Home | History | Annotate | Line # | Download | only in tmpfs
tmpfs_vnops.c revision 1.51.6.1
      1 /*	$NetBSD: tmpfs_vnops.c,v 1.51.6.1 2009/02/16 03:29:01 snj 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 1.51.6.1 2009/02/16 03:29:01 snj 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 <fs/tmpfs/tmpfs_vnops.h>
     58 #include <fs/tmpfs/tmpfs.h>
     59 
     60 /* --------------------------------------------------------------------- */
     61 
     62 /*
     63  * vnode operations vector used for files stored in a tmpfs file system.
     64  */
     65 int (**tmpfs_vnodeop_p)(void *);
     66 const struct vnodeopv_entry_desc tmpfs_vnodeop_entries[] = {
     67 	{ &vop_default_desc,		vn_default_error },
     68 	{ &vop_lookup_desc,		tmpfs_lookup },
     69 	{ &vop_create_desc,		tmpfs_create },
     70 	{ &vop_mknod_desc,		tmpfs_mknod },
     71 	{ &vop_open_desc,		tmpfs_open },
     72 	{ &vop_close_desc,		tmpfs_close },
     73 	{ &vop_access_desc,		tmpfs_access },
     74 	{ &vop_getattr_desc,		tmpfs_getattr },
     75 	{ &vop_setattr_desc,		tmpfs_setattr },
     76 	{ &vop_read_desc,		tmpfs_read },
     77 	{ &vop_write_desc,		tmpfs_write },
     78 	{ &vop_ioctl_desc,		tmpfs_ioctl },
     79 	{ &vop_fcntl_desc,		tmpfs_fcntl },
     80 	{ &vop_poll_desc,		tmpfs_poll },
     81 	{ &vop_kqfilter_desc,		tmpfs_kqfilter },
     82 	{ &vop_revoke_desc,		tmpfs_revoke },
     83 	{ &vop_mmap_desc,		tmpfs_mmap },
     84 	{ &vop_fsync_desc,		tmpfs_fsync },
     85 	{ &vop_seek_desc,		tmpfs_seek },
     86 	{ &vop_remove_desc,		tmpfs_remove },
     87 	{ &vop_link_desc,		tmpfs_link },
     88 	{ &vop_rename_desc,		tmpfs_rename },
     89 	{ &vop_mkdir_desc,		tmpfs_mkdir },
     90 	{ &vop_rmdir_desc,		tmpfs_rmdir },
     91 	{ &vop_symlink_desc,		tmpfs_symlink },
     92 	{ &vop_readdir_desc,		tmpfs_readdir },
     93 	{ &vop_readlink_desc,		tmpfs_readlink },
     94 	{ &vop_abortop_desc,		tmpfs_abortop },
     95 	{ &vop_inactive_desc,		tmpfs_inactive },
     96 	{ &vop_reclaim_desc,		tmpfs_reclaim },
     97 	{ &vop_lock_desc,		tmpfs_lock },
     98 	{ &vop_unlock_desc,		tmpfs_unlock },
     99 	{ &vop_bmap_desc,		tmpfs_bmap },
    100 	{ &vop_strategy_desc,		tmpfs_strategy },
    101 	{ &vop_print_desc,		tmpfs_print },
    102 	{ &vop_pathconf_desc,		tmpfs_pathconf },
    103 	{ &vop_islocked_desc,		tmpfs_islocked },
    104 	{ &vop_advlock_desc,		tmpfs_advlock },
    105 	{ &vop_bwrite_desc,		tmpfs_bwrite },
    106 	{ &vop_getpages_desc,		tmpfs_getpages },
    107 	{ &vop_putpages_desc,		tmpfs_putpages },
    108 	{ NULL, NULL }
    109 };
    110 const struct vnodeopv_desc tmpfs_vnodeop_opv_desc =
    111 	{ &tmpfs_vnodeop_p, tmpfs_vnodeop_entries };
    112 
    113 /* --------------------------------------------------------------------- */
    114 
    115 int
    116 tmpfs_lookup(void *v)
    117 {
    118 	struct vnode *dvp = ((struct vop_lookup_args *)v)->a_dvp;
    119 	struct vnode **vpp = ((struct vop_lookup_args *)v)->a_vpp;
    120 	struct componentname *cnp = ((struct vop_lookup_args *)v)->a_cnp;
    121 
    122 	int error;
    123 	struct tmpfs_dirent *de;
    124 	struct tmpfs_node *dnode;
    125 
    126 	KASSERT(VOP_ISLOCKED(dvp));
    127 
    128 	dnode = VP_TO_TMPFS_DIR(dvp);
    129 	*vpp = NULL;
    130 
    131 	/* Check accessibility of requested node as a first step. */
    132 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
    133 	if (error != 0)
    134 		goto out;
    135 
    136 	/* If requesting the last path component on a read-only file system
    137 	 * with a write operation, deny it. */
    138 	if ((cnp->cn_flags & ISLASTCN) &&
    139 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
    140 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    141 		error = EROFS;
    142 		goto out;
    143 	}
    144 
    145 	/* Avoid doing a linear scan of the directory if the requested
