
 174          * and we have to run I/O completion as the last bio.
 175          *
 176          * Nobody is waiting for our device I/O to complete, we are
 177          * responsible for bqrelse()ing it which means we also have to do
 178          * the equivalent of biowait() and clear BIO_DONE (which breadcb()
 179          * may have set).
 180          *
 181          * Any preexisting device buffer should match the requested size,
 182          * but due to bigblock recycling and other factors there is some
 183          * fragility there, so we assert that the device buffer covers
 184          * the request.
 185          */
 186         obp = obio->bio_buf;
 187
 188         if (dio->bp->b_flags & B_ERROR) {
 189                 obp->b_flags |= B_ERROR;
 190                 obp->b_error = dio->bp->b_error;
 191 #if 0
 192         } else if (obio->bio_caller_info2.index &&
 193                    obio->bio_caller_info1.uvalue32 !=
 194                     crc32(bp->b_data, bp->b_bufsize)) {
 195                 obp->b_flags |= B_ERROR;
 196                 obp->b_error = EIO;
 197 #endif
 198         } else {
 199                 char *compressed_buffer;
 200                 z_stream strm_decompress;
 201                 int result;
 202                 int ret;
 203
 204                 KKASSERT(obp->b_bufsize <= HAMMER2_PBUFSIZE);
 205                 strm_decompress.avail_in = 0;
 206                 strm_decompress.next_in = Z_NULL;
 207
 208                 ret = inflateInit(&strm_decompress);
 209
 210                 if (ret != Z_OK)
 211                         kprintf("HAMMER2 ZLIB: Fatal error in inflateInit.\n");
 212
 213                 bdata = hammer2_io_data(dio, arg_o);
 214                 compressed_buffer = objcache_get(cache_buffer_read, M_INTWAIT);
 215                 strm_decompress.next_in = bdata;
 216
 217                 /* XXX supply proper size, subset of device bp */
 218                 strm_decompress.avail_in = bytes;
 219                 strm_decompress.next_out = compressed_buffer;
 220                 strm_decompress.avail_out = obp->b_bufsize;
 221
 222                 ret = inflate(&strm_decompress, Z_FINISH);
 223                 if (ret != Z_STREAM_END) {
 224                         kprintf("HAMMER2 ZLIB: Fatar error during decompression.\n");
 225                         bzero(compressed_buffer, obp->b_bufsize);
 226                 }
 227                 bcopy(compressed_buffer, obp->b_data, obp->b_bufsize);
 228                 result = obp->b_bufsize - strm_decompress.avail_out;
 229                 if (result < obp->b_bufsize)
 230                         bzero(obp->b_data + result, strm_decompress.avail_out);
 231                 objcache_put(cache_buffer_read, compressed_buffer);
 232                 obp->b_resid = 0;
 233                 obp->b_flags |= B_AGE;
 234                 ret = inflateEnd(&strm_decompress);
 235         }
 236         biodone(obio);
 237 }
 238
 239 static __inline
 240 void
 241 hammer2_knote(struct vnode *vp, int flags)
 242 {
 243         if (flags)
 244                 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
 245 }
 246
 247 /*
 248  * Last reference to a vnode is going away but it is still cached.
 249  */
 250 static
 251 int
 252 hammer2_vop_inactive(struct vop_inactive_args *ap)
 253 {
 254         hammer2_inode_t *ip;
 255         hammer2_chain_t *parent;
 256         struct vnode *vp;
 257
 258         vp = ap->a_vp;
 259         ip = VTOI(vp);
 260
 261         /*
 262          * Degenerate case
 263          */
 264         if (ip == NULL) {
 265                 vrecycle(vp);
 266                 return (0);
 267         }
 268
 269         /*
 270          * Detect updates to the embedded data which may be synchronized by
 271          * the strategy code.  Simply mark the inode modified so it gets
 272          * picked up by our normal flush.
 273          */
 274         parent = hammer2_inode_lock_ex(ip);
 275         KKASSERT(parent);
 276
 277         /*
 278          * Check for deleted inodes and recycle immediately.
 279          */
 280         if (parent->flags & HAMMER2_CHAIN_DELETED) {
 281                 hammer2_inode_unlock_ex(ip, parent);
 282                 vrecycle(vp);
 283         } else {
 284                 hammer2_inode_unlock_ex(ip, parent);
 285         }
 286         return (0);
 287 }
 288
 289 /*
 290  * Reclaim a vnode so that it can be reused; after the inode is
 291  * disassociated, the filesystem must manage it alone.
 292  */
 293 static
 294 int
 295 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
 296 {
 297         hammer2_chain_t *chain;
 298         hammer2_inode_t *ip;
 299         hammer2_trans_t trans;
 300         struct vnode *vp;
 301
 302         vp = ap->a_vp;
 303         ip = VTOI(vp);
 304         if (ip == NULL)
 305                 return(0);
 306
 307         /*
 308          * Set update_hi so we can detect and propagate the DESTROYED
 309          * bit in the flush code.
 310          *
 311          * ip->chain might be stale, correct it before checking as older
 312          * versions of the chain are likely marked deleted even if the
 313          * file hasn't been.  XXX ip->chain should never be stale on
 314          * reclaim.
 315          */
 316         chain = hammer2_inode_lock_ex(ip);
 317 #if 0
 318         if (chain->next_parent)
 319                 kprintf("RECLAIM DUPLINKED IP: %p ip->ch=%p ch=%p np=%p\n",
 320                         ip, ip->chain, chain, chain->next_parent);
 321 #endif
 322
 323         /*
 324          * The final close of a deleted file or directory marks it for
 325          * destruction.  The DESTROYED flag allows the flusher to shortcut
 326          * any modified blocks still unflushed (that is, just ignore them).
 327          *
 328          * HAMMER2 usually does not try to optimize the freemap by returning
 329          * deleted blocks to it as it does not usually know how many snapshots
 330          * might be referencing portions of the file/dir.  XXX TODO.
 331          *
 332          * XXX TODO - However, any modified file as-of when a snapshot is made
 333          *            cannot use this optimization as some of the modifications
 334          *            may wind up being part of the snapshot.
 335          */
 336         vp->v_data = NULL;
 337         ip->vp = NULL;
 338         if (chain->flags & HAMMER2_CHAIN_DELETED) {
 339                 atomic_set_int(&chain->flags, HAMMER2_CHAIN_DESTROYED);
 340                 hammer2_trans_init(&trans, ip->pmp, NULL,
 341                                    HAMMER2_TRANS_BUFCACHE);
 342                 hammer2_chain_setsubmod(&trans, chain);
 343                 spin_lock(&chain->core->cst.spin);
 344                 if (chain->core->update_hi < trans.sync_tid)
 345                         chain->core->update_hi = trans.sync_tid; /* needed? */
 346                 spin_unlock(&chain->core->cst.spin);
 347                 hammer2_trans_done(&trans);
 348         }
 349
 350         /*
 351          * NOTE! We do not attempt to flush chains here, flushing is
 352          *       really fragile and could also deadlock.
 353          */
 354         vclrisdirty(vp);
 355         hammer2_inode_unlock_ex(ip, chain);             /* unlock */
 356         hammer2_inode_drop(ip);                         /* vp ref */
 357         /* chain no longer referenced */
 358         /* chain = NULL; not needed */
 359
 360         /*
 361          * XXX handle background sync when ip dirty, kernel will no longer
 362          * notify us regarding this inode because there is no longer a
 363          * vnode attached to it.
 364          */
 365
 366         return (0);
 367 }
 368
 369 static
 370 int
 371 hammer2_vop_fsync(struct vop_fsync_args *ap)
 372 {
 373         hammer2_inode_t *ip;
 374         hammer2_trans_t trans;
 375         hammer2_chain_t *chain;
 376         struct vnode *vp;
 377
 378         vp = ap->a_vp;
 379         ip = VTOI(vp);
 380
 381 #if 0
 382         /* XXX can't do this yet */
 383         hammer2_trans_init(&trans, ip->pmp, NULL, HAMMER2_TRANS_ISFLUSH);
 384         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 385         hammer2_trans_clear_invfsync(&trans);
 386 #endif
 387         hammer2_trans_init(&trans, ip->pmp, NULL, 0);
 388         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 389
 390         /*
 391          * Calling chain_flush here creates a lot of duplicative
 392          * COW operations due to non-optimal vnode ordering.
 393          *
 394          * Only do it for an actual fsync() syscall.  The other forms
 395          * which call this function will eventually call chain_flush
 396          * on the volume root as a catch-all, which is far more optimal.
