Home | History | Annotate | Line # | Download | only in ext2fs
ext2fs_rename.c revision 1.6
      1 /*	$NetBSD: ext2fs_rename.c,v 1.6 2014/01/28 13:02:56 martin Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2012 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Taylor R Campbell.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Ext2fs Rename
     34  */
     35 
     36 #include <sys/cdefs.h>
     37 __KERNEL_RCSID(0, "$NetBSD: ext2fs_rename.c,v 1.6 2014/01/28 13:02:56 martin Exp $");
     38 
     39 #include <sys/param.h>
     40 #include <sys/buf.h>
     41 #include <sys/errno.h>
     42 #include <sys/kauth.h>
     43 #include <sys/mount.h>
     44 #include <sys/namei.h>
     45 #include <sys/vnode.h>
     46 #include <sys/vnode_if.h>
     47 
     48 #include <miscfs/genfs/genfs.h>
     49 
     50 #include <ufs/ext2fs/ext2fs.h>
     51 #include <ufs/ext2fs/ext2fs_dir.h>
     52 #include <ufs/ext2fs/ext2fs_extern.h>
     53 #include <ufs/ufs/inode.h>
     54 #include <ufs/ufs/ufs_extern.h>
     55 #include <ufs/ufs/ufsmount.h>
     56 
     57 /*
     58  * Forward declarations
     59  */
     60 static int ext2fs_sane_rename(struct vnode *, struct componentname *,
     61     struct vnode *, struct componentname *,
     62     kauth_cred_t, bool);
     63 static bool ext2fs_rename_ulr_overlap_p(const struct ufs_lookup_results *,
     64     const struct ufs_lookup_results *);
     65 static int ext2fs_rename_recalculate_fulr(struct vnode *,
     66     struct ufs_lookup_results *, const struct ufs_lookup_results *,
     67     const struct componentname *);
     68 static bool ext2fs_rmdired_p(struct vnode *);
     69 static int ext2fs_read_dotdot(struct vnode *, kauth_cred_t, ino_t *);
     70 static int ext2fs_rename_replace_dotdot(struct vnode *,
     71     struct vnode *, struct vnode *, kauth_cred_t);
     72 static int ext2fs_gro_lock_directory(struct mount *, struct vnode *);
     73 
     74 static const struct genfs_rename_ops ext2fs_genfs_rename_ops;
     75 
     76 /*
     77  * ext2fs_sane_rename: The hairiest vop, with the saner API.
     78  *
     79  * Arguments:
     80  *
     81  * . fdvp (from directory vnode),
     82  * . fcnp (from component name),
     83  * . tdvp (to directory vnode),
     84  * . tcnp (to component name),
     85  * . cred (credentials structure), and
     86  * . posixly_correct (flag for behaviour if target & source link same file).
     87  *
     88  * fdvp and tdvp may be the same, and must be referenced and unlocked.
     89  */
     90 static int
     91 ext2fs_sane_rename(
     92     struct vnode *fdvp, struct componentname *fcnp,
     93     struct vnode *tdvp, struct componentname *tcnp,
     94     kauth_cred_t cred, bool posixly_correct)
     95 {
     96 	struct ufs_lookup_results fulr, tulr;
     97 
     98 	return genfs_sane_rename(&ext2fs_genfs_rename_ops,
     99 	    fdvp, fcnp, &fulr, tdvp, tcnp, &tulr,
    100 	    cred, posixly_correct);
    101 }
    102 
    103 /*
    104  * ext2fs_rename: The hairiest vop, with the insanest API.  Defer to
    105  * genfs_insane_rename immediately.
    106  */
    107 int
    108 ext2fs_rename(void *v)
    109 {
    110 
    111 	return genfs_insane_rename(v, &ext2fs_sane_rename);
    112 }
    113 
    114 /*
    115  * ext2fs_gro_directory_empty_p: Return true if the directory vp is
    116  * empty.  dvp is its parent.
    117  *
    118  * vp and dvp must be locked and referenced.
    119  */
    120 static bool
    121 ext2fs_gro_directory_empty_p(struct mount *mp, kauth_cred_t cred,
    122     struct vnode *vp, struct vnode *dvp)
    123 {
    124 
    125 	(void)mp;
    126 	KASSERT(mp != NULL);
    127 	KASSERT(vp != NULL);
    128 	KASSERT(dvp != NULL);
    129 	KASSERT(vp != dvp);
    130 	KASSERT(vp->v_mount == mp);
    131 	KASSERT(dvp->v_mount == mp);
    132 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    133 	KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
    134 
    135 	return ext2fs_dirempty(VTOI(vp), VTOI(dvp)->i_number, cred);
    136 }
    137 
    138 /*
    139  * ext2fs_gro_rename_check_possible: Check whether a rename is possible
    140  * independent of credentials.