    146 	 * directory/name couple is already in the cache. */
    147 	error = cache_lookup(dvp, vpp, cnp);
    148 	if (error >= 0)
    149 		goto out;
    150 
    151 	/* We cannot be requesting the parent directory of the root node. */
    152 	KASSERT(IMPLIES(dnode->tn_type == VDIR &&
    153 	    dnode->tn_spec.tn_dir.tn_parent == dnode,
    154 	    !(cnp->cn_flags & ISDOTDOT)));
    155 
    156 	if (cnp->cn_flags & ISDOTDOT) {
    157 		VOP_UNLOCK(dvp, 0);
    158 
    159 		/* Allocate a new vnode on the matching entry. */
    160 		error = tmpfs_alloc_vp(dvp->v_mount,
    161 		    dnode->tn_spec.tn_dir.tn_parent, vpp);
    162 
    163 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
    164 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
    165 		VREF(dvp);
    166 		*vpp = dvp;
    167 		error = 0;
    168 	} else {
    169 		de = tmpfs_dir_lookup(dnode, cnp);
    170 		if (de == NULL) {
    171 			/* The entry was not found in the directory.
    172 			 * This is OK iff we are creating or renaming an
    173 			 * entry and are working on the last component of
    174 			 * the path name. */
    175 			if ((cnp->cn_flags & ISLASTCN) &&
    176 			    (cnp->cn_nameiop == CREATE || \
    177 			    cnp->cn_nameiop == RENAME)) {
    178 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
    179 				if (error != 0)
    180 					goto out;
    181 
    182 				/* Keep the component name in the buffer for
    183 				 * future uses. */
    184 				cnp->cn_flags |= SAVENAME;
    185 
    186 				error = EJUSTRETURN;
    187 			} else
    188 				error = ENOENT;
    189 		} else {
    190 			struct tmpfs_node *tnode;
    191 
    192 			/* The entry was found, so get its associated
    193 			 * tmpfs_node. */
    194 			tnode = de->td_node;
    195 
    196 			/* If we are not at the last path component and
    197 			 * found a non-directory or non-link entry (which
    198 			 * may itself be pointing to a directory), raise
    199 			 * an error. */
    200 			if ((tnode->tn_type != VDIR &&
    201 			    tnode->tn_type != VLNK) &&
    202 			    !(cnp->cn_flags & ISLASTCN)) {
    203 				error = ENOTDIR;
    204 				goto out;
    205 			}
    206 
    207 			/* If we are deleting or renaming the entry, keep
    208 			 * track of its tmpfs_dirent so that it can be
    209 			 * easily deleted later. */
    210 			if ((cnp->cn_flags & ISLASTCN) &&
    211 			    (cnp->cn_nameiop == DELETE ||
    212 			    cnp->cn_nameiop == RENAME)) {
    213 				if ((dnode->tn_mode & S_ISTXT) != 0 &&
    214 				    kauth_authorize_generic(cnp->cn_cred,
    215 				     KAUTH_GENERIC_ISSUSER, NULL) != 0 &&
    216 				    kauth_cred_geteuid(cnp->cn_cred) != dnode->tn_uid &&
    217 				    kauth_cred_geteuid(cnp->cn_cred) != tnode->tn_uid)
    218 					return EPERM;
    219 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
    220 				if (error != 0)
    221 					goto out;
    222 			} else
    223 				de = NULL;
    224 
    225 			/* Allocate a new vnode on the matching entry. */
    226 			error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp);
    227 		}
    228 	}
    229 
    230 	/* Store the result of this lookup in the cache.  Avoid this if the
    231 	 * request was for creation, as it does not improve timings on
    232 	 * emprical tests. */
    233 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
    234 	    (cnp->cn_flags & ISDOTDOT) == 0)
    235 		cache_enter(dvp, *vpp, cnp);
    236 
    237 out:
    238 	/* If there were no errors, *vpp cannot be null and it must be
    239 	 * locked. */
    240 	KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
    241 
    242 	/* dvp must always be locked. */
    243 	KASSERT(VOP_ISLOCKED(dvp));
    244 
    245 	return error;
    246 }
    247 
    248 /* --------------------------------------------------------------------- */
    249 
    250 int
    251 tmpfs_create(void *v)
    252 {
