sys/fs/unionfs/unionfs_subr.c

   1 /*-
   2  * Copyright (c) 1994 Jan-Simon Pendry
   3  * Copyright (c) 1994
   4  *      The Regents of the University of California.  All rights reserved.
   5  * Copyright (c) 2005, 2006 Masanori Ozawa <ozawa@ongs.co.jp>, ONGS Inc.
   6  * Copyright (c) 2006 Daichi Goto <daichi@freebsd.org>
   7  *
   8  * This code is derived from software contributed to Berkeley by
   9  * Jan-Simon Pendry.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 4. Neither the name of the University nor the names of its contributors
  20  *    may be used to endorse or promote products derived from this software
  21  *    without specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  33  * SUCH DAMAGE.
  34  *
  35  *      @(#)union_subr.c        8.20 (Berkeley) 5/20/95
  36  * $FreeBSD: src/sys/fs/unionfs/union_subr.c,v 1.99 2008/01/24 12:34:27 attilio Exp $
  37  */
  38
  39 #include <sys/param.h>
  40 #include <sys/systm.h>
  41 #include <sys/kernel.h>
  42 #include <sys/lock.h>
  43 #include <sys/mutex.h>
  44 #include <sys/malloc.h>
  45 #include <sys/mount.h>
  46 #include <sys/namei.h>
  47 #include <sys/proc.h>
  48 #include <sys/vnode.h>
  49 #include <sys/dirent.h>
  50 #include <sys/fcntl.h>
  51 #include <sys/filedesc.h>
  52 #include <sys/stat.h>
  53 #include <sys/kauth.h>
  54 #include <sys/resourcevar.h>
  55
  56 #include <fs/unionfs/unionfs.h>
  57
  58 MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part");
  59
  60 /*
  61  * Make a new or get existing unionfs node.
  62  *
  63  * uppervp and lowervp should be unlocked. Because if new unionfs vnode is
  64  * locked, uppervp or lowervp is locked too. In order to prevent dead lock,
  65  * you should not lock plurality simultaneously.
  66  */
  67 int
  68 unionfs_nodeget(struct mount *mp, struct vnode *uppervp,
  69                 struct vnode *lowervp, struct vnode *dvp,
  70                 struct vnode **vpp, struct componentname *cnp)
  71 {
  72         struct unionfs_mount *ump;
  73         struct unionfs_node *unp;
  74         struct vnode   *vp;
  75         int             error;
  76         const char      *path;
  77
  78         ump = MOUNTTOUNIONFSMOUNT(mp);
  79         path = (cnp ? cnp->cn_nameptr : NULL);
  80
  81         if (uppervp == NULLVP && lowervp == NULLVP)
  82                 panic("unionfs_nodeget: upper and lower is null");
  83
  84         /* If it has no ISLASTCN flag, path check is skipped. */
  85         if (cnp && !(cnp->cn_flags & ISLASTCN))
  86                 path = NULL;
  87
  88         if ((uppervp == NULLVP || ump->um_uppervp != uppervp) ||
  89             (lowervp == NULLVP || ump->um_lowervp != lowervp)) {
  90                 if (dvp == NULLVP)
  91                         return (EINVAL);
  92         }
  93
  94         unp = kmem_zalloc(sizeof(*unp), KM_SLEEP);
  95         if (unp == NULL)
  96                 return (ENOMEM);
  97         error = getnewvnode(VT_UNION, mp, unionfs_vnodeop_p, &vp);
  98         if (error != 0) {
  99                 kmem_free(unp, sizeof(*unp));
 100                 return (error);
 101         }
 102         if (dvp != NULLVP)
 103                 vref(dvp);
 104         if (uppervp != NULLVP)
 105                 vref(uppervp);
 106         if (lowervp != NULLVP)
 107                 vref(lowervp);
 108
 109         unp->un_vnode = vp;
 110         unp->un_uppervp = uppervp;
 111         unp->un_lowervp = lowervp;
 112         unp->un_dvp = dvp;
 113         if (uppervp != NULLVP)
 114                 vp->v_vnlock = uppervp->v_vnlock;
 115         else
 116                 vp->v_vnlock = lowervp->v_vnlock;
 117
 118         if (path != NULL) {
 119                 unp->un_path = (char *)
 120                     malloc(cnp->cn_namelen +1, M_UNIONFSPATH, M_WAITOK|M_ZERO);
 121                 memcpy(unp->un_path, cnp->cn_nameptr, cnp->cn_namelen);
 122                 unp->un_path[cnp->cn_namelen] = '\0';
 123         }
 124         vp->v_type = (uppervp != NULLVP ? uppervp->v_type : lowervp->v_type);
 125         vp->v_data = unp;
 126         uvm_vnp_setsize(vp, 0);
 127
 128         if ((uppervp != NULLVP && ump->um_uppervp == uppervp) &&
 129             (lowervp != NULLVP && ump->um_lowervp == lowervp))
 130                 vp->v_vflag |= VV_ROOT;
 131
 132         *vpp = vp;
 133
 134         return (0);
 135 }
 136
 137 /*
 138  * Clean up the unionfs node.