 397          */
 398         chain = hammer2_inode_lock_ex(ip);
 399         atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
 400         vclrisdirty(vp);
 401         if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MTIME))
 402                 hammer2_inode_fsync(&trans, ip, &chain);
 403
 404 #if 0
 405         /*
 406          * XXX creates discontinuity w/modify_tid
 407          */
 408         if (ap->a_flags & VOP_FSYNC_SYSCALL) {
 409                 hammer2_chain_flush(&trans, &chain);
 410         }
 411 #endif
 412         hammer2_inode_unlock_ex(ip, chain);
 413         hammer2_trans_done(&trans);
 414
 415         return (0);
 416 }
 417
 418 static
 419 int
 420 hammer2_vop_access(struct vop_access_args *ap)
 421 {
 422         hammer2_inode_t *ip = VTOI(ap->a_vp);
 423         hammer2_inode_data_t *ipdata;
 424         hammer2_chain_t *chain;
 425         uid_t uid;
 426         gid_t gid;
 427         int error;
 428
 429         chain = hammer2_inode_lock_sh(ip);
 430         ipdata = &chain->data->ipdata;
 431         uid = hammer2_to_unix_xid(&ipdata->uid);
 432         gid = hammer2_to_unix_xid(&ipdata->gid);
 433         error = vop_helper_access(ap, uid, gid, ipdata->mode, ipdata->uflags);
 434         hammer2_inode_unlock_sh(ip, chain);
 435
 436         return (error);
 437 }
 438
 439 static
 440 int
 441 hammer2_vop_getattr(struct vop_getattr_args *ap)
 442 {
 443         hammer2_inode_data_t *ipdata;
 444         hammer2_chain_t *chain;
 445         hammer2_pfsmount_t *pmp;
 446         hammer2_inode_t *ip;
 447         struct vnode *vp;
 448         struct vattr *vap;
 449
 450         vp = ap->a_vp;
 451         vap = ap->a_vap;
 452
 453         ip = VTOI(vp);
 454         pmp = ip->pmp;
 455
 456         chain = hammer2_inode_lock_sh(ip);
 457         ipdata = &chain->data->ipdata;
 458
 459         vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
 460         vap->va_fileid = ipdata->inum;
 461         vap->va_mode = ipdata->mode;
 462         vap->va_nlink = ipdata->nlinks;
 463         vap->va_uid = hammer2_to_unix_xid(&ipdata->uid);
 464         vap->va_gid = hammer2_to_unix_xid(&ipdata->gid);
 465         vap->va_rmajor = 0;
 466         vap->va_rminor = 0;
 467         vap->va_size = ip->size;        /* protected by shared lock */
 468         vap->va_blocksize = HAMMER2_PBUFSIZE;
 469         vap->va_flags = ipdata->uflags;
 470         hammer2_time_to_timespec(ipdata->ctime, &vap->va_ctime);
 471         hammer2_time_to_timespec(ipdata->mtime, &vap->va_mtime);
 472         hammer2_time_to_timespec(ipdata->mtime, &vap->va_atime);
 473         vap->va_gen = 1;
 474         vap->va_bytes = vap->va_size;   /* XXX */
 475         vap->va_type = hammer2_get_vtype(chain);
 476         vap->va_filerev = 0;
 477         vap->va_uid_uuid = ipdata->uid;
 478         vap->va_gid_uuid = ipdata->gid;
 479         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 480                           VA_FSID_UUID_VALID;
 481
 482         hammer2_inode_unlock_sh(ip, chain);
 483
 484         return (0);
 485 }
 486
 487 static
 488 int
 489 hammer2_vop_setattr(struct vop_setattr_args *ap)
 490 {
 491         hammer2_inode_data_t *ipdata;
 492         hammer2_inode_t *ip;
 493         hammer2_chain_t *chain;
 494         hammer2_trans_t trans;
 495         struct vnode *vp;
 496         struct vattr *vap;
 497         int error;
 498         int kflags = 0;
 499         int domtime = 0;
 500         uint64_t ctime;
 501
 502         vp = ap->a_vp;
 503         vap = ap->a_vap;
 504         hammer2_update_time(&ctime);
 505
 506         ip = VTOI(vp);
 507
 508         if (ip->pmp->ronly)
 509                 return(EROFS);
 510
 511         hammer2_chain_memory_wait(ip->pmp);
 512         hammer2_trans_init(&trans, ip->pmp, NULL, 0);
 513         chain = hammer2_inode_lock_ex(ip);
 514         ipdata = &chain->data->ipdata;
 515         error = 0;
 516
 517         if (vap->va_flags != VNOVAL) {
 518                 u_int32_t flags;
 519
 520                 flags = ipdata->uflags;
 521                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
 522                                          hammer2_to_unix_xid(&ipdata->uid),
 523                                          ap->a_cred);
 524                 if (error == 0) {
 525                         if (ipdata->uflags != flags) {
 526                                 ipdata = hammer2_chain_modify_ip(&trans, ip,
 527                                                                  &chain, 0);
 528                                 ipdata->uflags = flags;
 529                                 ipdata->ctime = ctime;
 530                                 kflags |= NOTE_ATTRIB;
 531                         }
 532                         if (ipdata->uflags & (IMMUTABLE | APPEND)) {
 533                                 error = 0;
 534                                 goto done;
 535                         }
 536                 }
 537                 goto done;
 538         }
 539         if (ipdata->uflags & (IMMUTABLE | APPEND)) {
 540                 error = EPERM;
 541                 goto done;
 542         }
 543         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
 544                 mode_t cur_mode = ipdata->mode;
 545                 uid_t cur_uid = hammer2_to_unix_xid(&ipdata->uid);
 546                 gid_t cur_gid = hammer2_to_unix_xid(&ipdata->gid);
 547                 uuid_t uuid_uid;
 548                 uuid_t uuid_gid;
 549
 550                 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
 551                                          ap->a_cred,
 552                                          &cur_uid, &cur_gid, &cur_mode);
 553                 if (error == 0) {
 554                         hammer2_guid_to_uuid(&uuid_uid, cur_uid);
 555                         hammer2_guid_to_uuid(&uuid_gid, cur_gid);
 556                         if (bcmp(&uuid_uid, &ipdata->uid, sizeof(uuid_uid)) ||
 557                             bcmp(&uuid_gid, &ipdata->gid, sizeof(uuid_gid)) ||
 558                             ipdata->mode != cur_mode
 559                         ) {
 560                                 ipdata = hammer2_chain_modify_ip(&trans, ip,
 561                                                                  &chain, 0);
 562                                 ipdata->uid = uuid_uid;
 563                                 ipdata->gid = uuid_gid;
 564                                 ipdata->mode = cur_mode;
 565                                 ipdata->ctime = ctime;
 566                         }
 567                         kflags |= NOTE_ATTRIB;
 568                 }
 569         }
 570
 571         /*
 572          * Resize the file
 573          */
 574         if (vap->va_size != VNOVAL && ip->size != vap->va_size) {
 575                 switch(vp->v_type) {
 576                 case VREG:
 577                         if (vap->va_size == ip->size)
 578                                 break;
 579                         hammer2_inode_unlock_ex(ip, chain);
 580                         if (vap->va_size < ip->size) {
 581                                 hammer2_truncate_file(ip, vap->va_size);
 582                         } else {
 583                                 hammer2_extend_file(ip, vap->va_size);
 584                         }
 585                         chain = hammer2_inode_lock_ex(ip);
 586                         ipdata = &chain->data->ipdata; /* RELOAD */
 587                         domtime = 1;
 588                         break;
 589                 default:
 590                         error = EINVAL;
 591                         goto done;
 592                 }
 593         }
 594 #if 0
 595         /* atime not supported */
 596         if (vap->va_atime.tv_sec != VNOVAL) {
 597                 ipdata = hammer2_chain_modify_ip(&trans, ip, &chain, 0);
 598                 ipdata->atime = hammer2_timespec_to_time(&vap->va_atime);
 599                 kflags |= NOTE_ATTRIB;
 600         }
 601 #endif
 602         if (vap->va_mtime.tv_sec != VNOVAL) {
 603                 ipdata = hammer2_chain_modify_ip(&trans, ip, &chain, 0);
 604                 ipdata->mtime = hammer2_timespec_to_time(&vap->va_mtime);
 605                 kflags |= NOTE_ATTRIB;
 606                 domtime = 0;
 607         }
 608         if (vap->va_mode != (mode_t)VNOVAL) {
 609                 mode_t cur_mode = ipdata->mode;
 610                 uid_t cur_uid = hammer2_to_unix_xid(&ipdata->uid);
 611                 gid_t cur_gid = hammer2_to_unix_xid(&ipdata->gid);
 612
 613                 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
 614                                          cur_uid, cur_gid, &cur_mode);
 615                 if (error == 0 && ipdata->mode != cur_mode) {
 616                         ipdata = hammer2_chain_modify_ip(&trans, ip, &chain, 0);
 617                         ipdata->mode = cur_mode;
 618                         ipdata->ctime = ctime;
 619                         kflags |= NOTE_ATTRIB;
 620                 }
 621         }
 622
 623         /*
 624          * If a truncation occurred we must call inode_fsync() now in order
 625          * to trim the related data chains, otherwise a later expansion can
 626          * cause havoc.
 627          */
 628         hammer2_inode_fsync(&trans, ip, &chain);
 629
 630         /*
 631          * Cleanup.  If domtime is set an additional inode modification
 632          * must be flagged.  All other modifications will have already
 633          * set INODE_MODIFIED and called vsetisdirty().
 634          */
 635 done:
 636         if (domtime) {
 637                 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED |
 638                                            HAMMER2_INODE_MTIME);
 639                 vsetisdirty(ip->vp);
 640         }
 641         hammer2_inode_unlock_ex(ip, chain);
 642         hammer2_trans_done(&trans);
 643         hammer2_knote(ip->vp, kflags);
 644
 645         return (error);
 646 }
 647
 648 static
 649 int
 650 hammer2_vop_readdir(struct vop_readdir_args *ap)
 651 {
 652         hammer2_inode_data_t *ipdata;
 653         hammer2_inode_t *ip;
 654         hammer2_inode_t *xip;
 655         hammer2_chain_t *parent;
 656         hammer2_chain_t *chain;
 657         hammer2_chain_t *xchain;
 658         hammer2_tid_t inum;
 659         hammer2_key_t key_next;
 660         hammer2_key_t lkey;
 661         struct uio *uio;
 662         off_t *cookies;
 663         off_t saveoff;
 664         int cookie_index;
 665         int cache_index = -1;
 666         int ncookies;
 667         int error;
 668         int dtype;
 669         int r;
 670
 671         ip = VTOI(ap->a_vp);
 672         uio = ap->a_uio;
 673         saveoff = uio->uio_offset;
 674
 675         /*
 676          * Setup cookies directory entry cookies if requested
 677          */
 678         if (ap->a_ncookies) {
 679                 ncookies = uio->uio_resid / 16 + 1;
 680                 if (ncookies > 1024)
 681                         ncookies = 1024;
 682                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 683         } else {
 684                 ncookies = -1;
 685                 cookies = NULL;
 686         }
 687         cookie_index = 0;
 688
 689         parent = hammer2_inode_lock_sh(ip);
 690         ipdata = &parent->data->ipdata;
 691
 692         /*
 693          * Handle artificial entries.  To ensure that only positive 64 bit
 694          * quantities are returned to userland we always strip off bit 63.
