Home | History | Annotate | Line # | Download | only in tmpfs
tmpfs_vnops.c revision 1.39
      1 /*	$NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2005, 2006 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  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *        This product includes software developed by the NetBSD
     22  *        Foundation, Inc. and its contributors.
     23  * 4. Neither the name of The NetBSD Foundation nor the names of its
     24  *    contributors may be used to endorse or promote products derived
     25  *    from this software without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     37  * POSSIBILITY OF SUCH DAMAGE.
     38  */
     39 
     40 /*
     41  * tmpfs vnode interface.
     42  */
     43 
     44 #include <sys/cdefs.h>
     45 __KERNEL_RCSID(0, "$NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $");
     46 
     47 #include <sys/param.h>
     48 #include <sys/dirent.h>
     49 #include <sys/fcntl.h>
     50 #include <sys/event.h>
     51 #include <sys/malloc.h>
     52 #include <sys/namei.h>
     53 #include <sys/proc.h>
     54 #include <sys/stat.h>
     55 #include <sys/uio.h>
     56 #include <sys/unistd.h>
     57 #include <sys/vnode.h>
     58 #include <sys/lockf.h>
     59 #include <sys/kauth.h>
     60 
     61 #include <uvm/uvm.h>
     62 
     63 #include <miscfs/fifofs/fifo.h>
     64 #include <fs/tmpfs/tmpfs_vnops.h>
     65 #include <fs/tmpfs/tmpfs.h>
     66 
     67 /* --------------------------------------------------------------------- */
     68 
     69 /*
     70  * vnode operations vector used for files stored in a tmpfs file system.
     71  */
     72 int (**tmpfs_vnodeop_p)(void *);
     73 const struct vnodeopv_entry_desc tmpfs_vnodeop_entries[] = {
     74 	{ &vop_default_desc,		vn_default_error },
     75 	{ &vop_lookup_desc,		tmpfs_lookup },
     76 	{ &vop_create_desc,		tmpfs_create },
     77 	{ &vop_mknod_desc,		tmpfs_mknod },
     78 	{ &vop_open_desc,		tmpfs_open },
     79 	{ &vop_close_desc,		tmpfs_close },
     80 	{ &vop_access_desc,		tmpfs_access },
     81 	{ &vop_getattr_desc,		tmpfs_getattr },
     82 	{ &vop_setattr_desc,		tmpfs_setattr },
     83 	{ &vop_read_desc,		tmpfs_read },
     84 	{ &vop_write_desc,		tmpfs_write },
     85 	{ &vop_ioctl_desc,		tmpfs_ioctl },
     86 	{ &vop_fcntl_desc,		tmpfs_fcntl },
     87 	{ &vop_poll_desc,		tmpfs_poll },
     88 	{ &vop_kqfilter_desc,		tmpfs_kqfilter },
     89 	{ &vop_revoke_desc,		tmpfs_revoke },
     90 	{ &vop_mmap_desc,		tmpfs_mmap },
     91 	{ &vop_fsync_desc,		tmpfs_fsync },
     92 	{ &vop_seek_desc,		tmpfs_seek },
     93 	{ &vop_remove_desc,		tmpfs_remove },
     94 	{ &vop_link_desc,		tmpfs_link },
     95 	{ &vop_rename_desc,		tmpfs_rename },
     96 	{ &vop_mkdir_desc,		tmpfs_mkdir },
     97 	{ &vop_rmdir_desc,		tmpfs_rmdir },
     98 	{ &vop_symlink_desc,		tmpfs_symlink },
     99 	{ &vop_readdir_desc,		tmpfs_readdir },
    100 	{ &vop_readlink_desc,		tmpfs_readlink },
    101 	{ &vop_abortop_desc,		tmpfs_abortop },
    102 	{ &vop_inactive_desc,		tmpfs_inactive },
    103 	{ &vop_reclaim_desc,		tmpfs_reclaim },
    104 	{ &vop_lock_desc,		tmpfs_lock },
    105 	{ &vop_unlock_desc,		tmpfs_unlock },
    106 	{ &vop_bmap_desc,		tmpfs_bmap },
    107 	{ &vop_strategy_desc,		tmpfs_strategy },
    108 	{ &vop_print_desc,		tmpfs_print },
    109 	{ &vop_pathconf_desc,		tmpfs_pathconf },
    110 	{ &vop_islocked_desc,		tmpfs_islocked },
    111 	{ &vop_advlock_desc,		tmpfs_advlock },
    112 	{ &vop_lease_desc,		tmpfs_lease },
    113 	{ &vop_bwrite_desc,		tmpfs_bwrite },
    114 	{ &vop_getpages_desc,		tmpfs_getpages },
    115 	{ &vop_putpages_desc,		tmpfs_putpages },
    116 	{ NULL, NULL }
    117 };
    118 const struct vnodeopv_desc tmpfs_vnodeop_opv_desc =
    119 	{ &tmpfs_vnodeop_p, tmpfs_vnodeop_entries };
    120 
    121 /* --------------------------------------------------------------------- */
    122 
    123 int
    124 tmpfs_lookup(void *v)
    125 {
    126 	struct vnode *dvp = ((struct vop_lookup_args *)v)->a_dvp;
    127 	struct vnode **vpp = ((struct vop_lookup_args *)v)->a_vpp;
    128 	struct componentname *cnp = ((struct vop_lookup_args *)v)->a_cnp;
    129 
    130 	int error;
    131 	struct tmpfs_dirent *de;
    132 	struct tmpfs_node *dnode;
    133 
    134 	KASSERT(VOP_ISLOCKED(dvp));
    135 
    136 	dnode = VP_TO_TMPFS_DIR(dvp);
    137 	*vpp = NULL;
    138 
    139 	/* Check accessibility of requested node as a first step. */
    140 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_lwp);
    141 	if (error != 0)
    142 		goto out;
    143 
    144 	/* If requesting the last path component on a read-only file system
    145 	 * with a write operation, deny it. */
    146 	if ((cnp->cn_flags & ISLASTCN) &&
    147 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
    148 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    149 		error = EROFS;
    150 		goto out;
    151 	}
    152 
    153 	/* Avoid doing a linear scan of the directory if the requested
    154 	 * directory/name couple is already in the cache. */
    155 	error = cache_lookup(dvp, vpp, cnp);
    156 	if (error >= 0)
    157 		goto out;
    158 
    159 	/* We cannot be requesting the parent directory of the root node. */
    160 	KASSERT(IMPLIES(dnode->tn_type == VDIR &&
    161 	    dnode->tn_spec.tn_dir.tn_parent == dnode,
    162 	    !(cnp->cn_flags & ISDOTDOT)));
    163 
    164 	if (cnp->cn_flags & ISDOTDOT) {
    165 		VOP_UNLOCK(dvp, 0);
    166 
    167 		/* Allocate a new vnode on the matching entry. */
    168 		error = tmpfs_alloc_vp(dvp->v_mount,
    169 		    dnode->tn_spec.tn_dir.tn_parent, vpp);
    170 
    171 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
    172 		dnode->tn_spec.tn_dir.tn_parent->tn_lookup_dirent = NULL;
    173 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
    174 		VREF(dvp);
    175 		*vpp = dvp;
    176 		dnode->tn_lookup_dirent = NULL;
    177 		error = 0;
    178 	} else {
    179 		de = tmpfs_dir_lookup(dnode, cnp);
    180 		if (de == NULL) {
    181 			/* The entry was not found in the directory.