    253 	struct vnode *dvp = ((struct vop_create_args *)v)->a_dvp;
    254 	struct vnode **vpp = ((struct vop_create_args *)v)->a_vpp;
    255 	struct componentname *cnp = ((struct vop_create_args *)v)->a_cnp;
    256 	struct vattr *vap = ((struct vop_create_args *)v)->a_vap;
    257 
    258 	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
    259 
    260 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    261 }
    262 /* --------------------------------------------------------------------- */
    263 
    264 int
    265 tmpfs_mknod(void *v)
    266 {
    267 	struct vnode *dvp = ((struct vop_mknod_args *)v)->a_dvp;
    268 	struct vnode **vpp = ((struct vop_mknod_args *)v)->a_vpp;
    269 	struct componentname *cnp = ((struct vop_mknod_args *)v)->a_cnp;
    270 	struct vattr *vap = ((struct vop_mknod_args *)v)->a_vap;
    271 
    272 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
    273 	    vap->va_type != VFIFO)
    274 		return EINVAL;
    275 
    276 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    277 }
    278 
    279 /* --------------------------------------------------------------------- */
    280 
    281 int
    282 tmpfs_open(void *v)
    283 {
    284 	struct vnode *vp = ((struct vop_open_args *)v)->a_vp;
    285 	int mode = ((struct vop_open_args *)v)->a_mode;
    286 
    287 	int error;
    288 	struct tmpfs_node *node;
    289 
    290 	KASSERT(VOP_ISLOCKED(vp));
    291 
    292 	node = VP_TO_TMPFS_NODE(vp);
    293 
    294 	/* The file is still active but all its names have been removed
    295 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
    296 	 * it is about to die. */
    297 	if (node->tn_links < 1) {
    298 		error = ENOENT;
    299 		goto out;
    300 	}
    301 
    302 	/* If the file is marked append-only, deny write requests. */
    303 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
    304 		error = EPERM;
    305 	else
    306 		error = 0;
    307 
    308 out:
    309 	KASSERT(VOP_ISLOCKED(vp));
    310 
    311 	return error;
    312 }
    313 
    314 /* --------------------------------------------------------------------- */
    315 
    316 int
    317 tmpfs_close(void *v)
    318 {
    319 	struct vnode *vp = ((struct vop_close_args *)v)->a_vp;
    320 
    321 	struct tmpfs_node *node;
    322 
    323 	KASSERT(VOP_ISLOCKED(vp));
    324 
    325 	node = VP_TO_TMPFS_NODE(vp);
    326 
    327 	if (node->tn_links > 0) {
    328 		/* Update node times.  No need to do it if the node has
    329 		 * been deleted, because it will vanish after we return. */
    330 		tmpfs_update(vp, NULL, NULL, NULL, UPDATE_CLOSE);
    331 	}
    332 
    333 	return 0;
    334 }
    335 
    336 /* --------------------------------------------------------------------- */
    337 
    338 int
    339 tmpfs_access(void *v)
    340 {
    341 	struct vnode *vp = ((struct vop_access_args *)v)->a_vp;
    342 	int mode = ((struct vop_access_args *)v)->a_mode;
    343 	kauth_cred_t cred = ((struct vop_access_args *)v)->a_cred;
    344 
    345 	int error;
    346 	struct tmpfs_node *node;
    347 
    348 	KASSERT(VOP_ISLOCKED(vp));
    349 
    350 	node = VP_TO_TMPFS_NODE(vp);
    351 
    352 	switch (vp->v_type) {
    353 	case VDIR:
    354 		/* FALLTHROUGH */
    355 	case VLNK:
    356 		/* FALLTHROUGH */
    357 	case VREG:
    358 		if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
    359 			error = EROFS;
    360 			goto out;
    361 		}
    362 		break;
    363 
    364 	case VBLK:
    365 		/* FALLTHROUGH */
    366 	case VCHR:
    367 		/* FALLTHROUGH */
    368 	case VSOCK:
    369 		/* FALLTHROUGH */
    370 	case VFIFO:
    371 		break;
    372 
    373 	default:
    374 		error = EINVAL;
    375 		goto out;
    376 	}
    377 
    378 	if (mode & VWRITE && node->tn_flags & IMMUTABLE) {
    379 		error = EPERM;
    380 		goto out;
    381 	}
    382 
    383 	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