 139  */
 140 void
 141 unionfs_noderem(struct vnode *vp)
 142 {
 143         struct unionfs_mount *ump;
 144         struct unionfs_node *unp;
 145         struct unionfs_node_status *unsp;
 146         struct vnode   *lvp;
 147         struct vnode   *uvp;
 148
 149         ump = MOUNTTOUNIONFSMOUNT(vp->v_mount);
 150
 151         /*
 152          * Use the interlock to protect the clearing of v_data to
 153          * prevent faults in unionfs_lock().
 154          */
 155         unp = VTOUNIONFS(vp);
 156         lvp = unp->un_lowervp;
 157         uvp = unp->un_uppervp;
 158         unp->un_lowervp = unp->un_uppervp = NULLVP;
 159         vp->v_vnlock = &(vp->v_lock);
 160         vp->v_data = NULL;
 161
 162         if (lvp != NULLVP)
 163                 vrele(lvp);
 164         if (uvp != NULLVP)
 165                 vrele(uvp);
 166         if (unp->un_dvp != NULLVP) {
 167                 vrele(unp->un_dvp);
 168                 unp->un_dvp = NULLVP;
 169         }
 170         if (unp->un_path) {
 171                 free(unp->un_path, M_UNIONFSPATH);
 172                 unp->un_path = NULL;
 173         }
 174
 175         while ((unsp = LIST_FIRST(&(unp->un_unshead))) != NULL) {
 176                 LIST_REMOVE(unsp, uns_list);
 177                 free(unsp, M_TEMP);
 178         }
 179         kmem_free(unp, sizeof(*unp));
 180 }
 181
 182 /*
 183  * Get the unionfs node status.
 184  * You need exclusive lock this vnode.
 185  */
 186 void
 187 unionfs_get_node_status(struct unionfs_node *unp,
 188                         struct unionfs_node_status **unspp)
 189 {
 190         struct unionfs_node_status *unsp;
 191         pid_t pid;
 192         lwpid_t lid;
 193
 194         KASSERT(NULL != unspp);
 195         KASSERT(VOP_ISLOCKED(UNIONFSTOV(unp)) == LK_EXCLUSIVE);
 196
 197         pid = curproc->p_pid;
 198         lid = curlwp->l_lid;
 199
 200         LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) {
 201                 if (unsp->uns_pid == pid && unsp->uns_lid == lid) {
 202                         *unspp = unsp;
 203                         return;
 204                 }
 205         }
 206
 207         /* create a new unionfs node status */
 208         unsp = kmem_zalloc(sizeof(*unsp), KM_SLEEP);
 209         unsp->uns_pid = pid;
 210         unsp->uns_lid = lid;
 211         LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list);
 212
 213         *unspp = unsp;
 214 }
 215
 216 /*
 217  * Remove the unionfs node status, if you can.
 218  * You need exclusive lock this vnode.
 219  */
 220 void
 221 unionfs_tryrem_node_status(struct unionfs_node *unp,
 222                            struct unionfs_node_status *unsp)
 223 {
 224         KASSERT(NULL != unsp);
 225         KASSERT(VOP_ISLOCKED(UNIONFSTOV(unp)) == LK_EXCLUSIVE);
 226
 227         if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt)
 228                 return;
 229
 230         LIST_REMOVE(unsp, uns_list);
 231         kmem_free(unsp, sizeof(*unsp));
 232 }
 233
 234 /*
 235  * Create upper node attr.