    182 			 * This is OK iff we are creating or renaming an
    183 			 * entry and are working on the last component of
    184 			 * the path name. */
    185 			if ((cnp->cn_flags & ISLASTCN) &&
    186 			    (cnp->cn_nameiop == CREATE || \
    187 			    cnp->cn_nameiop == RENAME)) {
    188 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
    189 				    cnp->cn_lwp);
    190 				if (error != 0)
    191 					goto out;
    192 
    193 				/* Keep the component name in the buffer for
    194 				 * future uses. */
    195 				cnp->cn_flags |= SAVENAME;
    196 
    197 				error = EJUSTRETURN;
    198 			} else
    199 				error = ENOENT;
    200 		} else {
    201 			struct tmpfs_node *tnode;
    202 
    203 			/* The entry was found, so get its associated
    204 			 * tmpfs_node. */
    205 			tnode = de->td_node;
    206 
    207 			/* If we are not at the last path component and
    208 			 * found a non-directory or non-link entry (which
    209 			 * may itself be pointing to a directory), raise
    210 			 * an error. */
    211 			if ((tnode->tn_type != VDIR &&
    212 			    tnode->tn_type != VLNK) &&
    213 			    !(cnp->cn_flags & ISLASTCN)) {
    214 				error = ENOTDIR;
    215 				goto out;
    216 			}
    217 
    218 			/* If we are deleting or renaming the entry, keep
    219 			 * track of its tmpfs_dirent so that it can be
    220 			 * easily deleted later. */
    221 			if ((cnp->cn_flags & ISLASTCN) &&
    222 			    (cnp->cn_nameiop == DELETE ||
    223 			    cnp->cn_nameiop == RENAME)) {
    224 				if ((dnode->tn_mode & S_ISTXT) != 0 &&
    225 				    kauth_authorize_generic(cnp->cn_cred,
    226 				     KAUTH_GENERIC_ISSUSER, NULL) != 0 &&
    227 				    kauth_cred_geteuid(cnp->cn_cred) != dnode->tn_uid &&
    228 				    kauth_cred_geteuid(cnp->cn_cred) != tnode->tn_uid)
    229 					return EPERM;
    230 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
    231 				    cnp->cn_lwp);
    232 				if (error != 0)
    233 					goto out;
    234 				tnode->tn_lookup_dirent = de;
    235 			}
    236 
    237 			/* Allocate a new vnode on the matching entry. */
    238 			error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp);
    239 		}
    240 	}
    241 
    242 	/* Store the result of this lookup in the cache.  Avoid this if the
    243 	 * request was for creation, as it does not improve timings on
    244 	 * emprical tests. */
    245 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
    246 		cache_enter(dvp, *vpp, cnp);
    247 
    248 out:
    249 	/* If there were no errors, *vpp cannot be null and it must be
    250 	 * locked. */
    251 	KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
    252 
    253 	/* dvp must always be locked. */
    254 	KASSERT(VOP_ISLOCKED(dvp));
    255 
    256 	return error;
    257 }
    258 
    259 /* --------------------------------------------------------------------- */
    260 
    261 int
    262 tmpfs_create(void *v)
    263 {
    264 	struct vnode *dvp = ((struct vop_create_args *)v)->a_dvp;
    265 	struct vnode **vpp = ((struct vop_create_args *)v)->a_vpp;
    266 	struct componentname *cnp = ((struct vop_create_args *)v)->a_cnp;
    267 	struct vattr *vap = ((struct vop_create_args *)v)->a_vap;
    268 
    269 	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
    270 
    271 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    272 }
    273 /* --------------------------------------------------------------------- */
    274 
    275 int
    276 tmpfs_mknod(void *v)
    277 {
    278 	struct vnode *dvp = ((struct vop_mknod_args *)v)->a_dvp;
    279 	struct vnode **vpp = ((struct vop_mknod_args *)v)->a_vpp;
    280 	struct componentname *cnp = ((struct vop_mknod_args *)v)->a_cnp;
    281 	struct vattr *vap = ((struct vop_mknod_args *)v)->a_vap;
    282 
    283 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
    284 	    vap->va_type != VFIFO)
    285 		return EINVAL;
    286 
    287 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    288 }
    289 
    290 /* --------------------------------------------------------------------- */
    291 
    292 int
    293 tmpfs_open(void *v)
    294 {
    295 	struct vnode *vp = ((struct vop_open_args *)v)->a_vp;
    296 	int mode = ((struct vop_open_args *)v)->a_mode;
    297 
    298 	int error;
    299 	struct tmpfs_node *node;
    300 
    301 	KASSERT(VOP_ISLOCKED(vp));
    302 
    303 	node = VP_TO_TMPFS_NODE(vp);
    304 
    305 	/* The file is still active but all its names have been removed
    306 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
    307 	 * it is about to die. */
    308 	if (node->tn_links < 1) {
    309 		error = ENOENT;
    310 		goto out;
    311 	}
    312 
    313 	/* If the file is marked append-only, deny write requests. */
    314 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
    315 		error = EPERM;
    316 	else
    317 		error = 0;
    318 
    319 out:
    320 	KASSERT(VOP_ISLOCKED(vp));
    321 
    322 	return error;
    323 }
    324 
    325 /* --------------------------------------------------------------------- */
    326 
    327 int
    328 tmpfs_close(void *v)
    329 {
    330 	struct vnode *vp = ((struct vop_close_args *)v)->a_vp;
    331 
    332 	struct tmpfs_node *node;
    333 
    334 	KASSERT(VOP_ISLOCKED(vp));
    335 
    336 	node = VP_TO_TMPFS_NODE(vp);
    337 
    338 	if (node->tn_links > 0) {
    339 		/* Update node times.  No need to do it if the node has
    340 		 * been deleted, because it will vanish after we return. */
    341 		tmpfs_update(vp, NULL, NULL, UPDATE_CLOSE);
    342 	}
    343 
    344 	return 0;
    345 }
    346 
    347 /* --------------------------------------------------------------------- */
    348 
    349 int
    350 tmpfs_access(void *v)
    351 {
    352 	struct vnode *vp = ((struct vop_access_args *)v)->a_vp;
    353 	int mode = ((struct vop_access_args *)v)->a_mode;
    354 	kauth_cred_t cred = ((struct vop_access_args *)v)->a_cred;
    355 
    356 	int error;