    384 	    node->tn_gid, mode, cred);
    385 
    386 out:
    387 	KASSERT(VOP_ISLOCKED(vp));
    388 
    389 	return error;
    390 }
    391 
    392 /* --------------------------------------------------------------------- */
    393 
    394 int
    395 tmpfs_getattr(void *v)
    396 {
    397 	struct vnode *vp = ((struct vop_getattr_args *)v)->a_vp;
    398 	struct vattr *vap = ((struct vop_getattr_args *)v)->a_vap;
    399 
    400 	struct tmpfs_node *node;
    401 
    402 	node = VP_TO_TMPFS_NODE(vp);
    403 
    404 	VATTR_NULL(vap);
    405 
    406 	tmpfs_itimes(vp, NULL, NULL, NULL);
    407 
    408 	vap->va_type = vp->v_type;
    409 	vap->va_mode = node->tn_mode;
    410 	vap->va_nlink = node->tn_links;
    411 	vap->va_uid = node->tn_uid;
    412 	vap->va_gid = node->tn_gid;
    413 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
    414 	vap->va_fileid = node->tn_id;
    415 	vap->va_size = node->tn_size;
    416 	vap->va_blocksize = PAGE_SIZE;
    417 	vap->va_atime = node->tn_atime;
    418 	vap->va_mtime = node->tn_mtime;
    419 	vap->va_ctime = node->tn_ctime;
    420 	vap->va_birthtime = node->tn_birthtime;
    421 	vap->va_gen = node->tn_gen;
    422 	vap->va_flags = node->tn_flags;
    423 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
    424 		node->tn_spec.tn_dev.tn_rdev : VNOVAL;
    425 	vap->va_bytes = round_page(node->tn_size);
    426 	vap->va_filerev = VNOVAL;
    427 	vap->va_vaflags = 0;
    428 	vap->va_spare = VNOVAL; /* XXX */
    429 
    430 	return 0;
    431 }
    432 
    433 /* --------------------------------------------------------------------- */
    434 
    435 #define GOODTIME(tv)	((tv)->tv_sec != VNOVAL || (tv)->tv_nsec != VNOVAL)
    436 /* XXX Should this operation be atomic?  I think it should, but code in
    437  * XXX other places (e.g., ufs) doesn't seem to be... */
    438 int
    439 tmpfs_setattr(void *v)
    440 {
    441 	struct vnode *vp = ((struct vop_setattr_args *)v)->a_vp;
    442 	struct vattr *vap = ((struct vop_setattr_args *)v)->a_vap;
    443 	kauth_cred_t cred = ((struct vop_setattr_args *)v)->a_cred;
    444 	struct lwp *l = curlwp;
    445 
    446 	int error;
    447 
    448 	KASSERT(VOP_ISLOCKED(vp));
    449 
    450 	error = 0;
    451 
    452 	/* Abort if any unsettable attribute is given. */
    453 	if (vap->va_type != VNON ||
    454 	    vap->va_nlink != VNOVAL ||
    455 	    vap->va_fsid != VNOVAL ||
    456 	    vap->va_fileid != VNOVAL ||
    457 	    vap->va_blocksize != VNOVAL ||
    458 	    GOODTIME(&vap->va_ctime) ||
    459 	    vap->va_gen != VNOVAL ||
    460 	    vap->va_rdev != VNOVAL ||
    461 	    vap->va_bytes != VNOVAL)
    462 		error = EINVAL;
    463 
    464 	if (error == 0 && (vap->va_flags != VNOVAL))
    465 		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
    466 
    467 	if (error == 0 && (vap->va_size != VNOVAL))
    468 		error = tmpfs_chsize(vp, vap->va_size, cred, l);
    469 
    470 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
    471 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
    472 
    473 	if (error == 0 && (vap->va_mode != VNOVAL))
    474 		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
    475 
    476 	if (error == 0 && (GOODTIME(&vap->va_atime) || GOODTIME(&vap->va_mtime)
    477 	    || GOODTIME(&vap->va_birthtime)))
    478 		if ((error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
    479 		    &vap->va_birthtime, vap->va_vaflags, cred, l)) == 0)
    480 			return 0;
    481 
    482 	/* Update the node times.  We give preference to the error codes
    483 	 * generated by this function rather than the ones that may arise
    484 	 * from tmpfs_update. */
    485 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    486 
    487 	KASSERT(VOP_ISLOCKED(vp));
    488 
    489 	return error;
    490 }
    491 
    492 /* --------------------------------------------------------------------- */