 695          * The hash code is designed such that codes 0x0000-0x7FFF are not
 696          * used, allowing us to use these codes for articial entries.
 697          *
 698          * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
 699          * allow '..' to cross the mount point into (e.g.) the super-root.
 700          */
 701         error = 0;
 702         chain = (void *)(intptr_t)-1;   /* non-NULL for early goto done case */
 703
 704         if (saveoff == 0) {
 705                 inum = ipdata->inum & HAMMER2_DIRHASH_USERMSK;
 706                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
 707                 if (r)
 708                         goto done;
 709                 if (cookies)
 710                         cookies[cookie_index] = saveoff;
 711                 ++saveoff;
 712                 ++cookie_index;
 713                 if (cookie_index == ncookies)
 714                         goto done;
 715         }
 716
 717         if (saveoff == 1) {
 718                 /*
 719                  * Be careful with lockorder when accessing ".."
 720                  *
 721                  * (ip is the current dir. xip is the parent dir).
 722                  */
 723                 inum = ipdata->inum & HAMMER2_DIRHASH_USERMSK;
 724                 while (ip->pip != NULL && ip != ip->pmp->iroot) {
 725                         xip = ip->pip;
 726                         hammer2_inode_ref(xip);
 727                         hammer2_inode_unlock_sh(ip, parent);
 728                         xchain = hammer2_inode_lock_sh(xip);
 729                         parent = hammer2_inode_lock_sh(ip);
 730                         hammer2_inode_drop(xip);
 731                         if (xip == ip->pip) {
 732                                 inum = xchain->data->ipdata.inum &
 733                                        HAMMER2_DIRHASH_USERMSK;
 734                                 hammer2_inode_unlock_sh(xip, xchain);
 735                                 break;
 736                         }
 737                         hammer2_inode_unlock_sh(xip, xchain);
 738                 }
 739                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
 740                 if (r)
 741                         goto done;
 742                 if (cookies)
 743                         cookies[cookie_index] = saveoff;
 744                 ++saveoff;
 745                 ++cookie_index;
 746                 if (cookie_index == ncookies)
 747                         goto done;
 748         }
 749
 750         lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
 751
 752         /*
 753          * parent is the inode chain, already locked for us.  Don't
 754          * double lock shared locks as this will screw up upgrades.
 755          */
 756         if (error) {
 757                 goto done;
 758         }
 759         chain = hammer2_chain_lookup(&parent, &key_next, lkey, lkey,
 760                                      &cache_index, HAMMER2_LOOKUP_SHARED);
 761         if (chain == NULL) {
 762                 chain = hammer2_chain_lookup(&parent, &key_next,
 763                                              lkey, (hammer2_key_t)-1,
 764                                              &cache_index,
 765                                              HAMMER2_LOOKUP_SHARED);
 766         }
 767         while (chain) {
 768                 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
 769                         dtype = hammer2_get_dtype(chain);
 770                         saveoff = chain->bref.key & HAMMER2_DIRHASH_USERMSK;
 771                         r = vop_write_dirent(&error, uio,
 772                                              chain->data->ipdata.inum &
 773                                               HAMMER2_DIRHASH_USERMSK,
 774                                              dtype,
 775                                              chain->data->ipdata.name_len,
 776                                              chain->data->ipdata.filename);
 777                         if (r)
 778                                 break;
 779                         if (cookies)
 780                                 cookies[cookie_index] = saveoff;
 781                         ++cookie_index;
 782                 } else {
 783                         /* XXX chain error */
 784                         kprintf("bad chain type readdir %d\n",
 785                                 chain->bref.type);
 786                 }
 787
 788                 /*
 789                  * Keys may not be returned in order so once we have a
 790                  * placemarker (chain) the scan must allow the full range
 791                  * or some entries will be missed.
 792                  */
 793                 chain = hammer2_chain_next(&parent, chain, &key_next,
 794                                            key_next, (hammer2_key_t)-1,
 795                                            &cache_index, HAMMER2_LOOKUP_SHARED);
 796                 if (chain) {
 797                         saveoff = (chain->bref.key &
 798                                    HAMMER2_DIRHASH_USERMSK) + 1;
 799                 } else {
 800                         saveoff = (hammer2_key_t)-1;
 801                 }
 802                 if (cookie_index == ncookies)
 803                         break;
 804         }
 805         if (chain)
 806                 hammer2_chain_unlock(chain);
 807 done:
 808         hammer2_inode_unlock_sh(ip, parent);
 809         if (ap->a_eofflag)
 810                 *ap->a_eofflag = (chain == NULL);
 811         uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
 812         if (error && cookie_index == 0) {
 813                 if (cookies) {
 814                         kfree(cookies, M_TEMP);
 815                         *ap->a_ncookies = 0;
 816                         *ap->a_cookies = NULL;
 817                 }
 818         } else {
 819                 if (cookies) {
 820                         *ap->a_ncookies = cookie_index;
 821                         *ap->a_cookies = cookies;
 822                 }
 823         }
 824         return (error);
 825 }
 826
 827 /*
 828  * hammer2_vop_readlink { vp, uio, cred }
 829  */
 830 static
 831 int
 832 hammer2_vop_readlink(struct vop_readlink_args *ap)
 833 {
 834         struct vnode *vp;
 835         hammer2_inode_t *ip;
 836         int error;
 837
 838         vp = ap->a_vp;
 839         if (vp->v_type != VLNK)
 840                 return (EINVAL);
 841         ip = VTOI(vp);
 842
 843         error = hammer2_read_file(ip, ap->a_uio, 0);
 844         return (error);
 845 }
 846
 847 static
 848 int
 849 hammer2_vop_read(struct vop_read_args *ap)
 850 {
 851         struct vnode *vp;
 852         hammer2_inode_t *ip;
 853         struct uio *uio;
 854         int error;
 855         int seqcount;
 856         int bigread;
 857
 858         /*
 859          * Read operations supported on this vnode?
 860          */
 861         vp = ap->a_vp;
 862         if (vp->v_type != VREG)
 863                 return (EINVAL);
 864
 865         /*
 866          * Misc
 867          */
 868         ip = VTOI(vp);
 869         uio = ap->a_uio;
 870         error = 0;
 871
 872         seqcount = ap->a_ioflag >> 16;
 873         bigread = (uio->uio_resid > 100 * 1024 * 1024);
 874
 875         error = hammer2_read_file(ip, uio, seqcount);
 876         return (error);
 877 }
 878
 879 static
 880 int
 881 hammer2_vop_write(struct vop_write_args *ap)
 882 {
 883         hammer2_inode_t *ip;
 884         hammer2_trans_t trans;
 885         thread_t td;
 886         struct vnode *vp;
 887         struct uio *uio;
 888         int error;
 889         int seqcount;
 890         int bigwrite;
 891
 892         /*
 893          * Read operations supported on this vnode?
 894          */
 895         vp = ap->a_vp;
 896         if (vp->v_type != VREG)
 897                 return (EINVAL);
 898
 899         /*
 900          * Misc
 901          */
 902         ip = VTOI(vp);
 903         uio = ap->a_uio;
 904         error = 0;
 905         if (ip->pmp->ronly)
 906                 return (EROFS);
 907
 908         seqcount = ap->a_ioflag >> 16;
 909         bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
 910
 911         /*
 912          * Check resource limit
 913          */
 914         if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
 915             uio->uio_offset + uio->uio_resid >
 916              td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 917                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 918                 return (EFBIG);
 919         }
 920
 921         bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
 922
 923         /*
 924          * The transaction interlocks against flushes initiations
 925          * (note: but will run concurrently with the actual flush).
 926          */
 927         hammer2_trans_init(&trans, ip->pmp, NULL, 0);
 928         error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
 929         hammer2_trans_done(&trans);
 930
 931         return (error);
 932 }
 933
 934 /*
 935  * Perform read operations on a file or symlink given an UNLOCKED
 936  * inode and uio.
 937  *
 938  * The passed ip is not locked.
 939  */
 940 static
 941 int
 942 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
 943 {
 944         hammer2_off_t size;
 945         struct buf *bp;
 946         int error;
 947
 948         error = 0;
 949
 950         /*
 951          * UIO read loop.
 952          */
 953         ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE);
 954         size = ip->size;
 955         ccms_thread_unlock(&ip->topo_cst);
 956
 957         while (uio->uio_resid > 0 && uio->uio_offset < size) {
 958                 hammer2_key_t lbase;
 959                 hammer2_key_t leof;
 960                 int lblksize;
 961                 int loff;
 962                 int n;
 963
 964                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 965                                                 &lbase, &leof);
 966
 967                 error = cluster_read(ip->vp, leof, lbase, lblksize,
 968                                      uio->uio_resid, seqcount * BKVASIZE,
 969                                      &bp);
 970
 971                 if (error)
 972                         break;
 973                 loff = (int)(uio->uio_offset - lbase);
 974                 n = lblksize - loff;
 975                 if (n > uio->uio_resid)
 976                         n = uio->uio_resid;
 977                 if (n > size - uio->uio_offset)
 978                         n = (int)(size - uio->uio_offset);
 979                 bp->b_flags |= B_AGE;
 980                 uiomove((char *)bp->b_data + loff, n, uio);
 981                 bqrelse(bp);
 982         }
 983         return (error);
 984 }
 985
 986 /*
 987  * Write to the file represented by the inode via the logical buffer cache.