    357 	struct tmpfs_node *node;
    358 
    359 	KASSERT(VOP_ISLOCKED(vp));
    360 
    361 	node = VP_TO_TMPFS_NODE(vp);
    362 
    363 	switch (vp->v_type) {
    364 	case VDIR:
    365 		/* FALLTHROUGH */
    366 	case VLNK:
    367 		/* FALLTHROUGH */
    368 	case VREG:
    369 		if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
    370 			error = EROFS;
    371 			goto out;
    372 		}
    373 		break;
    374 
    375 	case VBLK:
    376 		/* FALLTHROUGH */
    377 	case VCHR:
    378 		/* FALLTHROUGH */
    379 	case VSOCK:
    380 		/* FALLTHROUGH */
    381 	case VFIFO:
    382 		break;
    383 
    384 	default:
    385 		error = EINVAL;
    386 		goto out;
    387 	}
    388 
    389 	if (mode & VWRITE && node->tn_flags & IMMUTABLE) {
    390 		error = EPERM;
    391 		goto out;
    392 	}
    393 
    394 	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
    395 	    node->tn_gid, mode, cred);
    396 
    397 out:
    398 	KASSERT(VOP_ISLOCKED(vp));
    399 
    400 	return error;
    401 }
    402 
    403 /* --------------------------------------------------------------------- */
    404 
    405 int
    406 tmpfs_getattr(void *v)
    407 {
    408 	struct vnode *vp = ((struct vop_getattr_args *)v)->a_vp;
    409 	struct vattr *vap = ((struct vop_getattr_args *)v)->a_vap;
    410 
    411 	struct tmpfs_node *node;
    412 
    413 	node = VP_TO_TMPFS_NODE(vp);
    414 
    415 	VATTR_NULL(vap);
    416 
    417 	tmpfs_itimes(vp, NULL, NULL);
    418 
    419 	vap->va_type = vp->v_type;
    420 	vap->va_mode = node->tn_mode;
    421 	vap->va_nlink = node->tn_links;
    422 	vap->va_uid = node->tn_uid;
    423 	vap->va_gid = node->tn_gid;
    424 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
    425 	vap->va_fileid = node->tn_id;
    426 	vap->va_size = node->tn_size;
    427 	vap->va_blocksize = PAGE_SIZE;
    428 	vap->va_atime = node->tn_atime;
    429 	vap->va_mtime = node->tn_mtime;
    430 	vap->va_ctime = node->tn_ctime;
    431 	vap->va_birthtime = node->tn_birthtime;
    432 	vap->va_gen = node->tn_gen;
    433 	vap->va_flags = node->tn_flags;
    434 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
    435 		node->tn_spec.tn_dev.tn_rdev : VNOVAL;
    436 	vap->va_bytes = round_page(node->tn_size);
    437 	vap->va_filerev = VNOVAL;
    438 	vap->va_vaflags = 0;
    439 	vap->va_spare = VNOVAL; /* XXX */
    440 
    441 	return 0;
    442 }
    443 
    444 /* --------------------------------------------------------------------- */
    445 
    446 /* XXX Should this operation be atomic?  I think it should, but code in
    447  * XXX other places (e.g., ufs) doesn't seem to be... */
    448 int
    449 tmpfs_setattr(void *v)
    450 {
    451 	struct vnode *vp = ((struct vop_setattr_args *)v)->a_vp;
    452 	struct vattr *vap = ((struct vop_setattr_args *)v)->a_vap;
    453 	kauth_cred_t cred = ((struct vop_setattr_args *)v)->a_cred;
    454 	struct lwp *l = ((struct vop_setattr_args *)v)->a_l;
    455 
    456 	int error;
    457 
    458 	KASSERT(VOP_ISLOCKED(vp));
    459 
    460 	error = 0;
    461 
    462 	/* Abort if any unsettable attribute is given. */
    463 	if (vap->va_type != VNON ||
    464 	    vap->va_nlink != VNOVAL ||
    465 	    vap->va_fsid != VNOVAL ||
    466 	    vap->va_fileid != VNOVAL ||
    467 	    vap->va_blocksize != VNOVAL ||
    468 	    vap->va_ctime.tv_sec != VNOVAL ||
    469 	    vap->va_ctime.tv_nsec != VNOVAL ||
    470 	    vap->va_birthtime.tv_sec != VNOVAL ||
    471 	    vap->va_birthtime.tv_nsec != VNOVAL ||
    472 	    vap->va_gen != VNOVAL ||
    473 	    vap->va_rdev != VNOVAL ||
    474 	    vap->va_bytes != VNOVAL)
    475 		error = EINVAL;
    476 
    477 	if (error == 0 && (vap->va_flags != VNOVAL))
    478 		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
    479 
    480 	if (error == 0 && (vap->va_size != VNOVAL))
    481 		error = tmpfs_chsize(vp, vap->va_size, cred, l);
    482 
    483 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
    484 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
    485 
    486 	if (error == 0 && (vap->va_mode != VNOVAL))
    487 		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
    488 
    489 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
    490 	    vap->va_atime.tv_nsec != VNOVAL) ||
    491 	    (vap->va_mtime.tv_sec != VNOVAL &&
    492 	    vap->va_mtime.tv_nsec != VNOVAL)))
    493 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
    494 		    vap->va_vaflags, cred, l);
    495 
    496 	/* Update the node times.  We give preference to the error codes
    497 	 * generated by this function rather than the ones that may arise
    498 	 * from tmpfs_update. */
    499 	tmpfs_update(vp, NULL, NULL, 0);
    500 
    501 	KASSERT(VOP_ISLOCKED(vp));
    502 
    503 	return error;
    504 }
    505 
    506 /* --------------------------------------------------------------------- */
    507 
    508 int
    509 tmpfs_read(void *v)
    510 {
    511 	struct vnode *vp = ((struct vop_read_args *)v)->a_vp;
    512 	struct uio *uio = ((struct vop_read_args *)v)->a_uio;
    513 
    514 	int error;
    515 	int flags;
    516 	struct tmpfs_node *node;
    517 	struct uvm_object *uobj;
    518 
    519 	KASSERT(VOP_ISLOCKED(vp));
    520 
    521 	node = VP_TO_TMPFS_NODE(vp);
    522 
    523 	if (vp->v_type != VREG) {
    524 		error = EISDIR;
    525 		goto out;
    526 	}
    527 
    528 	if (uio->uio_offset < 0) {
    529 		error = EINVAL;
    530 		goto out;
    531 	}
    532 
    533 	node->tn_status |= TMPFS_NODE_ACCESSED;
    534 
    535 	uobj = node->tn_spec.tn_reg.tn_aobj;
    536 	flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
    537 	error = 0;
    538 	while (error == 0 && uio->uio_resid > 0) {
    539 		vsize_t len;
    540 		void *win;
    541 
    542 		if (node->tn_size <= uio->uio_offset)
    543 			break;
    544 
    545 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    546 		if (len == 0)
    547 			break;
    548 
    549 		win = ubc_alloc(uobj, uio->uio_offset, &len, UVM_ADV_NORMAL,