    141  */
    142 static int
    143 ext2fs_gro_rename_check_possible(struct mount *mp,
    144     struct vnode *fdvp, struct vnode *fvp,
    145     struct vnode *tdvp, struct vnode *tvp)
    146 {
    147 
    148 	(void)mp;
    149 	KASSERT(mp != NULL);
    150 	KASSERT(fdvp != NULL);
    151 	KASSERT(fvp != NULL);
    152 	KASSERT(tdvp != NULL);
    153 	KASSERT(fdvp != fvp);
    154 	KASSERT(fdvp != tvp);
    155 	KASSERT(tdvp != fvp);
    156 	KASSERT(tdvp != tvp);
    157 	KASSERT(fvp != tvp);
    158 	KASSERT(fdvp->v_type == VDIR);
    159 	KASSERT(tdvp->v_type == VDIR);
    160 	KASSERT(fdvp->v_mount == mp);
    161 	KASSERT(fvp->v_mount == mp);
    162 	KASSERT(tdvp->v_mount == mp);
    163 	KASSERT((tvp == NULL) || (tvp->v_mount == mp));
    164 	KASSERT(VOP_ISLOCKED(fdvp) == LK_EXCLUSIVE);
    165 	KASSERT(VOP_ISLOCKED(fvp) == LK_EXCLUSIVE);
    166 	KASSERT(VOP_ISLOCKED(tdvp) == LK_EXCLUSIVE);
    167 	KASSERT((tvp == NULL) || (VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
    168 
    169 	return genfs_ufslike_rename_check_possible(
    170 	    VTOI(fdvp)->i_e2fs_flags, VTOI(fvp)->i_e2fs_flags,
    171 	    VTOI(tdvp)->i_e2fs_flags, (tvp? VTOI(tvp)->i_e2fs_flags : 0),
    172 	    (tvp != NULL),
    173 	    EXT2_IMMUTABLE, EXT2_APPEND);
    174 }
    175 
    176 /*
    177  * ext2fs_gro_rename_check_permitted: Check whether a rename is
    178  * permitted given our credentials.
    179  */
    180 static int
    181 ext2fs_gro_rename_check_permitted(struct mount *mp, kauth_cred_t cred,
    182     struct vnode *fdvp, struct vnode *fvp,
    183     struct vnode *tdvp, struct vnode *tvp)
    184 {
    185 
    186 	(void)mp;
    187 	KASSERT(mp != NULL);
    188 	KASSERT(fdvp != NULL);
    189 	KASSERT(fvp != NULL);
    190 	KASSERT(tdvp != NULL);
    191 	KASSERT(fdvp != fvp);
    192 	KASSERT(fdvp != tvp);
    193 	KASSERT(tdvp != fvp);
    194 	KASSERT(tdvp != tvp);
    195 	KASSERT(fvp != tvp);
    196 	KASSERT(fdvp->v_type == VDIR);
    197 	KASSERT(tdvp->v_type == VDIR);
    198 	KASSERT(fdvp->v_mount == mp);
    199 	KASSERT(fvp->v_mount == mp);
    200 	KASSERT(tdvp->v_mount == mp);
    201 	KASSERT((tvp == NULL) || (tvp->v_mount == mp));
    202 	KASSERT(VOP_ISLOCKED(fdvp) == LK_EXCLUSIVE);
    203 	KASSERT(VOP_ISLOCKED(fvp) == LK_EXCLUSIVE);
    204 	KASSERT(VOP_ISLOCKED(tdvp) == LK_EXCLUSIVE);
    205 	KASSERT((tvp == NULL) || (VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
    206 
    207 	return genfs_ufslike_rename_check_permitted(cred,
    208 	    fdvp, VTOI(fdvp)->i_e2fs_mode, VTOI(fdvp)->i_uid,
    209 	    fvp, VTOI(fvp)->i_uid,
    210 	    tdvp, VTOI(tdvp)->i_e2fs_mode, VTOI(tdvp)->i_uid,
    211 	    tvp, (tvp? VTOI(tvp)->i_uid : 0));
    212 }
    213 
    214 /*
    215  * ext2fs_gro_remove_check_possible: Check whether a remove is possible
    216  * independent of credentials.
    217  */
    218 static int
    219 ext2fs_gro_remove_check_possible(struct mount *mp,
    220     struct vnode *dvp, struct vnode *vp)
    221 {
    222 
    223 	(void)mp;
    224 	KASSERT(mp != NULL);
    225 	KASSERT(dvp != NULL);
    226 	KASSERT(vp != NULL);
    227 	KASSERT(dvp != vp);
    228 	KASSERT(dvp->v_type == VDIR);
    229 	KASSERT(vp->v_type != VDIR);
    230 	KASSERT(dvp->v_mount == mp);
    231 	KASSERT(vp->v_mount == mp);
    232 	KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
    233 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    234 
    235 	return genfs_ufslike_remove_check_possible(
    236 	    VTOI(dvp)->i_e2fs_flags, VTOI(vp)->i_e2fs_flags,
    237 	    EXT2_IMMUTABLE, EXT2_APPEND);
    238 }
    239 
    240 /*
    241  * ext2fs_gro_remove_check_permitted: Check whether a remove is
    242  * permitted given our credentials.
    243  */
    244 static int
    245 ext2fs_gro_remove_check_permitted(struct mount *mp, kauth_cred_t cred,
    246     struct vnode *dvp, struct vnode *vp)
    247 {
    248 
    249 	(void)mp;
    250 	KASSERT(mp != NULL);
    251 	KASSERT(dvp != NULL);
    252 	KASSERT(vp != NULL);
    253 	KASSERT(dvp != vp);
    254 	KASSERT(dvp->v_type == VDIR);
    255 	KASSERT(vp->v_type != VDIR);
    256 	KASSERT(dvp->v_mount == mp);
    257 	KASSERT(vp->v_mount == mp);
    258 	KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
    259 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    260 
    261 	return genfs_ufslike_remove_check_permitted(cred,
    262 	    dvp, VTOI(dvp)->i_e2fs_mode, VTOI(dvp)->i_uid,
    263 	    vp, VTOI(vp)->i_uid);
    264 }
    265 
    266 /*
    267  * ext2fs_gro_rename: Actually perform the rename operation.