 236  */
 237 void
 238 unionfs_create_uppervattr_core(struct unionfs_mount *ump,
 239                                struct vattr *lva,
 240                                struct vattr *uva)
 241 {
 242         VATTR_NULL(uva);
 243         uva->va_type = lva->va_type;
 244         uva->va_atime = lva->va_atime;
 245         uva->va_mtime = lva->va_mtime;
 246         uva->va_ctime = lva->va_ctime;
 247
 248         switch (ump->um_copymode) {
 249         case UNIONFS_TRANSPARENT:
 250                 uva->va_mode = lva->va_mode;
 251                 uva->va_uid = lva->va_uid;
 252                 uva->va_gid = lva->va_gid;
 253                 break;
 254         case UNIONFS_MASQUERADE:
 255                 if (ump->um_uid == lva->va_uid) {
 256                         uva->va_mode = lva->va_mode & 077077;
 257                         uva->va_mode |= (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile) & 0700;
 258                         uva->va_uid = lva->va_uid;
 259                         uva->va_gid = lva->va_gid;
 260                 } else {
 261                         uva->va_mode = (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile);
 262                         uva->va_uid = ump->um_uid;
 263                         uva->va_gid = ump->um_gid;
 264                 }
 265                 break;
 266         default:                /* UNIONFS_TRADITIONAL */
 267                 uva->va_mode = 0777 & ~curproc->p_cwdi->cwdi_cmask;
 268                 uva->va_uid = ump->um_uid;
 269                 uva->va_gid = ump->um_gid;
 270                 break;
 271         }
 272 }
 273
 274 /*
 275  * Create upper node attr.
 276  */
 277 int
 278 unionfs_create_uppervattr(struct unionfs_mount *ump,
 279                           struct vnode *lvp,
 280                           struct vattr *uva,
 281                           kauth_cred_t cred)
 282 {
 283         int             error;
 284         struct vattr    lva;
 285
 286         if ((error = VOP_GETATTR(lvp, &lva, cred)))
 287                 return (error);
 288
 289         unionfs_create_uppervattr_core(ump, &lva, uva);
 290
 291         return (error);
 292 }
 293
 294 /*
 295  * relookup
 296  *
 297  * dvp should be locked on entry and will be locked on return.
 298  *
 299  * If an error is returned, *vpp will be invalid, otherwise it will hold a
 300  * locked, referenced vnode. If *vpp == dvp then remember that only one
 301  * LK_EXCLUSIVE lock is held.
 302  */
 303 static int
 304 unionfs_relookup(struct vnode *dvp, struct vnode **vpp,
 305                  struct componentname *cnp, struct componentname *cn,
 306                  const char *path, int pathlen, u_long nameiop)
 307 {
 308         int     error;
 309
 310         cn->cn_namelen = pathlen;
 311         cn->cn_pnbuf = PNBUF_GET();
 312         memcpy(cn->cn_pnbuf, path, pathlen);
 313         cn->cn_pnbuf[pathlen] = '\0';
 314
 315         cn->cn_nameiop = nameiop;
 316         cn->cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
 317         cn->cn_cred = cnp->cn_cred;
 318
 319         cn->cn_nameptr = cn->cn_pnbuf;
 320         cn->cn_consume = cnp->cn_consume;
 321
 322         if (nameiop == DELETE)
 323                 cn->cn_flags |= (cnp->cn_flags & (DOWHITEOUT | SAVESTART));
 324         else if (RENAME == nameiop)
 325                 cn->cn_flags |= (cnp->cn_flags & SAVESTART);
 326
 327         vref(dvp);
 328         VOP_UNLOCK(dvp, 0);
 329
 330         if ((error = relookup(dvp, vpp, cn))) {
 331                 PNBUF_PUT(cn->cn_pnbuf);
 332                 cn->cn_flags &= ~HASBUF;
 333                 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
 334         } else
 335                 vrele(dvp);
 336
 337         return (error);
 338 }
 339
 340 /*
 341  * relookup for CREATE namei operation.