 988  * The inode may represent a regular file or a symlink.
 989  *
 990  * The inode must not be locked.
 991  */
 992 static
 993 int
 994 hammer2_write_file(hammer2_inode_t *ip,
 995                    struct uio *uio, int ioflag, int seqcount)
 996 {
 997         hammer2_key_t old_eof;
 998         hammer2_key_t new_eof;
 999         struct buf *bp;
1000         int kflags;
1001         int error;
1002         int modified;
1003
1004         /*
1005          * Setup if append
1006          */
1007         ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE);
1008         if (ioflag & IO_APPEND)
1009                 uio->uio_offset = ip->size;
1010         old_eof = ip->size;
1011         ccms_thread_unlock(&ip->topo_cst);
1012
1013         /*
1014          * Extend the file if necessary.  If the write fails at some point
1015          * we will truncate it back down to cover as much as we were able
1016          * to write.
1017          *
1018          * Doing this now makes it easier to calculate buffer sizes in
1019          * the loop.
1020          */
1021         kflags = 0;
1022         error = 0;
1023         modified = 0;
1024
1025         if (uio->uio_offset + uio->uio_resid > old_eof) {
1026                 new_eof = uio->uio_offset + uio->uio_resid;
1027                 modified = 1;
1028                 hammer2_extend_file(ip, new_eof);
1029                 kflags |= NOTE_EXTEND;
1030         } else {
1031                 new_eof = old_eof;
1032         }
1033
1034         /*
1035          * UIO write loop
1036          */
1037         while (uio->uio_resid > 0) {
1038                 hammer2_key_t lbase;
1039                 int trivial;
1040                 int endofblk;
1041                 int lblksize;
1042                 int loff;
1043                 int n;
1044
1045                 /*
1046                  * Don't allow the buffer build to blow out the buffer
1047                  * cache.
1048                  */
1049                 if ((ioflag & IO_RECURSE) == 0)
1050                         bwillwrite(HAMMER2_PBUFSIZE);
1051
1052                 /*
1053                  * This nominally tells us how much we can cluster and
1054                  * what the logical buffer size needs to be.  Currently
1055                  * we don't try to cluster the write and just handle one
1056                  * block at a time.
1057                  */
1058                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
1059                                                 &lbase, NULL);
1060                 loff = (int)(uio->uio_offset - lbase);
1061
1062                 KKASSERT(lblksize <= 65536);
1063
1064                 /*
1065                  * Calculate bytes to copy this transfer and whether the
1066                  * copy completely covers the buffer or not.
1067                  */
1068                 trivial = 0;
1069                 n = lblksize - loff;
1070                 if (n > uio->uio_resid) {
1071                         n = uio->uio_resid;
1072                         if (loff == lbase && uio->uio_offset + n == new_eof)
1073                                 trivial = 1;
1074                         endofblk = 0;
1075                 } else {
1076                         if (loff == 0)
1077                                 trivial = 1;
1078                         endofblk = 1;
1079                 }
1080
1081                 /*
1082                  * Get the buffer
1083                  */
1084                 if (uio->uio_segflg == UIO_NOCOPY) {
1085                         /*
1086                          * Issuing a write with the same data backing the
1087                          * buffer.  Instantiate the buffer to collect the
1088                          * backing vm pages, then read-in any missing bits.
1089                          *
1090                          * This case is used by vop_stdputpages().
1091                          */
1092                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1093                         if ((bp->b_flags & B_CACHE) == 0) {
1094                                 bqrelse(bp);
1095                                 error = bread(ip->vp, lbase, lblksize, &bp);
1096                         }
1097                 } else if (trivial) {
1098                         /*
1099                          * Even though we are entirely overwriting the buffer
1100                          * we may still have to zero it out to avoid a
1101                          * mmap/write visibility issue.
1102                          */
1103                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1104                         if ((bp->b_flags & B_CACHE) == 0)
1105                                 vfs_bio_clrbuf(bp);
1106                 } else {
1107                         /*
1108                          * Partial overwrite, read in any missing bits then
1109                          * replace the portion being written.
1110                          *
1111                          * (The strategy code will detect zero-fill physical
1112                          * blocks for this case).
1113                          */
1114                         error = bread(ip->vp, lbase, lblksize, &bp);
1115                         if (error == 0)
1116                                 bheavy(bp);
1117                 }
1118
1119                 if (error) {
1120                         brelse(bp);
1121                         break;
1122                 }
1123
1124                 /*
1125                  * Ok, copy the data in
1126                  */
1127                 error = uiomove(bp->b_data + loff, n, uio);
1128                 kflags |= NOTE_WRITE;
1129                 modified = 1;
1130                 if (error) {
1131                         brelse(bp);
1132                         break;
1133                 }
1134
1135                 /*
1136                  * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1137                  *          with IO_SYNC or IO_ASYNC set.  These writes
1138                  *          must be handled as the pageout daemon expects.
1139                  */
1140                 if (ioflag & IO_SYNC) {
1141                         bwrite(bp);
1142                 } else if ((ioflag & IO_DIRECT) && endofblk) {
1143                         bawrite(bp);
1144                 } else if (ioflag & IO_ASYNC) {
1145                         bawrite(bp);
1146                 } else {
1147                         bdwrite(bp);
1148                 }
1149         }
1150
1151         /*
1152          * Cleanup.  If we extended the file EOF but failed to write through
1153          * the entire write is a failure and we have to back-up.
1154          */
1155         if (error && new_eof != old_eof) {
1156                 hammer2_truncate_file(ip, old_eof);
1157         } else if (modified) {
1158                 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE);
1159                 hammer2_update_time(&ip->mtime);
1160                 atomic_set_int(&ip->flags, HAMMER2_INODE_MTIME);
1161                 ccms_thread_unlock(&ip->topo_cst);
1162         }
1163         atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1164         hammer2_knote(ip->vp, kflags);
1165         vsetisdirty(ip->vp);
1166
1167         return error;
1168 }
1169
1170 /*
1171  * Truncate the size of a file.  The inode must not be locked.
1172  *
1173  * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED
1174  */
1175 static
1176 void
1177 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1178 {
1179         hammer2_key_t lbase;
1180         int nblksize;
1181
1182         if (ip->vp) {
1183                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1184                 nvtruncbuf(ip->vp, nsize,
1185                            nblksize, (int)nsize & (nblksize - 1),
1186                            0);
1187         }
1188         ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE);
1189         ip->size = nsize;
1190         atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1191         ccms_thread_unlock(&ip->topo_cst);
1192 }
1193
1194 /*
1195  * Extend the size of a file.  The inode must not be locked.
1196  *
1197  * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED
1198  */
1199 static
1200 void
1201 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1202 {
1203         hammer2_key_t lbase;
1204         hammer2_key_t osize;
1205         int oblksize;
1206         int nblksize;
1207
1208         ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE);
1209         osize = ip->size;
1210         ip->size = nsize;
1211         ccms_thread_unlock(&ip->topo_cst);
1212
1213         if (ip->vp) {
1214                 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1215                 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1216                 nvextendbuf(ip->vp,
1217                             osize, nsize,
1218                             oblksize, nblksize,
1219                             -1, -1, 0);
1220         }
1221         atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1222 }
1223
1224 static
1225 int
1226 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1227 {
1228         hammer2_inode_t *ip;
1229         hammer2_inode_t *dip;
1230         hammer2_chain_t *parent;
1231         hammer2_chain_t *chain;
1232         hammer2_chain_t *ochain;
1233         hammer2_trans_t trans;
1234         hammer2_key_t key_next;
1235         hammer2_key_t lhc;
1236         struct namecache *ncp;
1237         const uint8_t *name;
1238         size_t name_len;
1239         int error = 0;
1240         int cache_index = -1;
1241         struct vnode *vp;
1242
1243         dip = VTOI(ap->a_dvp);
1244         ncp = ap->a_nch->ncp;
1245         name = ncp->nc_name;
1246         name_len = ncp->nc_nlen;
1247         lhc = hammer2_dirhash(name, name_len);
1248
1249         /*
1250          * Note: In DragonFly the kernel handles '.' and '..'.
1251          */
1252         parent = hammer2_inode_lock_sh(dip);
1253         chain = hammer2_chain_lookup(&parent, &key_next,
1254                                      lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1255                                      &cache_index, HAMMER2_LOOKUP_SHARED);
1256         while (chain) {
1257                 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
1258                     name_len == chain->data->ipdata.name_len &&
1259                     bcmp(name, chain->data->ipdata.filename, name_len) == 0) {
1260                         break;
1261                 }
1262                 chain = hammer2_chain_next(&parent, chain, &key_next,
1263                                            key_next,
1264                                            lhc + HAMMER2_DIRHASH_LOMASK,
1265                                            &cache_index, HAMMER2_LOOKUP_SHARED);
1266         }
1267         hammer2_inode_unlock_sh(dip, parent);
1268
1269         /*
1270          * If the inode represents a forwarding entry for a hardlink we have
1271          * to locate the actual inode.  The original ip is saved for possible
1272          * deconsolidation.  (ip) will only be set to non-NULL when we have
1273          * to locate the real file via a hardlink.  ip will be referenced but
1274          * not locked in that situation.  chain is passed in locked and
1275          * returned locked.
1276          *
1277          * XXX what kind of chain lock?
