Home | History | Annotate | Line # | Download | only in tmpfs
tmpfs_vnops.c revision 1.76
      1 /*	$NetBSD: tmpfs_vnops.c,v 1.76 2011/01/13 13:35:12 pooka Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2005, 2006, 2007 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
      9  * 2005 program.
     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  *
     20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     30  * POSSIBILITY OF SUCH DAMAGE.
     31  */
     32 
     33 /*
     34  * tmpfs vnode interface.
     35  */
     36 
     37 #include <sys/cdefs.h>
     38 __KERNEL_RCSID(0, "$NetBSD: tmpfs_vnops.c,v 1.76 2011/01/13 13:35:12 pooka Exp $");
     39 
     40 #include <sys/param.h>
     41 #include <sys/dirent.h>
     42 #include <sys/fcntl.h>
     43 #include <sys/event.h>
     44 #include <sys/malloc.h>
     45 #include <sys/namei.h>
     46 #include <sys/proc.h>
     47 #include <sys/stat.h>
     48 #include <sys/uio.h>
     49 #include <sys/unistd.h>
     50 #include <sys/vnode.h>
     51 #include <sys/lockf.h>
     52 #include <sys/kauth.h>
     53 
     54 #include <uvm/uvm.h>
     55 
     56 #include <miscfs/fifofs/fifo.h>
     57 #include <miscfs/genfs/genfs.h>
     58 #include <fs/tmpfs/tmpfs_vnops.h>
     59 #include <fs/tmpfs/tmpfs.h>
     60 
     61 /* --------------------------------------------------------------------- */
     62 
     63 /*
     64  * vnode operations vector used for files stored in a tmpfs file system.
     65  */
     66 int (**tmpfs_vnodeop_p)(void *);
     67 const struct vnodeopv_entry_desc tmpfs_vnodeop_entries[] = {
     68 	{ &vop_default_desc,		vn_default_error },
     69 	{ &vop_lookup_desc,		tmpfs_lookup },
     70 	{ &vop_create_desc,		tmpfs_create },
     71 	{ &vop_mknod_desc,		tmpfs_mknod },
     72 	{ &vop_open_desc,		tmpfs_open },
     73 	{ &vop_close_desc,		tmpfs_close },
     74 	{ &vop_access_desc,		tmpfs_access },
     75 	{ &vop_getattr_desc,		tmpfs_getattr },
     76 	{ &vop_setattr_desc,		tmpfs_setattr },
     77 	{ &vop_read_desc,		tmpfs_read },
     78 	{ &vop_write_desc,		tmpfs_write },
     79 	{ &vop_ioctl_desc,		tmpfs_ioctl },
     80 	{ &vop_fcntl_desc,		tmpfs_fcntl },
     81 	{ &vop_poll_desc,		tmpfs_poll },
     82 	{ &vop_kqfilter_desc,		tmpfs_kqfilter },
     83 	{ &vop_revoke_desc,		tmpfs_revoke },
     84 	{ &vop_mmap_desc,		tmpfs_mmap },
     85 	{ &vop_fsync_desc,		tmpfs_fsync },
     86 	{ &vop_seek_desc,		tmpfs_seek },
     87 	{ &vop_remove_desc,		tmpfs_remove },
     88 	{ &vop_link_desc,		tmpfs_link },
     89 	{ &vop_rename_desc,		tmpfs_rename },
     90 	{ &vop_mkdir_desc,		tmpfs_mkdir },
     91 	{ &vop_rmdir_desc,		tmpfs_rmdir },
     92 	{ &vop_symlink_desc,		tmpfs_symlink },
     93 	{ &vop_readdir_desc,		tmpfs_readdir },
     94 	{ &vop_readlink_desc,		tmpfs_readlink },
     95 	{ &vop_abortop_desc,		tmpfs_abortop },
     96 	{ &vop_inactive_desc,		tmpfs_inactive },
     97 	{ &vop_reclaim_desc,		tmpfs_reclaim },
     98 	{ &vop_lock_desc,		tmpfs_lock },
     99 	{ &vop_unlock_desc,		tmpfs_unlock },
    100 	{ &vop_bmap_desc,		tmpfs_bmap },
    101 	{ &vop_strategy_desc,		tmpfs_strategy },
    102 	{ &vop_print_desc,		tmpfs_print },
    103 	{ &vop_pathconf_desc,		tmpfs_pathconf },
    104 	{ &vop_islocked_desc,		tmpfs_islocked },
    105 	{ &vop_advlock_desc,		tmpfs_advlock },
    106 	{ &vop_bwrite_desc,		tmpfs_bwrite },
    107 	{ &vop_getpages_desc,		tmpfs_getpages },
    108 	{ &vop_putpages_desc,		tmpfs_putpages },
    109 #if TMPFS_WHITEOUT
    110 	{ &vop_whiteout_desc,		tmpfs_whiteout },
    111 #endif
    112 	{ NULL, NULL }
    113 };
    114 const struct vnodeopv_desc tmpfs_vnodeop_opv_desc =
    115 	{ &tmpfs_vnodeop_p, tmpfs_vnodeop_entries };
    116 
    117 /*
    118  * tmpfs_lookup: lookup routine.
    119  *
    120  * Arguments: dvp (directory being searched), vpp (result),
    121  * cnp (component name - path).
    122  *
    123  * => Caller holds a reference and lock on dvp.
    124  * => We return looked-up vnode (vpp) locked, with a reference held.
    125  */
    126 int
    127 tmpfs_lookup(void *v)
    128 {
    129 	struct vop_lookup_args /* {
    130 		struct vnode *a_dvp;
    131 		struct vnode **a_vpp;
    132 		struct componentname *a_cnp;
    133 	} */ *ap = v;
    134 	struct vnode *dvp = ap->a_dvp, **vpp = ap->a_vpp;
    135 	struct componentname *cnp = ap->a_cnp;
    136 	struct tmpfs_dirent *de;
    137 	struct tmpfs_node *dnode;
    138 	int error;
    139 
    140 	KASSERT(VOP_ISLOCKED(dvp));
    141 
    142 	dnode = VP_TO_TMPFS_DIR(dvp);
    143 	*vpp = NULL;
    144 
    145 	/* Check accessibility of requested node as a first step. */
    146 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
    147 	if (error != 0)
    148 		goto out;
    149 
    150 	/*
    151 	 * If requesting the last path component on a read-only file system
    152 	 * with a write operation, deny it.
    153 	 */
    154 	if ((cnp->cn_flags & ISLASTCN) &&
    155 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
    156 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    157 		error = EROFS;
    158 		goto out;
    159 	}
    160 
    161 	/*
    162 	 * Avoid doing a linear scan of the directory if the requested
    163 	 * directory/name couple is already in the cache.
