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  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  *
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in
  16  *    the documentation and/or other materials provided with the
  17  *    distribution.
  18  * 3. Neither the name of The DragonFly Project nor the names of its
  19  *    contributors may be used to endorse or promote products derived
  20  *    from this software without specific, prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  25  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  26  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  27  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  28  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  29  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  30  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  31  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  32  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  33  * SUCH DAMAGE.
  34  */
  35 #include <sys/param.h>
  36 #include <sys/systm.h>
  37 #include <sys/kernel.h>
  38 #include <sys/fcntl.h>
  39 #include <sys/buf.h>
  40 #include <sys/proc.h>
  41 #include <sys/namei.h>
  42 #include <sys/mount.h>
  43 #include <sys/vnode.h>
  44 #include <sys/mountctl.h>
  45 #include <sys/dirent.h>
  46 #include <sys/uio.h>
  47
  48 #include "hammer2.h"
  49
  50 #define ZFOFFSET        (-2LL)
  51
  52 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
  53                                 int seqcount);
  54 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
  55                                 int ioflag, int seqcount);
  56 static hammer2_off_t hammer2_assign_physical(hammer2_trans_t *trans,
  57                                 hammer2_inode_t *ip,
  58                                 hammer2_key_t lbase, int lblksize,
  59                                 int *errorp);
  60 static void hammer2_extend_file(hammer2_trans_t *trans,
  61                                 hammer2_inode_t *ip, hammer2_key_t nsize);
  62 static void hammer2_truncate_file(hammer2_trans_t *trans,
  63                                 hammer2_inode_t *ip, hammer2_key_t nsize);
  64
  65 static __inline
  66 void
  67 hammer2_knote(struct vnode *vp, int flags)
  68 {
  69         if (flags)
  70                 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
  71 }
  72
  73 /*
  74  * Last reference to a vnode is going away but it is still cached.
  75  */
  76 static
  77 int
  78 hammer2_vop_inactive(struct vop_inactive_args *ap)
  79 {
  80         hammer2_inode_t *ip;
  81         hammer2_trans_t trans;
  82         struct vnode *vp;
  83 #if 0
  84         struct hammer2_mount *hmp;
  85 #endif
  86
  87         vp = ap->a_vp;
  88         ip = VTOI(vp);
  89
  90         /*
  91          * Degenerate case
  92          */
  93         if (ip == NULL) {
  94                 vrecycle(vp);
  95                 return (0);
  96         }
  97
  98         /*
  99          * Detect updates to the embedded data which may be synchronized by
 100          * the strategy code.  Simply mark the inode modified so it gets
 101          * picked up by our normal flush.
 102          */
 103         hammer2_inode_lock_ex(ip);
 104         KKASSERT(ip->chain);
 105         if (ip->flags & HAMMER2_INODE_DIRTYEMBED) {
 106                 atomic_clear_int(&ip->flags, HAMMER2_INODE_DIRTYEMBED);
 107                 hammer2_trans_init(&trans, ip->hmp);
 108                 hammer2_chain_modify(&trans, ip->chain, 0);
 109                 hammer2_trans_done(&trans);
 110         }
 111
 112         /*
 113          * Check for deleted inodes and recycle immediately.
 114          */
 115         if (ip->chain->flags & HAMMER2_CHAIN_DELETED) {
 116                 hammer2_inode_unlock_ex(ip);
 117                 vrecycle(vp);
 118         } else {
 119                 hammer2_inode_unlock_ex(ip);
 120         }
 121         return (0);
 122 }
 123
 124 /*
 125  * Reclaim a vnode so that it can be reused; after the inode is
 126  * disassociated, the filesystem must manage it alone.
 127  */
 128 static
 129 int
 130 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
 131 {
 132         hammer2_chain_t *chain;
 133         hammer2_inode_t *ip;
 134         hammer2_mount_t *hmp;
 135         hammer2_trans_t trans;
 136         struct vnode *vp;
 137
 138         vp = ap->a_vp;
 139         ip = VTOI(vp);
 140         if (ip == NULL)
 141                 return(0);
 142         hmp = ip->hmp;
 143
 144         /*
 145          * Set SUBMODIFIED so we can detect and propagate the DESTROYED
 146          * bit in the flush code.
 147          */
 148         hammer2_inode_lock_ex(ip);
 149         chain = ip->chain;
 150         vp->v_data = NULL;
 151         ip->vp = NULL;
 152         if (chain->flags & HAMMER2_CHAIN_DELETED) {
 153                 KKASSERT(chain->flags & HAMMER2_CHAIN_DELETED);
 154                 atomic_set_int(&chain->flags, HAMMER2_CHAIN_DESTROYED |
 155                                               HAMMER2_CHAIN_SUBMODIFIED);
 156         }
 157         if (chain->flags & (HAMMER2_CHAIN_MODIFIED |
 158                             HAMMER2_CHAIN_DELETED |
 159                             HAMMER2_CHAIN_SUBMODIFIED)) {
 160                 hammer2_trans_init(&trans, ip->hmp);
 161                 hammer2_chain_flush(&trans, chain);
 162                 hammer2_trans_done(&trans);
 163         }
 164         if (ip->refs > 2)                           /* (our lock + vp ref) */
 165                 hammer2_inode_unlock_ex(ip);        /* unlock */
 166         else
 167                 hammer2_inode_put(ip);              /* unlock & disconnect */
 168         /* chain no longer referenced */
 169         /* chain = NULL; not needed */
 170         hammer2_inode_drop(ip);                     /* vp ref */
 171
 172         /*
 173          * XXX handle background sync when ip dirty, kernel will no longer
 174          * notify us regarding this inode because there is no longer a
 175          * vnode attached to it.
 176          */
 177
 178         return (0);
 179 }
 180
 181 static
 182 int
 183 hammer2_vop_fsync(struct vop_fsync_args *ap)
 184 {
 185         hammer2_inode_t *ip;
 186         hammer2_trans_t trans;
 187         struct vnode *vp;
 188
 189         vp = ap->a_vp;
 190         ip = VTOI(vp);
 191
 192         hammer2_trans_init(&trans, ip->hmp);
 193         hammer2_inode_lock_ex(ip);
 194
 195         vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
 196
 197         /*
 198          * Detect updates to the embedded data which may be synchronized by
 199          * the strategy code.  Simply mark the inode modified so it gets
 200          * picked up by our normal flush.
 201          */
 202         if (ip->flags & HAMMER2_INODE_DIRTYEMBED) {
 203                 atomic_clear_int(&ip->flags, HAMMER2_INODE_DIRTYEMBED);
 204                 hammer2_chain_modify(&trans, ip->chain, 0);
 205         }
 206
 207         /*
 208          * Calling chain_flush here creates a lot of duplicative
 209          * COW operations due to non-optimal vnode ordering.
 210          *
 211          * Only do it for an actual fsync() syscall.  The other forms
 212          * which call this function will eventually call chain_flush
 213          * on the volume root as a catch-all, which is far more optimal.
 214          */
 215         atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
 216         if (ap->a_flags & VOP_FSYNC_SYSCALL)
 217                 hammer2_chain_flush(&trans, ip->chain);
 218         hammer2_inode_unlock_ex(ip);
 219         hammer2_trans_done(&trans);
 220         return (0);
 221 }
 222
 223 static
 224 int
 225 hammer2_vop_access(struct vop_access_args *ap)
 226 {
 227         hammer2_inode_t *ip = VTOI(ap->a_vp);
 228         hammer2_inode_data_t *ipdata;
 229         uid_t uid;
 230         gid_t gid;
 231         int error;
 232
 233         hammer2_inode_lock_sh(ip);
 234         ipdata = &ip->chain->data->ipdata;
 235         uid = hammer2_to_unix_xid(&ipdata->uid);
 236         gid = hammer2_to_unix_xid(&ipdata->gid);
 237         error = vop_helper_access(ap, uid, gid, ipdata->mode, ipdata->uflags);
 238         hammer2_inode_unlock_sh(ip);
 239
 240         return (error);
 241 }
 242
 243 static
 244 int
 245 hammer2_vop_getattr(struct vop_getattr_args *ap)
 246 {
 247         hammer2_inode_data_t *ipdata;
 248         hammer2_pfsmount_t *pmp;
 249         hammer2_inode_t *ip;
 250         struct vnode *vp;
 251         struct vattr *vap;
 252
 253         vp = ap->a_vp;
 254         vap = ap->a_vap;
 255
 256         ip = VTOI(vp);
 257         pmp = ip->pmp;
 258
 259         hammer2_inode_lock_sh(ip);
 260         ipdata = &ip->chain->data->ipdata;
 261
 262         vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
 263         vap->va_fileid = ipdata->inum;
 264         vap->va_mode = ipdata->mode;
 265         vap->va_nlink = ipdata->nlinks;
 266         vap->va_uid = hammer2_to_unix_xid(&ipdata->uid);
 267         vap->va_gid = hammer2_to_unix_xid(&ipdata->gid);
 268         vap->va_rmajor = 0;
 269         vap->va_rminor = 0;
 270         vap->va_size = ipdata->size;
 271         vap->va_blocksize = HAMMER2_PBUFSIZE;
 272         vap->va_flags = ipdata->uflags;
 273         hammer2_time_to_timespec(ipdata->ctime, &vap->va_ctime);
 274         hammer2_time_to_timespec(ipdata->mtime, &vap->va_mtime);
 275         hammer2_time_to_timespec(ipdata->mtime, &vap->va_atime);
 276         vap->va_gen = 1;
 277         vap->va_bytes = vap->va_size;   /* XXX */
 278         vap->va_type = hammer2_get_vtype(ip->chain);
 279         vap->va_filerev = 0;
 280         vap->va_uid_uuid = ipdata->uid;
 281         vap->va_gid_uuid = ipdata->gid;
 282         vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
 283                           VA_FSID_UUID_VALID;
 284
 285         hammer2_inode_unlock_sh(ip);
 286
 287         return (0);
 288 }
 289
 290 static
 291 int
 292 hammer2_vop_setattr(struct vop_setattr_args *ap)
 293 {
 294         hammer2_inode_data_t *ipdata;
 295         hammer2_inode_t *ip;
 296         hammer2_mount_t *hmp;
 297         hammer2_trans_t trans;
 298         struct vnode *vp;
 299         struct vattr *vap;
 300         int error;
 301         int kflags = 0;
 302         int domtime = 0;
 303         uint64_t ctime;
 304
 305         vp = ap->a_vp;
 306         vap = ap->a_vap;
 307         hammer2_update_time(&ctime);
 308
 309         ip = VTOI(vp);
 310         hmp = ip->hmp;
 311
 312         if (hmp->ronly)
 313                 return(EROFS);
 314
 315         hammer2_trans_init(&trans, hmp);
 316         hammer2_inode_lock_ex(ip);
 317         ipdata = &ip->chain->data->ipdata;
 318         error = 0;
 319
 320         if (vap->va_flags != VNOVAL) {
 321                 u_int32_t flags;
 322