    493 
    494 int
    495 tmpfs_read(void *v)
    496 {
    497 	struct vnode *vp = ((struct vop_read_args *)v)->a_vp;
    498 	struct uio *uio = ((struct vop_read_args *)v)->a_uio;
    499 
    500 	int error;
    501 	int flags;
    502 	struct tmpfs_node *node;
    503 	struct uvm_object *uobj;
    504 
    505 	KASSERT(VOP_ISLOCKED(vp));
    506 
    507 	node = VP_TO_TMPFS_NODE(vp);
    508 
    509 	if (vp->v_type != VREG) {
    510 		error = EISDIR;
    511 		goto out;
    512 	}
    513 
    514 	if (uio->uio_offset < 0) {
    515 		error = EINVAL;
    516 		goto out;
    517 	}
    518 
    519 	node->tn_status |= TMPFS_NODE_ACCESSED;
    520 
    521 	uobj = node->tn_spec.tn_reg.tn_aobj;
    522 	flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
    523 	error = 0;
    524 	while (error == 0 && uio->uio_resid > 0) {
    525 		vsize_t len;
    526 		void *win;
    527 
    528 		if (node->tn_size <= uio->uio_offset)
    529 			break;
    530 
    531 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    532 		if (len == 0)
    533 			break;
    534 
    535 		win = ubc_alloc(uobj, uio->uio_offset, &len, UVM_ADV_NORMAL,
    536 		    UBC_READ);
    537 		error = uiomove(win, len, uio);
    538 		ubc_release(win, flags);
    539 	}
    540 
    541 out:
    542 	KASSERT(VOP_ISLOCKED(vp));
    543 
    544 	return error;
    545 }
    546 
    547 /* --------------------------------------------------------------------- */
    548 
    549 int
    550 tmpfs_write(void *v)
    551 {
    552 	struct vnode *vp = ((struct vop_write_args *)v)->a_vp;
    553 	struct uio *uio = ((struct vop_write_args *)v)->a_uio;
    554 	int ioflag = ((struct vop_write_args *)v)->a_ioflag;
    555 
    556 	bool extended;
    557 	int error;
    558 	int flags;
    559 	off_t oldsize;
    560 	struct tmpfs_node *node;
    561 	struct uvm_object *uobj;
    562 
    563 	KASSERT(VOP_ISLOCKED(vp));
    564 
    565 	node = VP_TO_TMPFS_NODE(vp);
    566 	oldsize = node->tn_size;
    567 
    568 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
    569 		error = EINVAL;
    570 		goto out;
    571 	}
    572 
    573 	if (uio->uio_resid == 0) {
    574 		error = 0;
    575 		goto out;
    576 	}
    577 
    578 	if (ioflag & IO_APPEND)
    579 		uio->uio_offset = node->tn_size;
    580 
    581 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
    582 	if (extended) {
    583 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
    584 		if (error != 0)
    585 			goto out;
    586 	}
    587 
    588 	uobj = node->tn_spec.tn_reg.tn_aobj;
    589 	flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
    590 	error = 0;
    591 	while (error == 0 && uio->uio_resid > 0) {
    592 		vsize_t len;
    593 		void *win;
    594 
    595 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    596 		if (len == 0)
    597 			break;
    598 
    599 		win = ubc_alloc(uobj, uio->uio_offset, &len, UVM_ADV_NORMAL,
    600 		    UBC_WRITE);
    601 		error = uiomove(win, len, uio);
    602 		ubc_release(win, flags);
    603 	}
    604 
    605 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
    606 	    (extended ? TMPFS_NODE_CHANGED : 0);
    607 
    608 	if (error != 0)
    609 		(void)tmpfs_reg_resize(vp, oldsize);
    610 
    611 	VN_KNOTE(vp, NOTE_WRITE);
    612 
    613 out:
    614 	KASSERT(VOP_ISLOCKED(vp));
    615 	KASSERT(IMPLIES(error == 0, uio->uio_resid == 0));
    616 	KASSERT(IMPLIES(error != 0, oldsize == node->tn_size));
    617 
    618 	return error;
    619 }
    620 
    621 /* --------------------------------------------------------------------- */
    622 
    623 int
    624 tmpfs_fsync(void *v)
    625 {
    626 	struct vnode *vp = ((struct vop_fsync_args *)v)->a_vp;
    627 
    628 	KASSERT(VOP_ISLOCKED(vp));
    629 
    630 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    631 
    632 	return 0;
    633 }
    634 