    164 	 */
    165 	error = cache_lookup(dvp, vpp, cnp);
    166 	if (error >= 0)
    167 		goto out;
    168 
    169 	/* We cannot be requesting the parent directory of the root node. */
    170 	KASSERT(IMPLIES(dnode->tn_type == VDIR &&
    171 	    dnode->tn_spec.tn_dir.tn_parent == dnode,
    172 	    !(cnp->cn_flags & ISDOTDOT)));
    173 
    174 	if (cnp->cn_flags & ISDOTDOT) {
    175 		VOP_UNLOCK(dvp);
    176 
    177 		/* Allocate a new vnode on the matching entry. */
    178 		error = tmpfs_alloc_vp(dvp->v_mount,
    179 		    dnode->tn_spec.tn_dir.tn_parent, vpp);
    180 
    181 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
    182 		goto done;
    183 
    184 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
    185 		if ((cnp->cn_flags & ISLASTCN) &&
    186 		    (cnp->cn_nameiop == RENAME)) {
    187 			error = EISDIR;
    188 			goto out;
    189 		}
    190 		vref(dvp);
    191 		*vpp = dvp;
    192 		error = 0;
    193 		goto done;
    194 	}
    195 
    196 	de = tmpfs_dir_lookup(dnode, cnp);
    197 	if (de == NULL || de->td_node == TMPFS_NODE_WHITEOUT) {
    198 		/*
    199 		 * The entry was not found in the directory.  This is valid
    200 		 * if we are creating or renaming an entry and are working
    201 		 * on the last component of the path name.
    202 		 */
    203 		if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == CREATE ||
    204 		    cnp->cn_nameiop == RENAME)) {
    205 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
    206 			if (error) {
    207 				goto out;
    208 			}
    209 			error = EJUSTRETURN;
    210 		} else {
    211 			error = ENOENT;
    212 		}
    213 		if (de) {
    214 			KASSERT(de->td_node == TMPFS_NODE_WHITEOUT);
    215 			cnp->cn_flags |= ISWHITEOUT;
    216 		}
    217 	} else {
    218 		struct tmpfs_node *tnode = de->td_node;
    219 
    220 		/*
    221 		 * If we are not at the last path component and found a
    222 		 * non-directory or non-link entry (which may itself be
    223 		 * pointing to a directory), raise an error.
    224 		 */
    225 		if ((tnode->tn_type != VDIR && tnode->tn_type != VLNK) &&
    226 		    (cnp->cn_flags & ISLASTCN) == 0) {
    227 			error = ENOTDIR;
    228 			goto out;
    229 		}
    230 
    231 		/* Check permissions. */
    232 		if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == DELETE ||
    233 		    cnp->cn_nameiop == RENAME)) {
    234 			kauth_action_t action = 0;
    235 
    236 			/* This is the file-system's decision. */
    237 			if ((dnode->tn_mode & S_ISTXT) != 0 &&
    238 			    kauth_cred_geteuid(cnp->cn_cred) != dnode->tn_uid &&
    239 			    kauth_cred_geteuid(cnp->cn_cred) != tnode->tn_uid)
    240 				error = EPERM;
    241 			else
    242 				error = 0;
    243 
    244 			/* Only bother if we are not already failing it. */
    245 			if (!error) {
    246 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
    247 			}
    248 
    249 			if (cnp->cn_nameiop == DELETE) {
    250 				action |= KAUTH_VNODE_DELETE;
    251 			} else {
    252 				KASSERT(cnp->cn_nameiop == RENAME);
    253 				action |= KAUTH_VNODE_RENAME;
    254 			}
    255 			error = kauth_authorize_vnode(cnp->cn_cred,
    256 			    action, *vpp, dvp, error);
    257 			if (error) {
    258 				goto out;
    259 			}
    260 		}
    261 		/* Allocate a new vnode on the matching entry. */
    262 		error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp);
    263 	}
    264 done:
    265 	/*
    266 	 * Store the result of this lookup in the cache.  Avoid this if the
    267 	 * request was for creation, as it does not improve timings on
    268 	 * emprical tests.
    269 	 */
    270 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
    271 	    (cnp->cn_flags & ISDOTDOT) == 0)
    272 		cache_enter(dvp, *vpp, cnp);
    273 
    274 out:
    275 	KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
    276 	KASSERT(VOP_ISLOCKED(dvp));
    277 
    278 	return error;
    279 }
    280 
    281 int
    282 tmpfs_create(void *v)
    283 {
    284 	struct vnode *dvp = ((struct vop_create_args *)v)->a_dvp;
    285 	struct vnode **vpp = ((struct vop_create_args *)v)->a_vpp;
    286 	struct componentname *cnp = ((struct vop_create_args *)v)->a_cnp;
    287 	struct vattr *vap = ((struct vop_create_args *)v)->a_vap;
    288 
    289 	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
    290 
    291 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    292 }
    293 /* --------------------------------------------------------------------- */
    294 
    295 int
    296 tmpfs_mknod(void *v)
    297 {
    298 	struct vnode *dvp = ((struct vop_mknod_args *)v)->a_dvp;
    299 	struct vnode **vpp = ((struct vop_mknod_args *)v)->a_vpp;
    300 	struct componentname *cnp = ((struct vop_mknod_args *)v)->a_cnp;
    301 	struct vattr *vap = ((struct vop_mknod_args *)v)->a_vap;
    302 
    303 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
    304 	    vap->va_type != VFIFO) {
    305 		vput(dvp);
    306 		return EINVAL;
    307 	}
    308 
    309 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    310 }
    311 
    312 /* --------------------------------------------------------------------- */
    313 
    314 int
    315 tmpfs_open(void *v)
    316 {
    317 	struct vnode *vp = ((struct vop_open_args *)v)->a_vp;
    318 	int mode = ((struct vop_open_args *)v)->a_mode;
    319 
    320 	int error;
    321 	struct tmpfs_node *node;
    322 
    323 	KASSERT(VOP_ISLOCKED(vp));
    324 
    325 	node = VP_TO_TMPFS_NODE(vp);
    326 
    327 	/* The file is still active but all its names have been removed
    328 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
    329 	 * it is about to die. */
    330 	if (node->tn_links < 1) {
    331 		error = ENOENT;
    332 		goto out;
    333 	}
    334 
    335 	/* If the file is marked append-only, deny write requests. */
    336 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
    337 		error = EPERM;
    338 	else
    339 		error = 0;
    340 
    341 out:
    342 	KASSERT(VOP_ISLOCKED(vp));
    343 
    344 	return error;
    345 }
    346 
    347 /* --------------------------------------------------------------------- */
    348 
    349 int
    350 tmpfs_close(void *v)
    351 {
    352 	struct vnode *vp = ((struct vop_close_args *)v)->a_vp;
    353 
    354 	struct tmpfs_node *node;
    355 
    356 	KASSERT(VOP_ISLOCKED(vp));
    357 
    358 	node = VP_TO_TMPFS_NODE(vp);
    359 
    360 	if (node->tn_links > 0) {
    361 		/* Update node times.  No need to do it if the node has
    362 		 * been deleted, because it will vanish after we return. */
    363 		tmpfs_update(vp, NULL, NULL, NULL, UPDATE_CLOSE);
    364 	}
    365 
    366 	return 0;
    367 }
    368 
    369 /* --------------------------------------------------------------------- */