1278          */
1279         ochain = NULL;
1280         if (chain && chain->data->ipdata.type == HAMMER2_OBJTYPE_HARDLINK) {
1281                 error = hammer2_hardlink_find(dip, &chain, &ochain);
1282                 if (error) {
1283                         kprintf("hammer2: unable to find hardlink\n");
1284                         if (chain) {
1285                                 hammer2_chain_unlock(chain);
1286                                 chain = NULL;
1287                         }
1288                         goto failed;
1289                 }
1290         }
1291
1292         /*
1293          * Deconsolidate any hardlink whos nlinks == 1.  Ignore errors.
1294          * If an error occurs chain and ip are left alone.
1295          *
1296          * XXX upgrade shared lock?
1297          */
1298         if (ochain && chain &&
1299             chain->data->ipdata.nlinks == 1 && !dip->pmp->ronly) {
1300                 kprintf("hammer2: need to unconsolidate hardlink for %s\n",
1301                         chain->data->ipdata.filename);
1302                 /* XXX retain shared lock on dip? (currently not held) */
1303                 hammer2_trans_init(&trans, dip->pmp, NULL, 0);
1304                 hammer2_hardlink_deconsolidate(&trans, dip, &chain, &ochain);
1305                 hammer2_trans_done(&trans);
1306         }
1307
1308         /*
1309          * Acquire the related vnode
1310          *
1311          * NOTE: For error processing, only ENOENT resolves the namecache
1312          *       entry to NULL, otherwise we just return the error and
1313          *       leave the namecache unresolved.
1314          *
1315          * NOTE: multiple hammer2_inode structures can be aliased to the
1316          *       same chain element, for example for hardlinks.  This
1317          *       use case does not 'reattach' inode associations that
1318          *       might already exist, but always allocates a new one.
1319          *
1320          * WARNING: inode structure is locked exclusively via inode_get
1321          *          but chain was locked shared.  inode_unlock_ex()
1322          *          will handle it properly.
1323          */
1324         if (chain) {
1325                 ip = hammer2_inode_get(dip->pmp, dip, chain);
1326                 vp = hammer2_igetv(ip, &error);
1327                 if (error == 0) {
1328                         vn_unlock(vp);
1329                         cache_setvp(ap->a_nch, vp);
1330                 } else if (error == ENOENT) {
1331                         cache_setvp(ap->a_nch, NULL);
1332                 }
1333                 hammer2_inode_unlock_ex(ip, chain);
1334
1335                 /*
1336                  * The vp should not be released until after we've disposed
1337                  * of our locks, because it might cause vop_inactive() to
1338                  * be called.
1339                  */
1340                 if (vp)
1341                         vrele(vp);
1342         } else {
1343                 error = ENOENT;
1344                 cache_setvp(ap->a_nch, NULL);
1345         }
1346 failed:
1347         KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1348                 ("resolve error %d/%p chain %p ap %p\n",
1349                  error, ap->a_nch->ncp->nc_vp, chain, ap));
1350         if (ochain)
1351                 hammer2_chain_drop(ochain);
1352         return error;
1353 }
1354
1355 static
1356 int
1357 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1358 {
1359         hammer2_inode_t *dip;
1360         hammer2_inode_t *ip;
1361         hammer2_chain_t *parent;
1362         int error;
1363
1364         dip = VTOI(ap->a_dvp);
1365
1366         if ((ip = dip->pip) == NULL) {
1367                 *ap->a_vpp = NULL;
1368                 return ENOENT;
1369         }
1370         parent = hammer2_inode_lock_ex(ip);
1371         *ap->a_vpp = hammer2_igetv(ip, &error);
1372         hammer2_inode_unlock_ex(ip, parent);
1373
1374         return error;
1375 }
1376
1377 static
1378 int
1379 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1380 {
1381         hammer2_inode_t *dip;
1382         hammer2_inode_t *nip;
1383         hammer2_trans_t trans;
1384         hammer2_chain_t *chain;
1385         struct namecache *ncp;
1386         const uint8_t *name;
1387         size_t name_len;
1388         int error;
1389
1390         dip = VTOI(ap->a_dvp);
1391         if (dip->pmp->ronly)
1392                 return (EROFS);
1393
1394         ncp = ap->a_nch->ncp;
1395         name = ncp->nc_name;
1396         name_len = ncp->nc_nlen;
1397
1398         hammer2_chain_memory_wait(dip->pmp);
1399         hammer2_trans_init(&trans, dip->pmp, NULL, HAMMER2_TRANS_NEWINODE);
1400         nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1401                                    name, name_len, &chain, &error);
1402         if (error) {
1403                 KKASSERT(nip == NULL);
1404                 *ap->a_vpp = NULL;
1405         } else {
1406                 *ap->a_vpp = hammer2_igetv(nip, &error);
1407                 hammer2_inode_unlock_ex(nip, chain);
1408         }
1409         hammer2_trans_done(&trans);
1410
1411         if (error == 0) {
1412                 cache_setunresolved(ap->a_nch);
1413                 cache_setvp(ap->a_nch, *ap->a_vpp);
1414         }
1415         return error;
1416 }
1417
1418 /*
1419  * Return the largest contiguous physical disk range for the logical
1420  * request, in bytes.
1421  *
1422  * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
1423  *
1424  * Basically disabled, the logical buffer write thread has to deal with
1425  * buffers one-at-a-time.
1426  */
1427 static
1428 int
1429 hammer2_vop_bmap(struct vop_bmap_args *ap)
1430 {
1431         *ap->a_doffsetp = NOOFFSET;
1432         if (ap->a_runp)
1433                 *ap->a_runp = 0;
1434         if (ap->a_runb)
1435                 *ap->a_runb = 0;
1436         return (EOPNOTSUPP);
1437 }
1438
1439 static
1440 int
1441 hammer2_vop_open(struct vop_open_args *ap)
1442 {
1443         return vop_stdopen(ap);
1444 }
1445
1446 /*
1447  * hammer2_vop_advlock { vp, id, op, fl, flags }
1448  */
1449 static
1450 int
1451 hammer2_vop_advlock(struct vop_advlock_args *ap)
1452 {
1453         hammer2_inode_t *ip = VTOI(ap->a_vp);
1454         hammer2_chain_t *parent;
1455         hammer2_off_t size;
1456
1457         parent = hammer2_inode_lock_sh(ip);
1458         size = parent->data->ipdata.size;
1459         hammer2_inode_unlock_sh(ip, parent);
1460         return (lf_advlock(ap, &ip->advlock, size));
1461 }
1462
1463
1464 static
1465 int
1466 hammer2_vop_close(struct vop_close_args *ap)
1467 {
1468         return vop_stdclose(ap);
1469 }
1470
1471 /*
1472  * hammer2_vop_nlink { nch, dvp, vp, cred }
1473  *
1474  * Create a hardlink from (vp) to {dvp, nch}.
1475  */
1476 static
1477 int
1478 hammer2_vop_nlink(struct vop_nlink_args *ap)
1479 {
1480         hammer2_inode_t *dip;   /* target directory to create link in */
1481         hammer2_inode_t *ip;    /* inode we are hardlinking to */
1482         hammer2_chain_t *chain;
1483         hammer2_trans_t trans;
1484         struct namecache *ncp;
1485         const uint8_t *name;
1486         size_t name_len;
1487         int error;
1488
1489         dip = VTOI(ap->a_dvp);
1490         if (dip->pmp->ronly)
1491                 return (EROFS);
1492
1493         ncp = ap->a_nch->ncp;
1494         name = ncp->nc_name;
1495         name_len = ncp->nc_nlen;
1496
1497         /*
1498          * ip represents the file being hardlinked.  The file could be a
1499          * normal file or a hardlink target if it has already been hardlinked.
1500          * If ip is a hardlinked target then ip->pip represents the location
1501          * of the hardlinked target, NOT the location of the hardlink pointer.
1502          *
1503          * Bump nlinks and potentially also create or move the hardlink
1504          * target in the parent directory common to (ip) and (dip).  The
1505          * consolidation code can modify ip->chain and ip->pip.  The
1506          * returned chain is locked.
1507          */
1508         ip = VTOI(ap->a_vp);
1509         hammer2_chain_memory_wait(ip->pmp);
1510         hammer2_trans_init(&trans, ip->pmp, NULL, HAMMER2_TRANS_NEWINODE);
1511
1512         chain = hammer2_inode_lock_ex(ip);
1513         error = hammer2_hardlink_consolidate(&trans, ip, &chain, dip, 1);
1514         if (error)
1515                 goto done;
1516
1517         /*
1518          * Create a directory entry connected to the specified chain.
1519          * The hardlink consolidation code has already adjusted ip->pip
1520          * to the common parent directory containing the actual hardlink
1521          *
1522          * (which may be different from dip where we created our hardlink
1523          * entry. ip->chain always represents the actual hardlink and not
1524          * any of the pointers to the actual hardlink).
1525          */
1526         error = hammer2_inode_connect(&trans, 1,
1527                                       dip, &chain,
1528                                       name, name_len);
1529         if (error == 0) {
1530                 cache_setunresolved(ap->a_nch);
1531                 cache_setvp(ap->a_nch, ap->a_vp);
1532         }
1533 done:
1534         hammer2_inode_unlock_ex(ip, chain);
1535         hammer2_trans_done(&trans);
1536
1537         return error;
1538 }
1539
1540 /*
1541  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1542  *
1543  * The operating system has already ensured that the directory entry
1544  * does not exist and done all appropriate namespace locking.