 342  *
 343  * dvp is unionfs vnode. dvp should be locked.
 344  *
 345  * If it called 'unionfs_copyfile' function by unionfs_link etc,
 346  * VOP_LOOKUP information is broken.
 347  * So it need relookup in order to create link etc.
 348  */
 349 int
 350 unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp)
 351 {
 352         int     error;
 353         struct vnode *udvp;
 354         struct vnode *vp;
 355         struct componentname cn;
 356
 357         udvp = UNIONFSVPTOUPPERVP(dvp);
 358         vp = NULLVP;
 359
 360         error = unionfs_relookup(udvp, &vp, cnp, &cn, cnp->cn_nameptr,
 361             strlen(cnp->cn_nameptr), CREATE);
 362         if (error)
 363                 return (error);
 364
 365         if (vp != NULLVP) {
 366                 if (udvp == vp)
 367                         vrele(vp);
 368                 else
 369                         vput(vp);
 370
 371                 error = EEXIST;
 372         }
 373
 374         if (cn.cn_flags & HASBUF) {
 375                 PNBUF_PUT(cn.cn_pnbuf);
 376                 cn.cn_flags &= ~HASBUF;
 377         }
 378
 379         if (!error) {
 380                 cn.cn_flags |= (cnp->cn_flags & HASBUF);
 381                 cnp->cn_flags = cn.cn_flags;
 382         }
 383
 384         return (error);
 385 }
 386
 387 /*
 388  * relookup for DELETE namei operation.
 389  *
 390  * dvp is unionfs vnode. dvp should be locked.
 391  */
 392 int
 393 unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp)
 394 {
 395         int     error;
 396         struct vnode *udvp;
 397         struct vnode *vp;
 398         struct componentname cn;
 399
 400         udvp = UNIONFSVPTOUPPERVP(dvp);
 401         vp = NULLVP;
 402
 403         error = unionfs_relookup(udvp, &vp, cnp, &cn, cnp->cn_nameptr,
 404             strlen(cnp->cn_nameptr), DELETE);
 405         if (error)
 406                 return (error);
 407
 408         if (vp == NULLVP)
 409                 error = ENOENT;
 410         else {
 411                 if (udvp == vp)
 412                         vrele(vp);
 413                 else
 414                         vput(vp);
 415         }
 416
 417         if (cn.cn_flags & HASBUF) {
 418                 PNBUF_PUT(cn.cn_pnbuf);
 419                 cn.cn_flags &= ~HASBUF;
 420         }
 421
 422         if (!error) {
 423                 cn.cn_flags |= (cnp->cn_flags & HASBUF);
 424                 cnp->cn_flags = cn.cn_flags;
 425         }
 426
 427         return (error);
 428 }
 429
 430 /*
 431  * relookup for RENAME namei operation.
 432  *
 433  * dvp is unionfs vnode. dvp should be locked.
 434  */
 435 int
 436 unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp)
 437 {
 438         int error;
 439         struct vnode *udvp;
 440         struct vnode *vp;
 441         struct componentname cn;
 442
 443         udvp = UNIONFSVPTOUPPERVP(dvp);
 444         vp = NULLVP;
 445
 446         error = unionfs_relookup(udvp, &vp, cnp, &cn, cnp->cn_nameptr,
 447             strlen(cnp->cn_nameptr), RENAME);
 448         if (error)
 449                 return (error);
 450
 451         if (vp != NULLVP) {
 452                 if (udvp == vp)
 453                         vrele(vp);
 454                 else
 455                         vput(vp);
 456         }
 457
 458         if (cn.cn_flags & HASBUF) {
 459                 PNBUF_PUT(cn.cn_pnbuf);
 460                 cn.cn_flags &= ~HASBUF;
 461         }
 462
 463         if (!error) {
 464                 cn.cn_flags |= (cnp->cn_flags & HASBUF);
 465                 cnp->cn_flags = cn.cn_flags;
 466         }
 467
 468         return (error);
 469
 470 }
 471
 472 /*
 473  * Update the unionfs_node.
 474  *
 475  * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the
 476  * uvp's lock and lower's lock will be unlocked.