    370 
    371 static int
    372 tmpfs_check_possible(struct vnode *vp, struct tmpfs_node *node, mode_t mode)
    373 {
    374 	int error = 0;
    375 
    376 	switch (vp->v_type) {
    377 	case VDIR:
    378 		/* FALLTHROUGH */
    379 	case VLNK:
    380 		/* FALLTHROUGH */
    381 	case VREG:
    382 		if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
    383 			error = EROFS;
    384 			goto out;
    385 		}
    386 		break;
    387 
    388 	case VBLK:
    389 		/* FALLTHROUGH */
    390 	case VCHR:
    391 		/* FALLTHROUGH */
    392 	case VSOCK:
    393 		/* FALLTHROUGH */
    394 	case VFIFO:
    395 		break;
    396 
    397 	default:
    398 		error = EINVAL;
    399 		goto out;
    400 	}
    401 
    402 	if (mode & VWRITE && node->tn_flags & IMMUTABLE) {
    403 		error = EPERM;
    404 		goto out;
    405 	}
    406 
    407  out:
    408 	return error;
    409 }
    410 
    411 static int
    412 tmpfs_check_permitted(struct vnode *vp, struct tmpfs_node *node, mode_t mode,
    413     kauth_cred_t cred)
    414 {
    415 
    416 	return genfs_can_access(vp->v_type, node->tn_mode, node->tn_uid,
    417 	    node->tn_gid, mode, cred);
    418 }
    419 
    420 int
    421 tmpfs_access(void *v)
    422 {
    423 	struct vnode *vp = ((struct vop_access_args *)v)->a_vp;
    424 	int mode = ((struct vop_access_args *)v)->a_mode;
    425 	kauth_cred_t cred = ((struct vop_access_args *)v)->a_cred;
    426 
    427 	int error;
    428 	struct tmpfs_node *node;
    429 
    430 	KASSERT(VOP_ISLOCKED(vp));
    431 
    432 	node = VP_TO_TMPFS_NODE(vp);
    433 
    434 	error = tmpfs_check_possible(vp, node, mode);
    435 	if (error)
    436 		goto out;
    437 
    438 	error = tmpfs_check_permitted(vp, node, mode, cred);
    439 
    440 	error = kauth_authorize_vnode(cred, kauth_mode_to_action(mode), vp,
    441 	    NULL, error);
    442 
    443 out:
    444 	KASSERT(VOP_ISLOCKED(vp));
    445 
    446 	return error;
    447 }
    448 
    449 /* --------------------------------------------------------------------- */
    450 
    451 int
    452 tmpfs_getattr(void *v)
    453 {
    454 	struct vnode *vp = ((struct vop_getattr_args *)v)->a_vp;
    455 	struct vattr *vap = ((struct vop_getattr_args *)v)->a_vap;
    456 
    457 	struct tmpfs_node *node;
    458 
    459 	node = VP_TO_TMPFS_NODE(vp);
    460 
    461 	vattr_null(vap);
    462 
    463 	tmpfs_itimes(vp, NULL, NULL, NULL);
    464 
    465 	vap->va_type = vp->v_type;
    466 	vap->va_mode = node->tn_mode;
    467 	vap->va_nlink = node->tn_links;
    468 	vap->va_uid = node->tn_uid;
    469 	vap->va_gid = node->tn_gid;
    470 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
    471 	vap->va_fileid = node->tn_id;
    472 	vap->va_size = node->tn_size;
    473 	vap->va_blocksize = PAGE_SIZE;
    474 	vap->va_atime = node->tn_atime;
    475 	vap->va_mtime = node->tn_mtime;
    476 	vap->va_ctime = node->tn_ctime;
    477 	vap->va_birthtime = node->tn_birthtime;
    478 	vap->va_gen = node->tn_gen;
    479 	vap->va_flags = node->tn_flags;
    480 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
    481 		node->tn_spec.tn_dev.tn_rdev : VNOVAL;
    482 	vap->va_bytes = round_page(node->tn_size);
    483 	vap->va_filerev = VNOVAL;
    484 	vap->va_vaflags = 0;
    485 	vap->va_spare = VNOVAL; /* XXX */
    486 
    487 	return 0;
    488 }
    489 
    490 /* --------------------------------------------------------------------- */
    491 
    492 #define GOODTIME(tv)	((tv)->tv_sec != VNOVAL || (tv)->tv_nsec != VNOVAL)
    493 /* XXX Should this operation be atomic?  I think it should, but code in
    494  * XXX other places (e.g., ufs) doesn't seem to be... */
    495 int
    496 tmpfs_setattr(void *v)
    497 {
    498 	struct vnode *vp = ((struct vop_setattr_args *)v)->a_vp;
    499 	struct vattr *vap = ((struct vop_setattr_args *)v)->a_vap;
    500 	kauth_cred_t cred = ((struct vop_setattr_args *)v)->a_cred;
    501 	struct lwp *l = curlwp;
    502 
    503 	int error;
    504 
    505 	KASSERT(VOP_ISLOCKED(vp));
    506 
    507 	error = 0;
    508 
    509 	/* Abort if any unsettable attribute is given. */
    510 	if (vap->va_type != VNON ||
    511 	    vap->va_nlink != VNOVAL ||
    512 	    vap->va_fsid != VNOVAL ||
    513 	    vap->va_fileid != VNOVAL ||
    514 	    vap->va_blocksize != VNOVAL ||
    515 	    GOODTIME(&vap->va_ctime) ||
    516 	    vap->va_gen != VNOVAL ||
    517 	    vap->va_rdev != VNOVAL ||
    518 	    vap->va_bytes != VNOVAL)
    519 		error = EINVAL;
    520 
    521 	if (error == 0 && (vap->va_flags != VNOVAL))
    522 		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
    523 
    524 	if (error == 0 && (vap->va_size != VNOVAL))
    525 		error = tmpfs_chsize(vp, vap->va_size, cred, l);
    526 
    527 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
    528 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
    529 
    530 	if (error == 0 && (vap->va_mode != VNOVAL))
    531 		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
    532 
    533 	if (error == 0 && (GOODTIME(&vap->va_atime) || GOODTIME(&vap->va_mtime)
    534 	    || GOODTIME(&vap->va_birthtime)))
    535 		if ((error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
    536 		    &vap->va_birthtime, vap->va_vaflags, cred, l)) == 0)
    537 			return 0;
    538 
    539 	/* Update the node times.  We give preference to the error codes
    540 	 * generated by this function rather than the ones that may arise
    541 	 * from tmpfs_update. */
    542 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    543 
    544 	KASSERT(VOP_ISLOCKED(vp));
    545 
    546 	return error;
    547 }
    548 
    549 /* --------------------------------------------------------------------- */
    550 
    551 int
    552 tmpfs_read(void *v)
    553 {
    554 	struct vnode *vp = ((struct vop_read_args *)v)->a_vp;
    555 	struct uio *uio = ((struct vop_read_args *)v)->a_uio;
    556 	int ioflag = ((struct vop_read_args *)v)->a_ioflag;
    557 
    558 	int error;
    559 	struct tmpfs_node *node;
    560 	struct uvm_object *uobj;
    561 
    562 	KASSERT(VOP_ISLOCKED(vp));
    563 
    564 	node = VP_TO_TMPFS_NODE(vp);
    565 
    566 	if (vp->v_type != VREG) {
    567 		error = EISDIR;
    568 		goto out;
    569 	}
    570 
    571 	if (uio->uio_offset < 0) {
    572 		error = EINVAL;
    573 		goto out;
    574 	}
    575 
    576 	node->tn_status |= TMPFS_NODE_ACCESSED;
    577 
    578 	uobj = node->tn_spec.tn_reg.tn_aobj;
    579 	error = 0;
    580 	while (error == 0 && uio->uio_resid > 0) {
    581 		vsize_t len;
    582 
    583 		if (node->tn_size <= uio->uio_offset)
    584 			break;
    585 
    586 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    587 		if (len == 0)
    588 			break;