    268  */
    269 static int
    270 ext2fs_gro_rename(struct mount *mp, kauth_cred_t cred,
    271     struct vnode *fdvp, struct componentname *fcnp,
    272     void *fde, struct vnode *fvp,
    273     struct vnode *tdvp, struct componentname *tcnp,
    274     void *tde, struct vnode *tvp)
    275 {
    276 	struct ufs_lookup_results *fulr = fde;
    277 	struct ufs_lookup_results *tulr = tde;
    278 	bool directory_p, reparent_p;
    279 	int error;
    280 
    281 	(void)mp;
    282 	KASSERT(mp != NULL);
    283 	KASSERT(fdvp != NULL);
    284 	KASSERT(fcnp != NULL);
    285 	KASSERT(fulr != NULL);
    286 	KASSERT(fvp != NULL);
    287 	KASSERT(tdvp != NULL);
    288 	KASSERT(tcnp != NULL);
    289 	KASSERT(tulr != NULL);
    290 	KASSERT(fulr != tulr);
    291 	KASSERT(fdvp != fvp);
    292 	KASSERT(fdvp != tvp);
    293 	KASSERT(tdvp != fvp);
    294 	KASSERT(tdvp != tvp);
    295 	KASSERT(fvp != tvp);
    296 	KASSERT(fdvp->v_mount == mp);
    297 	KASSERT(fvp->v_mount == mp);
    298 	KASSERT(tdvp->v_mount == mp);
    299 	KASSERT((tvp == NULL) || (tvp->v_mount == mp));
    300 	KASSERT(VOP_ISLOCKED(fdvp) == LK_EXCLUSIVE);
    301 	KASSERT(VOP_ISLOCKED(fvp) == LK_EXCLUSIVE);
    302 	KASSERT(VOP_ISLOCKED(tdvp) == LK_EXCLUSIVE);
    303 	KASSERT((tvp == NULL) || (VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
    304 
    305 	/*
    306 	 * We shall need to temporarily bump the link count, so make
    307 	 * sure there is room to do so.
    308 	 */
    309 	if ((nlink_t)VTOI(fvp)->i_e2fs_nlink >= LINK_MAX)
    310 		return EMLINK;
    311 
    312 	directory_p = (fvp->v_type == VDIR);
    313 	KASSERT(directory_p == ((VTOI(fvp)->i_e2fs_mode & IFMT) == IFDIR));
    314 	KASSERT((tvp == NULL) || (directory_p == (tvp->v_type == VDIR)));
    315 	KASSERT((tvp == NULL) || (directory_p ==
    316 		((VTOI(tvp)->i_e2fs_mode & IFMT) == IFDIR)));
    317 
    318 	reparent_p = (fdvp != tdvp);
    319 	KASSERT(reparent_p == (VTOI(fdvp)->i_number != VTOI(tdvp)->i_number));
    320 
    321 	/*
    322 	 * Commence hacking of the data on disk.
    323 	 */
    324 
    325 	/*
    326 	 * 1) Bump link count while we're moving stuff
    327 	 *    around.  If we crash somewhere before
    328 	 *    completing our work, the link count
    329 	 *    may be wrong, but correctable.
    330 	 */
    331 
    332 	KASSERT((nlink_t)VTOI(fvp)->i_e2fs_nlink < LINK_MAX);
    333 	VTOI(fvp)->i_e2fs_nlink++;
    334 	VTOI(fvp)->i_flag |= IN_CHANGE;
    335 	error = ext2fs_update(fvp, NULL, NULL, UPDATE_WAIT);
    336 	if (error)
    337 		goto whymustithurtsomuch;
    338 
    339 	/*
    340 	 * 2) If target doesn't exist, link the target
    341 	 *    to the source and unlink the source.
    342 	 *    Otherwise, rewrite the target directory
    343 	 *    entry to reference the source inode and
    344 	 *    expunge the original entry's existence.
    345 	 */
    346 
    347 	if (tvp == NULL) {
    348 		/*
    349 		 * Account for ".." in new directory.
    350 		 * When source and destination have the same
    351 		 * parent we don't fool with the link count.
    352 		 */
    353 		if (directory_p && reparent_p) {
    354 			if ((nlink_t)VTOI(tdvp)->i_e2fs_nlink >= LINK_MAX) {
    355 				error = EMLINK;
    356 				goto whymustithurtsomuch;
    357 			}
    358 			KASSERT((nlink_t)VTOI(tdvp)->i_e2fs_nlink < LINK_MAX);
    359 			VTOI(tdvp)->i_e2fs_nlink++;
    360 			VTOI(tdvp)->i_flag |= IN_CHANGE;
    361 			error = ext2fs_update(tdvp, NULL, NULL, UPDATE_WAIT);
    362 			if (error) {
    363 				/*
    364 				 * Link count update didn't take --
    365 				 * back out the in-memory link count.