 323                 flags = ipdata->uflags;
 324                 error = vop_helper_setattr_flags(&flags, vap->va_flags,
 325                                          hammer2_to_unix_xid(&ipdata->uid),
 326                                          ap->a_cred);
 327                 if (error == 0) {
 328                         if (ipdata->uflags != flags) {
 329                                 hammer2_chain_modify(&trans, ip->chain, 0);
 330                                 ipdata->uflags = flags;
 331                                 ipdata->ctime = ctime;
 332                                 kflags |= NOTE_ATTRIB;
 333                         }
 334                         if (ipdata->uflags & (IMMUTABLE | APPEND)) {
 335                                 error = 0;
 336                                 goto done;
 337                         }
 338                 }
 339                 goto done;
 340         }
 341         if (ipdata->uflags & (IMMUTABLE | APPEND)) {
 342                 error = EPERM;
 343                 goto done;
 344         }
 345         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
 346                 mode_t cur_mode = ipdata->mode;
 347                 uid_t cur_uid = hammer2_to_unix_xid(&ipdata->uid);
 348                 gid_t cur_gid = hammer2_to_unix_xid(&ipdata->gid);
 349                 uuid_t uuid_uid;
 350                 uuid_t uuid_gid;
 351
 352                 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
 353                                          ap->a_cred,
 354                                          &cur_uid, &cur_gid, &cur_mode);
 355                 if (error == 0) {
 356                         hammer2_guid_to_uuid(&uuid_uid, cur_uid);
 357                         hammer2_guid_to_uuid(&uuid_gid, cur_gid);
 358                         if (bcmp(&uuid_uid, &ipdata->uid, sizeof(uuid_uid)) ||
 359                             bcmp(&uuid_gid, &ipdata->gid, sizeof(uuid_gid)) ||
 360                             ipdata->mode != cur_mode
 361                         ) {
 362                                 hammer2_chain_modify(&trans, ip->chain, 0);
 363                                 ipdata->uid = uuid_uid;
 364                                 ipdata->gid = uuid_gid;
 365                                 ipdata->mode = cur_mode;
 366                                 ipdata->ctime = ctime;
 367                         }
 368                         kflags |= NOTE_ATTRIB;
 369                 }
 370         }
 371
 372         /*
 373          * Resize the file
 374          */
 375         if (vap->va_size != VNOVAL && ipdata->size != vap->va_size) {
 376                 switch(vp->v_type) {
 377                 case VREG:
 378                         if (vap->va_size == ipdata->size)
 379                                 break;
 380                         if (vap->va_size < ipdata->size) {
 381                                 hammer2_truncate_file(&trans, ip, vap->va_size);
 382                         } else {
 383                                 hammer2_extend_file(&trans, ip, vap->va_size);
 384                         }
 385                         domtime = 1;
 386                         break;
 387                 default:
 388                         error = EINVAL;
 389                         goto done;
 390                 }
 391         }
 392 #if 0
 393         /* atime not supported */
 394         if (vap->va_atime.tv_sec != VNOVAL) {
 395                 hammer2_chain_modify(&trans, ip->chain, 0);
 396                 ipdata->atime = hammer2_timespec_to_time(&vap->va_atime);
 397                 kflags |= NOTE_ATTRIB;
 398         }
 399 #endif
 400         if (vap->va_mtime.tv_sec != VNOVAL) {
 401                 hammer2_chain_modify(&trans, ip->chain, 0);
 402                 ipdata->mtime = hammer2_timespec_to_time(&vap->va_mtime);
 403                 kflags |= NOTE_ATTRIB;
 404         }
 405         if (vap->va_mode != (mode_t)VNOVAL) {
 406                 mode_t cur_mode = ipdata->mode;
 407                 uid_t cur_uid = hammer2_to_unix_xid(&ipdata->uid);
 408                 gid_t cur_gid = hammer2_to_unix_xid(&ipdata->gid);
 409
 410                 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
 411                                          cur_uid, cur_gid, &cur_mode);
 412                 if (error == 0 && ipdata->mode != cur_mode) {
 413                         hammer2_chain_modify(&trans, ip->chain, 0);
 414                         ipdata->mode = cur_mode;
 415                         ipdata->ctime = ctime;
 416                         kflags |= NOTE_ATTRIB;
 417                 }
 418         }
 419 done:
 420         hammer2_inode_unlock_ex(ip);
 421         hammer2_trans_done(&trans);
 422         return (error);
 423 }
 424
 425 static
 426 int
 427 hammer2_vop_readdir(struct vop_readdir_args *ap)
 428 {
 429         hammer2_inode_data_t *ipdata;
 430         hammer2_mount_t *hmp;
 431         hammer2_inode_t *ip;
 432         hammer2_inode_t *xip;
 433         hammer2_chain_t *parent;
 434         hammer2_chain_t *chain;
 435         hammer2_tid_t inum;
 436         hammer2_key_t lkey;
 437         struct uio *uio;
 438         off_t *cookies;
 439         off_t saveoff;
 440         int cookie_index;
 441         int ncookies;
 442         int error;
 443         int dtype;
 444         int r;
 445
 446         ip = VTOI(ap->a_vp);
 447         hmp = ip->hmp;
 448         uio = ap->a_uio;
 449         saveoff = uio->uio_offset;
 450
 451         /*
 452          * Setup cookies directory entry cookies if requested
 453          */
 454         if (ap->a_ncookies) {
 455                 ncookies = uio->uio_resid / 16 + 1;
 456                 if (ncookies > 1024)
 457                         ncookies = 1024;
 458                 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
 459         } else {
 460                 ncookies = -1;
 461                 cookies = NULL;
 462         }
 463         cookie_index = 0;
 464
 465         hammer2_inode_lock_sh(ip);
 466         ipdata = &ip->chain->data->ipdata;
 467
 468         /*
 469          * Handle artificial entries.  To ensure that only positive 64 bit
 470          * quantities are returned to userland we always strip off bit 63.
 471          * The hash code is designed such that codes 0x0000-0x7FFF are not
 472          * used, allowing us to use these codes for articial entries.
 473          *
 474          * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
 475          * allow '..' to cross the mount point into (e.g.) the super-root.
 476          */
 477         error = 0;
 478         chain = (void *)(intptr_t)-1;   /* non-NULL for early goto done case */
 479
 480         if (saveoff == 0) {
 481                 inum = ipdata->inum & HAMMER2_DIRHASH_USERMSK;
 482                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
 483                 if (r)
 484                         goto done;
 485                 if (cookies)
 486                         cookies[cookie_index] = saveoff;
 487                 ++saveoff;
 488                 ++cookie_index;
 489                 if (cookie_index == ncookies)
 490                         goto done;
 491         }
 492
 493         if (saveoff == 1) {
 494                 /*
 495                  * Be careful with lockorder when accessing ".."
 496                  *
 497                  * (ip is the current dir. xip is the parent dir).
 498                  */
 499                 inum = ipdata->inum & HAMMER2_DIRHASH_USERMSK;
 500                 while (ip->pip != NULL && ip != ip->pmp->iroot) {
 501                         xip = ip->pip;
 502                         hammer2_inode_ref(xip);
 503                         hammer2_inode_unlock_sh(ip);
 504                         hammer2_inode_lock_sh(xip);
 505                         hammer2_inode_lock_sh(ip);
 506                         hammer2_inode_drop(xip);
 507                         if (xip == ip->pip) {
 508                                 inum = xip->chain->data->ipdata.inum &
 509                                        HAMMER2_DIRHASH_USERMSK;
 510                                 hammer2_inode_unlock_sh(xip);
 511                                 break;
 512                         }
 513                         hammer2_inode_unlock_sh(xip);
 514                 }
 515                 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
 516                 if (r)
 517                         goto done;
 518                 if (cookies)
 519                         cookies[cookie_index] = saveoff;
 520                 ++saveoff;
 521                 ++cookie_index;
 522                 if (cookie_index == ncookies)
 523                         goto done;
 524         }
 525
 526         lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
 527
 528         /*
 529          * parent is the inode chain, already locked for us.  Don't
 530          * double lock shared locks as this will screw up upgrades.
 531          */
 532         if (error) {
 533                 goto done;
 534         }
 535         parent = hammer2_chain_lookup_init(ip->chain, HAMMER2_LOOKUP_SHARED);
 536         chain = hammer2_chain_lookup(&parent, lkey, lkey,
 537                                      HAMMER2_LOOKUP_SHARED);
 538         if (chain == NULL) {
 539                 chain = hammer2_chain_lookup(&parent,
 540                                              lkey, (hammer2_key_t)-1,
 541                                              HAMMER2_LOOKUP_SHARED);
 542         }
 543         while (chain) {
 544                 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
 545                         dtype = hammer2_get_dtype(chain);
 546                         saveoff = chain->bref.key & HAMMER2_DIRHASH_USERMSK;
 547                         r = vop_write_dirent(&error, uio,
 548                                              chain->data->ipdata.inum &
 549                                               HAMMER2_DIRHASH_USERMSK,
 550                                              dtype,
 551                                              chain->data->ipdata.name_len,
 552                                              chain->data->ipdata.filename);
 553                         if (r)
 554                                 break;
 555                         if (cookies)
 556                                 cookies[cookie_index] = saveoff;
 557                         ++cookie_index;
 558                 } else {
 559                         /* XXX chain error */
 560                         kprintf("bad chain type readdir %d\n",
 561                                 chain->bref.type);
 562                 }
 563
 564                 /*
 565                  * Keys may not be returned in order so once we have a
 566                  * placemarker (chain) the scan must allow the full range
 567                  * or some entries will be missed.