    550 		    UBC_READ);
    551 		error = uiomove(win, len, uio);
    552 		ubc_release(win, flags);
    553 	}
    554 
    555 out:
    556 	KASSERT(VOP_ISLOCKED(vp));
    557 
    558 	return error;
    559 }
    560 
    561 /* --------------------------------------------------------------------- */
    562 
    563 int
    564 tmpfs_write(void *v)
    565 {
    566 	struct vnode *vp = ((struct vop_write_args *)v)->a_vp;
    567 	struct uio *uio = ((struct vop_write_args *)v)->a_uio;
    568 	int ioflag = ((struct vop_write_args *)v)->a_ioflag;
    569 
    570 	bool extended;
    571 	int error;
    572 	int flags;
    573 	off_t oldsize;
    574 	struct tmpfs_node *node;
    575 	struct uvm_object *uobj;
    576 
    577 	KASSERT(VOP_ISLOCKED(vp));
    578 
    579 	node = VP_TO_TMPFS_NODE(vp);
    580 	oldsize = node->tn_size;
    581 
    582 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
    583 		error = EINVAL;
    584 		goto out;
    585 	}
    586 
    587 	if (uio->uio_resid == 0) {
    588 		error = 0;
    589 		goto out;
    590 	}
    591 
    592 	if (ioflag & IO_APPEND)
    593 		uio->uio_offset = node->tn_size;
    594 
    595 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
    596 	if (extended) {
    597 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
    598 		if (error != 0)
    599 			goto out;
    600 	}
    601 
    602 	uobj = node->tn_spec.tn_reg.tn_aobj;
    603 	flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
    604 	error = 0;
    605 	while (error == 0 && uio->uio_resid > 0) {
    606 		vsize_t len;
    607 		void *win;
    608 
    609 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    610 		if (len == 0)
    611 			break;
    612 
    613 		win = ubc_alloc(uobj, uio->uio_offset, &len, UVM_ADV_NORMAL,
    614 		    UBC_WRITE);
    615 		error = uiomove(win, len, uio);
    616 		ubc_release(win, flags);
    617 	}
    618 
    619 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
    620 	    (extended ? TMPFS_NODE_CHANGED : 0);
    621 
    622 	if (error != 0)
    623 		(void)tmpfs_reg_resize(vp, oldsize);
    624 
    625 	VN_KNOTE(vp, NOTE_WRITE);
    626 
    627 out:
    628 	KASSERT(VOP_ISLOCKED(vp));
    629 	KASSERT(IMPLIES(error == 0, uio->uio_resid == 0));
    630 	KASSERT(IMPLIES(error != 0, oldsize == node->tn_size));
    631 
    632 	return error;
    633 }
    634 
    635 /* --------------------------------------------------------------------- */
    636 
    637 int
    638 tmpfs_fsync(void *v)
    639 {
    640 	struct vnode *vp = ((struct vop_fsync_args *)v)->a_vp;
    641 
    642 	KASSERT(VOP_ISLOCKED(vp));
    643 
    644 	tmpfs_update(vp, NULL, NULL, 0);
    645 
    646 	return 0;
    647 }
    648 
    649 /* --------------------------------------------------------------------- */
    650 
    651 int
    652 tmpfs_remove(void *v)
    653 {
    654 	struct vnode *dvp = ((struct vop_remove_args *)v)->a_dvp;
    655 	struct vnode *vp = ((struct vop_remove_args *)v)->a_vp;
    656 
    657 	int error;
    658 	struct tmpfs_dirent *de;
    659 	struct tmpfs_mount *tmp;
    660 	struct tmpfs_node *dnode;
    661 	struct tmpfs_node *node;
    662 
    663 	KASSERT(VOP_ISLOCKED(dvp));
    664 	KASSERT(VOP_ISLOCKED(vp));
    665 
    666 	if (vp->v_type == VDIR) {
    667 		error = EPERM;
    668 		goto out;
    669 	}
    670 
    671 	dnode = VP_TO_TMPFS_DIR(dvp);
    672 	node = VP_TO_TMPFS_NODE(vp);
    673 	tmp = VFS_TO_TMPFS(vp->v_mount);
    674 	de = node->tn_lookup_dirent;
    675 	KASSERT(de != NULL);
    676 
    677 	/* Files marked as immutable or append-only cannot be deleted. */
    678 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    679 		error = EPERM;
    680 		goto out;
    681 	}
    682 
    683 	/* Remove the entry from the directory; as it is a file, we do not
    684 	 * have to change the number of hard links of the directory. */
    685 	tmpfs_dir_detach(dvp, de);
    686 
    687 	/* Free the directory entry we just deleted.  Note that the node
    688 	 * referred by it will not be removed until the vnode is really
    689 	 * reclaimed. */
    690 	tmpfs_free_dirent(tmp, de, true);
    691 
    692 	error = 0;
    693 
    694 out:
    695 	vput(vp);
    696 	if (dvp == vp)
    697 		vrele(dvp);
    698 	else
    699 		vput(dvp);
    700 
    701 	KASSERT(!VOP_ISLOCKED(dvp));
    702 
    703 	return error;
    704 }
    705 
    706 /* --------------------------------------------------------------------- */
    707 
    708 int
    709 tmpfs_link(void *v)
    710 {
    711 	struct vnode *dvp = ((struct vop_link_args *)v)->a_dvp;
    712 	struct vnode *vp = ((struct vop_link_args *)v)->a_vp;
    713 	struct componentname *cnp = ((struct vop_link_args *)v)->a_cnp;
    714 
    715 	int error;
    716 	struct tmpfs_dirent *de;
    717 	struct tmpfs_node *dnode;
    718 	struct tmpfs_node *node;
    719 
    720 	KASSERT(VOP_ISLOCKED(dvp));
    721 	KASSERT(!VOP_ISLOCKED(vp));
    722 	KASSERT(cnp->cn_flags & HASBUF);
    723 	KASSERT(dvp != vp); /* XXX When can this be false? */
    724 
    725 	dnode = VP_TO_TMPFS_DIR(dvp);
    726 	node = VP_TO_TMPFS_NODE(vp);
    727 
    728 	/* Lock vp because we will need to run tmpfs_update over it, which
    729 	 * needs the vnode to be locked. */
    730 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    731 	if (error != 0)
    732 		goto out;
    733 
    734 	/* XXX: Why aren't the following two tests done by the caller? */
    735 
    736 	/* Hard links of directories are forbidden. */
    737 	if (vp->v_type == VDIR) {
    738 		error = EPERM;
    739 		goto out;
    740 	}
    741 
    742 	/* Cannot create cross-device links. */
    743 	if (dvp->v_mount != vp->v_mount) {
    744 		error = EXDEV;
    745 		goto out;
    746 	}
    747 
    748 	/* Ensure that we do not overflow the maximum number of links imposed
    749 	 * by the system. */
    750 	KASSERT(node->tn_links <= LINK_MAX);
    751 	if (node->tn_links == LINK_MAX) {