    589 
    590 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
    591 		    UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
    592 	}
    593 
    594 out:
    595 	KASSERT(VOP_ISLOCKED(vp));
    596 
    597 	return error;
    598 }
    599 
    600 /* --------------------------------------------------------------------- */
    601 
    602 int
    603 tmpfs_write(void *v)
    604 {
    605 	struct vnode *vp = ((struct vop_write_args *)v)->a_vp;
    606 	struct uio *uio = ((struct vop_write_args *)v)->a_uio;
    607 	int ioflag = ((struct vop_write_args *)v)->a_ioflag;
    608 
    609 	bool extended;
    610 	int error;
    611 	off_t oldsize;
    612 	struct tmpfs_node *node;
    613 	struct uvm_object *uobj;
    614 
    615 	KASSERT(VOP_ISLOCKED(vp));
    616 
    617 	node = VP_TO_TMPFS_NODE(vp);
    618 	oldsize = node->tn_size;
    619 
    620 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
    621 		error = EINVAL;
    622 		goto out;
    623 	}
    624 
    625 	if (uio->uio_resid == 0) {
    626 		error = 0;
    627 		goto out;
    628 	}
    629 
    630 	if (ioflag & IO_APPEND)
    631 		uio->uio_offset = node->tn_size;
    632 
    633 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
    634 	if (extended) {
    635 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
    636 		if (error != 0)
    637 			goto out;
    638 	}
    639 
    640 	uobj = node->tn_spec.tn_reg.tn_aobj;
    641 	error = 0;
    642 	while (error == 0 && uio->uio_resid > 0) {
    643 		vsize_t len;
    644 
    645 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    646 		if (len == 0)
    647 			break;
    648 
    649 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
    650 		    UBC_WRITE | UBC_UNMAP_FLAG(vp));
    651 	}
    652 
    653 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
    654 	    (extended ? TMPFS_NODE_CHANGED : 0);
    655 
    656 	if (error != 0)
    657 		(void)tmpfs_reg_resize(vp, oldsize);
    658 
    659 	VN_KNOTE(vp, NOTE_WRITE);
    660 
    661 out:
    662 	KASSERT(VOP_ISLOCKED(vp));
    663 	KASSERT(IMPLIES(error == 0, uio->uio_resid == 0));
    664 	KASSERT(IMPLIES(error != 0, oldsize == node->tn_size));
    665 
    666 	return error;
    667 }
    668 
    669 /* --------------------------------------------------------------------- */
    670 
    671 int
    672 tmpfs_fsync(void *v)
    673 {
    674 	struct vnode *vp = ((struct vop_fsync_args *)v)->a_vp;
    675 
    676 	KASSERT(VOP_ISLOCKED(vp));
    677 
    678 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    679 
    680 	return 0;
    681 }
    682 
    683 /* --------------------------------------------------------------------- */
    684 
    685 int
    686 tmpfs_remove(void *v)
    687 {
    688 	struct vnode *dvp = ((struct vop_remove_args *)v)->a_dvp;
    689 	struct vnode *vp = ((struct vop_remove_args *)v)->a_vp;
    690 	struct componentname *cnp = (((struct vop_remove_args *)v)->a_cnp);
    691 
    692 	int error;
    693 	struct tmpfs_dirent *de;
    694 	struct tmpfs_mount *tmp;
    695 	struct tmpfs_node *dnode;
    696 	struct tmpfs_node *node;
    697 
    698 	KASSERT(VOP_ISLOCKED(dvp));
    699 	KASSERT(VOP_ISLOCKED(vp));
    700 
    701 	if (vp->v_type == VDIR) {
    702 		error = EPERM;
    703 		goto out;
    704 	}
    705 
    706 	dnode = VP_TO_TMPFS_DIR(dvp);
    707 	node = VP_TO_TMPFS_NODE(vp);
    708 	tmp = VFS_TO_TMPFS(vp->v_mount);
    709 	de = tmpfs_dir_lookup(dnode, cnp);
    710 	KASSERT(de);
    711 	KASSERT(de->td_node == node);
    712 
    713 	/* Files marked as immutable or append-only cannot be deleted. */
    714 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    715 		error = EPERM;
    716 		goto out;
    717 	}
    718 
    719 	/* Remove the entry from the directory; as it is a file, we do not
    720 	 * have to change the number of hard links of the directory. */
    721 	tmpfs_dir_detach(dvp, de);
    722 
    723 	/* Free the directory entry we just deleted.  Note that the node
    724 	 * referred by it will not be removed until the vnode is really
    725 	 * reclaimed. */
    726 	tmpfs_free_dirent(tmp, de, true);
    727 
    728 	error = 0;
    729 
    730 out:
    731 	vput(vp);
    732 	if (dvp == vp)
    733 		vrele(dvp);
    734 	else
    735 		vput(dvp);
    736 
    737 	return error;
    738 }
    739 
    740 /* --------------------------------------------------------------------- */
    741 
    742 int
    743 tmpfs_link(void *v)
    744 {
    745 	struct vnode *dvp = ((struct vop_link_args *)v)->a_dvp;
    746 	struct vnode *vp = ((struct vop_link_args *)v)->a_vp;
    747 	struct componentname *cnp = ((struct vop_link_args *)v)->a_cnp;
    748 
    749 	int error;
    750 	struct tmpfs_dirent *de;
    751 	struct tmpfs_node *dnode;
    752 	struct tmpfs_node *node;
    753 
    754 	KASSERT(VOP_ISLOCKED(dvp));
    755 	KASSERT(dvp != vp); /* XXX When can this be false? */
    756 
    757 	dnode = VP_TO_TMPFS_DIR(dvp);
    758 	node = VP_TO_TMPFS_NODE(vp);
    759 
    760 	/* Lock vp because we will need to run tmpfs_update over it, which
    761 	 * needs the vnode to be locked. */
    762 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    763 
    764 	/* XXX: Why aren't the following two tests done by the caller? */
    765 
    766 	/* Hard links of directories are forbidden. */
    767 	if (vp->v_type == VDIR) {
    768 		error = EPERM;
    769 		goto out;
    770 	}
    771 
    772 	/* Cannot create cross-device links. */
    773 	if (dvp->v_mount != vp->v_mount) {
    774 		error = EXDEV;
    775 		goto out;
    776 	}
    777 
    778 	/* Ensure that we do not overflow the maximum number of links imposed
    779 	 * by the system. */
    780 	KASSERT(node->tn_links <= LINK_MAX);
    781 	if (node->tn_links == LINK_MAX) {
    782 		error = EMLINK;
    783 		goto out;
    784 	}
    785 
    786 	/* We cannot create links of files marked immutable or append-only. */
    787 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    788 		error = EPERM;
    789 		goto out;
    790 	}
    791 
    792 	/* Allocate a new directory entry to represent the node. */
    793 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
    794 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
    795 	if (error != 0)
    796 		goto out;
    797 
    798 	/* Insert the new directory entry into the appropriate directory. */
    799 	tmpfs_dir_attach(dvp, de);
    800 
    801 	/* vp link count has changed, so update node times. */
    802 	node->tn_status |= TMPFS_NODE_CHANGED;
    803 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    804 
    805 	error = 0;
    806 
    807 out:
    808 	VOP_UNLOCK(vp);
    809 	vput(dvp);
    810 
    811 	return error;
    812 }
    813 
    814 /*
    815  * tmpfs_rename: rename routine.