1545  */
1546 static
1547 int
1548 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1549 {
1550         hammer2_inode_t *dip;
1551         hammer2_inode_t *nip;
1552         hammer2_trans_t trans;
1553         hammer2_chain_t *nchain;
1554         struct namecache *ncp;
1555         const uint8_t *name;
1556         size_t name_len;
1557         int error;
1558
1559         dip = VTOI(ap->a_dvp);
1560         if (dip->pmp->ronly)
1561                 return (EROFS);
1562
1563         ncp = ap->a_nch->ncp;
1564         name = ncp->nc_name;
1565         name_len = ncp->nc_nlen;
1566         hammer2_chain_memory_wait(dip->pmp);
1567         hammer2_trans_init(&trans, dip->pmp, NULL, HAMMER2_TRANS_NEWINODE);
1568
1569         nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1570                                    name, name_len, &nchain, &error);
1571         if (error) {
1572                 KKASSERT(nip == NULL);
1573                 *ap->a_vpp = NULL;
1574         } else {
1575                 *ap->a_vpp = hammer2_igetv(nip, &error);
1576                 hammer2_inode_unlock_ex(nip, nchain);
1577         }
1578         hammer2_trans_done(&trans);
1579
1580         if (error == 0) {
1581                 cache_setunresolved(ap->a_nch);
1582                 cache_setvp(ap->a_nch, *ap->a_vpp);
1583         }
1584         return error;
1585 }
1586
1587 /*
1588  *
1589  */
1590 static
1591 int
1592 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1593 {
1594         hammer2_inode_t *dip;
1595         hammer2_inode_t *nip;
1596         hammer2_trans_t trans;
1597         hammer2_chain_t *nchain;
1598         struct namecache *ncp;
1599         const uint8_t *name;
1600         size_t name_len;
1601         int error;
1602
1603         dip = VTOI(ap->a_dvp);
1604         if (dip->pmp->ronly)
1605                 return (EROFS);
1606
1607         ncp = ap->a_nch->ncp;
1608         name = ncp->nc_name;
1609         name_len = ncp->nc_nlen;
1610         hammer2_chain_memory_wait(dip->pmp);
1611         hammer2_trans_init(&trans, dip->pmp, NULL, HAMMER2_TRANS_NEWINODE);
1612
1613         nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1614                                    name, name_len, &nchain, &error);
1615         if (error) {
1616                 KKASSERT(nip == NULL);
1617                 *ap->a_vpp = NULL;
1618         } else {
1619                 *ap->a_vpp = hammer2_igetv(nip, &error);
1620                 hammer2_inode_unlock_ex(nip, nchain);
1621         }
1622         hammer2_trans_done(&trans);
1623
1624         if (error == 0) {
1625                 cache_setunresolved(ap->a_nch);
1626                 cache_setvp(ap->a_nch, *ap->a_vpp);
1627         }
1628         return error;
1629 }
1630
1631 /*
1632  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1633  */
1634 static
1635 int
1636 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1637 {
1638         hammer2_inode_t *dip;
1639         hammer2_inode_t *nip;
1640         hammer2_chain_t *nparent;
1641         hammer2_trans_t trans;
1642         struct namecache *ncp;
1643         const uint8_t *name;
1644         size_t name_len;
1645         int error;
1646
1647         dip = VTOI(ap->a_dvp);
1648         if (dip->pmp->ronly)
1649                 return (EROFS);
1650
1651         ncp = ap->a_nch->ncp;
1652         name = ncp->nc_name;
1653         name_len = ncp->nc_nlen;
1654         hammer2_chain_memory_wait(dip->pmp);
1655         hammer2_trans_init(&trans, dip->pmp, NULL, HAMMER2_TRANS_NEWINODE);
1656
1657         ap->a_vap->va_type = VLNK;      /* enforce type */
1658
1659         nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1660                                    name, name_len, &nparent, &error);
1661         if (error) {
1662                 KKASSERT(nip == NULL);
1663                 *ap->a_vpp = NULL;
1664                 hammer2_trans_done(&trans);
1665                 return error;
1666         }
1667         *ap->a_vpp = hammer2_igetv(nip, &error);
1668
1669         /*
1670          * Build the softlink (~like file data) and finalize the namecache.
1671          */
1672         if (error == 0) {
1673                 size_t bytes;
1674                 struct uio auio;
1675                 struct iovec aiov;
1676                 hammer2_inode_data_t *nipdata;
1677
1678                 nipdata = &nip->chain->data->ipdata;
1679                 bytes = strlen(ap->a_target);
1680
1681                 if (bytes <= HAMMER2_EMBEDDED_BYTES) {
1682                         KKASSERT(nipdata->op_flags &
1683                                  HAMMER2_OPFLAG_DIRECTDATA);
1684                         bcopy(ap->a_target, nipdata->u.data, bytes);
1685                         nipdata->size = bytes;
1686                         nip->size = bytes;
1687                         hammer2_inode_unlock_ex(nip, nparent);
1688                 } else {
1689                         hammer2_inode_unlock_ex(nip, nparent);
1690                         bzero(&auio, sizeof(auio));
1691                         bzero(&aiov, sizeof(aiov));
1692                         auio.uio_iov = &aiov;
1693                         auio.uio_segflg = UIO_SYSSPACE;
1694                         auio.uio_rw = UIO_WRITE;
1695                         auio.uio_resid = bytes;
1696                         auio.uio_iovcnt = 1;
1697                         auio.uio_td = curthread;
1698                         aiov.iov_base = ap->a_target;
1699                         aiov.iov_len = bytes;
1700                         error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1701                         nipdata = &nip->chain->data->ipdata; /* RELOAD */
1702                         /* XXX handle error */
1703                         error = 0;
1704                 }
1705         } else {
1706                 hammer2_inode_unlock_ex(nip, nparent);
1707         }
1708         hammer2_trans_done(&trans);
1709
1710         /*
1711          * Finalize namecache
1712          */
1713         if (error == 0) {
1714                 cache_setunresolved(ap->a_nch);
1715                 cache_setvp(ap->a_nch, *ap->a_vpp);
1716                 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1717         }
1718         return error;
1719 }
1720
1721 /*
1722  * hammer2_vop_nremove { nch, dvp, cred }
1723  */
1724 static
1725 int
1726 hammer2_vop_nremove(struct vop_nremove_args *ap)
1727 {
1728         hammer2_inode_t *dip;
1729         hammer2_trans_t trans;
1730         struct namecache *ncp;
1731         const uint8_t *name;
1732         size_t name_len;
1733         int error;
1734
1735         dip = VTOI(ap->a_dvp);
1736         if (dip->pmp->ronly)
1737                 return(EROFS);
1738
1739         ncp = ap->a_nch->ncp;
1740         name = ncp->nc_name;
1741         name_len = ncp->nc_nlen;
1742         hammer2_chain_memory_wait(dip->pmp);
1743         hammer2_trans_init(&trans, dip->pmp, NULL, 0);
1744         error = hammer2_unlink_file(&trans, dip, name, name_len, 0, NULL);
1745         hammer2_trans_done(&trans);
1746         if (error == 0) {
1747                 cache_unlink(ap->a_nch);
1748         }
1749         return (error);
1750 }
1751
1752 /*
1753  * hammer2_vop_nrmdir { nch, dvp, cred }
1754  */
1755 static
1756 int
1757 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1758 {
1759         hammer2_inode_t *dip;
1760         hammer2_trans_t trans;
1761         struct namecache *ncp;
1762         const uint8_t *name;
1763         size_t name_len;
1764         int error;
1765
1766         dip = VTOI(ap->a_dvp);
1767         if (dip->pmp->ronly)
1768                 return(EROFS);
1769
1770         ncp = ap->a_nch->ncp;
1771         name = ncp->nc_name;
1772         name_len = ncp->nc_nlen;
1773
1774         hammer2_chain_memory_wait(dip->pmp);
1775         hammer2_trans_init(&trans, dip->pmp, NULL, 0);
1776         error = hammer2_unlink_file(&trans, dip, name, name_len, 1, NULL);
1777         hammer2_trans_done(&trans);
1778         if (error == 0) {
1779                 cache_unlink(ap->a_nch);
1780         }
1781         return (error);
1782 }
1783
1784 /*
1785  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1786  */
1787 static
1788 int
1789 hammer2_vop_nrename(struct vop_nrename_args *ap)
1790 {
1791         struct namecache *fncp;
1792         struct namecache *tncp;
1793         hammer2_inode_t *fdip;
1794         hammer2_inode_t *tdip;
1795         hammer2_inode_t *ip;
1796         hammer2_chain_t *chain;
1797         hammer2_trans_t trans;
1798         const uint8_t *fname;
1799         size_t fname_len;
1800         const uint8_t *tname;
1801         size_t tname_len;
1802         int error;
1803         int hlink;
1804
1805         if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1806                 return(EXDEV);
1807         if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1808                 return(EXDEV);
1809
1810         fdip = VTOI(ap->a_fdvp);        /* source directory */
1811         tdip = VTOI(ap->a_tdvp);        /* target directory */
1812
1813         if (fdip->pmp->ronly)
1814                 return(EROFS);
1815
1816         fncp = ap->a_fnch->ncp;         /* entry name in source */
1817         fname = fncp->nc_name;
1818         fname_len = fncp->nc_nlen;
1819
1820         tncp = ap->a_tnch->ncp;         /* entry name in target */
1821         tname = tncp->nc_name;
1822         tname_len = tncp->nc_nlen;
1823
1824         hammer2_chain_memory_wait(tdip->pmp);
1825         hammer2_trans_init(&trans, tdip->pmp, NULL, 0);
1826
1827         /*
1828          * ip is the inode being renamed.  If this is a hardlink then
1829          * ip represents the actual file and not the hardlink marker.
1830          */
1831         ip = VTOI(fncp->nc_vp);
1832         chain = NULL;
1833
1834         /*
1835          * Keep a tight grip on the inode so the temporary unlinking from
1836          * the source location prior to linking to the target location
1837          * does not cause the chain to be destroyed.
1838          *
1839          * NOTE: To avoid deadlocks we cannot lock (ip) while we are
1840          *       unlinking elements from their directories.  Locking
1841          *       the nlinks field does not lock the whole inode.