    752 		error = EMLINK;
    753 		goto out;
    754 	}
    755 
    756 	/* We cannot create links of files marked immutable or append-only. */
    757 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    758 		error = EPERM;
    759 		goto out;
    760 	}
    761 
    762 	/* Allocate a new directory entry to represent the node. */
    763 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
    764 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
    765 	if (error != 0)
    766 		goto out;
    767 
    768 	/* Insert the new directory entry into the appropriate directory. */
    769 	tmpfs_dir_attach(dvp, de);
    770 
    771 	/* vp link count has changed, so update node times. */
    772 	node->tn_status |= TMPFS_NODE_CHANGED;
    773 	tmpfs_update(vp, NULL, NULL, 0);
    774 
    775 	error = 0;
    776 
    777 out:
    778 	if (VOP_ISLOCKED(vp))
    779 		VOP_UNLOCK(vp, 0);
    780 
    781 	PNBUF_PUT(cnp->cn_pnbuf);
    782 
    783 	vput(dvp);
    784 
    785 	/* XXX Locking status of dvp does not match manual page. */
    786 	KASSERT(!VOP_ISLOCKED(dvp));
    787 	KASSERT(!VOP_ISLOCKED(vp));
    788 
    789 	return error;
    790 }
    791 
    792 /* --------------------------------------------------------------------- */
    793 
    794 int
    795 tmpfs_rename(void *v)
    796 {
    797 	struct vnode *fdvp = ((struct vop_rename_args *)v)->a_fdvp;
    798 	struct vnode *fvp = ((struct vop_rename_args *)v)->a_fvp;
    799 	struct componentname *fcnp = ((struct vop_rename_args *)v)->a_fcnp;
    800 	struct vnode *tdvp = ((struct vop_rename_args *)v)->a_tdvp;
    801 	struct vnode *tvp = ((struct vop_rename_args *)v)->a_tvp;
    802 	struct componentname *tcnp = ((struct vop_rename_args *)v)->a_tcnp;
    803 
    804 	char *newname;
    805 	int error;
    806 	struct tmpfs_dirent *de;
    807 	struct tmpfs_mount *tmp;
    808 	struct tmpfs_node *fdnode;
    809 	struct tmpfs_node *fnode;
    810 	struct tmpfs_node *tnode;
    811 	struct tmpfs_node *tdnode;
    812 
    813 	KASSERT(VOP_ISLOCKED(tdvp));
    814 	KASSERT(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
    815 	KASSERT(fcnp->cn_flags & HASBUF);
    816 	KASSERT(tcnp->cn_flags & HASBUF);
    817 
    818 	fdnode = VP_TO_TMPFS_DIR(fdvp);
    819 	fnode = VP_TO_TMPFS_NODE(fvp);
    820 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
    821 	de = fnode->tn_lookup_dirent;
    822 
    823 	/* Disallow cross-device renames.
    824 	 * XXX Why isn't this done by the caller? */
    825 	if (fvp->v_mount != tdvp->v_mount ||
    826 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
    827 		error = EXDEV;
    828 		goto out;
    829 	}
    830 
    831 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
    832 	tdnode = VP_TO_TMPFS_DIR(tdvp);
    833 
    834 	/* If source and target are the same file, there is nothing to do. */
    835 	if (fvp == tvp) {
    836 		error = 0;
    837 		goto out;
    838 	}
    839 
    840 	/* Avoid manipulating '.' and '..' entries. */
    841 	if (de == NULL) {
    842 		KASSERT(fvp->v_type == VDIR);
    843 		error = EINVAL;
    844 		goto out;
    845 	}
    846 	KASSERT(de->td_node == fnode);
    847 
    848 	/* If replacing an existing entry, ensure we can do the operation. */
    849 	if (tvp != NULL) {
    850 		KASSERT(tnode != NULL);
    851 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
    852 			if (tnode->tn_size > 0) {
    853 				error = ENOTEMPTY;
    854 				goto out;
    855 			}
    856 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
    857 			error = ENOTDIR;
    858 			goto out;
    859 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
    860 			error = EISDIR;
    861 			goto out;
    862 		} else {
    863 			KASSERT(fnode->tn_type != VDIR &&
    864 			        tnode->tn_type != VDIR);
    865 		}
    866 	}
    867 
    868 	/* If we need to move the directory between entries, lock the
    869 	 * source so that we can safely operate on it. */
    870 	if (fdnode != tdnode) {
    871 		error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
    872 		if (error != 0)
    873 			goto out;
    874 	}
    875 
    876 	/* Ensure that we have enough memory to hold the new name, if it
    877 	 * has to be changed. */
    878 	if (fcnp->cn_namelen != tcnp->cn_namelen ||
    879 	    memcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
    880 		newname = tmpfs_str_pool_get(&tmp->tm_str_pool,
    881 		    tcnp->cn_namelen, 0);
    882 		if (newname == NULL) {
    883 			error = ENOSPC;
    884 			goto out_locked;
    885 		}
    886 	} else
    887 		newname = NULL;
    888 
    889 	/* If the node is being moved to another directory, we have to do
    890 	 * the move. */
    891 	if (fdnode != tdnode) {
    892 		/* In case we are moving a directory, we have to adjust its
    893 		 * parent to point to the new parent. */
    894 		if (de->td_node->tn_type == VDIR) {
    895 			struct tmpfs_node *n;
    896 
    897 			/* Ensure the target directory is not a child of the
    898 			 * directory being moved.  Otherwise, we'd end up
    899 			 * with stale nodes. */
    900 			n = tdnode;
    901 			while (n != n->tn_spec.tn_dir.tn_parent) {
    902 				if (n == fnode) {
    903 					error = EINVAL;
    904 					goto out_locked;
    905 				}
    906 				n = n->tn_spec.tn_dir.tn_parent;
    907 			}
    908 
    909 			/* Adjust the parent pointer. */
    910 			TMPFS_VALIDATE_DIR(fnode);
    911 			de->td_node->tn_spec.tn_dir.tn_parent = tdnode;
    912 
    913 			/* As a result of changing the target of the '..'
    914 			 * entry, the link count of the source and target
    915 			 * directories has to be adjusted. */
    916 			fdnode->tn_links--;
    917 			tdnode->tn_links++;
    918 		}
    919 
    920 		/* Do the move: just remove the entry from the source directory
    921 		 * and insert it into the target one. */
    922 		tmpfs_dir_detach(fdvp, de);
    923 		tmpfs_dir_attach(tdvp, de);
    924 
    925 		/* Notify listeners of fdvp about the change in the directory.