    816  *
    817  * Arguments: fdvp (from-parent vnode), fvp (from-leaf), tdvp (to-parent)
    818  * and tvp (to-leaf), if exists (NULL if not).
    819  *
    820  * => Caller holds a reference on fdvp and fvp, they are unlocked.
    821  *    Note: fdvp and fvp can refer to the same object (i.e. when it is root).
    822  *
    823  * => Both tdvp and tvp are referenced and locked.  It is our responsibility
    824  *    to release the references and unlock them (or destroy).
    825  */
    826 int
    827 tmpfs_rename(void *v)
    828 {
    829 	struct vnode *fdvp = ((struct vop_rename_args *)v)->a_fdvp;
    830 	struct vnode *fvp = ((struct vop_rename_args *)v)->a_fvp;
    831 	struct componentname *fcnp = ((struct vop_rename_args *)v)->a_fcnp;
    832 	struct vnode *tdvp = ((struct vop_rename_args *)v)->a_tdvp;
    833 	struct vnode *tvp = ((struct vop_rename_args *)v)->a_tvp;
    834 	struct componentname *tcnp = ((struct vop_rename_args *)v)->a_tcnp;
    835 
    836 	char *newname;
    837 	int error;
    838 	struct tmpfs_dirent *de, *de2;
    839 	struct tmpfs_mount *tmp;
    840 	struct tmpfs_node *fdnode;
    841 	struct tmpfs_node *fnode;
    842 	struct tmpfs_node *tnode;
    843 	struct tmpfs_node *tdnode;
    844 	size_t namelen;
    845 
    846 	KASSERT(VOP_ISLOCKED(tdvp));
    847 	KASSERT(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
    848 
    849 	newname = NULL;
    850 	namelen = 0;
    851 	tmp = NULL;
    852 
    853 	/* Disallow cross-device renames. */
    854 	if (fvp->v_mount != tdvp->v_mount ||
    855 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
    856 		error = EXDEV;
    857 		goto out_unlocked;
    858 	}
    859 
    860 	fnode = VP_TO_TMPFS_NODE(fvp);
    861 	fdnode = VP_TO_TMPFS_DIR(fdvp);
    862 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
    863 	tdnode = VP_TO_TMPFS_DIR(tdvp);
    864 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
    865 
    866 	if (fdvp == tvp) {
    867 		error = 0;
    868 		goto out_unlocked;
    869 	}
    870 
    871 	/* Allocate memory, if necessary, for a new name. */
    872 	namelen = tcnp->cn_namelen;
    873 	if (tmpfs_strname_neqlen(fcnp, tcnp)) {
    874 		newname = tmpfs_strname_alloc(tmp, namelen);
    875 		if (newname == NULL) {
    876 			error = ENOSPC;
    877 			goto out_unlocked;
    878 		}
    879 	}
    880 
    881 	/* If we need to move the directory between entries, lock the
    882 	 * source so that we can safely operate on it. */
    883 
    884 	/* XXX: this is a potential locking order violation! */
    885 	if (fdnode != tdnode) {
    886 		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
    887 	}
    888 
    889 	/*
    890 	 * If the node we were renaming has scarpered, just give up.
    891 	 */
    892 	de = tmpfs_dir_lookup(fdnode, fcnp);
    893 	if (de == NULL || de->td_node != fnode) {
    894 		error = ENOENT;
    895 		goto out;
    896 	}
    897 
    898 	/* If source and target is the same vnode, remove the source link. */
    899 	if (fvp == tvp) {
    900 		/*
    901 		 * Detach and free the directory entry.  Drops the link
    902 		 * count on the node.
    903 		 */
    904 		tmpfs_dir_detach(fdvp, de);
    905 		tmpfs_free_dirent(VFS_TO_TMPFS(fvp->v_mount), de, true);
    906 		VN_KNOTE(fdvp, NOTE_WRITE);
    907 		goto out_ok;
    908 	}
    909 
    910 	/* If replacing an existing entry, ensure we can do the operation. */
    911 	if (tvp != NULL) {
    912 		KASSERT(tnode != NULL);
    913 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
    914 			if (tnode->tn_size > 0) {
    915 				error = ENOTEMPTY;
    916 				goto out;
    917 			}
    918 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
    919 			error = ENOTDIR;
    920 			goto out;
    921 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
    922 			error = EISDIR;
    923 			goto out;
    924 		} else {
    925 			KASSERT(fnode->tn_type != VDIR &&
    926 			        tnode->tn_type != VDIR);
    927 		}
    928 	}
    929 
    930 	/* If the node is being moved to another directory, we have to do
    931 	 * the move. */
    932 	if (fdnode != tdnode) {
    933 		/* In case we are moving a directory, we have to adjust its
    934 		 * parent to point to the new parent. */
    935 		if (de->td_node->tn_type == VDIR) {
    936 			struct tmpfs_node *n;
    937 
    938 			/* Ensure the target directory is not a child of the
    939 			 * directory being moved.  Otherwise, we'd end up
    940 			 * with stale nodes. */
    941 			n = tdnode;
    942 			while (n != n->tn_spec.tn_dir.tn_parent) {
    943 				if (n == fnode) {
    944 					error = EINVAL;
    945 					goto out;
    946 				}
    947 				n = n->tn_spec.tn_dir.tn_parent;
    948 			}
    949 
    950 			/* Adjust the parent pointer. */
    951 			TMPFS_VALIDATE_DIR(fnode);
    952 			de->td_node->tn_spec.tn_dir.tn_parent = tdnode;
    953 
    954 			/* As a result of changing the target of the '..'
    955 			 * entry, the link count of the source and target
    956 			 * directories has to be adjusted. */
    957 			fdnode->tn_links--;
    958 			tdnode->tn_links++;
    959 		}
    960 
    961 		/* Do the move: just remove the entry from the source directory
    962 		 * and insert it into the target one. */
    963 		tmpfs_dir_detach(fdvp, de);
    964 		tmpfs_dir_attach(tdvp, de);
    965 
    966 		/* Notify listeners of fdvp about the change in the directory.
    967 		 * We can do it at this point because we aren't touching fdvp
    968 		 * any more below. */
    969 		VN_KNOTE(fdvp, NOTE_WRITE);
    970 	}
    971 
    972 	/* If we are overwriting an entry, we have to remove the old one
    973 	 * from the target directory. */
    974 	if (tvp != NULL) {
    975 		KASSERT(tnode != NULL);
    976 
    977 		/* Remove the old entry from the target directory.