 568                  */
 569                 chain = hammer2_chain_next(&parent, chain,
 570                                            HAMMER2_DIRHASH_VISIBLE,
 571                                            (hammer2_key_t)-1,
 572                                            HAMMER2_LOOKUP_SHARED);
 573                 if (chain) {
 574                         saveoff = (chain->bref.key &
 575                                    HAMMER2_DIRHASH_USERMSK) + 1;
 576                 } else {
 577                         saveoff = (hammer2_key_t)-1;
 578                 }
 579                 if (cookie_index == ncookies)
 580                         break;
 581         }
 582         if (chain)
 583                 hammer2_chain_unlock(chain);
 584         hammer2_chain_lookup_done(parent);
 585 done:
 586         hammer2_inode_unlock_sh(ip);
 587         if (ap->a_eofflag)
 588                 *ap->a_eofflag = (chain == NULL);
 589         uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
 590         if (error && cookie_index == 0) {
 591                 if (cookies) {
 592                         kfree(cookies, M_TEMP);
 593                         *ap->a_ncookies = 0;
 594                         *ap->a_cookies = NULL;
 595                 }
 596         } else {
 597                 if (cookies) {
 598                         *ap->a_ncookies = cookie_index;
 599                         *ap->a_cookies = cookies;
 600                 }
 601         }
 602         return (error);
 603 }
 604
 605 /*
 606  * hammer2_vop_readlink { vp, uio, cred }
 607  */
 608 static
 609 int
 610 hammer2_vop_readlink(struct vop_readlink_args *ap)
 611 {
 612         struct vnode *vp;
 613         hammer2_mount_t *hmp;
 614         hammer2_inode_t *ip;
 615         int error;
 616
 617         vp = ap->a_vp;
 618         if (vp->v_type != VLNK)
 619                 return (EINVAL);
 620         ip = VTOI(vp);
 621         hmp = ip->hmp;
 622
 623         error = hammer2_read_file(ip, ap->a_uio, 0);
 624         return (error);
 625 }
 626
 627 static
 628 int
 629 hammer2_vop_read(struct vop_read_args *ap)
 630 {
 631         struct vnode *vp;
 632         hammer2_mount_t *hmp;
 633         hammer2_inode_t *ip;
 634         struct uio *uio;
 635         int error;
 636         int seqcount;
 637         int bigread;
 638
 639         /*
 640          * Read operations supported on this vnode?
 641          */
 642         vp = ap->a_vp;
 643         if (vp->v_type != VREG)
 644                 return (EINVAL);
 645
 646         /*
 647          * Misc
 648          */
 649         ip = VTOI(vp);
 650         hmp = ip->hmp;
 651         uio = ap->a_uio;
 652         error = 0;
 653
 654         seqcount = ap->a_ioflag >> 16;
 655         bigread = (uio->uio_resid > 100 * 1024 * 1024);
 656
 657         error = hammer2_read_file(ip, uio, seqcount);
 658         return (error);
 659 }
 660
 661 static
 662 int
 663 hammer2_vop_write(struct vop_write_args *ap)
 664 {
 665         hammer2_mount_t *hmp;
 666         hammer2_inode_t *ip;
 667         thread_t td;
 668         struct vnode *vp;
 669         struct uio *uio;
 670         int error;
 671         int seqcount;
 672         int bigwrite;
 673
 674         /*
 675          * Read operations supported on this vnode?
 676          */
 677         vp = ap->a_vp;
 678         if (vp->v_type != VREG)
 679                 return (EINVAL);
 680
 681         /*
 682          * Misc
 683          */
 684         ip = VTOI(vp);
 685         hmp = ip->hmp;
 686         uio = ap->a_uio;
 687         error = 0;
 688         if (hmp->ronly)
 689                 return (EROFS);
 690
 691         seqcount = ap->a_ioflag >> 16;
 692         bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
 693
 694         /*
 695          * Check resource limit
 696          */
 697         if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
 698             uio->uio_offset + uio->uio_resid >
 699              td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
 700                 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
 701                 return (EFBIG);
 702         }
 703
 704         bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
 705
 706         /*
 707          * ip must be locked if extending the file.
 708          * ip must be locked to avoid racing a truncation.
 709          *
 710          * ip must be marked modified, particularly because the write
 711          * might wind up being copied into the embedded data area.
 712          */
 713         hammer2_inode_lock_ex(ip);
 714         error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
 715         hammer2_inode_unlock_ex(ip);
 716         return (error);
 717 }
 718
 719 /*
 720  * Perform read operations on a file or symlink given an UNLOCKED
 721  * inode and uio.
 722  *
 723  * The passed ip is not locked.
 724  */
 725 static
 726 int
 727 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
 728 {
 729         hammer2_off_t size;
 730         struct buf *bp;
 731         int error;
 732
 733         error = 0;
 734
 735         /*
 736          * UIO read loop.
 737          */
 738         hammer2_inode_lock_sh(ip);
 739         size = ip->chain->data->ipdata.size;
 740
 741         while (uio->uio_resid > 0 && uio->uio_offset < size) {
 742                 hammer2_key_t lbase;
 743                 hammer2_key_t leof;
 744                 int lblksize;
 745                 int loff;
 746                 int n;
 747
 748                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 749                                                 &lbase, &leof);
 750
 751                 error = cluster_read(ip->vp, leof, lbase, lblksize,
 752                                      uio->uio_resid, seqcount * BKVASIZE,
 753                                      &bp);
 754
 755                 if (error)
 756                         break;
 757                 loff = (int)(uio->uio_offset - lbase);
 758                 n = lblksize - loff;
 759                 if (n > uio->uio_resid)
 760                         n = uio->uio_resid;
 761                 if (n > size - uio->uio_offset)
 762                         n = (int)(size - uio->uio_offset);
 763                 bp->b_flags |= B_AGE;
 764                 uiomove((char *)bp->b_data + loff, n, uio);
 765                 bqrelse(bp);
 766         }
 767         hammer2_inode_unlock_sh(ip);
 768         return (error);
 769 }
 770
 771 /*
 772  * Called with a locked (ip) to do the underlying write to a file or
 773  * to build the symlink target.
 774  */
 775 static
 776 int
 777 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
 778                    int ioflag, int seqcount)
 779 {
 780         hammer2_trans_t trans;
 781         hammer2_inode_data_t *ipdata;
 782         hammer2_key_t old_eof;
 783         struct buf *bp;
 784         int kflags;
 785         int error;
 786         int modified = 0;
 787
 788         /*
 789          * Setup if append
 790          */
 791         ipdata = &ip->chain->data->ipdata;
 792         if (ioflag & IO_APPEND)
 793                 uio->uio_offset = ipdata->size;
 794         kflags = 0;
 795         error = 0;
 796
 797         hammer2_trans_init(&trans, ip->hmp);
 798
 799         /*
 800          * Extend the file if necessary.  If the write fails at some point
 801          * we will truncate it back down to cover as much as we were able
 802          * to write.
 803          *
 804          * Doing this now makes it easier to calculate buffer sizes in
 805          * the loop.
 806          */
 807         old_eof = ipdata->size;
 808         if (uio->uio_offset + uio->uio_resid > ipdata->size) {
 809                 modified = 1;
 810                 hammer2_extend_file(&trans, ip,
 811                                     uio->uio_offset + uio->uio_resid);
 812                 kflags |= NOTE_EXTEND;
 813         }
 814
 815         /*
 816          * UIO write loop
 817          */
 818         while (uio->uio_resid > 0) {
 819                 hammer2_key_t lbase;
 820                 hammer2_key_t leof;
 821                 int trivial;
 822                 int lblksize;
 823                 int loff;
 824                 int n;
 825
 826                 /*
 827                  * Don't allow the buffer build to blow out the buffer
 828                  * cache.
 829                  */
 830                 if ((ioflag & IO_RECURSE) == 0) {
 831                         /*
 832                          * XXX should try to leave this unlocked through
 833                          *      the whole loop
 834                          */
 835                         hammer2_inode_unlock_ex(ip);
 836                         bwillwrite(HAMMER2_PBUFSIZE);
 837                         hammer2_inode_lock_ex(ip);
 838                         ipdata = &ip->chain->data->ipdata;      /* reload */
 839                 }
 840
 841                 /* XXX bigwrite & signal check test */
 842
 843                 /*
 844                  * This nominally tells us how much we can cluster and
 845                  * what the logical buffer size needs to be.  Currently
 846                  * we don't try to cluster the write and just handle one
 847                  * block at a time.
 848                  */
 849                 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
 850                                                 &lbase, &leof);
 851                 loff = (int)(uio->uio_offset - lbase);
 852
 853                 /*
 854                  * Calculate bytes to copy this transfer and whether the
 855                  * copy completely covers the buffer or not.
 856                  */
 857                 trivial = 0;
 858                 n = lblksize - loff;
 859                 if (n > uio->uio_resid) {
 860                         n = uio->uio_resid;
 861                         if (uio->uio_offset + n == ipdata->size)
 862                                 trivial = 1;
 863                 } else if (loff == 0) {
 864                         trivial = 1;
 865                 }
 866
 867                 /*
 868                  * Get the buffer
 869                  */
 870                 if (uio->uio_segflg == UIO_NOCOPY) {
 871                         /*
 872                          * Issuing a write with the same data backing the
 873                          * buffer.  Instantiate the buffer to collect the
 874                          * backing vm pages, then read-in any missing bits.
 875                          *
 876                          * This case is used by vop_stdputpages().
 877                          */
 878                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 879                         if ((bp->b_flags & B_CACHE) == 0) {
 880                                 bqrelse(bp);
 881                                 error = bread(ip->vp, lbase, lblksize, &bp);
 882                         }
 883                 } else if (trivial) {
 884                         /*
 885                          * Even though we are entirely overwriting the buffer
 886                          * we may still have to zero it out to avoid a
 887                          * mmap/write visibility issue.
 888                          */
 889                         bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
 890                         if ((bp->b_flags & B_CACHE) == 0)
 891                                 vfs_bio_clrbuf(bp);
 892                 } else {
 893                         /*
 894                          * Partial overwrite, read in any missing bits then
 895                          * replace the portion being written.
 896                          *
 897                          * (The strategy code will detect zero-fill physical
 898                          * blocks for this case).
 899                          */
 900                         error = bread(ip->vp, lbase, lblksize, &bp);
 901                         if (error == 0)
 902                                 bheavy(bp);
 903                 }
 904
 905                 if (error) {
 906                         brelse(bp);
 907                         break;
 908                 }
 909
 910                 /*
 911                  * We have to assign physical storage to the buffer we intend
 912                  * to dirty or write now to avoid deadlocks in the strategy
 913                  * code later.
 914                  *
 915                  * This can return NOOFFSET for inode-embedded data.  The
 916                  * strategy code will take care of it in that case.
 917                  */
 918                 bp->b_bio2.bio_offset =
 919                         hammer2_assign_physical(&trans, ip,
 920                                                 lbase, lblksize, &error);
 921                 if (error) {
 922                         brelse(bp);
 923                         break;
 924                 }
 925
 926                 /*
 927                  * Ok, copy the data in
 928                  */
 929                 hammer2_inode_unlock_ex(ip);
 930                 error = uiomove(bp->b_data + loff, n, uio);
 931                 hammer2_inode_lock_ex(ip);
 932                 ipdata = &ip->chain->data->ipdata;      /* reload */
 933                 kflags |= NOTE_WRITE;
 934                 modified = 1;
 935
 936                 if (error) {
 937                         brelse(bp);
 938                         break;
 939                 }
 940
 941                 /* XXX update ip_data.mtime */
 942
 943                 /*
 944                  * Once we dirty a buffer any cached offset becomes invalid.