    635 /* --------------------------------------------------------------------- */
    636 
    637 int
    638 tmpfs_remove(void *v)
    639 {
    640 	struct vnode *dvp = ((struct vop_remove_args *)v)->a_dvp;
    641 	struct vnode *vp = ((struct vop_remove_args *)v)->a_vp;
    642 	struct componentname *cnp = (((struct vop_remove_args *)v)->a_cnp);
    643 
    644 	int error;
    645 	struct tmpfs_dirent *de;
    646 	struct tmpfs_mount *tmp;
    647 	struct tmpfs_node *dnode;
    648 	struct tmpfs_node *node;
    649 
    650 	KASSERT(VOP_ISLOCKED(dvp));
    651 	KASSERT(VOP_ISLOCKED(vp));
    652 
    653 	if (vp->v_type == VDIR) {
    654 		error = EPERM;
    655 		goto out;
    656 	}
    657 
    658 	dnode = VP_TO_TMPFS_DIR(dvp);
    659 	node = VP_TO_TMPFS_NODE(vp);
    660 	tmp = VFS_TO_TMPFS(vp->v_mount);
    661 	de = tmpfs_dir_lookup(dnode, cnp);
    662 	if (de == NULL) {
    663 		error = ENOENT;
    664 		goto out;
    665 	}
    666 	KASSERT(de->td_node == node);
    667 
    668 	/* Files marked as immutable or append-only cannot be deleted. */
    669 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    670 		error = EPERM;
    671 		goto out;
    672 	}
    673 
    674 	/* Remove the entry from the directory; as it is a file, we do not
    675 	 * have to change the number of hard links of the directory. */
    676 	tmpfs_dir_detach(dvp, de);
    677 
    678 	/* Free the directory entry we just deleted.  Note that the node
    679 	 * referred by it will not be removed until the vnode is really
    680 	 * reclaimed. */
    681 	tmpfs_free_dirent(tmp, de, true);
    682 
    683 	error = 0;
    684 
    685 out:
    686 	vput(vp);
    687 	if (dvp == vp)
    688 		vrele(dvp);
    689 	else
    690 		vput(dvp);
    691 
    692 	return error;
    693 }
    694 
    695 /* --------------------------------------------------------------------- */
    696 
    697 int
    698 tmpfs_link(void *v)
    699 {
    700 	struct vnode *dvp = ((struct vop_link_args *)v)->a_dvp;
    701 	struct vnode *vp = ((struct vop_link_args *)v)->a_vp;
    702 	struct componentname *cnp = ((struct vop_link_args *)v)->a_cnp;
    703 
    704 	int error;
    705 	struct tmpfs_dirent *de;
    706 	struct tmpfs_node *dnode;
    707 	struct tmpfs_node *node;
    708 
    709 	KASSERT(VOP_ISLOCKED(dvp));
    710 	KASSERT(cnp->cn_flags & HASBUF);
    711 	KASSERT(dvp != vp); /* XXX When can this be false? */
    712 
    713 	dnode = VP_TO_TMPFS_DIR(dvp);
    714 	node = VP_TO_TMPFS_NODE(vp);
    715 
    716 	/* Lock vp because we will need to run tmpfs_update over it, which
    717 	 * needs the vnode to be locked. */
    718 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    719 	if (error != 0)
    720 		goto out1;
    721 
    722 	/* XXX: Why aren't the following two tests done by the caller? */
    723 
    724 	/* Hard links of directories are forbidden. */
    725 	if (vp->v_type == VDIR) {
    726 		error = EPERM;
    727 		goto out;
    728 	}
    729 
    730 	/* Cannot create cross-device links. */
    731 	if (dvp->v_mount != vp->v_mount) {
    732 		error = EXDEV;
    733 		goto out;
    734 	}
    735 
    736 	/* Ensure that we do not overflow the maximum number of links imposed
    737 	 * by the system. */
    738 	KASSERT(node->tn_links <= LINK_MAX);
    739 	if (node->tn_links == LINK_MAX) {
    740 		error = EMLINK;
    741 		goto out;
    742 	}
    743 
    744 	/* We cannot create links of files marked immutable or append-only. */
    745 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    746 		error = EPERM;
    747 		goto out;
    748 	}
    749 
    750 	/* Allocate a new directory entry to represent the node. */
    751 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
    752 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
    753 	if (error != 0)
    754 		goto out;
    755 
    756 	/* Insert the new directory entry into the appropriate directory. */