    926 		 * We can do it at this point because we aren't touching fdvp
    927 		 * any more below. */
    928 		VN_KNOTE(fdvp, NOTE_WRITE);
    929 	}
    930 
    931 	/* If the name has changed, we need to make it effective by changing
    932 	 * it in the directory entry. */
    933 	if (newname != NULL) {
    934 		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
    935 		KASSERT(tcnp->cn_namelen < 0xffff);
    936 
    937 		tmpfs_str_pool_put(&tmp->tm_str_pool, de->td_name,
    938 		    de->td_namelen);
    939 		de->td_namelen = (uint16_t)tcnp->cn_namelen;
    940 		memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
    941 		de->td_name = newname;
    942 
    943 		fnode->tn_status |= TMPFS_NODE_CHANGED;
    944 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
    945 	}
    946 
    947 	/* If we are overwriting an entry, we have to remove the old one
    948 	 * from the target directory. */
    949 	if (tvp != NULL) {
    950 		KASSERT(tnode != NULL);
    951 
    952 		/* Remove the old entry from the target directory. */
    953 		de = tnode->tn_lookup_dirent;
    954 		tmpfs_dir_detach(tdvp, de);
    955 
    956 		/* Free the directory entry we just deleted.  Note that the
    957 		 * node referred by it will not be removed until the vnode is
    958 		 * really reclaimed. */
    959 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, true);
    960 	}
    961 
    962 	/* Notify listeners of tdvp about the change in the directory (either
    963 	 * because a new entry was added or because one was removed) and
    964 	 * listeners of fvp about the rename. */
    965 	VN_KNOTE(tdvp, NOTE_WRITE);
    966 	VN_KNOTE(fvp, NOTE_RENAME);
    967 
    968 	error = 0;
    969 
    970 out_locked:
    971 	if (fdnode != tdnode)
    972 		VOP_UNLOCK(fdvp, 0);
    973 
    974 out:
    975 	/* Release target nodes. */
    976 	/* XXX: I don't understand when tdvp can be the same as tvp, but
    977 	 * other code takes care of this... */
    978 	if (tdvp == tvp)
    979 		vrele(tdvp);
    980 	else
    981 		vput(tdvp);
    982 	if (tvp != NULL)
    983 		vput(tvp);
    984 
    985 	/* Release source nodes. */
    986 	vrele(fdvp);
    987 	vrele(fvp);
    988 
    989 	return error;
    990 }
    991 
    992 /* --------------------------------------------------------------------- */
    993 
    994 int
    995 tmpfs_mkdir(void *v)
    996 {
    997 	struct vnode *dvp = ((struct vop_mkdir_args *)v)->a_dvp;
    998 	struct vnode **vpp = ((struct vop_mkdir_args *)v)->a_vpp;
    999 	struct componentname *cnp = ((struct vop_mkdir_args *)v)->a_cnp;
   1000 	struct vattr *vap = ((struct vop_mkdir_args *)v)->a_vap;
   1001 
   1002 	KASSERT(vap->va_type == VDIR);
   1003 
   1004 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
   1005 }
   1006 
   1007 /* --------------------------------------------------------------------- */
   1008 
   1009 int
   1010 tmpfs_rmdir(void *v)
   1011 {
   1012 	struct vnode *dvp = ((struct vop_rmdir_args *)v)->a_dvp;
   1013 	struct vnode *vp = ((struct vop_rmdir_args *)v)->a_vp;
   1014 
   1015 	int error;
   1016 	struct tmpfs_dirent *de;
   1017 	struct tmpfs_mount *tmp;
   1018 	struct tmpfs_node *dnode;
   1019 	struct tmpfs_node *node;
   1020 
   1021 	KASSERT(VOP_ISLOCKED(dvp));
   1022 	KASSERT(VOP_ISLOCKED(vp));
   1023 
   1024 	tmp = VFS_TO_TMPFS(dvp->v_mount);
   1025 	dnode = VP_TO_TMPFS_DIR(dvp);
   1026 	node = VP_TO_TMPFS_DIR(vp);
   1027 
   1028 	/* Directories with more than two entries ('.' and '..') cannot be
   1029 	 * removed. */
   1030 	if (node->tn_size > 0) {
   1031 		error = ENOTEMPTY;
   1032 		goto out;
   1033 	}
   1034 
   1035 	/* This invariant holds only if we are not trying to remove "..".
   1036 	 * We checked for that above so this is safe now. */
   1037 	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
   1038 
   1039 	/* Get the directory entry associated with node (vp).  This was
   1040 	 * filled by tmpfs_lookup while looking up the entry. */
   1041 	de = node->tn_lookup_dirent;
   1042 	KASSERT(TMPFS_DIRENT_MATCHES(de,
   1043 	    ((struct vop_rmdir_args *)v)->a_cnp->cn_nameptr,
   1044 	    ((struct vop_rmdir_args *)v)->a_cnp->cn_namelen));
   1045 
   1046 	/* Check flags to see if we are allowed to remove the directory. */
   1047 	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
   1048 		error = EPERM;
   1049 		goto out;
   1050 	}
   1051 
   1052 	/* Detach the directory entry from the directory (dnode). */
   1053 	tmpfs_dir_detach(dvp, de);
   1054 
   1055 	node->tn_links--;
   1056 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
   1057 	    TMPFS_NODE_MODIFIED;
   1058 	node->tn_spec.tn_dir.tn_parent->tn_links--;
   1059 	node->tn_spec.tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \
   1060 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
   1061 
   1062 	/* Release the parent. */
   1063 	cache_purge(dvp); /* XXX Is this needed? */
   1064 	vput(dvp);
   1065 
   1066 	/* Free the directory entry we just deleted.  Note that the node
   1067 	 * referred by it will not be removed until the vnode is really
   1068 	 * reclaimed. */
   1069 	tmpfs_free_dirent(tmp, de, true);
   1070 
   1071 	/* Release the deleted vnode (will destroy the node, notify
   1072 	 * interested parties and clean it from the cache). */
   1073 	vput(vp);
   1074 
   1075 	error = 0;
   1076 
   1077 out:
   1078 	if (error != 0) {
   1079 		vput(dvp);
   1080 		vput(vp);
   1081 	}
   1082 
   1083 	return error;
   1084 }
   1085 
   1086 /* --------------------------------------------------------------------- */
   1087 
   1088 int
   1089 tmpfs_symlink(void *v)
   1090 {
   1091 	struct vnode *dvp = ((struct vop_symlink_args *)v)->a_dvp;
   1092 	struct vnode **vpp = ((struct vop_symlink_args *)v)->a_vpp;
   1093 	struct componentname *cnp = ((struct vop_symlink_args *)v)->a_cnp;
   1094 	struct vattr *vap = ((struct vop_symlink_args *)v)->a_vap;
   1095 	char *target = ((struct vop_symlink_args *)v)->a_target;