 945                  *
 946                  * NOTE: For cluster_write() always use the trailing block
 947                  *       size, which is HAMMER2_PBUFSIZE.  lblksize is the
 948                  *       eof-straddling blocksize and is incorrect.
 949                  */
 950                 bp->b_flags |= B_AGE;
 951                 if (ioflag & IO_SYNC) {
 952                         bwrite(bp);
 953                 } else if ((ioflag & IO_DIRECT) && loff + n == lblksize) {
 954                         if (bp->b_bcount == HAMMER2_PBUFSIZE)
 955                                 bp->b_flags |= B_CLUSTEROK;
 956                         bdwrite(bp);
 957                 } else if (ioflag & IO_ASYNC) {
 958                         bawrite(bp);
 959                 } else if (hammer2_cluster_enable) {
 960                         if (bp->b_bcount == HAMMER2_PBUFSIZE)
 961                                 bp->b_flags |= B_CLUSTEROK;
 962                         cluster_write(bp, leof, HAMMER2_PBUFSIZE, seqcount);
 963                 } else {
 964                         if (bp->b_bcount == HAMMER2_PBUFSIZE)
 965                                 bp->b_flags |= B_CLUSTEROK;
 966                         bdwrite(bp);
 967                 }
 968         }
 969
 970         /*
 971          * Cleanup.  If we extended the file EOF but failed to write through
 972          * the entire write is a failure and we have to back-up.
 973          */
 974         if (error && ipdata->size != old_eof) {
 975                 hammer2_truncate_file(&trans, ip, old_eof);
 976         } else if (modified) {
 977                 hammer2_chain_modify(&trans, ip->chain, 0);
 978                 hammer2_update_time(&ipdata->mtime);
 979         }
 980         hammer2_knote(ip->vp, kflags);
 981         hammer2_trans_done(&trans);
 982         return error;
 983 }
 984
 985 /*
 986  * Assign physical storage to a logical block.
 987  *
 988  * NOOFFSET is returned if the data is inode-embedded.  In this case the
 989  * strategy code will simply bcopy() the data into the inode.
 990  *
 991  * The inode's delta_dcount is adjusted.
 992  */
 993 static
 994 hammer2_off_t
 995 hammer2_assign_physical(hammer2_trans_t *trans, hammer2_inode_t *ip,
 996                         hammer2_key_t lbase, int lblksize, int *errorp)
 997 {
 998         hammer2_mount_t *hmp;
 999         hammer2_chain_t *parent;
1000         hammer2_chain_t *chain;
1001         hammer2_off_t pbase;
1002
1003         /*
1004          * Locate the chain associated with lbase, return a locked chain.
1005          * However, do not instantiate any data reference (which utilizes a
1006          * device buffer) because we will be using direct IO via the
1007          * logical buffer cache buffer.
1008          */
1009         hmp = ip->hmp;
1010         *errorp = 0;
1011 retry:
1012         hammer2_inode_lock_ex(ip);
1013         parent = hammer2_chain_lookup_init(ip->chain, 0);
1014         chain = hammer2_chain_lookup(&parent,
1015                                      lbase, lbase,
1016                                      HAMMER2_LOOKUP_NODATA);
1017
1018         if (chain == NULL) {
1019                 /*
1020                  * We found a hole, create a new chain entry.
1021                  *
1022                  * NOTE: DATA chains are created without device backing
1023                  *       store (nor do we want any).
1024                  */
1025                 *errorp = hammer2_chain_create(trans, parent, &chain,
1026                                                lbase, HAMMER2_PBUFRADIX,
1027                                                HAMMER2_BREF_TYPE_DATA,
1028                                                lblksize);
1029                 if (chain == NULL) {
1030                         hammer2_inode_unlock_ex(ip);
1031                         hammer2_chain_lookup_done(parent);
1032                         panic("hammer2_chain_create: par=%p error=%d\n",
1033                                 parent, *errorp);
1034                         goto retry;
1035                 }
1036
1037                 pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
1038                 /*ip->delta_dcount += lblksize;*/
1039         } else {
1040                 switch (chain->bref.type) {
1041                 case HAMMER2_BREF_TYPE_INODE:
1042                         /*
1043                          * The data is embedded in the inode.  The
1044                          * caller is responsible for marking the inode
1045                          * modified and copying the data to the embedded
1046                          * area.
1047                          */
1048                         pbase = NOOFFSET;
1049                         break;
1050                 case HAMMER2_BREF_TYPE_DATA:
1051                         if (chain->bytes != lblksize) {
1052                                 panic("hammer2_assign_physical: "
1053                                       "size mismatch %d/%d\n",
1054                                       lblksize, chain->bytes);
1055                         }
1056                         hammer2_chain_modify(trans, chain,
1057                                              HAMMER2_MODIFY_OPTDATA);
1058                         pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
1059                         break;
1060                 default:
1061                         panic("hammer2_assign_physical: bad type");
1062                         /* NOT REACHED */
1063                         pbase = NOOFFSET;
1064                         break;
1065                 }
1066         }
1067         if (chain)
1068                 hammer2_chain_unlock(chain);
1069         hammer2_chain_lookup_done(parent);
1070
1071         hammer2_inode_unlock_ex(ip);
1072
1073         return (pbase);
1074 }
1075
1076 /*
1077  * Truncate the size of a file.
1078  *
1079  * This routine adjusts ipdata->size smaller, destroying any related
1080  * data beyond the new EOF and potentially resizing the block straddling
1081  * the EOF.
1082  *
1083  * The inode must be locked.
1084  */
1085 static
1086 void
1087 hammer2_truncate_file(hammer2_trans_t *trans,
1088                       hammer2_inode_t *ip, hammer2_key_t nsize)
1089 {
1090         hammer2_inode_data_t *ipdata;
1091         hammer2_chain_t *parent;
1092         hammer2_chain_t *chain;
1093         hammer2_key_t lbase;
1094         hammer2_key_t leof;
1095         struct buf *bp;
1096         int loff;
1097         int error;
1098         int oblksize;
1099         int nblksize;
1100
1101         hammer2_chain_modify(trans, ip->chain, 0);
1102         bp = NULL;
1103         ipdata = &ip->chain->data->ipdata;
1104         error = 0;
1105
1106         /*
1107          * Destroy any logical buffer cache buffers beyond the file EOF.
1108          *
1109          * We call nvtruncbuf() w/ trivial == 1 to prevent it from messing
1110          * around with the buffer straddling EOF, because we need to assign
1111          * a new physical offset to it.
1112          */
1113         if (ip->vp) {
1114                 nvtruncbuf(ip->vp, nsize,
1115                            HAMMER2_PBUFSIZE, (int)nsize & HAMMER2_PBUFMASK,
1116                            1);
1117         }
1118
1119         /*
1120          * Setup for lookup/search
1121          */
1122         parent = hammer2_chain_lookup_init(ip->chain, 0);
1123
1124         /*
1125          * Handle the case where a chain/logical-buffer straddles the new
1126          * EOF.  We told nvtruncbuf() above not to mess with the logical
1127          * buffer straddling the EOF because we need to reassign its storage
1128          * and can't let the strategy code do it for us.
1129          */
1130         loff = (int)nsize & HAMMER2_PBUFMASK;
1131         if (loff && ip->vp) {
1132                 oblksize = hammer2_calc_logical(ip, nsize, &lbase, &leof);
1133                 error = bread(ip->vp, lbase, oblksize, &bp);
1134                 KKASSERT(error == 0);
1135         }
1136         ipdata->size = nsize;
1137         nblksize = hammer2_calc_logical(ip, nsize, &lbase, &leof);
1138
1139         /*
1140          * Fixup the chain element.  If we have a logical buffer in-hand
1141          * we don't want to create a conflicting device buffer.
1142          */
1143         if (loff && bp) {
1144                 chain = hammer2_chain_lookup(&parent, lbase, lbase,
1145                                              HAMMER2_LOOKUP_NODATA);
1146                 if (chain) {
1147                         switch(chain->bref.type) {
1148                         case HAMMER2_BREF_TYPE_DATA:
1149                                 hammer2_chain_resize(trans, ip, bp,
1150                                              parent, &chain,
1151                                              hammer2_allocsize(nblksize),
1152                                              HAMMER2_MODIFY_OPTDATA);
1153                                 allocbuf(bp, nblksize);
1154                                 bzero(bp->b_data + loff, nblksize - loff);
1155                                 bp->b_bio2.bio_offset = chain->bref.data_off &
1156                                                         HAMMER2_OFF_MASK;
1157                                 break;
1158                         case HAMMER2_BREF_TYPE_INODE:
1159                                 allocbuf(bp, nblksize);
1160                                 bzero(bp->b_data + loff, nblksize - loff);
1161                                 bp->b_bio2.bio_offset = NOOFFSET;
1162                                 break;
1163                         default:
1164                                 panic("hammer2_truncate_file: bad type");
1165                                 break;
1166                         }
1167                         hammer2_chain_unlock(chain);
1168                         if (bp->b_bcount == HAMMER2_PBUFSIZE)
1169                                 bp->b_flags |= B_CLUSTEROK;
1170                         bdwrite(bp);
1171                 } else {
1172                         /*
1173                          * Destroy clean buffer w/ wrong buffer size.  Retain
1174                          * backing store.
1175                          */
1176                         bp->b_flags |= B_RELBUF;
1177                         KKASSERT(bp->b_bio2.bio_offset == NOOFFSET);
1178                         KKASSERT((bp->b_flags & B_DIRTY) == 0);
1179                         bqrelse(bp);
1180                 }
1181         } else if (loff) {
1182                 /*
1183                  * WARNING: This utilizes a device buffer for the data.
1184                  *
1185                  * This case should not occur because file truncations without
1186                  * a vnode (and hence no logical buffer cache) should only
1187                  * always truncate to 0-length.