    978 		 * Note! This relies on tmpfs_dir_attach() putting the new
    979 		 * node on the end of the target's node list. */
    980 		de2 = tmpfs_dir_lookup(tdnode, tcnp);
    981 		KASSERT(de2 != NULL);
    982 		KASSERT(de2->td_node == tnode);
    983 		tmpfs_dir_detach(tdvp, de2);
    984 
    985 		/* Free the directory entry we just deleted.  Note that the
    986 		 * node referred by it will not be removed until the vnode is
    987 		 * really reclaimed. */
    988 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de2, true);
    989 	}
    990 
    991 	/* If the name has changed, we need to make it effective by changing
    992 	 * it in the directory entry. */
    993 	if (newname != NULL) {
    994 		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
    995 		KASSERT(tcnp->cn_namelen < 0xffff);
    996 
    997 		tmpfs_strname_free(tmp, de->td_name, de->td_namelen);
    998 		de->td_namelen = (uint16_t)namelen;
    999 		memcpy(newname, tcnp->cn_nameptr, namelen);
   1000 		de->td_name = newname;
   1001 		newname = NULL;
   1002 
   1003 		fnode->tn_status |= TMPFS_NODE_CHANGED;
   1004 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
   1005 	}
   1006  out_ok:
   1007 	/* Notify listeners of tdvp about the change in the directory (either
   1008 	 * because a new entry was added or because one was removed) and
   1009 	 * listeners of fvp about the rename. */
   1010 	VN_KNOTE(tdvp, NOTE_WRITE);
   1011 	VN_KNOTE(fvp, NOTE_RENAME);
   1012 
   1013 	error = 0;
   1014 
   1015  out:
   1016 	if (fdnode != tdnode)
   1017 		VOP_UNLOCK(fdvp);
   1018 
   1019  out_unlocked:
   1020 	/* Release target nodes. */
   1021 	if (tdvp == tvp)
   1022 		vrele(tdvp);
   1023 	else
   1024 		vput(tdvp);
   1025 	if (tvp != NULL)
   1026 		vput(tvp);
   1027 
   1028 	/* Release source nodes. */
   1029 	vrele(fdvp);
   1030 	vrele(fvp);
   1031 
   1032 	if (newname != NULL) {
   1033 		tmpfs_strname_free(tmp, newname, namelen);
   1034 	}
   1035 	return error;
   1036 }
   1037 
   1038 /* --------------------------------------------------------------------- */
   1039 
   1040 int
   1041 tmpfs_mkdir(void *v)
   1042 {
   1043 	struct vnode *dvp = ((struct vop_mkdir_args *)v)->a_dvp;
   1044 	struct vnode **vpp = ((struct vop_mkdir_args *)v)->a_vpp;
   1045 	struct componentname *cnp = ((struct vop_mkdir_args *)v)->a_cnp;
   1046 	struct vattr *vap = ((struct vop_mkdir_args *)v)->a_vap;
   1047 
   1048 	KASSERT(vap->va_type == VDIR);
   1049 
   1050 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
   1051 }
   1052 
   1053 /* --------------------------------------------------------------------- */
   1054 
   1055 int
   1056 tmpfs_rmdir(void *v)
   1057 {
   1058 	struct vnode *dvp = ((struct vop_rmdir_args *)v)->a_dvp;
   1059 	struct vnode *vp = ((struct vop_rmdir_args *)v)->a_vp;
   1060 	struct componentname *cnp = ((struct vop_rmdir_args *)v)->a_cnp;
   1061 
   1062 	int error;
   1063 	struct tmpfs_dirent *de;
   1064 	struct tmpfs_mount *tmp;
   1065 	struct tmpfs_node *dnode;
   1066 	struct tmpfs_node *node;
   1067 
   1068 	KASSERT(VOP_ISLOCKED(dvp));
   1069 	KASSERT(VOP_ISLOCKED(vp));
   1070 
   1071 	tmp = VFS_TO_TMPFS(dvp->v_mount);
   1072 	dnode = VP_TO_TMPFS_DIR(dvp);
   1073 	node = VP_TO_TMPFS_DIR(vp);
   1074 	error = 0;
   1075 
   1076 	/* Directories with more than two entries ('.' and '..') cannot be
   1077 	 * removed. */
   1078 	if (node->tn_size > 0) {
   1079 		error = ENOTEMPTY;
   1080 		goto out;
   1081 	}
   1082 
   1083 	/* This invariant holds only if we are not trying to remove "..".
   1084 	 * We checked for that above so this is safe now. */
   1085 	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
   1086 
   1087 	/* Get the directory entry associated with node (vp). */
   1088 	de = tmpfs_dir_lookup(dnode, cnp);
   1089 	KASSERT(de);
   1090 	KASSERT(de->td_node == node);
   1091 
   1092 	/* Check flags to see if we are allowed to remove the directory. */
   1093 	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
   1094 		error = EPERM;
   1095 		goto out;
   1096 	}
   1097 
   1098 	/* Detach the directory entry from the directory (dnode). */
   1099 	tmpfs_dir_detach(dvp, de);
   1100 
   1101 	node->tn_links--;
   1102 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
   1103 	    TMPFS_NODE_MODIFIED;
   1104 	node->tn_spec.tn_dir.tn_parent->tn_links--;
   1105 	node->tn_spec.tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \
   1106 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
   1107 
   1108 	/* Release the parent. */
   1109 	cache_purge(dvp); /* XXX Is this needed? */
   1110 
   1111 	/* Free the directory entry we just deleted.  Note that the node
   1112 	 * referred by it will not be removed until the vnode is really
   1113 	 * reclaimed. */
   1114 	tmpfs_free_dirent(tmp, de, true);
   1115 
   1116 	KASSERT(node->tn_links == 0);
   1117  out:
   1118 	/* Release the nodes. */
   1119 	vput(dvp);
   1120 	vput(vp);
   1121 
   1122 	return error;
   1123 }
   1124 
   1125 /* --------------------------------------------------------------------- */
   1126 
   1127 int
   1128 tmpfs_symlink(void *v)
   1129 {
   1130 	struct vnode *dvp = ((struct vop_symlink_args *)v)->a_dvp;
   1131 	struct vnode **vpp = ((struct vop_symlink_args *)v)->a_vpp;
   1132 	struct componentname *cnp = ((struct vop_symlink_args *)v)->a_cnp;
   1133 	struct vattr *vap = ((struct vop_symlink_args *)v)->a_vap;
   1134 	char *target = ((struct vop_symlink_args *)v)->a_target;
   1135 
   1136 	KASSERT(vap->va_type == VLNK);
   1137 
   1138 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
   1139 }
   1140 
   1141 /* --------------------------------------------------------------------- */
   1142 
   1143 int
   1144 tmpfs_readdir(void *v)
   1145 {
   1146 	struct vnode *vp = ((struct vop_readdir_args *)v)->a_vp;
   1147 	struct uio *uio = ((struct vop_readdir_args *)v)->a_uio;
   1148 	int *eofflag = ((struct vop_readdir_args *)v)->a_eofflag;
   1149 	off_t **cookies = ((struct vop_readdir_args *)v)->a_cookies;
   1150 	int *ncookies = ((struct vop_readdir_args *)v)->a_ncookies;
   1151 
   1152 	int error;
   1153 	off_t startoff;
   1154 	off_t cnt;
   1155 	struct tmpfs_node *node;
   1156 
   1157 	KASSERT(VOP_ISLOCKED(vp));
   1158 
   1159 	/* This operation only makes sense on directory nodes. */
   1160 	if (vp->v_type != VDIR) {
   1161 		error = ENOTDIR;
   1162 		goto out;
   1163 	}
   1164 
   1165 	node = VP_TO_TMPFS_DIR(vp);
   1166 
   1167 	startoff = uio->uio_offset;
   1168 
   1169 	cnt = 0;
   1170 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
   1171 		error = tmpfs_dir_getdotdent(node, uio);
   1172 		if (error == -1) {
   1173 			error = 0;
   1174 			goto outok;
   1175 		} else if (error != 0)
   1176 			goto outok;
   1177 		cnt++;
   1178 	}
   1179 
   1180 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
   1181 		error = tmpfs_dir_getdotdotdent(node, uio);
   1182 		if (error == -1) {
   1183 			error = 0;
   1184 			goto outok;
   1185 		} else if (error != 0)
   1186 			goto outok;
   1187 		cnt++;
   1188 	}
   1189 
   1190 	error = tmpfs_dir_getdents(node, uio, &cnt);
   1191 	if (error == -1)
   1192 		error = 0;
   1193 	KASSERT(error >= 0);
   1194 
   1195 outok:
   1196 	/* This label assumes that startoff has been
   1197 	 * initialized.  If the compiler didn't spit out warnings, we'd
   1198 	 * simply make this one be 'out' and drop 'outok'. */
   1199 
   1200 	if (eofflag != NULL)
   1201 		*eofflag =
   1202 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
   1203 
   1204 	/* Update NFS-related variables. */
   1205 	if (error == 0 && cookies != NULL && ncookies != NULL) {
   1206 		off_t i;
   1207 		off_t off = startoff;
   1208 		struct tmpfs_dirent *de = NULL;
   1209 
   1210 		*ncookies = cnt;
   1211 		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
   1212 
   1213 		for (i = 0; i < cnt; i++) {
   1214 			KASSERT(off != TMPFS_DIRCOOKIE_EOF);
   1215 			if (off == TMPFS_DIRCOOKIE_DOT) {
   1216 				off = TMPFS_DIRCOOKIE_DOTDOT;
   1217 			} else {
   1218 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
   1219 					de = TAILQ_FIRST(&node->tn_spec.