    366 				 */
    367 				KASSERT(0 < VTOI(tdvp)->i_e2fs_nlink);
    368 				VTOI(tdvp)->i_e2fs_nlink--;
    369 				VTOI(tdvp)->i_flag |= IN_CHANGE;
    370 				goto whymustithurtsomuch;
    371 			}
    372 		}
    373 
    374 		error = ext2fs_direnter(VTOI(fvp), tdvp, tulr, tcnp);
    375 		if (error) {
    376 			if (directory_p && reparent_p) {
    377 				/*
    378 				 * Directory update didn't take, but
    379 				 * the link count update did -- back
    380 				 * out the in-memory link count and the
    381 				 * on-disk link count.
    382 				 */
    383 				KASSERT(0 < VTOI(tdvp)->i_e2fs_nlink);
    384 				VTOI(tdvp)->i_e2fs_nlink--;
    385 				VTOI(tdvp)->i_flag |= IN_CHANGE;
    386 				(void)ext2fs_update(tdvp, NULL, NULL,
    387 				    UPDATE_WAIT);
    388 			}
    389 			goto whymustithurtsomuch;
    390 		}
    391 	} else {
    392 		if (directory_p)
    393 			/* XXX WTF?  Why purge here?  Why not purge others?  */
    394 			cache_purge(tdvp);
    395 
    396 		/*
    397 		 * Make the target directory's entry for tcnp point at
    398 		 * the source node.
    399 		 */
    400 		error = ext2fs_dirrewrite(VTOI(tdvp), tulr, VTOI(fvp), tcnp);
    401 		if (error)
    402 			goto whymustithurtsomuch;
    403 
    404 		/*
    405 		 * If the source and target are directories, and the
    406 		 * target is in the same directory as the source,
    407 		 * decrement the link count of the common parent
    408 		 * directory, since we are removing the target from
    409 		 * that directory.
    410 		 */
    411 		if (directory_p && !reparent_p) {
    412 			KASSERT(fdvp == tdvp);
    413 			/* XXX check, don't kassert */
    414 			KASSERT(0 < VTOI(tdvp)->i_e2fs_nlink);
    415 			VTOI(tdvp)->i_e2fs_nlink--;
    416 			VTOI(tdvp)->i_flag |= IN_CHANGE;
    417 		}
    418 
    419 		/*
    420 		 * Adjust the link count of the target to
    421 		 * reflect the dirrewrite above.  If this is
    422 		 * a directory it is empty and there are
    423 		 * no links to it, so we can squash the inode and
    424 		 * any space associated with it.  We disallowed
    425 		 * renaming over top of a directory with links to
    426 		 * it above, as the remaining link would point to
    427 		 * a directory without "." or ".." entries.
    428 		 */
    429 		/* XXX check, don't kassert */
    430 		KASSERT(0 < VTOI(tvp)->i_e2fs_nlink);
    431 		VTOI(tvp)->i_e2fs_nlink--;
    432 		if (directory_p) {
    433 			/*
    434 			 * XXX The ext2fs_dirempty call earlier does
    435 			 * not guarantee anything about nlink.
    436 			 */
    437 			if (VTOI(tvp)->i_e2fs_nlink != 1)
    438 				ufs_dirbad(VTOI(tvp), (doff_t)0,
    439 				    "hard-linked directory");
    440 			VTOI(tvp)->i_e2fs_nlink = 0;
    441 			error = ext2fs_truncate(tvp, (off_t)0, IO_SYNC, cred);
    442 #if 0			/* XXX This branch was not in ext2fs_rename!  */
    443 			if (error)
    444 				goto whymustithurtsomuch;
    445 #endif
    446 		}
    447 		/*
    448 		 * XXX Why is this here, and not above the preceding
    449 		 * conditional?
    450 		 */
    451 		VTOI(tvp)->i_flag |= IN_CHANGE;
    452 	}
    453 
    454 	/*
    455 	 * If the source is a directory with a new parent, the link
    456 	 * count of the old parent directory must be decremented and
    457 	 * ".." set to point to the new parent.
    458 	 */
    459 	if (directory_p && reparent_p) {
    460 		error = ext2fs_rename_replace_dotdot(fvp, fdvp, tdvp, cred);
    461 		if (error)
    462 			goto whymustithurtsomuch;
    463 
    464 		/* XXX WTF?  Why purge here?  Why not purge others?  */
    465 		cache_purge(fdvp);
    466 	}
    467 
    468 	/*
    469 	 * 3) Unlink the source.
    470 	 */
    471 
    472 	/*
    473 	 * ext2fs_direnter may compact the directory in the process of
    474 	 * inserting a new entry.  That may invalidate fulr, which we
    475 	 * need in order to remove the old entry.  In that case, we
    476 	 * need to recalculate what fulr should be.
    477 	 */
    478 	if (!reparent_p && (tvp == NULL) &&
    479 	    ext2fs_rename_ulr_overlap_p(fulr, tulr)) {
    480 		error = ext2fs_rename_recalculate_fulr(fdvp, fulr, tulr, fcnp);
    481 #if 0				/* XXX */
    482 		if (error)	/* XXX Try to back out changes?  */
    483 			goto whymustithurtsomuch;
    484 #endif
    485 	}
    486 
    487 	error = ext2fs_dirremove(fdvp, fulr, fcnp);
    488 	if (error)
    489 		goto whymustithurtsomuch;
    490 
    491 	/*
    492 	 * XXX Perhaps this should go at the top, in case the file
    493 	 * system is modified but incompletely so because of an
    494 	 * intermediate error.