 477  */
 478 static void
 479 unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp)
 480 {
 481         int             count, lockcnt;
 482         struct vnode   *vp;
 483         struct vnode   *lvp;
 484
 485         vp = UNIONFSTOV(unp);
 486         lvp = unp->un_lowervp;
 487
 488         /*
 489          * lock update
 490          */
 491         mutex_enter(&vp->v_interlock);
 492         unp->un_uppervp = uvp;
 493         vp->v_vnlock = uvp->v_vnlock;
 494         lockcnt = lvp->v_vnlock->vl_recursecnt +
 495             rw_write_held(&lvp->v_vnlock->vl_lock);
 496         if (lockcnt <= 0)
 497                 panic("unionfs: no exclusive lock");
 498         mutex_exit(&vp->v_interlock);
 499         for (count = 1; count < lockcnt; count++)
 500                 vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
 501 }
 502
 503 /*
 504  * Create a new shadow dir.
 505  *
 506  * udvp should be locked on entry and will be locked on return.
 507  *
 508  * If no error returned, unp will be updated.
 509  */
 510 int
 511 unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp,
 512                     struct unionfs_node *unp, struct componentname *cnp)
 513 {
 514         int             error;
 515         struct vnode   *lvp;
 516         struct vnode   *uvp;
 517         struct vattr    va;
 518         struct vattr    lva;
 519         struct componentname cn;
 520
 521         if (unp->un_uppervp != NULLVP)
 522                 return (EEXIST);
 523
 524         lvp = unp->un_lowervp;
 525         uvp = NULLVP;
 526
 527         memset(&cn, 0, sizeof(cn));
 528
 529         if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred)))
 530                 goto unionfs_mkshadowdir_abort;
 531
 532         if ((error = unionfs_relookup(udvp, &uvp, cnp, &cn, cnp->cn_nameptr, cnp->cn_namelen, CREATE)))
 533                 goto unionfs_mkshadowdir_abort;
 534         if (uvp != NULLVP) {
 535                 if (udvp == uvp)
 536                         vrele(uvp);
 537                 else
 538                         vput(uvp);
 539
 540                 error = EEXIST;
 541                 goto unionfs_mkshadowdir_free_out;
 542         }
 543
 544         unionfs_create_uppervattr_core(ump, &lva, &va);
 545
 546         error = VOP_MKDIR(udvp, &uvp, &cn, &va);
 547
 548         if (!error) {
 549                 unionfs_node_update(unp, uvp);
 550
 551                 /*
 552                  * XXX The bug which cannot set uid/gid was corrected.
 553                  * Ignore errors.   XXXNETBSD Why is this done as root?
 554                  */
 555                 va.va_type = VNON;
 556                 VOP_SETATTR(uvp, &va, lwp0.l_cred);
 557         }
 558
 559 unionfs_mkshadowdir_free_out:
 560         if (cn.cn_flags & HASBUF) {
 561                 PNBUF_PUT(cn.cn_pnbuf);
 562                 cn.cn_flags &= ~HASBUF;
 563         }
 564
 565 unionfs_mkshadowdir_abort:
 566
 567         return (error);
 568 }
 569
 570 /*
 571  * Create a new whiteout.
 572  *
 573  * dvp should be locked on entry and will be locked on return.
 574  */
 575 int
 576 unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp, const char *path)
 577 {
 578         int             error;
 579         struct vnode   *wvp;
 580         struct componentname cn;
 581
 582         if (path == NULL)
 583                 path = cnp->cn_nameptr;
 584
 585         wvp = NULLVP;
 586         if ((error = unionfs_relookup(dvp, &wvp, cnp, &cn, path, strlen(path), CREATE)))
 587                 return (error);
 588         if (wvp != NULLVP) {
 589                 if (cn.cn_flags & HASBUF) {
 590                         PNBUF_PUT(cn.cn_pnbuf);
 591                         cn.cn_flags &= ~HASBUF;
 592                 }
 593                 if (dvp == wvp)
 594                         vrele(wvp);
 595                 else
 596                         vput(wvp);
 597
 598                 return (EEXIST);
 599         }
 600
 601         if (cn.cn_flags & HASBUF) {
 602                 PNBUF_PUT(cn.cn_pnbuf);
 603                 cn.cn_flags &= ~HASBUF;
 604         }
 605
 606         return (error);
 607 }
 608
 609 /*
 610  * Create a new vnode for create a new shadow file.