   1220 					    tn_dir.tn_dir);
   1221 				} else if (de != NULL) {
   1222 					de = TAILQ_NEXT(de, td_entries);
   1223 				} else {
   1224 					de = tmpfs_dir_lookupbycookie(node,
   1225 					    off);
   1226 					KASSERT(de != NULL);
   1227 					de = TAILQ_NEXT(de, td_entries);
   1228 				}
   1229 				if (de == NULL) {
   1230 					off = TMPFS_DIRCOOKIE_EOF;
   1231 				} else {
   1232 					off = tmpfs_dircookie(de);
   1233 				}
   1234 			}
   1235 
   1236 			(*cookies)[i] = off;
   1237 		}
   1238 		KASSERT(uio->uio_offset == off);
   1239 	}
   1240 
   1241 out:
   1242 	KASSERT(VOP_ISLOCKED(vp));
   1243 
   1244 	return error;
   1245 }
   1246 
   1247 /* --------------------------------------------------------------------- */
   1248 
   1249 int
   1250 tmpfs_readlink(void *v)
   1251 {
   1252 	struct vnode *vp = ((struct vop_readlink_args *)v)->a_vp;
   1253 	struct uio *uio = ((struct vop_readlink_args *)v)->a_uio;
   1254 
   1255 	int error;
   1256 	struct tmpfs_node *node;
   1257 
   1258 	KASSERT(VOP_ISLOCKED(vp));
   1259 	KASSERT(uio->uio_offset == 0);
   1260 	KASSERT(vp->v_type == VLNK);
   1261 
   1262 	node = VP_TO_TMPFS_NODE(vp);
   1263 
   1264 	error = uiomove(node->tn_spec.tn_lnk.tn_link,
   1265 	    MIN(node->tn_size, uio->uio_resid), uio);
   1266 	node->tn_status |= TMPFS_NODE_ACCESSED;
   1267 
   1268 	KASSERT(VOP_ISLOCKED(vp));
   1269 
   1270 	return error;
   1271 }
   1272 
   1273 /* --------------------------------------------------------------------- */
   1274 
   1275 int
   1276 tmpfs_inactive(void *v)
   1277 {
   1278 	struct vnode *vp = ((struct vop_inactive_args *)v)->a_vp;
   1279 
   1280 	struct tmpfs_node *node;
   1281 
   1282 	KASSERT(VOP_ISLOCKED(vp));
   1283 
   1284 	node = VP_TO_TMPFS_NODE(vp);
   1285 	*((struct vop_inactive_args *)v)->a_recycle = (node->tn_links == 0);
   1286 	VOP_UNLOCK(vp);
   1287 
   1288 	return 0;
   1289 }
   1290 
   1291 /* --------------------------------------------------------------------- */
   1292 
   1293 int
   1294 tmpfs_reclaim(void *v)
   1295 {
   1296 	struct vnode *vp = ((struct vop_reclaim_args *)v)->a_vp;
   1297 
   1298 	struct tmpfs_mount *tmp;
   1299 	struct tmpfs_node *node;
   1300 
   1301 	node = VP_TO_TMPFS_NODE(vp);
   1302 	tmp = VFS_TO_TMPFS(vp->v_mount);
   1303 
   1304 	cache_purge(vp);
   1305 	tmpfs_free_vp(vp);
   1306 
   1307 	/* If the node referenced by this vnode was deleted by the user,
   1308 	 * we must free its associated data structures (now that the vnode
   1309 	 * is being reclaimed). */
   1310 	if (node->tn_links == 0)
   1311 		tmpfs_free_node(tmp, node);
   1312 
   1313 	KASSERT(vp->v_data == NULL);
   1314 
   1315 	return 0;
   1316 }
   1317 
   1318 /* --------------------------------------------------------------------- */
   1319 
   1320 int
   1321 tmpfs_print(void *v)
   1322 {
   1323 	struct vnode *vp = ((struct vop_print_args *)v)->a_vp;
   1324 
   1325 	struct tmpfs_node *node;
   1326 
   1327 	node = VP_TO_TMPFS_NODE(vp);
   1328 
   1329 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
   1330 	    node, node->tn_flags, node->tn_links);
   1331 	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
   1332 	    ", status 0x%x",
   1333 	    node->tn_mode, node->tn_uid, node->tn_gid,
   1334 	    (uintmax_t)node->tn_size, node->tn_status);
   1335 	if (vp->v_type == VFIFO)
   1336 		VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
   1337 	printf("\n");
   1338 
   1339 	return 0;
   1340 }
   1341 
   1342 /* --------------------------------------------------------------------- */
   1343 
   1344 int
   1345 tmpfs_pathconf(void *v)
   1346 {
   1347 	int name = ((struct vop_pathconf_args *)v)->a_name;
   1348 	register_t *retval = ((struct vop_pathconf_args *)v)->a_retval;
   1349 
   1350 	int error;
   1351 
   1352 	error = 0;
   1353 
   1354 	switch (name) {
   1355 	case _PC_LINK_MAX:
   1356 		*retval = LINK_MAX;
   1357 		break;
   1358 
   1359 	case _PC_NAME_MAX:
   1360 		*retval = NAME_MAX;
   1361 		break;
   1362 
   1363 	case _PC_PATH_MAX:
   1364 		*retval = PATH_MAX;
   1365 		break;
   1366 
   1367 	case _PC_PIPE_BUF:
   1368 		*retval = PIPE_BUF;
   1369 		break;
   1370 
   1371 	case _PC_CHOWN_RESTRICTED:
   1372 		*retval = 1;
   1373 		break;
   1374 
   1375 	case _PC_NO_TRUNC:
   1376 		*retval = 1;
   1377 		break;
   1378 
   1379 	case _PC_SYNC_IO:
   1380 		*retval = 1;
   1381 		break;
   1382 
   1383 	case _PC_FILESIZEBITS:
   1384 		*retval = 0; /* XXX Don't know which value should I return. */
   1385 		break;
   1386 
   1387 	default:
   1388 		error = EINVAL;
   1389 	}
   1390 
   1391 	return error;
   1392 }
   1393 
   1394 /* --------------------------------------------------------------------- */
   1395 
   1396 int
   1397 tmpfs_advlock(void *v)
   1398 {
   1399 	struct vnode *vp = ((struct vop_advlock_args *)v)->a_vp;
   1400 
   1401 	struct tmpfs_node *node;
   1402 
   1403 	node = VP_TO_TMPFS_NODE(vp);
   1404 
   1405 	return lf_advlock(v, &node->tn_lockf, node->tn_size);
   1406 }
   1407 