    495 	 */
    496 	genfs_rename_knote(fdvp, fvp, tdvp, tvp,
    497 	    ((tvp != NULL) && (VTOI(tvp)->i_e2fs_nlink == 0)));
    498 #if 0				/* XXX */
    499 	genfs_rename_cache_purge(fdvp, fvp, tdvp, tvp);
    500 #endif
    501 
    502 whymustithurtsomuch:
    503 	KASSERT(0 < VTOI(fvp)->i_e2fs_nlink);
    504 	VTOI(fvp)->i_e2fs_nlink--;
    505 	VTOI(fvp)->i_flag |= IN_CHANGE;
    506 	return error;
    507 }
    508 
    509 /*
    510  * ext2fs_rename_ulr_overlap_p: True iff tulr overlaps with fulr so
    511  * that entering a directory entry at tulr may move fulr.
    512  */
    513 static bool
    514 ext2fs_rename_ulr_overlap_p(const struct ufs_lookup_results *fulr,
    515     const struct ufs_lookup_results *tulr)
    516 {
    517 	doff_t from_prev_start, from_prev_end, to_start, to_end;
    518 
    519 	KASSERT(fulr != NULL);
    520 	KASSERT(tulr != NULL);
    521 	KASSERT(fulr != tulr);
    522 
    523 	/*
    524 	 * fulr is from a DELETE lookup, so fulr->ulr_count is the size
    525 	 * of the preceding entry (d_reclen).
    526 	 */
    527 	from_prev_end = fulr->ulr_offset;
    528 	KASSERT(fulr->ulr_count <= from_prev_end);
    529 	from_prev_start = (from_prev_end - fulr->ulr_count);
    530 
    531 	/*
    532 	 * tulr is from a RENAME lookup, so tulr->ulr_count is the size
    533 	 * of the free space for an entry that we are about to fill.
    534 	 */
    535 	to_start = tulr->ulr_offset;
    536 	KASSERT(tulr->ulr_count < (EXT2FS_MAXDIRSIZE - to_start));
    537 	to_end = (to_start + tulr->ulr_count);
    538 
    539 	return
    540 	    (((to_start <= from_prev_start) && (from_prev_start < to_end)) ||
    541 		((to_start <= from_prev_end) && (from_prev_end < to_end)));
    542 }
    543 
    544 /*
    545  * ext2fs_rename_recalculate_fulr: If we have just entered a directory
    546  * into dvp at tulr, and we were about to remove one at fulr for an
    547  * entry named fcnp, fulr may be invalid.  So, if necessary,
    548  * recalculate it.
    549  */
    550 static int
    551 ext2fs_rename_recalculate_fulr(struct vnode *dvp,
    552     struct ufs_lookup_results *fulr, const struct ufs_lookup_results *tulr,
    553     const struct componentname *fcnp)
    554 {
    555 	struct mount *mp;
    556 	struct ufsmount *ump;
    557 	/* XXX int is a silly type for this; blame ufsmount::um_dirblksiz.  */
    558 	int dirblksiz;
    559 	doff_t search_start, search_end;
    560 	doff_t offset;		/* Offset of entry we're examining.  */
    561 	struct buf *bp;		/* I/O block we're examining.  */
    562 	char *dirbuf;		/* Pointer into directory at search_start.  */
    563 	struct ext2fs_direct *ep; /* Pointer to the entry we're examining.  */
    564 	/* XXX direct::d_reclen is 16-bit;
    565 	 * ufs_lookup_results::ulr_reclen is 32-bit.  Blah.  */
    566 	uint32_t reclen;	/* Length of the entry we're examining.  */
    567 	uint32_t prev_reclen;	/* Length of the preceding entry.  */
    568 	int error;
    569 
    570 	KASSERT(dvp != NULL);
    571 	KASSERT(dvp->v_mount != NULL);
    572 	KASSERT(VTOI(dvp) != NULL);
    573 	KASSERT(fulr != NULL);
    574 	KASSERT(tulr != NULL);
    575 	KASSERT(fulr != tulr);
    576 	KASSERT(ext2fs_rename_ulr_overlap_p(fulr, tulr));
    577 
    578 	mp = dvp->v_mount;
    579 	ump = VFSTOUFS(mp);
    580 	KASSERT(ump != NULL);
    581 	KASSERT(ump == VTOI(dvp)->i_ump);
    582 
    583 	dirblksiz = ump->um_dirblksiz;
    584 	KASSERT(0 < dirblksiz);
    585 	KASSERT((dirblksiz & (dirblksiz - 1)) == 0);
    586 
    587 	/* A directory block may not span across multiple I/O blocks.  */
    588 	KASSERT(dirblksiz <= mp->mnt_stat.f_iosize);
    589 
    590 	/* Find the bounds of the search.  */
    591 	search_start = tulr->ulr_offset;
    592 	KASSERT(fulr->ulr_reclen < (EXT2FS_MAXDIRSIZE - fulr->ulr_offset));
    593 	search_end = (fulr->ulr_offset + fulr->ulr_reclen);
    594 
    595 	/* Compaction must happen only within a directory block. (*)  */
    596 	KASSERT(search_start <= search_end);
    597 	KASSERT((search_end - (search_start &~ (dirblksiz - 1))) <= dirblksiz);
    598 
    599 	dirbuf = NULL;
    600 	bp = NULL;
    601 	error = ext2fs_blkatoff(dvp, (off_t)search_start, &dirbuf, &bp);
    602 	if (error)
    603 		return error;
    604 	KASSERT(dirbuf != NULL);
    605 	KASSERT(bp != NULL);
    606 
    607 	/*
    608 	 * Guarantee we sha'n't go past the end of the buffer we got.