1842          */
1843         hammer2_inode_ref(ip);
1844
1845         /*
1846          * Remove target if it exists
1847          */
1848         error = hammer2_unlink_file(&trans, tdip, tname, tname_len, -1, NULL);
1849         if (error && error != ENOENT)
1850                 goto done;
1851         cache_setunresolved(ap->a_tnch);
1852
1853         /*
1854          * When renaming a hardlinked file we may have to re-consolidate
1855          * the location of the hardlink target.  Since the element is simply
1856          * being moved, nlinks is not modified in this case.
1857          *
1858          * If ip represents a regular file the consolidation code essentially
1859          * does nothing other than return the same locked chain that was
1860          * passed in.
1861          *
1862          * The returned chain will be locked.
1863          *
1864          * WARNING!  We do not currently have a local copy of ipdata but
1865          *           we do use one later remember that it must be reloaded
1866          *           on any modification to the inode, including connects.
1867          */
1868         chain = hammer2_inode_lock_ex(ip);
1869         error = hammer2_hardlink_consolidate(&trans, ip, &chain, tdip, 0);
1870         if (error)
1871                 goto done;
1872
1873         /*
1874          * Disconnect (fdip, fname) from the source directory.  This will
1875          * disconnect (ip) if it represents a direct file.  If (ip) represents
1876          * a hardlink the HARDLINK pointer object will be removed but the
1877          * hardlink will stay intact.
1878          *
1879          * The target chain may be marked DELETED but will not be destroyed
1880          * since we retain our hold on ip and chain.
1881          */
1882         error = hammer2_unlink_file(&trans, fdip, fname, fname_len, -1, &hlink);
1883         KKASSERT(error != EAGAIN);
1884         if (error)
1885                 goto done;
1886
1887         /*
1888          * Reconnect ip to target directory using chain.  Chains cannot
1889          * actually be moved, so this will duplicate the chain in the new
1890          * spot and assign it to the ip, replacing the old chain.
1891          *
1892          * WARNING: Because recursive locks are allowed and we unlinked the
1893          *          file that we have a chain-in-hand for just above, the
1894          *          chain might have been delete-duplicated.  We must refactor
1895          *          the chain.
1896          *
1897          * WARNING: Chain locks can lock buffer cache buffers, to avoid
1898          *          deadlocks we want to unlock before issuing a cache_*()
1899          *          op (that might have to lock a vnode).
1900          */
1901         hammer2_chain_refactor(&chain);
1902         error = hammer2_inode_connect(&trans, hlink,
1903                                       tdip, &chain,
1904                                       tname, tname_len);
1905         if (error == 0) {
1906                 KKASSERT(chain != NULL);
1907                 hammer2_inode_repoint(ip, (hlink ? ip->pip : tdip), chain);
1908                 cache_rename(ap->a_fnch, ap->a_tnch);
1909         }
1910 done:
1911         hammer2_inode_unlock_ex(ip, chain);
1912         hammer2_inode_drop(ip);
1913         hammer2_trans_done(&trans);
1914
1915         return (error);
1916 }
1917
1918 /*
1919  * Strategy code
1920  *
1921  * WARNING: The strategy code cannot safely use hammer2 transactions
1922  *          as this can deadlock against vfs_sync's vfsync() call
1923  *          if multiple flushes are queued.
1924  */
1925 static int hammer2_strategy_read(struct vop_strategy_args *ap);
1926 static int hammer2_strategy_write(struct vop_strategy_args *ap);
1927 static void hammer2_strategy_read_callback(hammer2_io_t *dio,
1928                                 hammer2_chain_t *chain,
1929                                 void *arg_p, off_t arg_o);
1930
1931 static
1932 int
1933 hammer2_vop_strategy(struct vop_strategy_args *ap)
1934 {
1935         struct bio *biop;
1936         struct buf *bp;
1937         int error;
1938
1939         biop = ap->a_bio;
1940         bp = biop->bio_buf;
1941
1942         switch(bp->b_cmd) {
1943         case BUF_CMD_READ:
1944                 error = hammer2_strategy_read(ap);
1945                 ++hammer2_iod_file_read;
1946                 break;
1947         case BUF_CMD_WRITE:
1948                 error = hammer2_strategy_write(ap);
1949                 ++hammer2_iod_file_write;
1950                 break;
1951         default:
1952                 bp->b_error = error = EINVAL;
1953                 bp->b_flags |= B_ERROR;
1954                 biodone(biop);
1955                 break;
1956         }
1957
1958         return (error);
1959 }
1960
1961 static
1962 int
1963 hammer2_strategy_read(struct vop_strategy_args *ap)
1964 {
1965         struct buf *bp;
1966         struct bio *bio;
1967         struct bio *nbio;
1968         hammer2_inode_t *ip;
1969         hammer2_chain_t *parent;
1970         hammer2_chain_t *chain;
1971         hammer2_key_t key_dummy;
1972         hammer2_key_t lbase;
1973         int cache_index = -1;
1974
1975         bio = ap->a_bio;
1976         bp = bio->bio_buf;
1977         ip = VTOI(ap->a_vp);
1978         nbio = push_bio(bio);
1979
1980         lbase = bio->bio_offset;
1981         chain = NULL;
1982         KKASSERT(((int)lbase & HAMMER2_PBUFMASK) == 0);
1983
1984         parent = hammer2_inode_lock_sh(ip);
1985         chain = hammer2_chain_lookup(&parent, &key_dummy,
1986                                      lbase, lbase,
1987                                      &cache_index,
1988                                      HAMMER2_LOOKUP_NODATA |
1989                                      HAMMER2_LOOKUP_SHARED);
1990
1991         if (chain == NULL) {
1992                 /*
1993                  * Data is zero-fill
1994                  */
1995                 bp->b_resid = 0;
1996                 bp->b_error = 0;
1997                 bzero(bp->b_data, bp->b_bcount);
1998                 biodone(nbio);
1999         } else if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
2000                 /*
2001                  * Data is embedded in the inode (copy from inode).
2002                  */
2003                 hammer2_chain_load_async(chain,
2004                                          hammer2_strategy_read_callback,
2005                                          nbio, 0);
2006         } else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
2007                 /*
2008                  * Data is on-media, issue device I/O and copy.
2009                  *
2010                  * XXX direct-IO shortcut could go here XXX.
2011                  */
2012                 if (HAMMER2_DEC_COMP(chain->bref.methods) == HAMMER2_COMP_LZ4) {
2013                         /*
2014                          * Block compression is determined by bref.methods
2015                          */
2016                         hammer2_blockref_t *bref;
2017
2018                         bref = &chain->bref;
2019                         hammer2_io_breadcb(chain->hmp, bref->data_off,
2020                                            chain->bytes,
2021                                            hammer2_decompress_LZ4_callback,
2022                                            NULL, nbio, bref->data_off);
2023                         /* XXX async read dev blk not protected by chain lk */
2024                         hammer2_chain_unlock(chain);
2025                 } else if (HAMMER2_DEC_COMP(chain->bref.methods) ==
2026                            HAMMER2_COMP_ZLIB) {
2027                         hammer2_blockref_t *bref;
2028
2029                         bref = &chain->bref;
2030                         hammer2_io_breadcb(chain->hmp, bref->data_off,
2031                                            chain->bytes,
2032                                            hammer2_decompress_ZLIB_callback,
2033                                            NULL, nbio, bref->data_off);
2034                         /* XXX async read dev blk not protected by chain lk */
2035                         hammer2_chain_unlock(chain);
2036                 } else {
2037                         hammer2_chain_load_async(chain,
2038                                                  hammer2_strategy_read_callback,
2039                                                  nbio, 0);
2040                 }
2041         } else {
2042                 panic("READ PATH: hammer2_strategy_read: unknown bref type");
2043                 chain = NULL;
2044         }
2045         hammer2_inode_unlock_sh(ip, parent);
2046         return (0);
2047 }
2048
2049 /*
2050  * Read callback for block that is not compressed.
2051  */
2052 static
2053 void
2054 hammer2_strategy_read_callback(hammer2_io_t *dio, hammer2_chain_t *chain,
2055                                void *arg_p, off_t arg_o __unused)
2056 {
2057         struct bio *nbio = arg_p;
2058         struct buf *bp = nbio->bio_buf;
2059         char *data;
2060
2061         if (dio)
2062                 data = hammer2_io_data(dio, chain->bref.data_off);
2063         else
2064                 data = (void *)chain->data;
2065
2066         if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
2067                 /*
2068                  * Data is embedded in the inode (copy from inode).
2069                  */
2070                 bcopy(((hammer2_inode_data_t *)data)->u.data,
2071                       bp->b_data, HAMMER2_EMBEDDED_BYTES);
2072                 bzero(bp->b_data + HAMMER2_EMBEDDED_BYTES,
2073                       bp->b_bcount - HAMMER2_EMBEDDED_BYTES);
2074                 bp->b_resid = 0;
2075                 bp->b_error = 0;
2076                 hammer2_chain_unlock(chain);
2077                 biodone(nbio);
2078         } else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
2079                 /*
2080                  * Data is on-media, issue device I/O and copy.
2081                  *
2082                  * XXX direct-IO shortcut could go here XXX.