1188                  */
1189                 panic("hammer2_truncate_file: non-zero truncation, no-vnode");
1190 #if 0
1191                 chain = hammer2_chain_lookup(&parent, lbase, lbase, 0);
1192                 if (chain) {
1193                         switch(chain->bref.type) {
1194                         case HAMMER2_BREF_TYPE_DATA:
1195                                 chain = hammer2_chain_resize(trans, ip, bp,
1196                                              parent, chain,
1197                                              hammer2_allocsize(nblksize),
1198                                              0);
1199                                 hammer2_chain_modify(hmp, chain, 0);
1200                                 bzero(chain->data->buf + loff, nblksize - loff);
1201                                 break;
1202                         case HAMMER2_BREF_TYPE_INODE:
1203                                 if (loff < HAMMER2_EMBEDDED_BYTES) {
1204                                         hammer2_chain_modify(hmp, chain, 0);
1205                                         bzero(chain->data->ipdata.u.data + loff,
1206                                               HAMMER2_EMBEDDED_BYTES - loff);
1207                                 }
1208                                 break;
1209                         }
1210                         hammer2_chain_unlock(chain);
1211                 }
1212 #endif
1213         }
1214
1215         /*
1216          * Clean up any fragmentory VM pages now that we have properly
1217          * resized the straddling buffer.  These pages are no longer
1218          * part of the buffer.
1219          */
1220         if (ip->vp) {
1221                 nvtruncbuf(ip->vp, nsize,
1222                            nblksize, (int)nsize & (nblksize - 1),
1223                            1);
1224         }
1225
1226         /*
1227          * Destroy any physical blocks after the new EOF point.
1228          */
1229         lbase = (nsize + HAMMER2_PBUFMASK64) & ~HAMMER2_PBUFMASK64;
1230         chain = hammer2_chain_lookup(&parent,
1231                                      lbase, (hammer2_key_t)-1,
1232                                      HAMMER2_LOOKUP_NODATA);
1233         while (chain) {
1234                 /*
1235                  * Degenerate embedded data case, nothing to loop on.
1236                  */
1237                 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
1238                         hammer2_chain_unlock(chain);
1239                         break;
1240                 }
1241
1242                 /*
1243                  * Delete physical data blocks past the file EOF.
1244                  */
1245                 if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
1246                         /*ip->delta_dcount -= chain->bytes;*/
1247                         hammer2_chain_delete(trans, parent, chain);
1248                 }
1249                 /* XXX check parent if empty indirect block & delete */
1250                 chain = hammer2_chain_next(&parent, chain,
1251                                            lbase, (hammer2_key_t)-1,
1252                                            HAMMER2_LOOKUP_NODATA);
1253         }
1254         hammer2_chain_lookup_done(parent);
1255 }
1256
1257 /*
1258  * Extend the size of a file.  The inode must be locked.
1259  *
1260  * We may have to resize the block straddling the old EOF.
1261  */
1262 static
1263 void
1264 hammer2_extend_file(hammer2_trans_t *trans,
1265                     hammer2_inode_t *ip, hammer2_key_t nsize)
1266 {
1267         hammer2_inode_data_t *ipdata;
1268         hammer2_mount_t *hmp;
1269         hammer2_chain_t *parent;
1270         hammer2_chain_t *chain;
1271         struct buf *bp;
1272         hammer2_key_t osize;
1273         hammer2_key_t obase;
1274         hammer2_key_t nbase;
1275         hammer2_key_t leof;
1276         int oblksize;
1277         int nblksize;
1278         int nradix;
1279         int error;
1280
1281         KKASSERT(ip->vp);
1282         hmp = ip->hmp;
1283
1284         hammer2_chain_modify(trans, ip->chain, 0);
1285         ipdata = &ip->chain->data->ipdata;
1286
1287         /*
1288          * Nothing to do if the direct-data case is still intact
1289          */
1290         if ((ipdata->op_flags & HAMMER2_OPFLAG_DIRECTDATA) &&
1291             nsize <= HAMMER2_EMBEDDED_BYTES) {
1292                 ipdata->size = nsize;
1293                 nvextendbuf(ip->vp,
1294                             ipdata->size, nsize,
1295                             0, HAMMER2_EMBEDDED_BYTES,
1296                             0, (int)nsize,
1297                             1);
1298                 return;
1299         }
1300
1301         /*
1302          * Calculate the blocksize at the original EOF and resize the block
1303          * if necessary.  Adjust the file size in the inode.
1304          */
1305         osize = ipdata->size;
1306         oblksize = hammer2_calc_logical(ip, osize, &obase, &leof);
1307         ipdata->size = nsize;
1308         nblksize = hammer2_calc_logical(ip, osize, &nbase, &leof);
1309
1310         /*
1311          * Do all required vnode operations, but do not mess with the
1312          * buffer straddling the orignal EOF.
1313          */
1314         nvextendbuf(ip->vp,
1315                     ipdata->size, nsize,
1316                     0, nblksize,
1317                     0, (int)nsize & HAMMER2_PBUFMASK,
1318                     1);
1319
1320         /*
1321          * Early return if we have no more work to do.
1322          */
1323         if (obase == nbase && oblksize == nblksize &&
1324             (ipdata->op_flags & HAMMER2_OPFLAG_DIRECTDATA) == 0) {
1325                 return;
1326         }
1327
1328         /*
1329          * We have work to do, including possibly resizing the buffer
1330          * at the previous EOF point and turning off DIRECTDATA mode.
1331          */
1332         bp = NULL;
1333         if (((int)osize & HAMMER2_PBUFMASK)) {
1334                 error = bread(ip->vp, obase, oblksize, &bp);
1335                 KKASSERT(error == 0);
1336         }
1337
1338         /*
1339          * Disable direct-data mode by loading up a buffer cache buffer
1340          * with the data, then converting the inode data area into the
1341          * inode indirect block array area.
1342          */
1343         if (ipdata->op_flags & HAMMER2_OPFLAG_DIRECTDATA) {
1344                 ipdata->op_flags &= ~HAMMER2_OPFLAG_DIRECTDATA;
1345                 bzero(&ipdata->u.blockset, sizeof(ipdata->u.blockset));
1346         }
1347
1348         /*
1349          * Resize the chain element at the old EOF.
1350          */
1351         if (((int)osize & HAMMER2_PBUFMASK)) {
1352 retry:
1353                 error = 0;
1354                 parent = hammer2_chain_lookup_init(ip->chain, 0);
1355                 nradix = hammer2_allocsize(nblksize);
1356
1357                 chain = hammer2_chain_lookup(&parent,
1358                                              obase, obase,
1359                                              HAMMER2_LOOKUP_NODATA);
1360                 if (chain == NULL) {
1361                         error = hammer2_chain_create(trans, parent, &chain,
1362                                                      obase, nblksize,
1363                                                      HAMMER2_BREF_TYPE_DATA,
1364                                                      nblksize);
1365                         if (chain == NULL) {
1366                                 hammer2_chain_lookup_done(parent);
1367                                 panic("hammer2_chain_create: par=%p error=%d\n",
1368                                         parent, error);
1369                                 goto retry;
1370                         }
1371                         /*ip->delta_dcount += nblksize;*/
1372                 } else {
1373                         KKASSERT(chain->bref.type == HAMMER2_BREF_TYPE_DATA);
1374                         hammer2_chain_resize(trans, ip, bp,
1375                                              parent, &chain,
1376                                              nradix,
1377                                              HAMMER2_MODIFY_OPTDATA);
1378                 }
1379                 if (obase != nbase) {
1380                         if (oblksize != HAMMER2_PBUFSIZE)
1381                                 allocbuf(bp, HAMMER2_PBUFSIZE);
1382                 } else {
1383                         if (oblksize != nblksize)
1384                                 allocbuf(bp, nblksize);
1385                 }
1386                 bp->b_bio2.bio_offset = chain->bref.data_off &
1387                                         HAMMER2_OFF_MASK;
1388                 hammer2_chain_unlock(chain);
1389                 if (bp->b_bcount == HAMMER2_PBUFSIZE)
1390                         bp->b_flags |= B_CLUSTEROK;
1391                 bdwrite(bp);
1392                 hammer2_chain_lookup_done(parent);  /* must be after bdwrite */
1393         }
1394 }
1395
1396 static
1397 int
1398 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1399 {
1400         hammer2_inode_t *ip;
1401         hammer2_inode_t *dip;
1402         hammer2_mount_t *hmp;
1403         hammer2_chain_t *parent;
1404         hammer2_chain_t *chain;
1405         hammer2_chain_t *ochain;
1406         hammer2_trans_t trans;
1407         struct namecache *ncp;
1408         const uint8_t *name;
1409         size_t name_len;
1410         hammer2_key_t lhc;
1411         int error = 0;
1412         struct vnode *vp;
1413
1414         dip = VTOI(ap->a_dvp);
1415         hmp = dip->hmp;
1416         ncp = ap->a_nch->ncp;
1417         name = ncp->nc_name;
1418         name_len = ncp->nc_nlen;
1419         lhc = hammer2_dirhash(name, name_len);
1420
1421         /*
1422          * Note: In DragonFly the kernel handles '.' and '..'.
1423          */
1424         hammer2_inode_lock_sh(dip);
1425         parent = hammer2_chain_lookup_init(dip->chain, HAMMER2_LOOKUP_SHARED);
1426         chain = hammer2_chain_lookup(&parent,
1427                                      lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1428                                      HAMMER2_LOOKUP_SHARED);
1429         while (chain) {
1430                 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
1431                     name_len == chain->data->ipdata.name_len &&
1432                     bcmp(name, chain->data->ipdata.filename, name_len) == 0) {
1433                         break;
1434                 }
1435                 chain = hammer2_chain_next(&parent, chain,
1436                                            lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1437                                            HAMMER2_LOOKUP_SHARED);
1438         }
1439         hammer2_chain_lookup_done(parent);
1440         hammer2_inode_unlock_sh(dip);
1441
1442         /*
1443          * If the inode represents a forwarding entry for a hardlink we have
1444          * to locate the actual inode.  The original ip is saved for possible
1445          * deconsolidation.  (ip) will only be set to non-NULL when we have
1446          * to locate the real file via a hardlink.  ip will be referenced but
1447          * not locked in that situation.  chain is passed in locked and
1448          * returned locked.
1449          *
1450          * XXX what kind of chain lock?
1451          */
1452         ochain = NULL;
1453         if (chain && chain->data->ipdata.type == HAMMER2_OBJTYPE_HARDLINK) {
1454                 error = hammer2_hardlink_find(dip, &chain, &ochain);
1455                 if (error) {
1456                         kprintf("hammer2: unable to find hardlink\n");
1457                         if (chain) {
1458                                 hammer2_chain_unlock(chain);
1459                                 chain = NULL;
1460                         }
1461                         goto failed;
1462                 }
1463         }
1464
1465         /*
1466          * Deconsolidate any hardlink whos nlinks == 1.  Ignore errors.
1467          * If an error occurs chain and ip are left alone.
1468          *
1469          * XXX upgrade shared lock?