    609 	 * dirbuf is bp->b_data + (search_start & (iosize - 1)), and
    610 	 * the valid range is [bp->b_data, bp->b_data + bp->b_bcount).
    611 	 */
    612 	KASSERT((search_end - search_start) <=
    613 	    (bp->b_bcount - (search_start & (mp->mnt_stat.f_iosize - 1))));
    614 
    615 	prev_reclen = fulr->ulr_count;
    616 	offset = search_start;
    617 
    618 	/*
    619 	 * Search from search_start to search_end for the entry matching
    620 	 * fcnp, which must be there because we found it before and it
    621 	 * should only at most have moved earlier.
    622 	 */
    623 	for (;;) {
    624 		KASSERT(search_start <= offset);
    625 		KASSERT(offset < search_end);
    626 
    627 		/*
    628 		 * Examine the directory entry at offset.
    629 		 */
    630 		ep = (struct ext2fs_direct *)
    631 		    (dirbuf + (offset - search_start));
    632 		reclen = fs2h16(ep->e2d_reclen);
    633 
    634 		if (ep->e2d_ino == 0)
    635 			goto next;	/* Entry is unused.  */
    636 
    637 		if (fs2h32(ep->e2d_ino) == UFS_WINO)
    638 			goto next;	/* Entry is whiteout.  */
    639 
    640 		if (fcnp->cn_namelen != ep->e2d_namlen)
    641 			goto next;	/* Wrong name length.  */
    642 
    643 		if (memcmp(ep->e2d_name, fcnp->cn_nameptr, fcnp->cn_namelen))
    644 			goto next;	/* Wrong name.  */
    645 
    646 		/* Got it!  */
    647 		break;
    648 
    649 next:
    650 		if (! ((reclen < search_end) &&
    651 			(offset < (search_end - reclen)))) {
    652 			brelse(bp, 0);
    653 			return EIO;	/* XXX Panic?  What?  */
    654 		}
    655 
    656 		/* We may not move past the search end.  */
    657 		KASSERT(reclen < search_end);
    658 		KASSERT(offset < (search_end - reclen));
    659 
    660 		/*
    661 		 * We may not move across a directory block boundary;
    662 		 * see (*) above.
    663 		 */
    664 		KASSERT((offset &~ (dirblksiz - 1)) ==
    665 		    ((offset + reclen) &~ (dirblksiz - 1)));
    666 
    667 		prev_reclen = reclen;
    668 		offset += reclen;
    669 	}
    670 
    671 	/*
    672 	 * Found the entry.  Record where.
    673 	 */
    674 	fulr->ulr_offset = offset;
    675 	fulr->ulr_reclen = reclen;
    676 
    677 	/*
    678 	 * Record the preceding record length, but not if we're at the
    679 	 * start of a directory block.
    680 	 */
    681 	fulr->ulr_count = ((offset & (dirblksiz - 1))? prev_reclen : 0);
    682 
    683 	brelse(bp, 0);
    684 	return 0;
    685 }
    686 
    687 /*
    688  * ext2fs_gro_remove: Rename an object over another link to itself,
    689  * effectively removing just the original link.
    690  */
    691 static int
    692 ext2fs_gro_remove(struct mount *mp, kauth_cred_t cred,
    693     struct vnode *dvp, struct componentname *cnp, void *de, struct vnode *vp)
    694 {
    695 	struct ufs_lookup_results *ulr = de;
    696 	int error;
    697 
    698 	(void)mp;
    699 	KASSERT(mp != NULL);
    700 	KASSERT(dvp != NULL);
    701 	KASSERT(cnp != NULL);
    702 	KASSERT(ulr != NULL);
    703 	KASSERT(vp != NULL);
    704 	KASSERT(dvp != vp);
    705 	KASSERT(dvp->v_mount == mp);
    706 	KASSERT(vp->v_mount == mp);
    707 	KASSERT(dvp->v_type == VDIR);
    708 	KASSERT(vp->v_type != VDIR);
    709 	KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
    710 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    711 
    712 	error = ext2fs_dirremove(dvp, ulr, cnp);
    713 	if (error)
    714 		return error;
    715 
    716 	KASSERT(0 < VTOI(vp)->i_e2fs_nlink);
    717 	VTOI(vp)->i_e2fs_nlink--;
    718 	VTOI(vp)->i_flag |= IN_CHANGE;
    719 
    720 	VN_KNOTE(dvp, NOTE_WRITE);
    721 	VN_KNOTE(vp, (VTOI(vp)->i_e2fs_nlink? NOTE_LINK : NOTE_DELETE));
    722 
    723 	return 0;
    724 }
    725 
    726 /*
    727  * ext2fs_gro_lookup: Look up and save the lookup results.