   1096 
   1097 	KASSERT(vap->va_type == VLNK);
   1098 
   1099 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
   1100 }
   1101 
   1102 /* --------------------------------------------------------------------- */
   1103 
   1104 int
   1105 tmpfs_readdir(void *v)
   1106 {
   1107 	struct vnode *vp = ((struct vop_readdir_args *)v)->a_vp;
   1108 	struct uio *uio = ((struct vop_readdir_args *)v)->a_uio;
   1109 	int *eofflag = ((struct vop_readdir_args *)v)->a_eofflag;
   1110 	off_t **cookies = ((struct vop_readdir_args *)v)->a_cookies;
   1111 	int *ncookies = ((struct vop_readdir_args *)v)->a_ncookies;
   1112 
   1113 	int error;
   1114 	off_t startoff;
   1115 	off_t cnt;
   1116 	struct tmpfs_node *node;
   1117 
   1118 	KASSERT(VOP_ISLOCKED(vp));
   1119 
   1120 	/* This operation only makes sense on directory nodes. */
   1121 	if (vp->v_type != VDIR) {
   1122 		error = ENOTDIR;
   1123 		goto out;
   1124 	}
   1125 
   1126 	node = VP_TO_TMPFS_DIR(vp);
   1127 
   1128 	startoff = uio->uio_offset;
   1129 
   1130 	cnt = 0;
   1131 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
   1132 		error = tmpfs_dir_getdotdent(node, uio);
   1133 		if (error == -1) {
   1134 			error = 0;
   1135 			goto outok;
   1136 		} else if (error != 0)
   1137 			goto outok;
   1138 		cnt++;
   1139 	}
   1140 
   1141 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
   1142 		error = tmpfs_dir_getdotdotdent(node, uio);
   1143 		if (error == -1) {
   1144 			error = 0;
   1145 			goto outok;
   1146 		} else if (error != 0)
   1147 			goto outok;
   1148 		cnt++;
   1149 	}
   1150 
   1151 	error = tmpfs_dir_getdents(node, uio, &cnt);
   1152 	if (error == -1)
   1153 		error = 0;
   1154 	KASSERT(error >= 0);
   1155 
   1156 outok:
   1157 	/* This label assumes that startoff has been
   1158 	 * initialized.  If the compiler didn't spit out warnings, we'd
   1159 	 * simply make this one be 'out' and drop 'outok'. */
   1160 
   1161 	if (eofflag != NULL)
   1162 		*eofflag =
   1163 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
   1164 
   1165 	/* Update NFS-related variables. */
   1166 	if (error == 0 && cookies != NULL && ncookies != NULL) {
   1167 		off_t i;
   1168 		off_t off = startoff;
   1169 		struct tmpfs_dirent *de = NULL;
   1170 
   1171 		*ncookies = cnt;
   1172 		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
   1173 
   1174 		for (i = 0; i < cnt; i++) {
   1175 			KASSERT(off != TMPFS_DIRCOOKIE_EOF);
   1176 			if (off == TMPFS_DIRCOOKIE_DOT) {
   1177 				off = TMPFS_DIRCOOKIE_DOTDOT;
   1178 			} else {
   1179 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
   1180 					de = TAILQ_FIRST(&node->tn_spec.
   1181 					    tn_dir.tn_dir);
   1182 				} else if (de != NULL) {
   1183 					de = TAILQ_NEXT(de, td_entries);
   1184 				} else {
   1185 					de = tmpfs_dir_lookupbycookie(node,
   1186 					    off);
   1187 					KASSERT(de != NULL);
   1188 					de = TAILQ_NEXT(de, td_entries);
   1189 				}
   1190 				if (de == NULL) {
   1191 					off = TMPFS_DIRCOOKIE_EOF;
   1192 				} else {
   1193 					off = tmpfs_dircookie(de);
   1194 				}
   1195 			}
   1196 
   1197 			(*cookies)[i] = off;
   1198 		}
   1199 		KASSERT(uio->uio_offset == off);
   1200 	}
   1201 
   1202 out:
   1203 	KASSERT(VOP_ISLOCKED(vp));
   1204 
   1205 	return error;
   1206 }
   1207 
   1208 /* --------------------------------------------------------------------- */
   1209 
   1210 int
   1211 tmpfs_readlink(void *v)
   1212 {
   1213 	struct vnode *vp = ((struct vop_readlink_args *)v)->a_vp;
   1214 	struct uio *uio = ((struct vop_readlink_args *)v)->a_uio;
   1215 
   1216 	int error;
   1217 	struct tmpfs_node *node;
   1218 
   1219 	KASSERT(VOP_ISLOCKED(vp));
   1220 	KASSERT(uio->uio_offset == 0);
   1221 	KASSERT(vp->v_type == VLNK);
   1222 
   1223 	node = VP_TO_TMPFS_NODE(vp);
   1224 
   1225 	error = uiomove(node->tn_spec.tn_lnk.tn_link,
   1226 	    MIN(node->tn_size, uio->uio_resid), uio);
   1227 	node->tn_status |= TMPFS_NODE_ACCESSED;
   1228 
   1229 	KASSERT(VOP_ISLOCKED(vp));
   1230 
   1231 	return error;
   1232 }
   1233 
   1234 /* --------------------------------------------------------------------- */
   1235 
   1236 int
   1237 tmpfs_inactive(void *v)
   1238 {
   1239 	struct vnode *vp = ((struct vop_inactive_args *)v)->a_vp;
   1240 	struct lwp *l = ((struct vop_inactive_args *)v)->a_l;
   1241 	nlink_t links;
   1242 
   1243 	struct tmpfs_node *node;
   1244 
   1245 	KASSERT(VOP_ISLOCKED(vp));
   1246 
   1247 	node = VP_TO_TMPFS_NODE(vp);
   1248 	links = node->tn_links;
   1249 
   1250 	VOP_UNLOCK(vp, 0);
   1251 
   1252 	if (links == 0)
   1253 		vrecycle(vp, NULL, l);
   1254 
   1255 	return 0;
   1256 }
   1257 
   1258 /* --------------------------------------------------------------------- */
   1259 
   1260 int
   1261 tmpfs_reclaim(void *v)
   1262 {
   1263 	struct vnode *vp = ((struct vop_reclaim_args *)v)->a_vp;
   1264 
   1265 	struct tmpfs_mount *tmp;
   1266 	struct tmpfs_node *node;
   1267 
   1268 	KASSERT(!VOP_ISLOCKED(vp));
   1269 
   1270 	node = VP_TO_TMPFS_NODE(vp);
   1271 	tmp = VFS_TO_TMPFS(vp->v_mount);
   1272 
   1273 	cache_purge(vp);
   1274 	tmpfs_free_vp(vp);
   1275 
   1276 	/* If the node referenced by this vnode was deleted by the user,
   1277 	 * we must free its associated data structures (now that the vnode
   1278 	 * is being reclaimed). */
   1279 	if (node->tn_links == 0)
   1280 		tmpfs_free_node(tmp, node);
   1281 
   1282 	KASSERT(!VOP_ISLOCKED(vp));
   1283 	KASSERT(vp->v_data == NULL);
   1284 
   1285 	return 0;
   1286 }
   1287 
   1288 /* --------------------------------------------------------------------- */
   1289 
   1290 int
   1291 tmpfs_print(void *v)
   1292 {
   1293 	struct vnode *vp = ((struct vop_print_args *)v)->a_vp;
   1294 
   1295 	struct tmpfs_node *node;
   1296 
   1297 	node = VP_TO_TMPFS_NODE(vp);
   1298 