1470          */
1471         if (ochain && chain && chain->data->ipdata.nlinks == 1 && !hmp->ronly) {
1472                 kprintf("hammer2: need to unconsolidate hardlink for %s\n",
1473                         chain->data->ipdata.filename);
1474                 /* XXX retain shared lock on dip? (currently not held) */
1475                 hammer2_trans_init(&trans, dip->hmp);
1476                 hammer2_hardlink_deconsolidate(&trans, dip, &chain, &ochain);
1477                 hammer2_trans_done(&trans);
1478         }
1479
1480         /*
1481          * Acquire the related vnode
1482          *
1483          * NOTE: For error processing, only ENOENT resolves the namecache
1484          *       entry to NULL, otherwise we just return the error and
1485          *       leave the namecache unresolved.
1486          *
1487          * NOTE: multiple hammer2_inode structures can be aliased to the
1488          *       same chain element, for example for hardlinks.  This
1489          *       use case does not 'reattach' inode associations that
1490          *       might already exist, but always allocates a new one.
1491          *
1492          * WARNING: inode structure is locked exclusively via inode_get
1493          *          but chain was locked shared.  inode_unlock_ex()
1494          *          will handle it properly.
1495          */
1496         if (chain) {
1497                 ip = hammer2_inode_get(hmp, dip->pmp, dip, chain);
1498                 vp = hammer2_igetv(ip, &error);
1499                 if (error == 0) {
1500                         vn_unlock(vp);
1501                         cache_setvp(ap->a_nch, vp);
1502                 } else if (error == ENOENT) {
1503                         cache_setvp(ap->a_nch, NULL);
1504                 }
1505                 hammer2_inode_unlock_ex(ip);
1506
1507                 /*
1508                  * The vp should not be released until after we've disposed
1509                  * of our locks, because it might cause vop_inactive() to
1510                  * be called.
1511                  */
1512                 if (vp)
1513                         vrele(vp);
1514         } else {
1515                 error = ENOENT;
1516                 cache_setvp(ap->a_nch, NULL);
1517         }
1518 failed:
1519         KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1520                 ("resolve error %d/%p chain %p ap %p\n",
1521                  error, ap->a_nch->ncp->nc_vp, chain, ap));
1522         if (ochain)
1523                 hammer2_chain_drop(ochain);
1524         return error;
1525 }
1526
1527 static
1528 int
1529 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1530 {
1531         hammer2_inode_t *dip;
1532         hammer2_inode_t *ip;
1533         hammer2_mount_t *hmp;
1534         int error;
1535
1536         dip = VTOI(ap->a_dvp);
1537         hmp = dip->hmp;
1538
1539         if ((ip = dip->pip) == NULL) {
1540                 *ap->a_vpp = NULL;
1541                 return ENOENT;
1542         }
1543         hammer2_inode_lock_ex(ip);
1544         *ap->a_vpp = hammer2_igetv(ip, &error);
1545         hammer2_inode_unlock_ex(ip);
1546
1547         return error;
1548 }
1549
1550 static
1551 int
1552 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1553 {
1554         hammer2_mount_t *hmp;
1555         hammer2_inode_t *dip;
1556         hammer2_inode_t *nip;
1557         hammer2_trans_t trans;
1558         struct namecache *ncp;
1559         const uint8_t *name;
1560         size_t name_len;
1561         int error;
1562
1563         dip = VTOI(ap->a_dvp);
1564         hmp = dip->hmp;
1565         if (hmp->ronly)
1566                 return (EROFS);
1567
1568         ncp = ap->a_nch->ncp;
1569         name = ncp->nc_name;
1570         name_len = ncp->nc_nlen;
1571
1572         hammer2_trans_init(&trans, hmp);
1573         nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1574                                    name, name_len, &error);
1575         if (error) {
1576                 KKASSERT(nip == NULL);
1577                 *ap->a_vpp = NULL;
1578         } else {
1579                 *ap->a_vpp = hammer2_igetv(nip, &error);
1580                 hammer2_inode_unlock_ex(nip);
1581         }
1582         hammer2_trans_done(&trans);
1583
1584         if (error == 0) {
1585                 cache_setunresolved(ap->a_nch);
1586                 cache_setvp(ap->a_nch, *ap->a_vpp);
1587         }
1588         return error;
1589 }
1590
1591 /*
1592  * Return the largest contiguous physical disk range for the logical
1593  * request.
1594  *
1595  * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
1596  */
1597 static
1598 int
1599 hammer2_vop_bmap(struct vop_bmap_args *ap)
1600 {
1601         struct vnode *vp;
1602         hammer2_mount_t *hmp;
1603         hammer2_inode_t *ip;
1604         hammer2_chain_t *parent;
1605         hammer2_chain_t *chain;
1606         hammer2_key_t lbeg;
1607         hammer2_key_t lend;
1608         hammer2_off_t pbeg;
1609         hammer2_off_t pbytes;
1610         hammer2_off_t array[HAMMER2_BMAP_COUNT][2];
1611         int loff;
1612         int ai;
1613
1614         /*
1615          * Only supported on regular files
1616          *
1617          * Only supported for read operations (required for cluster_read).
1618          * The block allocation is delayed for write operations.
1619          */
1620         vp = ap->a_vp;
1621         if (vp->v_type != VREG)
1622                 return (EOPNOTSUPP);
1623         if (ap->a_cmd != BUF_CMD_READ)
1624                 return (EOPNOTSUPP);
1625
1626         ip = VTOI(vp);
1627         hmp = ip->hmp;
1628         bzero(array, sizeof(array));
1629
1630         /*
1631          * Calculate logical range
1632          */
1633         KKASSERT((ap->a_loffset & HAMMER2_LBUFMASK64) == 0);
1634         lbeg = ap->a_loffset & HAMMER2_OFF_MASK_HI;
1635         lend = lbeg + HAMMER2_BMAP_COUNT * HAMMER2_PBUFSIZE - 1;
1636         if (lend < lbeg)
1637                 lend = lbeg;
1638         loff = ap->a_loffset & HAMMER2_OFF_MASK_LO;
1639
1640         hammer2_inode_lock_sh(ip);
1641         parent = hammer2_chain_lookup_init(ip->chain, HAMMER2_LOOKUP_SHARED);
1642         chain = hammer2_chain_lookup(&parent,
1643                                      lbeg, lend,
1644                                      HAMMER2_LOOKUP_NODATA |
1645                                      HAMMER2_LOOKUP_SHARED);
1646         if (chain == NULL) {
1647                 *ap->a_doffsetp = ZFOFFSET;
1648                 hammer2_chain_lookup_done(parent);
1649                 hammer2_inode_unlock_sh(ip);
1650                 return (0);
1651         }
1652
1653         while (chain) {
1654                 if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
1655                         ai = (chain->bref.key - lbeg) / HAMMER2_PBUFSIZE;
1656                         KKASSERT(ai >= 0 && ai < HAMMER2_BMAP_COUNT);
1657                         array[ai][0] = chain->bref.data_off & HAMMER2_OFF_MASK;
1658                         array[ai][1] = chain->bytes;
1659                 }
1660                 chain = hammer2_chain_next(&parent, chain,
1661                                            lbeg, lend,
1662                                            HAMMER2_LOOKUP_NODATA |
1663                                            HAMMER2_LOOKUP_SHARED);
1664         }
1665         hammer2_chain_lookup_done(parent);
1666         hammer2_inode_unlock_sh(ip);
1667
1668         /*
1669          * If the requested loffset is not mappable physically we can't
1670          * bmap.  The caller will have to access the file data via a
1671          * device buffer.
1672          */
1673         if (array[0][0] == 0 || array[0][1] < loff + HAMMER2_LBUFSIZE) {
1674                 *ap->a_doffsetp = NOOFFSET;
1675                 return (0);
1676         }
1677
1678         /*
1679          * Calculate the physical disk offset range for array[0]
1680          */
1681         pbeg = array[0][0] + loff;
1682         pbytes = array[0][1] - loff;
1683
1684         for (ai = 1; ai < HAMMER2_BMAP_COUNT; ++ai) {
1685                 if (array[ai][0] != pbeg + pbytes)
1686                         break;
1687                 pbytes += array[ai][1];
1688         }
1689
1690         *ap->a_doffsetp = pbeg;
1691         if (ap->a_runp)
1692                 *ap->a_runp = pbytes;
1693         return (0);
1694 }
1695
1696 static
1697 int
1698 hammer2_vop_open(struct vop_open_args *ap)
1699 {
1700         return vop_stdopen(ap);
1701 }
1702
1703 /*
1704  * hammer2_vop_advlock { vp, id, op, fl, flags }
1705  */
1706 static
1707 int
1708 hammer2_vop_advlock(struct vop_advlock_args *ap)
1709 {
1710         hammer2_inode_t *ip = VTOI(ap->a_vp);
1711         hammer2_off_t size;
1712
1713         hammer2_inode_lock_sh(ip);
1714         size = ip->chain->data->ipdata.size;
1715         hammer2_inode_unlock_sh(ip);
1716         return (lf_advlock(ap, &ip->advlock, size));
1717 }
1718
1719
1720 static
1721 int
1722 hammer2_vop_close(struct vop_close_args *ap)
1723 {
1724         return vop_stdclose(ap);
1725 }
1726
1727 /*
1728  * hammer2_vop_nlink { nch, dvp, vp, cred }
1729  *
1730  * Create a hardlink from (vp) to {dvp, nch}.
1731  */
1732 static
1733 int
1734 hammer2_vop_nlink(struct vop_nlink_args *ap)
1735 {
1736         hammer2_inode_t *dip;   /* target directory to create link in */
1737         hammer2_inode_t *ip;    /* inode we are hardlinking to */
1738         hammer2_mount_t *hmp;
1739         hammer2_chain_t *chain;
1740         hammer2_trans_t trans;
1741         struct namecache *ncp;
1742         const uint8_t *name;
1743         size_t name_len;
1744         int error;
1745
1746         dip = VTOI(ap->a_dvp);
1747         hmp = dip->hmp;
1748         if (hmp->ronly)
1749                 return (EROFS);
1750
1751         ncp = ap->a_nch->ncp;
1752         name = ncp->nc_name;
1753         name_len = ncp->nc_nlen;
1754         hammer2_trans_init(&trans, hmp);
1755
1756         /*
1757          * ip represents the file being hardlinked.  The file could be a
1758          * normal file or a hardlink target if it has already been hardlinked.
1759          * If ip is a hardlinked target then ip->pip represents the location
1760          * of the hardlinked target, NOT the location of the hardlink pointer.
1761          *
1762          * Bump nlinks and potentially also create or move the hardlink
1763          * target in the parent directory common to (ip) and (dip).  The
1764          * consolidation code can modify ip->chain and ip->pip.  The
1765          * returned chain is locked.
1766          */
1767         ip = VTOI(ap->a_vp);
1768         hammer2_inode_ref(ip);
1769         error = hammer2_hardlink_consolidate(&trans, ip, &chain, dip, 1);
1770         if (error)
1771                 goto done;
1772
1773         /*
1774          * Create a directory entry connected to the specified chain.