 611  *
 612  * If an error is returned, *vpp will be invalid, otherwise it will hold a
 613  * locked, referenced and opened vnode.
 614  *
 615  * unp is never updated.
 616  */
 617 static int
 618 unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp,
 619                            struct unionfs_node *unp, struct vattr *uvap)
 620 {
 621         struct unionfs_mount *ump;
 622         struct vnode   *vp;
 623         struct vnode   *lvp;
 624         kauth_cred_t    cred;
 625         struct vattr    lva;
 626         int             fmode;
 627         int             error;
 628         struct componentname cn;
 629
 630         ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount);
 631         vp = NULLVP;
 632         lvp = unp->un_lowervp;
 633         cred = kauth_cred_get();
 634         fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL);
 635         error = 0;
 636
 637         if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0)
 638                 return (error);
 639         unionfs_create_uppervattr_core(ump, &lva, uvap);
 640
 641         if (unp->un_path == NULL)
 642                 panic("unionfs: un_path is null");
 643
 644         cn.cn_namelen = strlen(unp->un_path);
 645         cn.cn_pnbuf = PNBUF_GET();
 646         memcpy(cn.cn_pnbuf, unp->un_path, cn.cn_namelen + 1);
 647         cn.cn_nameiop = CREATE;
 648         cn.cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
 649         cn.cn_cred = cred;
 650         cn.cn_nameptr = cn.cn_pnbuf;
 651         cn.cn_consume = 0;
 652
 653         vref(udvp);
 654         if ((error = relookup(udvp, &vp, &cn)) != 0)
 655                 goto unionfs_vn_create_on_upper_free_out2;
 656         vrele(udvp);
 657
 658         if (vp != NULLVP) {
 659                 if (vp == udvp)
 660                         vrele(vp);
 661                 else
 662                         vput(vp);
 663                 error = EEXIST;
 664                 goto unionfs_vn_create_on_upper_free_out1;
 665         }
 666
 667         if ((error = VOP_CREATE(udvp, &vp, &cn, uvap)) != 0)
 668                 goto unionfs_vn_create_on_upper_free_out1;
 669
 670         if ((error = VOP_OPEN(vp, fmode, cred)) != 0) {
 671                 vput(vp);
 672                 goto unionfs_vn_create_on_upper_free_out1;
 673         }
 674         vp->v_writecount++;
 675         *vpp = vp;
 676
 677 unionfs_vn_create_on_upper_free_out1:
 678         VOP_UNLOCK(udvp, 0);
 679
 680 unionfs_vn_create_on_upper_free_out2:
 681         if (cn.cn_flags & HASBUF) {
 682                 PNBUF_PUT(cn.cn_pnbuf);
 683                 cn.cn_flags &= ~HASBUF;
 684         }
 685
 686         return (error);
 687 }
 688
 689 /*
 690  * Copy from lvp to uvp.
 691  *
 692  * lvp and uvp should be locked and opened on entry and will be locked and
 693  * opened on return.