    757 	tmpfs_dir_attach(dvp, de);
    758 
    759 	/* vp link count has changed, so update node times. */
    760 	node->tn_status |= TMPFS_NODE_CHANGED;
    761 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    762 
    763 	error = 0;
    764 
    765 out:
    766 	VOP_UNLOCK(vp, 0);
    767 out1:
    768 	PNBUF_PUT(cnp->cn_pnbuf);
    769 
    770 	vput(dvp);
    771 
    772 	return error;
    773 }
    774 
    775 /* --------------------------------------------------------------------- */
    776 
    777 int
    778 tmpfs_rename(void *v)
    779 {
    780 	struct vnode *fdvp = ((struct vop_rename_args *)v)->a_fdvp;
    781 	struct vnode *fvp = ((struct vop_rename_args *)v)->a_fvp;
    782 	struct componentname *fcnp = ((struct vop_rename_args *)v)->a_fcnp;
    783 	struct vnode *tdvp = ((struct vop_rename_args *)v)->a_tdvp;
    784 	struct vnode *tvp = ((struct vop_rename_args *)v)->a_tvp;
    785 	struct componentname *tcnp = ((struct vop_rename_args *)v)->a_tcnp;
    786 
    787 	char *newname;
    788 	int error;
    789 	struct tmpfs_dirent *de, *de2;
    790 	struct tmpfs_mount *tmp;
    791 	struct tmpfs_node *fdnode;
    792 	struct tmpfs_node *fnode;
    793 	struct tmpfs_node *tnode;
    794 	struct tmpfs_node *tdnode;
    795 	size_t namelen;
    796 
    797 	KASSERT(VOP_ISLOCKED(tdvp));
    798 	KASSERT(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
    799 	KASSERT(fcnp->cn_flags & HASBUF);
    800 	KASSERT(tcnp->cn_flags & HASBUF);
    801 
    802 	newname = NULL;
    803 	namelen = 0;
    804 	tmp = NULL;
    805 
    806 	/* Disallow cross-device renames. */
    807 	if (fvp->v_mount != tdvp->v_mount ||
    808 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
    809 		error = EXDEV;
    810 		goto out_unlocked;
    811 	}
    812 
    813 	fnode = VP_TO_TMPFS_NODE(fvp);
    814 	fdnode = VP_TO_TMPFS_DIR(fdvp);
    815 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
    816 	tdnode = VP_TO_TMPFS_DIR(tdvp);
    817 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
    818 
    819 	if (fdvp == tvp) {
    820 		error = 0;
    821 		goto out_unlocked;
    822 	}
    823 
    824 	/* If we need to move the directory between entries, lock the
    825 	 * source so that we can safely operate on it. */
    826 
    827 	/* XXX: this is a potential locking order violation! */
    828 	if (fdnode != tdnode) {
    829 		error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
    830 		if (error != 0)
    831 			goto out_unlocked;
    832 	}
    833 
    834 	de = tmpfs_dir_lookup(fdnode, fcnp);
    835 	if (de == NULL) {
    836 		error = ENOENT;
    837 		goto out;
    838 	}
    839 	KASSERT(de->td_node == fnode);
    840 
    841 	/* If source and target are the same file, there is nothing to do. */
    842 	if (fvp == tvp) {
    843 		error = 0;
    844 		goto out;
    845 	}
    846 
    847 	/* Avoid manipulating '.' and '..' entries. */
    848 	if (de == NULL) {
    849 		KASSERT(fvp->v_type == VDIR);
    850 		error = EINVAL;
    851 		goto out;
    852 	}
    853 	KASSERT(de->td_node == fnode);
    854 
    855 	/* If replacing an existing entry, ensure we can do the operation. */
    856 	if (tvp != NULL) {
    857 		KASSERT(tnode != NULL);
    858 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
    859 			if (tnode->tn_size > 0) {
    860 				error = ENOTEMPTY;
    861 				goto out;
    862 			}
    863 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
    864 			error = ENOTDIR;
    865 			goto out;
    866 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
    867 			error = EISDIR;
    868 			goto out;
    869 		} else {
    870 			KASSERT(fnode->tn_type != VDIR &&
    871 			        tnode->tn_type != VDIR);
    872 		}
    873 	}
    874 
    875 	/* Ensure that we have enough memory to hold the new name, if it
    876 	 * has to be changed. */
    877 	namelen = tcnp->cn_namelen;