1775          * This function unlocks and NULL's chain on return.
1776          */
1777         error = hammer2_inode_connect(&trans, dip, ip, &chain, name, name_len);
1778         if (chain) {
1779                 hammer2_chain_unlock(chain);
1780                 chain = NULL;
1781         }
1782         if (error == 0) {
1783                 cache_setunresolved(ap->a_nch);
1784                 cache_setvp(ap->a_nch, ap->a_vp);
1785         }
1786 done:
1787         hammer2_inode_drop(ip);
1788         hammer2_trans_done(&trans);
1789
1790         return error;
1791 }
1792
1793 /*
1794  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1795  *
1796  * The operating system has already ensured that the directory entry
1797  * does not exist and done all appropriate namespace locking.
1798  */
1799 static
1800 int
1801 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1802 {
1803         hammer2_mount_t *hmp;
1804         hammer2_inode_t *dip;
1805         hammer2_inode_t *nip;
1806         hammer2_trans_t trans;
1807         struct namecache *ncp;
1808         const uint8_t *name;
1809         size_t name_len;
1810         int error;
1811
1812         dip = VTOI(ap->a_dvp);
1813         hmp = dip->hmp;
1814         if (hmp->ronly)
1815                 return (EROFS);
1816
1817         ncp = ap->a_nch->ncp;
1818         name = ncp->nc_name;
1819         name_len = ncp->nc_nlen;
1820         hammer2_trans_init(&trans, hmp);
1821
1822         nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1823                                    name, name_len, &error);
1824         if (error) {
1825                 KKASSERT(nip == NULL);
1826                 *ap->a_vpp = NULL;
1827         } else {
1828                 *ap->a_vpp = hammer2_igetv(nip, &error);
1829                 hammer2_inode_unlock_ex(nip);
1830         }
1831         hammer2_trans_done(&trans);
1832
1833         if (error == 0) {
1834                 cache_setunresolved(ap->a_nch);
1835                 cache_setvp(ap->a_nch, *ap->a_vpp);
1836         }
1837         return error;
1838 }
1839
1840 /*
1841  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1842  */
1843 static
1844 int
1845 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1846 {
1847         hammer2_mount_t *hmp;
1848         hammer2_inode_t *dip;
1849         hammer2_inode_t *nip;
1850         hammer2_trans_t trans;
1851         struct namecache *ncp;
1852         const uint8_t *name;
1853         size_t name_len;
1854         int error;
1855
1856         dip = VTOI(ap->a_dvp);
1857         hmp = dip->hmp;
1858         if (hmp->ronly)
1859                 return (EROFS);
1860
1861         ncp = ap->a_nch->ncp;
1862         name = ncp->nc_name;
1863         name_len = ncp->nc_nlen;
1864         hammer2_trans_init(&trans, hmp);
1865
1866         ap->a_vap->va_type = VLNK;      /* enforce type */
1867
1868         nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1869                                    name, name_len, &error);
1870         if (error) {
1871                 KKASSERT(nip == NULL);
1872                 *ap->a_vpp = NULL;
1873                 hammer2_trans_done(&trans);
1874                 return error;
1875         }
1876         *ap->a_vpp = hammer2_igetv(nip, &error);
1877
1878         /*
1879          * Build the softlink (~like file data) and finalize the namecache.
1880          */
1881         if (error == 0) {
1882                 size_t bytes;
1883                 struct uio auio;
1884                 struct iovec aiov;
1885                 hammer2_inode_data_t *nipdata;
1886
1887                 nipdata = &nip->chain->data->ipdata;
1888                 bytes = strlen(ap->a_target);
1889
1890                 if (bytes <= HAMMER2_EMBEDDED_BYTES) {
1891                         KKASSERT(nipdata->op_flags &
1892                                  HAMMER2_OPFLAG_DIRECTDATA);
1893                         bcopy(ap->a_target, nipdata->u.data, bytes);
1894                         nipdata->size = bytes;
1895                 } else {
1896                         bzero(&auio, sizeof(auio));
1897                         bzero(&aiov, sizeof(aiov));
1898                         auio.uio_iov = &aiov;
1899                         auio.uio_segflg = UIO_SYSSPACE;
1900                         auio.uio_rw = UIO_WRITE;
1901                         auio.uio_resid = bytes;
1902                         auio.uio_iovcnt = 1;
1903                         auio.uio_td = curthread;
1904                         aiov.iov_base = ap->a_target;
1905                         aiov.iov_len = bytes;
1906                         error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1907                         /* XXX handle error */
1908                         error = 0;
1909                 }
1910         }
1911         hammer2_inode_unlock_ex(nip);
1912         hammer2_trans_done(&trans);
1913
1914         /*
1915          * Finalize namecache
1916          */
1917         if (error == 0) {
1918                 cache_setunresolved(ap->a_nch);
1919                 cache_setvp(ap->a_nch, *ap->a_vpp);
1920                 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1921         }
1922         return error;
1923 }
1924
1925 /*
1926  * hammer2_vop_nremove { nch, dvp, cred }
1927  */
1928 static
1929 int
1930 hammer2_vop_nremove(struct vop_nremove_args *ap)
1931 {
1932         hammer2_inode_t *dip;
1933         hammer2_mount_t *hmp;
1934         hammer2_trans_t trans;
1935         struct namecache *ncp;
1936         const uint8_t *name;
1937         size_t name_len;
1938         int error;
1939
1940         dip = VTOI(ap->a_dvp);
1941         hmp = dip->hmp;
1942         if (hmp->ronly)
1943                 return(EROFS);
1944
1945         ncp = ap->a_nch->ncp;
1946         name = ncp->nc_name;
1947         name_len = ncp->nc_nlen;
1948         hammer2_trans_init(&trans, hmp);
1949         error = hammer2_unlink_file(&trans, dip, name, name_len, 0);
1950         hammer2_trans_done(&trans);
1951         if (error == 0) {
1952                 cache_unlink(ap->a_nch);
1953         }
1954         return (error);
1955 }
1956
1957 /*
1958  * hammer2_vop_nrmdir { nch, dvp, cred }
1959  */
1960 static
1961 int
1962 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1963 {
1964         hammer2_inode_t *dip;
1965         hammer2_mount_t *hmp;
1966         hammer2_trans_t trans;
1967         struct namecache *ncp;
1968         const uint8_t *name;
1969         size_t name_len;
1970         int error;
1971
1972         dip = VTOI(ap->a_dvp);
1973         hmp = dip->hmp;
1974         if (hmp->ronly)
1975                 return(EROFS);
1976
1977         ncp = ap->a_nch->ncp;
1978         name = ncp->nc_name;
1979         name_len = ncp->nc_nlen;
1980
1981         hammer2_trans_init(&trans, hmp);
1982         error = hammer2_unlink_file(&trans, dip, name, name_len, 1);
1983         hammer2_trans_done(&trans);
1984         if (error == 0) {
1985                 cache_unlink(ap->a_nch);
1986         }
1987         return (error);
1988 }
1989
1990 /*
1991  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1992  */
1993 static
1994 int
1995 hammer2_vop_nrename(struct vop_nrename_args *ap)
1996 {
1997         struct namecache *fncp;
1998         struct namecache *tncp;
1999         hammer2_inode_t *fdip;
2000         hammer2_inode_t *tdip;
2001         hammer2_inode_t *ip;
2002         hammer2_chain_t *chain;
2003         hammer2_mount_t *hmp;
2004         hammer2_trans_t trans;
2005         const uint8_t *fname;
2006         size_t fname_len;
2007         const uint8_t *tname;
2008         size_t tname_len;
2009         int error;
2010
2011         if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
2012                 return(EXDEV);
2013         if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
2014                 return(EXDEV);
2015
2016         fdip = VTOI(ap->a_fdvp);        /* source directory */
2017         tdip = VTOI(ap->a_tdvp);        /* target directory */
2018
2019         hmp = fdip->hmp;                /* check read-only filesystem */
2020         if (hmp->ronly)
2021                 return(EROFS);
2022
2023         fncp = ap->a_fnch->ncp;         /* entry name in source */
2024         fname = fncp->nc_name;
2025         fname_len = fncp->nc_nlen;
2026
2027         tncp = ap->a_tnch->ncp;         /* entry name in target */
2028         tname = tncp->nc_name;
2029         tname_len = tncp->nc_nlen;
2030
2031         hammer2_trans_init(&trans, hmp);
2032
2033         /*
2034          * ip is the inode being removed.  If this is a hardlink then
2035          * ip represents the actual file and not the hardlink marker.
2036          */
2037         ip = VTOI(fncp->nc_vp);
2038         chain = NULL;
2039
2040         /*
2041          * Keep a tight grip on the inode so the temporary unlinking from
2042          * the source location prior to linking to the target location
2043          * does not cause the chain to be destroyed.
2044          *
2045          * NOTE: To avoid deadlocks we cannot lock (ip) while we are
2046          *       unlinking elements from their directories.  Locking
2047          *       the nlinks field does not lock the whole inode.
2048          */
2049         hammer2_inode_ref(ip);
2050
2051         /*
2052          * Remove target if it exists
2053          */
2054         error = hammer2_unlink_file(&trans, tdip, tname, tname_len, -1);
2055         if (error && error != ENOENT)
2056                 goto done;
2057         cache_setunresolved(ap->a_tnch);
2058
2059         /*
2060          * When renaming a hardlinked file we may have to re-consolidate
2061          * the location of the hardlink target.  Since the element is simply
2062          * being moved, nlinks is not modified in this case.
2063          *
2064          * If ip represents a regular file the consolidation code essentially
2065          * does nothing other than return the locked chain.
2066          *
2067          * The returned chain will be locked.
2068          */
2069         error = hammer2_hardlink_consolidate(&trans, ip, &chain, tdip, 0);
2070         if (error)
2071                 goto done;
2072
2073         /*
2074          * Disconnect (fdip, fname) from the source directory.  This will
2075          * disconnect (ip) if it represents a direct file.  If (ip) represents
2076          * a hardlink the HARDLINK pointer object will be removed but the
2077          * hardlink will stay intact.
2078          *
2079          * The target chain may be marked DELETED but will not be destroyed
2080          * since we retain our hold on ip and chain.
2081          */
2082         error = hammer2_unlink_file(&trans, fdip, fname, fname_len, -1);
2083         KKASSERT(error != EAGAIN);
2084         if (error)
2085                 goto done;
2086
2087         /*
2088          * Reconnect ip to target directory using chain.  Chains cannot
2089          * actually be moved, so this will duplicate the chain in the new
2090          * spot and assign it to the ip, replacing the old chain.