   1299 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
   1300 	    node, node->tn_flags, node->tn_links);
   1301 	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
   1302 	    ", status 0x%x\n",
   1303 	    node->tn_mode, node->tn_uid, node->tn_gid,
   1304 	    (uintmax_t)node->tn_size, node->tn_status);
   1305 
   1306 	if (vp->v_type == VFIFO)
   1307 		fifo_printinfo(vp);
   1308 	lockmgr_printinfo(&vp->v_lock);
   1309 
   1310 	printf("\n");
   1311 
   1312 	return 0;
   1313 }
   1314 
   1315 /* --------------------------------------------------------------------- */
   1316 
   1317 int
   1318 tmpfs_pathconf(void *v)
   1319 {
   1320 	int name = ((struct vop_pathconf_args *)v)->a_name;
   1321 	register_t *retval = ((struct vop_pathconf_args *)v)->a_retval;
   1322 
   1323 	int error;
   1324 
   1325 	error = 0;
   1326 
   1327 	switch (name) {
   1328 	case _PC_LINK_MAX:
   1329 		*retval = LINK_MAX;
   1330 		break;
   1331 
   1332 	case _PC_NAME_MAX:
   1333 		*retval = NAME_MAX;
   1334 		break;
   1335 
   1336 	case _PC_PATH_MAX:
   1337 		*retval = PATH_MAX;
   1338 		break;
   1339 
   1340 	case _PC_PIPE_BUF:
   1341 		*retval = PIPE_BUF;
   1342 		break;
   1343 
   1344 	case _PC_CHOWN_RESTRICTED:
   1345 		*retval = 1;
   1346 		break;
   1347 
   1348 	case _PC_NO_TRUNC:
   1349 		*retval = 1;
   1350 		break;
   1351 
   1352 	case _PC_SYNC_IO:
   1353 		*retval = 1;
   1354 		break;
   1355 
   1356 	case _PC_FILESIZEBITS:
   1357 		*retval = 0; /* XXX Don't know which value should I return. */
   1358 		break;
   1359 
   1360 	default:
   1361 		error = EINVAL;
   1362 	}
   1363 
   1364 	return error;
   1365 }
   1366 
   1367 /* --------------------------------------------------------------------- */
   1368 
   1369 int
   1370 tmpfs_advlock(void *v)
   1371 {
   1372 	struct vnode *vp = ((struct vop_advlock_args *)v)->a_vp;
   1373 
   1374 	struct tmpfs_node *node;
   1375 
   1376 	node = VP_TO_TMPFS_NODE(vp);
   1377 
   1378 	return lf_advlock(v, &node->tn_lockf, node->tn_size);
   1379 }
   1380 
   1381 /* --------------------------------------------------------------------- */
   1382 
   1383 int
   1384 tmpfs_getpages(void *v)
   1385 {
   1386 	struct vnode *vp = ((struct vop_getpages_args *)v)->a_vp;
   1387 	voff_t offset = ((struct vop_getpages_args *)v)->a_offset;
   1388 	struct vm_page **m = ((struct vop_getpages_args *)v)->a_m;
   1389 	int *count = ((struct vop_getpages_args *)v)->a_count;
   1390 	int centeridx = ((struct vop_getpages_args *)v)->a_centeridx;
   1391 	vm_prot_t access_type = ((struct vop_getpages_args *)v)->a_access_type;
   1392 	int advice = ((struct vop_getpages_args *)v)->a_advice;
   1393 	int flags = ((struct vop_getpages_args *)v)->a_flags;
   1394 
   1395 	int error;
   1396 	int i;
   1397 	struct tmpfs_node *node;
   1398 	struct uvm_object *uobj;
   1399 	int npages = *count;
   1400 
   1401 	KASSERT(vp->v_type == VREG);
   1402 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
   1403 
   1404 	node = VP_TO_TMPFS_NODE(vp);
   1405 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1406 
   1407 	/* We currently don't rely on PGO_PASTEOF. */
   1408 
   1409 	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
   1410 		if ((flags & PGO_LOCKED) == 0)
   1411 			simple_unlock(&vp->v_interlock);
   1412 		return EINVAL;
   1413 	}
   1414 
   1415 	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
   1416 		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
   1417 	}
   1418 
   1419 	if ((flags & PGO_LOCKED) != 0)
   1420 		return EBUSY;
   1421 
   1422 	if ((flags & PGO_NOTIMESTAMP) == 0) {
   1423 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
   1424 			node->tn_status |= TMPFS_NODE_ACCESSED;
   1425 
   1426 		if ((access_type & VM_PROT_WRITE) != 0)
   1427 			node->tn_status |= TMPFS_NODE_MODIFIED;
   1428 	}
   1429 
   1430 	simple_unlock(&vp->v_interlock);
   1431 
   1432 	/*
   1433 	 * Make sure that the array on which we will store the
   1434 	 * gotten pages is clean.  Otherwise uao_get (pointed to by
   1435 	 * the pgo_get below) gets confused and does not return the
   1436 	 * appropriate pages.
   1437 	 *
   1438 	 * XXX This shall be revisited when kern/32166 is addressed
   1439 	 * because the loop to clean m[i] will most likely be redundant
   1440 	 * as well as the PGO_ALLPAGES flag.
   1441 	 */
   1442 	if (m != NULL)
   1443 		for (i = 0; i < npages; i++)
   1444 			m[i] = NULL;
   1445 	simple_lock(&uobj->vmobjlock);
   1446 	error = (*uobj->pgops->pgo_get)(uobj, offset, m, &npages, centeridx,
   1447 	    access_type, advice, flags | PGO_ALLPAGES);
   1448 #if defined(DEBUG)
   1449 	{
   1450 		/* Make sure that all the pages we return are valid. */
   1451 		int dbgi;
   1452 		if (error == 0 && m != NULL)
   1453 			for (dbgi = 0; dbgi < npages; dbgi++)
   1454 				KASSERT(m[dbgi] != NULL);
   1455 	}
   1456 #endif
   1457 
   1458 	return error;
   1459 }
   1460 
   1461 /* --------------------------------------------------------------------- */
   1462 
   1463 int
   1464 tmpfs_putpages(void *v)
   1465 {
   1466 	struct vnode *vp = ((struct vop_putpages_args *)v)->a_vp;
   1467 	voff_t offlo = ((struct vop_putpages_args *)v)->a_offlo;
   1468 	voff_t offhi = ((struct vop_putpages_args *)v)->a_offhi;
   1469 	int flags = ((struct vop_putpages_args *)v)->a_flags;
   1470 
   1471 	int error;
   1472 	struct tmpfs_node *node;
   1473 	struct uvm_object *uobj;
   1474 
   1475 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
   1476 
   1477 	node = VP_TO_TMPFS_NODE(vp);
   1478 
   1479 	if (vp->v_type != VREG) {
   1480 		simple_unlock(&vp->v_interlock);
   1481 		return 0;
   1482 	}
   1483 
   1484 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1485 	simple_unlock(&vp->v_interlock);
   1486 
   1487 	simple_lock(&uobj->vmobjlock);
   1488 	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
   1489 
   1490 	/* XXX mtime */
   1491 
   1492 	return error;
   1493 }
   1494