 694  */
 695 static int
 696 unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp,
 697                       kauth_cred_t cred)
 698 {
 699         int             error;
 700         off_t           offset;
 701         int             count;
 702         int             bufoffset;
 703         char           *buf;
 704         struct uio      uio;
 705         struct iovec    iov;
 706
 707         error = 0;
 708         memset(&uio, 0, sizeof(uio));
 709         UIO_SETUP_SYSSPACE(&uio);
 710         uio.uio_offset = 0;
 711
 712         buf = kmem_alloc(MAXBSIZE, KM_SLEEP);
 713         if (buf == NULL)
 714                 return ENOMEM;
 715
 716         while (error == 0) {
 717                 offset = uio.uio_offset;
 718
 719                 uio.uio_iov = &iov;
 720                 uio.uio_iovcnt = 1;
 721                 iov.iov_base = buf;
 722                 iov.iov_len = MAXBSIZE;
 723                 uio.uio_resid = iov.iov_len;
 724                 uio.uio_rw = UIO_READ;
 725
 726                 if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0)
 727                         break;
 728                 if ((count = MAXBSIZE - uio.uio_resid) == 0)
 729                         break;
 730
 731                 bufoffset = 0;
 732                 while (bufoffset < count) {
 733                         uio.uio_iov = &iov;
 734                         uio.uio_iovcnt = 1;
 735                         iov.iov_base = buf + bufoffset;
 736                         iov.iov_len = count - bufoffset;
 737                         uio.uio_offset = offset + bufoffset;
 738                         uio.uio_resid = iov.iov_len;
 739                         uio.uio_rw = UIO_WRITE;
 740
 741                         if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0)
 742                                 break;
 743
 744                         bufoffset += (count - bufoffset) - uio.uio_resid;
 745                 }
 746
 747                 uio.uio_offset = offset + bufoffset;
 748         }
 749
 750         kmem_free(buf, MAXBSIZE);
 751
 752         return (error);
 753 }
 754
 755 /*
 756  * Copy file from lower to upper.
 757  *
 758  * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to
 759  * docopy.
 760  *
 761  * If no error returned, unp will be updated.
 762  */
 763 int
 764 unionfs_copyfile(struct unionfs_node *unp, int docopy, kauth_cred_t cred)
 765 {
 766         int             error;
 767         struct vnode   *udvp;
 768         struct vnode   *lvp;
 769         struct vnode   *uvp;
 770         struct vattr    uva;
 771
 772         lvp = unp->un_lowervp;
 773         uvp = NULLVP;
 774
 775         if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY))
 776                 return (EROFS);
 777         if (unp->un_dvp == NULLVP)
 778                 return (EINVAL);
 779         if (unp->un_uppervp != NULLVP)
 780                 return (EEXIST);
 781         udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp;
 782         if (udvp == NULLVP)
 783                 return (EROFS);
 784         if ((udvp->v_mount->mnt_flag & MNT_RDONLY))
 785                 return (EROFS);
 786
 787         error = VOP_ACCESS(lvp, VREAD, cred);
 788         if (error != 0)
 789                 return (error);
 790
 791         error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva);
 792         if (error != 0)
 793                 return (error);
 794
 795         if (docopy != 0) {
 796                 error = VOP_OPEN(lvp, FREAD, cred);
 797                 if (error == 0) {
 798                         error = unionfs_copyfile_core(lvp, uvp, cred);
 799                         VOP_CLOSE(lvp, FREAD, cred);
 800                 }
 801         }
 802         VOP_CLOSE(uvp, FWRITE, cred);
 803         uvp->v_writecount--;
 804
 805         if (error == 0) {
 806                 /* Reset the attributes. Ignore errors. */
 807                 uva.va_type = VNON;
 808                 VOP_SETATTR(uvp, &uva, cred);
 809         }
 810
 811         unionfs_node_update(unp, uvp);
 812
 813         return (error);
 814 }
 815
 816 /*
 817  * It checks whether vp can rmdir. (check empty)
 818  *
 819  * vp is unionfs vnode.
 820  * vp should be locked.
 821  */
 822 int
 823 unionfs_check_rmdir(struct vnode *vp, kauth_cred_t cred)
 824 {
 825         int             error;
 826         int             eofflag;
 827         int             lookuperr;
 828         struct vnode   *uvp;
 829         struct vnode   *lvp;
 830         struct vnode   *tvp;
 831         struct vattr    va;
 832         struct componentname cn;
 833         /*
 834          * The size of buf needs to be larger than DIRBLKSIZ.