    728  */
    729 static int
    730 ext2fs_gro_lookup(struct mount *mp, struct vnode *dvp,
    731     struct componentname *cnp, void *de_ret, struct vnode **vp_ret)
    732 {
    733 	struct ufs_lookup_results *ulr_ret = de_ret;
    734 	struct vnode *vp;
    735 	int error;
    736 
    737 	(void)mp;
    738 	KASSERT(mp != NULL);
    739 	KASSERT(dvp != NULL);
    740 	KASSERT(cnp != NULL);
    741 	KASSERT(ulr_ret != NULL);
    742 	KASSERT(vp_ret != NULL);
    743 	KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
    744 
    745 	/* Kludge cargo-culted from dholland's ufs_rename.  */
    746 	cnp->cn_flags &=~ MODMASK;
    747 	cnp->cn_flags |= (LOCKPARENT | LOCKLEAF);
    748 
    749 	error = relookup(dvp, &vp, cnp, 0 /* dummy */);
    750 	if ((error == 0) && (vp == NULL)) {
    751 		error = ENOENT;
    752 		goto out;
    753 	} else if (error) {
    754 		return error;
    755 	}
    756 
    757 	/*
    758 	 * Thanks to VFS insanity, relookup locks vp, which screws us
    759 	 * in various ways.
    760 	 */
    761 	KASSERT(vp != NULL);
    762 	VOP_UNLOCK(vp);
    763 
    764 out:	*ulr_ret = VTOI(dvp)->i_crap;
    765 	*vp_ret = vp;
    766 	return error;
    767 }
    768 
    769 /*
    770  * ext2fs_rmdired_p: Check whether the directory vp has been rmdired.
    771  *
    772  * vp must be locked and referenced.
    773  */
    774 static bool
    775 ext2fs_rmdired_p(struct vnode *vp)
    776 {
    777 
    778 	KASSERT(vp != NULL);
    779 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    780 	KASSERT(vp->v_type == VDIR);
    781 
    782 	/* XXX Is this correct?  */
    783 	return (ext2fs_size(VTOI(vp)) == 0);
    784 }
    785 
    786 /*
    787  * ext2fs_gro_genealogy: Analyze the genealogy of the source and target
    788  * directories.
    789  */
    790 static int
    791 ext2fs_gro_genealogy(struct mount *mp, kauth_cred_t cred,
    792     struct vnode *fdvp, struct vnode *tdvp,
    793     struct vnode **intermediate_node_ret)
    794 {
    795 	struct vnode *vp, *dvp;
    796 	ino_t dotdot_ino = -1;	/* XXX gcc 4.8.3: maybe-uninitialized */
    797 	int error;
    798 
    799 	KASSERT(mp != NULL);
    800 	KASSERT(fdvp != NULL);
    801 	KASSERT(tdvp != NULL);
    802 	KASSERT(fdvp != tdvp);
    803 	KASSERT(intermediate_node_ret != NULL);
    804 	KASSERT(fdvp->v_mount == mp);
    805 	KASSERT(tdvp->v_mount == mp);
    806 	KASSERT(fdvp->v_type == VDIR);
    807 	KASSERT(tdvp->v_type == VDIR);
    808 
    809 	/*
    810 	 * We need to provisionally lock tdvp to keep rmdir from
    811 	 * deleting it -- or any ancestor -- at an inopportune moment.
    812 	 */
    813 	error = ext2fs_gro_lock_directory(mp, tdvp);
    814 	if (error)
    815 		return error;
    816 
    817 	vp = tdvp;
    818 	vref(vp);
    819 
    820 	for (;;) {
    821 		KASSERT(vp != NULL);
    822 		KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    823 		KASSERT(vp->v_mount == mp);
    824 		KASSERT(vp->v_type == VDIR);
    825 		KASSERT(!ext2fs_rmdired_p(vp));
    826 
    827 		/* Did we hit the root without finding fdvp?  */
    828 		if (VTOI(vp)->i_number == UFS_ROOTINO) {
    829 			vput(vp);
    830 			*intermediate_node_ret = NULL;
    831 			return 0;
    832 		}
    833 
    834 		error = ext2fs_read_dotdot(vp, cred, &dotdot_ino);
    835 		if (error) {
    836 			vput(vp);
    837 			return error;
    838 		}
    839 
    840 		/* Did we find that fdvp is an ancestor of tdvp?  */
    841 		if (VTOI(fdvp)->i_number == dotdot_ino) {
    842 			/* Unlock vp, but keep it referenced.  */
    843 			VOP_UNLOCK(vp);
    844 			*intermediate_node_ret = vp;
    845 			return 0;
    846 		}
    847 
    848 		/* Neither -- keep ascending the family tree.  */
    849 
    850 		/*
    851 		 * Unlock vp so that we can lock the parent, but keep
    852 		 * vp referenced until after we have found the parent,
    853 		 * so that dotdot_ino will not be recycled.
    854 		 *
    855 		 * XXX This guarantees that vp's inode number will not
    856 		 * be recycled, but why can't dotdot_ino be recycled?
    857 		 */
    858 		VOP_UNLOCK(vp);
    859 		error = VFS_VGET(mp, dotdot_ino, &dvp);
    860 		vrele(vp);
    861 		if (error)
    862 			return error;
    863 
    864 		KASSERT(dvp != NULL);
    865 		KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
    866 		vp = dvp;
    867 
    868 		if (vp->v_type != VDIR) {
    869 			/*
    870 			 * XXX Panic?  Print a warning?  Can this
    871 			 * happen if we lose the race I suspect to
    872 			 * exist above, and the `..' inode number has
    873 			 * been recycled?