   1408 /* --------------------------------------------------------------------- */
   1409 
   1410 int
   1411 tmpfs_getpages(void *v)
   1412 {
   1413 	struct vnode *vp = ((struct vop_getpages_args *)v)->a_vp;
   1414 	voff_t offset = ((struct vop_getpages_args *)v)->a_offset;
   1415 	struct vm_page **m = ((struct vop_getpages_args *)v)->a_m;
   1416 	int *count = ((struct vop_getpages_args *)v)->a_count;
   1417 	int centeridx = ((struct vop_getpages_args *)v)->a_centeridx;
   1418 	vm_prot_t access_type = ((struct vop_getpages_args *)v)->a_access_type;
   1419 	int advice = ((struct vop_getpages_args *)v)->a_advice;
   1420 	int flags = ((struct vop_getpages_args *)v)->a_flags;
   1421 
   1422 	int error;
   1423 	int i;
   1424 	struct tmpfs_node *node;
   1425 	struct uvm_object *uobj;
   1426 	int npages = *count;
   1427 
   1428 	KASSERT(vp->v_type == VREG);
   1429 	KASSERT(mutex_owned(&vp->v_interlock));
   1430 
   1431 	node = VP_TO_TMPFS_NODE(vp);
   1432 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1433 
   1434 	/* We currently don't rely on PGO_PASTEOF. */
   1435 
   1436 	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
   1437 		if ((flags & PGO_LOCKED) == 0)
   1438 			mutex_exit(&vp->v_interlock);
   1439 		return EINVAL;
   1440 	}
   1441 
   1442 	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
   1443 		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
   1444 	}
   1445 
   1446 	if ((flags & PGO_LOCKED) != 0)
   1447 		return EBUSY;
   1448 
   1449 	if ((flags & PGO_NOTIMESTAMP) == 0) {
   1450 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
   1451 			node->tn_status |= TMPFS_NODE_ACCESSED;
   1452 
   1453 		if ((access_type & VM_PROT_WRITE) != 0)
   1454 			node->tn_status |= TMPFS_NODE_MODIFIED;
   1455 	}
   1456 
   1457 	mutex_exit(&vp->v_interlock);
   1458 
   1459 	/*
   1460 	 * Make sure that the array on which we will store the
   1461 	 * gotten pages is clean.  Otherwise uao_get (pointed to by
   1462 	 * the pgo_get below) gets confused and does not return the
   1463 	 * appropriate pages.
   1464 	 *
   1465 	 * XXX This shall be revisited when kern/32166 is addressed
   1466 	 * because the loop to clean m[i] will most likely be redundant
   1467 	 * as well as the PGO_ALLPAGES flag.
   1468 	 */
   1469 	if (m != NULL)
   1470 		for (i = 0; i < npages; i++)
   1471 			m[i] = NULL;
   1472 	mutex_enter(&uobj->vmobjlock);
   1473 	error = (*uobj->pgops->pgo_get)(uobj, offset, m, &npages, centeridx,
   1474 	    access_type, advice, flags | PGO_ALLPAGES);
   1475 #if defined(DEBUG)
   1476 	{
   1477 		/* Make sure that all the pages we return are valid. */
   1478 		int dbgi;
   1479 		if (error == 0 && m != NULL)
   1480 			for (dbgi = 0; dbgi < npages; dbgi++)
   1481 				KASSERT(m[dbgi] != NULL);
   1482 	}
   1483 #endif
   1484 
   1485 	return error;
   1486 }
   1487 
   1488 /* --------------------------------------------------------------------- */
   1489 
   1490 int
   1491 tmpfs_putpages(void *v)
   1492 {
   1493 	struct vnode *vp = ((struct vop_putpages_args *)v)->a_vp;
   1494 	voff_t offlo = ((struct vop_putpages_args *)v)->a_offlo;
   1495 	voff_t offhi = ((struct vop_putpages_args *)v)->a_offhi;
   1496 	int flags = ((struct vop_putpages_args *)v)->a_flags;
   1497 
   1498 	int error;
   1499 	struct tmpfs_node *node;
   1500 	struct uvm_object *uobj;
   1501 
   1502 	KASSERT(mutex_owned(&vp->v_interlock));
   1503 
   1504 	node = VP_TO_TMPFS_NODE(vp);
   1505 
   1506 	if (vp->v_type != VREG) {
   1507 		mutex_exit(&vp->v_interlock);
   1508 		return 0;
   1509 	}
   1510 
   1511 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1512 	mutex_exit(&vp->v_interlock);
   1513 
   1514 	mutex_enter(&uobj->vmobjlock);
   1515 	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
   1516 
   1517 	/* XXX mtime */
   1518 
   1519 	return error;
   1520 }
   1521 
   1522 /* --------------------------------------------------------------------- */
   1523 
   1524 #ifdef TMPFS_WHITEOUT
   1525 int
   1526 tmpfs_whiteout(void *v)
   1527 {
   1528 	struct vnode *dvp = ((struct vop_whiteout_args *)v)->a_dvp;
   1529 	struct componentname *cnp = ((struct vop_whiteout_args *)v)->a_cnp;
   1530 	int flags = ((struct vop_whiteout_args *)v)->a_flags;
   1531 	struct tmpfs_mount *tmp = VFS_TO_TMPFS(dvp->v_mount);
   1532 	struct tmpfs_dirent *de;
   1533 	int error;
   1534 
   1535 	switch (flags) {
   1536 	case LOOKUP:
   1537 		break;
   1538 	case CREATE:
   1539 		error = tmpfs_alloc_dirent(tmp, TMPFS_NODE_WHITEOUT,
   1540 		    cnp->cn_nameptr, cnp->cn_namelen, &de);
   1541 		if (error)
   1542 			return error;
   1543 		tmpfs_dir_attach(dvp, de);
   1544 		break;
   1545 	case DELETE:
   1546 		cnp->cn_flags &= ~DOWHITEOUT; /* when in doubt, cargo cult */
   1547 		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), cnp);
   1548 		if (de == NULL)
   1549 			return ENOENT;
   1550 		tmpfs_dir_detach(dvp, de);
   1551 		tmpfs_free_dirent(tmp, de, true);
   1552 		break;
   1553 	}
   1554 
   1555 	return 0;
   1556 }
   1557 #endif
   1558