 835          */
 836         char            buf[256 * 6];
 837         struct dirent  *dp;
 838         struct dirent  *edp;
 839         struct uio      uio;
 840         struct iovec    iov;
 841
 842         KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
 843
 844         eofflag = 0;
 845         uvp = UNIONFSVPTOUPPERVP(vp);
 846         lvp = UNIONFSVPTOLOWERVP(vp);
 847
 848         /* check opaque */
 849         if ((error = VOP_GETATTR(uvp, &va, cred)) != 0)
 850                 return (error);
 851         if (va.va_flags & OPAQUE)
 852                 return (0);
 853
 854         /* open vnode */
 855         if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred)) != 0)
 856                 return (error);
 857         if ((error = VOP_OPEN(vp, FREAD, cred)) != 0)
 858                 return (error);
 859
 860         UIO_SETUP_SYSSPACE(&uio);
 861         uio.uio_rw = UIO_READ;
 862         uio.uio_offset = 0;
 863
 864         while (!error && !eofflag) {
 865                 iov.iov_base = buf;
 866                 iov.iov_len = sizeof(buf);
 867                 uio.uio_iov = &iov;
 868                 uio.uio_iovcnt = 1;
 869                 uio.uio_resid = iov.iov_len;
 870
 871                 error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL);
 872                 if (error)
 873                         break;
 874
 875                 edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid];
 876                 for (dp = (struct dirent*)buf; !error && dp < edp;
 877                      dp = (struct dirent*)((char *)dp + dp->d_reclen)) {
 878                         if (dp->d_type == DT_WHT ||
 879                             (dp->d_namlen == 1 && dp->d_name[0] == '.') ||
 880                             (dp->d_namlen == 2 && !memcmp(dp->d_name, "..", 2)))
 881                                 continue;
 882
 883                         cn.cn_namelen = dp->d_namlen;
 884                         cn.cn_pnbuf = NULL;
 885                         cn.cn_nameptr = dp->d_name;
 886                         cn.cn_nameiop = LOOKUP;
 887                         cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN);
 888                         cn.cn_cred = cred;
 889                         cn.cn_consume = 0;
 890
 891                         /*
 892                          * check entry in lower.
 893                          * Sometimes, readdir function returns
 894                          * wrong entry.
 895                          */
 896                         lookuperr = VOP_LOOKUP(lvp, &tvp, &cn);
 897
 898                         if (!lookuperr)
 899                                 vput(tvp);
 900                         else
 901                                 continue; /* skip entry */
 902
 903                         /*
 904                          * check entry
 905                          * If it has no exist/whiteout entry in upper,
 906                          * directory is not empty.
 907                          */
 908                         cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN);
 909                         lookuperr = VOP_LOOKUP(uvp, &tvp, &cn);
 910
 911                         if (!lookuperr)
 912                                 vput(tvp);
 913
 914                         /* ignore exist or whiteout entry */
 915                         if (!lookuperr ||
 916                             (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT)))
 917                                 continue;
 918
 919                         error = ENOTEMPTY;
 920                 }
 921         }
 922
 923         /* close vnode */
 924         VOP_CLOSE(vp, FREAD, cred);
 925
 926         return (error);
 927 }
 928
 929 #ifdef DIAGNOSTIC
 930
 931 struct vnode   *
 932 unionfs_checkuppervp(struct vnode *vp, const char *fil, int lno)
 933 {
 934         struct unionfs_node *unp;
 935
 936         unp = VTOUNIONFS(vp);
 937
 938 #ifdef notyet
 939         if (vp->v_op != unionfs_vnodeop_p) {
 940                 printf("unionfs_checkuppervp: on non-unionfs-node.\n");
 941 #ifdef KDB
 942                 kdb_enter(KDB_WHY_UNIONFS,
 943                     "unionfs_checkuppervp: on non-unionfs-node.\n");
 944 #endif
 945                 panic("unionfs_checkuppervp");
 946         };
 947 #endif
 948         return (unp->un_uppervp);
 949 }
 950
 951 struct vnode   *
 952 unionfs_checklowervp(struct vnode *vp, const char *fil, int lno)
 953 {
 954         struct unionfs_node *unp;
 955
 956         unp = VTOUNIONFS(vp);
 957
 958 #ifdef notyet
 959         if (vp->v_op != unionfs_vnodeop_p) {
 960                 printf("unionfs_checklowervp: on non-unionfs-node.\n");
 961 #ifdef KDB
 962                 kdb_enter(KDB_WHY_UNIONFS,
 963                     "unionfs_checklowervp: on non-unionfs-node.\n");
 964 #endif
 965                 panic("unionfs_checklowervp");
 966         };
 967 #endif
 968         return (unp->un_lowervp);
 969 }
 970 #endif