    874 			 */
    875 			vput(vp);
    876 			return ENOTDIR;
    877 		}
    878 
    879 		if (ext2fs_rmdired_p(vp)) {
    880 			vput(vp);
    881 			return ENOENT;
    882 		}
    883 	}
    884 }
    885 
    886 /*
    887  * ext2fs_read_dotdot: Store in *ino_ret the inode number of the parent
    888  * of the directory vp.
    889  */
    890 static int
    891 ext2fs_read_dotdot(struct vnode *vp, kauth_cred_t cred, ino_t *ino_ret)
    892 {
    893 	struct ext2fs_dirtemplate dirbuf;
    894 	int error;
    895 
    896 	KASSERT(vp != NULL);
    897 	KASSERT(ino_ret != NULL);
    898 	KASSERT(vp->v_type == VDIR);
    899 
    900 	error = vn_rdwr(UIO_READ, vp, &dirbuf, sizeof dirbuf, (off_t)0,
    901 	    UIO_SYSSPACE, IO_NODELOCKED, cred, NULL, NULL);
    902 	if (error)
    903 		return error;
    904 
    905 	if (dirbuf.dotdot_namlen != 2 ||
    906 	    dirbuf.dotdot_name[0] != '.' ||
    907 	    dirbuf.dotdot_name[1] != '.')
    908 		/* XXX Panic?  Print warning?  */
    909 		return ENOTDIR;
    910 
    911 	*ino_ret = fs2h32(dirbuf.dotdot_ino);
    912 	return 0;
    913 }
    914 
    915 /*
    916  * ext2fs_rename_replace_dotdot: Change the target of the `..' entry of
    917  * the directory vp from fdvp to tdvp.
    918  */
    919 static int
    920 ext2fs_rename_replace_dotdot(struct vnode *vp,
    921     struct vnode *fdvp, struct vnode *tdvp,
    922     kauth_cred_t cred)
    923 {
    924 	struct ext2fs_dirtemplate dirbuf;
    925 	int error;
    926 
    927 	/* XXX Does it make sense to do this before the sanity checks below?  */
    928 	KASSERT(0 < VTOI(fdvp)->i_e2fs_nlink);
    929 	VTOI(fdvp)->i_e2fs_nlink--;
    930 	VTOI(fdvp)->i_flag |= IN_CHANGE;
    931 
    932 	error = vn_rdwr(UIO_READ, vp, &dirbuf, sizeof dirbuf, (off_t)0,
    933 	    UIO_SYSSPACE, IO_NODELOCKED, cred, NULL, NULL);
    934 	if (error)
    935 		return error;
    936 
    937 	if (dirbuf.dotdot_namlen != 2 ||
    938 	    dirbuf.dotdot_name[0] != '.' ||
    939 	    dirbuf.dotdot_name[1] != '.') {
    940 		ufs_dirbad(VTOI(vp), (doff_t)12, "bad `..' entry");
    941 		return 0;
    942 	}
    943 
    944 	if (fs2h32(dirbuf.dotdot_ino) != VTOI(fdvp)->i_number) {
    945 		ufs_dirbad(VTOI(vp), (doff_t)12,
    946 		    "`..' does not point at parent");
    947 		return 0;
    948 	}
    949 
    950 	dirbuf.dotdot_ino = h2fs32(VTOI(tdvp)->i_number);
    951 	/* XXX WTF?  Why not check error?  */
    952 	(void)vn_rdwr(UIO_WRITE, vp, &dirbuf, sizeof dirbuf, (off_t)0,
    953 	    UIO_SYSSPACE, (IO_NODELOCKED | IO_SYNC), cred, NULL, NULL);
    954 
    955 	return 0;
    956 }
    957 
    958 /*
    959  * ext2fs_gro_lock_directory: Lock the directory vp, but fail if it has
    960  * been rmdir'd.
    961  */
    962 static int
    963 ext2fs_gro_lock_directory(struct mount *mp, struct vnode *vp)
    964 {
    965 
    966 	(void)mp;
    967 	KASSERT(mp != NULL);
    968 	KASSERT(vp != NULL);
    969 	KASSERT(vp->v_mount == mp);
    970 
    971 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    972 
    973 	if (ext2fs_rmdired_p(vp)) {
    974 		VOP_UNLOCK(vp);
    975 		return ENOENT;
    976 	}
    977 
    978 	return 0;
    979 }
    980 
    981 static const struct genfs_rename_ops ext2fs_genfs_rename_ops = {
    982 	.gro_directory_empty_p		= ext2fs_gro_directory_empty_p,
    983 	.gro_rename_check_possible	= ext2fs_gro_rename_check_possible,
    984 	.gro_rename_check_permitted	= ext2fs_gro_rename_check_permitted,
    985 	.gro_remove_check_possible	= ext2fs_gro_remove_check_possible,
    986 	.gro_remove_check_permitted	= ext2fs_gro_remove_check_permitted,
    987 	.gro_rename			= ext2fs_gro_rename,
    988 	.gro_remove			= ext2fs_gro_remove,
    989 	.gro_lookup			= ext2fs_gro_lookup,
    990 	.gro_genealogy			= ext2fs_gro_genealogy,
    991 	.gro_lock_directory		= ext2fs_gro_lock_directory,
    992 };
    993