2091          *
2092          * WARNING: chain locks can lock buffer cache buffers, to avoid
2093          *          deadlocks we want to unlock before issuing a cache_*()
2094          *          op (that might have to lock a vnode).
2095          */
2096         error = hammer2_inode_connect(&trans, tdip,
2097                                       ip, &chain,
2098                                       tname, tname_len);
2099         if (error == 0) {
2100                 if (chain) {
2101                         hammer2_chain_unlock(chain);
2102                         chain = NULL;
2103                 }
2104                 cache_rename(ap->a_fnch, ap->a_tnch);
2105         }
2106 done:
2107         if (chain)
2108                 hammer2_chain_unlock(chain);
2109         hammer2_inode_drop(ip);
2110         hammer2_trans_done(&trans);
2111
2112         return (error);
2113 }
2114
2115 static int hammer2_strategy_read(struct vop_strategy_args *ap);
2116 static int hammer2_strategy_write(struct vop_strategy_args *ap);
2117
2118 static
2119 int
2120 hammer2_vop_strategy(struct vop_strategy_args *ap)
2121 {
2122         struct bio *biop;
2123         struct buf *bp;
2124         int error;
2125
2126         biop = ap->a_bio;
2127         bp = biop->bio_buf;
2128
2129         switch(bp->b_cmd) {
2130         case BUF_CMD_READ:
2131                 error = hammer2_strategy_read(ap);
2132                 ++hammer2_iod_file_read;
2133                 break;
2134         case BUF_CMD_WRITE:
2135                 error = hammer2_strategy_write(ap);
2136                 ++hammer2_iod_file_write;
2137                 break;
2138         default:
2139                 bp->b_error = error = EINVAL;
2140                 bp->b_flags |= B_ERROR;
2141                 biodone(biop);
2142                 break;
2143         }
2144
2145         return (error);
2146 }
2147
2148 static
2149 int
2150 hammer2_strategy_read(struct vop_strategy_args *ap)
2151 {
2152         struct buf *bp;
2153         struct bio *bio;
2154         struct bio *nbio;
2155         hammer2_mount_t *hmp;
2156         hammer2_inode_t *ip;
2157         hammer2_chain_t *parent;
2158         hammer2_chain_t *chain;
2159         hammer2_key_t lbase;
2160
2161         bio = ap->a_bio;
2162         bp = bio->bio_buf;
2163         ip = VTOI(ap->a_vp);
2164         hmp = ip->hmp;
2165         nbio = push_bio(bio);
2166
2167         lbase = bio->bio_offset;
2168         chain = NULL;
2169         KKASSERT(((int)lbase & HAMMER2_PBUFMASK) == 0);
2170
2171         /*
2172          * We must characterize the logical->physical translation if it
2173          * has not already been cached.
2174          *
2175          * Physical data references < LBUFSIZE are never cached.  This
2176          * includes both small-block allocations and inode-embedded data.
2177          */
2178         if (nbio->bio_offset == NOOFFSET) {
2179                 hammer2_inode_lock_sh(ip);
2180
2181                 parent = hammer2_chain_lookup_init(ip->chain,
2182                                                    HAMMER2_LOOKUP_SHARED);
2183
2184                 chain = hammer2_chain_lookup(&parent, lbase, lbase,
2185                                              HAMMER2_LOOKUP_NODATA |
2186                                              HAMMER2_LOOKUP_SHARED);
2187                 if (chain == NULL) {
2188                         /*
2189                          * Data is zero-fill
2190                          */
2191                         nbio->bio_offset = ZFOFFSET;
2192                 } else if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
2193                         /*
2194                          * Data is embedded in the inode (do nothing)
2195                          */
2196                         KKASSERT(chain == parent);
2197                         hammer2_chain_unlock(chain);
2198                 } else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
2199                         /*
2200                          * Data is on-media
2201                          */
2202                         KKASSERT(bp->b_bcount == chain->bytes);
2203                         nbio->bio_offset = chain->bref.data_off &
2204                                            HAMMER2_OFF_MASK;
2205                         hammer2_chain_unlock(chain);
2206                         KKASSERT(nbio->bio_offset != 0);
2207                 } else {
2208                         panic("hammer2_strategy_read: unknown bref type");
2209                 }
2210                 hammer2_chain_lookup_done(parent);
2211                 hammer2_inode_unlock_sh(ip);
2212         }
2213
2214         if (hammer2_debug & 0x0020) {
2215                 kprintf("read %016jx %016jx\n",
2216                         bio->bio_offset, nbio->bio_offset);
2217         }
2218
2219         if (nbio->bio_offset == ZFOFFSET) {
2220                 /*
2221                  * Data is zero-fill
2222                  */
2223                 bp->b_resid = 0;
2224                 bp->b_error = 0;
2225                 bzero(bp->b_data, bp->b_bcount);
2226                 biodone(nbio);
2227         } else if (nbio->bio_offset != NOOFFSET) {
2228                 /*
2229                  * Forward direct IO to the device
2230                  */
2231                 vn_strategy(hmp->devvp, nbio);
2232         } else {
2233                 /*
2234                  * Data is embedded in inode.
2235                  */
2236                 bcopy(chain->data->ipdata.u.data, bp->b_data,
2237                       HAMMER2_EMBEDDED_BYTES);
2238                 bzero(bp->b_data + HAMMER2_EMBEDDED_BYTES,
2239                       bp->b_bcount - HAMMER2_EMBEDDED_BYTES);
2240                 bp->b_resid = 0;
2241                 bp->b_error = 0;
2242                 biodone(nbio);
2243         }
2244         return (0);
2245 }
2246
2247 static
2248 int
2249 hammer2_strategy_write(struct vop_strategy_args *ap)
2250 {
2251         struct buf *bp;
2252         struct bio *bio;
2253         struct bio *nbio;
2254         hammer2_mount_t *hmp;
2255         hammer2_inode_t *ip;
2256
2257         bio = ap->a_bio;
2258         bp = bio->bio_buf;
2259         ip = VTOI(ap->a_vp);
2260         hmp = ip->hmp;
2261         nbio = push_bio(bio);
2262
2263         KKASSERT((bio->bio_offset & HAMMER2_PBUFMASK64) == 0);
2264         KKASSERT(nbio->bio_offset != 0 && nbio->bio_offset != ZFOFFSET);
2265
2266         if (nbio->bio_offset == NOOFFSET) {
2267                 /*
2268                  * Must be embedded in the inode.
2269                  *
2270                  * Because the inode is dirty, the chain must exist whether
2271                  * the inode is locked or not. XXX
2272                  */
2273                 KKASSERT(bio->bio_offset == 0);
2274                 KKASSERT(ip->chain && ip->chain->data);
2275                 bcopy(bp->b_data, ip->chain->data->ipdata.u.data,
2276                       HAMMER2_EMBEDDED_BYTES);
2277                 bp->b_resid = 0;
2278                 bp->b_error = 0;
2279                 biodone(nbio);
2280
2281                 /*
2282                  * This special flag does not follow the normal MODIFY rules
2283                  * because we might deadlock on ip.  Instead we depend on
2284                  * VOP_FSYNC() to detect the case.
2285                  */
2286                 atomic_set_int(&ip->flags, HAMMER2_INODE_DIRTYEMBED);
2287         } else {
2288                 /*
2289                  * Forward direct IO to the device
2290                  */
2291                 vn_strategy(hmp->devvp, nbio);
2292         }
2293         return (0);
2294 }
2295
2296 /*
2297  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2298  */
2299 static
2300 int
2301 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2302 {
2303         hammer2_mount_t *hmp;
2304         hammer2_inode_t *ip;
2305         int error;
2306
2307         ip = VTOI(ap->a_vp);
2308         hmp = ip->hmp;
2309
2310         error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2311                               ap->a_fflag, ap->a_cred);
2312         return (error);
2313 }
2314
2315 static
2316 int
2317 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2318 {
2319         struct mount *mp;
2320         hammer2_pfsmount_t *pmp;
2321         int rc;
2322
2323         switch (ap->a_op) {
2324         case (MOUNTCTL_SET_EXPORT):
2325                 mp = ap->a_head.a_ops->head.vv_mount;
2326                 pmp = MPTOPMP(mp);
2327
2328                 if (ap->a_ctllen != sizeof(struct export_args))
2329                         rc = (EINVAL);
2330                 else
2331                         rc = vfs_export(mp, &pmp->export,
2332                                         (const struct export_args *)ap->a_ctl);
2333                 break;
2334         default:
2335                 rc = vop_stdmountctl(ap);
2336                 break;
2337         }
2338         return (rc);
2339 }
2340
2341 struct vop_ops hammer2_vnode_vops = {
2342         .vop_default    = vop_defaultop,
2343         .vop_fsync      = hammer2_vop_fsync,
2344         .vop_getpages   = vop_stdgetpages,
2345         .vop_putpages   = vop_stdputpages,
2346         .vop_access     = hammer2_vop_access,
2347         .vop_advlock    = hammer2_vop_advlock,
2348         .vop_close      = hammer2_vop_close,
2349         .vop_nlink      = hammer2_vop_nlink,
2350         .vop_ncreate    = hammer2_vop_ncreate,
2351         .vop_nsymlink   = hammer2_vop_nsymlink,
2352         .vop_nremove    = hammer2_vop_nremove,
2353         .vop_nrmdir     = hammer2_vop_nrmdir,
2354         .vop_nrename    = hammer2_vop_nrename,
2355         .vop_getattr    = hammer2_vop_getattr,
2356         .vop_setattr    = hammer2_vop_setattr,
2357         .vop_readdir    = hammer2_vop_readdir,
2358         .vop_readlink   = hammer2_vop_readlink,
2359         .vop_getpages   = vop_stdgetpages,
2360         .vop_putpages   = vop_stdputpages,
2361         .vop_read       = hammer2_vop_read,
2362         .vop_write      = hammer2_vop_write,
2363         .vop_open       = hammer2_vop_open,
2364         .vop_inactive   = hammer2_vop_inactive,
2365         .vop_reclaim    = hammer2_vop_reclaim,
2366         .vop_nresolve   = hammer2_vop_nresolve,
2367         .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2368         .vop_nmkdir     = hammer2_vop_nmkdir,
2369         .vop_ioctl      = hammer2_vop_ioctl,
2370         .vop_mountctl   = hammer2_vop_mountctl,
2371         .vop_bmap       = hammer2_vop_bmap,
2372         .vop_strategy   = hammer2_vop_strategy,
2373 };
2374
2375 struct vop_ops hammer2_spec_vops = {
2376
2377 };
2378
2379 struct vop_ops hammer2_fifo_vops = {
2380
2381 };