2083                  */
2084                 KKASSERT(chain->bytes <= bp->b_bcount);
2085                 bcopy(data, bp->b_data, chain->bytes);
2086                 if (chain->bytes < bp->b_bcount); {
2087                         bzero(bp->b_data + chain->bytes,
2088                               bp->b_bcount - chain->bytes);
2089                 }
2090                 bp->b_flags |= B_NOTMETA;
2091                 bp->b_resid = 0;
2092                 bp->b_error = 0;
2093                 hammer2_chain_unlock(chain);
2094                 biodone(nbio);
2095         } else {
2096                 /* bqrelse the dio to help stabilize the call to panic() */
2097                 if (dio)
2098                         hammer2_io_bqrelse(&dio);
2099                 panic("hammer2_strategy_read: unknown bref type");
2100                 /*hammer2_chain_unlock(chain);*/
2101                 /*chain = NULL;*/
2102         }
2103 }
2104
2105 static
2106 int
2107 hammer2_strategy_write(struct vop_strategy_args *ap)
2108 {
2109         hammer2_pfsmount_t *pmp;
2110         struct bio *bio;
2111         struct buf *bp;
2112         hammer2_inode_t *ip;
2113
2114         bio = ap->a_bio;
2115         bp = bio->bio_buf;
2116         ip = VTOI(ap->a_vp);
2117         pmp = ip->pmp;
2118
2119         hammer2_lwinprog_ref(pmp);
2120         mtx_lock(&pmp->wthread_mtx);
2121         if (TAILQ_EMPTY(&pmp->wthread_bioq.queue)) {
2122                 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio);
2123                 mtx_unlock(&pmp->wthread_mtx);
2124                 wakeup(&pmp->wthread_bioq);
2125         } else {
2126                 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio);
2127                 mtx_unlock(&pmp->wthread_mtx);
2128         }
2129         hammer2_lwinprog_wait(pmp);
2130
2131         return(0);
2132 }
2133
2134 /*
2135  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2136  */
2137 static
2138 int
2139 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2140 {
2141         hammer2_inode_t *ip;
2142         int error;
2143
2144         ip = VTOI(ap->a_vp);
2145
2146         error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2147                               ap->a_fflag, ap->a_cred);
2148         return (error);
2149 }
2150
2151 static
2152 int
2153 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2154 {
2155         struct mount *mp;
2156         hammer2_pfsmount_t *pmp;
2157         int rc;
2158
2159         switch (ap->a_op) {
2160         case (MOUNTCTL_SET_EXPORT):
2161                 mp = ap->a_head.a_ops->head.vv_mount;
2162                 pmp = MPTOPMP(mp);
2163
2164                 if (ap->a_ctllen != sizeof(struct export_args))
2165                         rc = (EINVAL);
2166                 else
2167                         rc = vfs_export(mp, &pmp->export,
2168                                         (const struct export_args *)ap->a_ctl);
2169                 break;
2170         default:
2171                 rc = vop_stdmountctl(ap);
2172                 break;
2173         }
2174         return (rc);
2175 }
2176
2177 /*
2178  * KQFILTER
2179  */
2180 static void filt_hammer2detach(struct knote *kn);
2181 static int filt_hammer2read(struct knote *kn, long hint);
2182 static int filt_hammer2write(struct knote *kn, long hint);
2183 static int filt_hammer2vnode(struct knote *kn, long hint);
2184
2185 static struct filterops hammer2read_filtops =
2186         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2187           NULL, filt_hammer2detach, filt_hammer2read };
2188 static struct filterops hammer2write_filtops =
2189         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2190           NULL, filt_hammer2detach, filt_hammer2write };
2191 static struct filterops hammer2vnode_filtops =
2192         { FILTEROP_ISFD | FILTEROP_MPSAFE,
2193           NULL, filt_hammer2detach, filt_hammer2vnode };
2194
2195 static
2196 int
2197 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2198 {
2199         struct vnode *vp = ap->a_vp;
2200         struct knote *kn = ap->a_kn;
2201
2202         switch (kn->kn_filter) {
2203         case EVFILT_READ:
2204                 kn->kn_fop = &hammer2read_filtops;
2205                 break;
2206         case EVFILT_WRITE:
2207                 kn->kn_fop = &hammer2write_filtops;
2208                 break;
2209         case EVFILT_VNODE:
2210                 kn->kn_fop = &hammer2vnode_filtops;
2211                 break;
2212         default:
2213                 return (EOPNOTSUPP);
2214         }
2215
2216         kn->kn_hook = (caddr_t)vp;
2217
2218         knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2219
2220         return(0);
2221 }
2222
2223 static void
2224 filt_hammer2detach(struct knote *kn)
2225 {
2226         struct vnode *vp = (void *)kn->kn_hook;
2227
2228         knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2229 }
2230
2231 static int
2232 filt_hammer2read(struct knote *kn, long hint)
2233 {
2234         struct vnode *vp = (void *)kn->kn_hook;
2235         hammer2_inode_t *ip = VTOI(vp);
2236         off_t off;
2237
2238         if (hint == NOTE_REVOKE) {
2239                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2240                 return(1);
2241         }
2242         off = ip->size - kn->kn_fp->f_offset;
2243         kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2244         if (kn->kn_sfflags & NOTE_OLDAPI)
2245                 return(1);
2246         return (kn->kn_data != 0);
2247 }
2248
2249
2250 static int
2251 filt_hammer2write(struct knote *kn, long hint)
2252 {
2253         if (hint == NOTE_REVOKE)
2254                 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2255         kn->kn_data = 0;
2256         return (1);
2257 }
2258
2259 static int
2260 filt_hammer2vnode(struct knote *kn, long hint)
2261 {
2262         if (kn->kn_sfflags & hint)
2263                 kn->kn_fflags |= hint;
2264         if (hint == NOTE_REVOKE) {
2265                 kn->kn_flags |= (EV_EOF | EV_NODATA);
2266                 return (1);
2267         }
2268         return (kn->kn_fflags != 0);
2269 }
2270
2271 /*
2272  * FIFO VOPS
2273  */
2274 static
2275 int
2276 hammer2_vop_markatime(struct vop_markatime_args *ap)
2277 {
2278         hammer2_inode_t *ip;
2279         struct vnode *vp;
2280
2281         vp = ap->a_vp;
2282         ip = VTOI(vp);
2283
2284         if (ip->pmp->ronly)
2285                 return(EROFS);
2286         return(0);
2287 }
2288
2289 static
2290 int
2291 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2292 {
2293         int error;
2294
2295         error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2296         if (error)
2297                 error = hammer2_vop_kqfilter(ap);
2298         return(error);
2299 }
2300
2301 /*
2302  * VOPS vector
2303  */
2304 struct vop_ops hammer2_vnode_vops = {
2305         .vop_default    = vop_defaultop,
2306         .vop_fsync      = hammer2_vop_fsync,
2307         .vop_getpages   = vop_stdgetpages,
2308         .vop_putpages   = vop_stdputpages,
2309         .vop_access     = hammer2_vop_access,
2310         .vop_advlock    = hammer2_vop_advlock,
2311         .vop_close      = hammer2_vop_close,
2312         .vop_nlink      = hammer2_vop_nlink,
2313         .vop_ncreate    = hammer2_vop_ncreate,
2314         .vop_nsymlink   = hammer2_vop_nsymlink,
2315         .vop_nremove    = hammer2_vop_nremove,
2316         .vop_nrmdir     = hammer2_vop_nrmdir,
2317         .vop_nrename    = hammer2_vop_nrename,
2318         .vop_getattr    = hammer2_vop_getattr,
2319         .vop_setattr    = hammer2_vop_setattr,
2320         .vop_readdir    = hammer2_vop_readdir,
2321         .vop_readlink   = hammer2_vop_readlink,
2322         .vop_getpages   = vop_stdgetpages,
2323         .vop_putpages   = vop_stdputpages,
2324         .vop_read       = hammer2_vop_read,
2325         .vop_write      = hammer2_vop_write,
2326         .vop_open       = hammer2_vop_open,
2327         .vop_inactive   = hammer2_vop_inactive,
2328         .vop_reclaim    = hammer2_vop_reclaim,
2329         .vop_nresolve   = hammer2_vop_nresolve,
2330         .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2331         .vop_nmkdir     = hammer2_vop_nmkdir,
2332         .vop_nmknod     = hammer2_vop_nmknod,
2333         .vop_ioctl      = hammer2_vop_ioctl,
2334         .vop_mountctl   = hammer2_vop_mountctl,
2335         .vop_bmap       = hammer2_vop_bmap,
2336         .vop_strategy   = hammer2_vop_strategy,
2337         .vop_kqfilter   = hammer2_vop_kqfilter
2338 };
2339
2340 struct vop_ops hammer2_spec_vops = {
2341         .vop_default =          vop_defaultop,
2342         .vop_fsync =            hammer2_vop_fsync,
2343         .vop_read =             vop_stdnoread,
2344         .vop_write =            vop_stdnowrite,
2345         .vop_access =           hammer2_vop_access,
2346         .vop_close =            hammer2_vop_close,
2347         .vop_markatime =        hammer2_vop_markatime,
2348         .vop_getattr =          hammer2_vop_getattr,
2349         .vop_inactive =         hammer2_vop_inactive,
2350         .vop_reclaim =          hammer2_vop_reclaim,
2351         .vop_setattr =          hammer2_vop_setattr
2352 };
2353
2354 struct vop_ops hammer2_fifo_vops = {
2355         .vop_default =          fifo_vnoperate,
2356         .vop_fsync =            hammer2_vop_fsync,
2357 #if 0
2358         .vop_read =             hammer2_vop_fiforead,
2359         .vop_write =            hammer2_vop_fifowrite,
2360 #endif
2361         .vop_access =           hammer2_vop_access,
2362 #if 0
2363         .vop_close =            hammer2_vop_fifoclose,
2364 #endif
2365         .vop_markatime =        hammer2_vop_markatime,
2366         .vop_getattr =          hammer2_vop_getattr,
2367         .vop_inactive =         hammer2_vop_inactive,
2368         .vop_reclaim =          hammer2_vop_reclaim,
2369         .vop_setattr =          hammer2_vop_setattr,
2370         .vop_kqfilter =         hammer2_vop_fifokqfilter
2371 };
2372