    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, namelen, 0);
    881 		if (newname == NULL) {
    882 			error = ENOSPC;
    883 			goto out;
    884 		}
    885 	}
    886 
    887 	/* If the node is being moved to another directory, we have to do
    888 	 * the move. */
    889 	if (fdnode != tdnode) {
    890 		/* In case we are moving a directory, we have to adjust its
    891 		 * parent to point to the new parent. */
    892 		if (de->td_node->tn_type == VDIR) {
    893 			struct tmpfs_node *n;
    894 
    895 			/* Ensure the target directory is not a child of the
    896 			 * directory being moved.  Otherwise, we'd end up
    897 			 * with stale nodes. */
    898 			n = tdnode;
    899 			while (n != n->tn_spec.tn_dir.tn_parent) {
    900 				if (n == fnode) {
    901 					error = EINVAL;
    902 					goto out;
    903 				}
    904 				n = n->tn_spec.tn_dir.tn_parent;
    905 			}
    906 
    907 			/* Adjust the parent pointer. */
    908 			TMPFS_VALIDATE_DIR(fnode);
    909 			de->td_node->tn_spec.tn_dir.tn_parent = tdnode;
    910 
    911 			/* As a result of changing the target of the '..'
    912 			 * entry, the link count of the source and target
    913 			 * directories has to be adjusted. */
    914 			fdnode->tn_links--;
    915 			tdnode->tn_links++;
    916 		}
    917 
    918 		/* Do the move: just remove the entry from the source directory
    919 		 * and insert it into the target one. */
    920 		tmpfs_dir_detach(fdvp, de);
    921 		tmpfs_dir_attach(tdvp, de);
    922 
    923 		/* Notify listeners of fdvp about the change in the directory.
    924 		 * We can do it at this point because we aren't touching fdvp
    925 		 * any more below. */
    926 		VN_KNOTE(fdvp, NOTE_WRITE);
    927 	}
    928 
    929 	/* If we are overwriting an entry, we have to remove the old one
    930 	 * from the target directory. */
    931 	if (tvp != NULL) {
    932 		KASSERT(tnode != NULL);
    933 
    934 		/* Remove the old entry from the target directory.
    935 		 * Note! This relies on tmpfs_dir_attach() putting the new
    936 		 * node on the end of the target's node list. */
    937 		de2 = tmpfs_dir_lookup(tdnode, tcnp);
    938 		KASSERT(de2 != NULL);
    939 		KASSERT(de2->td_node == tnode);
    940 		tmpfs_dir_detach(tdvp, de2);
    941 
    942 		/* Free the directory entry we just deleted.  Note that the
    943 		 * node referred by it will not be removed until the vnode is
    944 		 * really reclaimed. */
    945 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de2, true);
    946 	}
    947 
    948 	/* If the name has changed, we need to make it effective by changing
    949 	 * it in the directory entry. */
    950 	if (newname != NULL) {
    951 		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
    952 		KASSERT(tcnp->cn_namelen < 0xffff);
    953 
    954 		tmpfs_str_pool_put(&tmp->tm_str_pool, de->td_name,
    955 		    de->td_namelen);
    956 		de->td_namelen = (uint16_t)namelen;
    957 		memcpy(newname, tcnp->cn_nameptr, namelen);
    958 		de->td_name = newname;
    959 		newname = NULL;
    960 
    961 		fnode->tn_status |= TMPFS_NODE_CHANGED;
    962 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
    963 	}
    964 
    965 	/* Notify listeners of tdvp about the change in the directory (either
    966 	 * because a new entry was added or because one was removed) and
    967 	 * listeners of fvp about the rename. */
    968 	VN_KNOTE(tdvp, NOTE_WRITE);
    969 	VN_KNOTE(fvp, NOTE_RENAME);
    970 
    971 	error = 0;
    972 
    973  out:
    974 	if (fdnode != tdnode)
    975 		VOP_UNLOCK(fdvp, 0);
    976 
    977  out_unlocked:
    978 	/* Release target nodes. */
    979 	if (tdvp == tvp)
    980 		vrele(tdvp);
    981 	else
    982 		vput(tdvp);
    983 	if (tvp != NULL)
    984 		vput(tvp);
    985 
    986 	/* Release source nodes. */
    987 	vrele(fdvp);
    988 	vrele(fvp);
    989 
    990 	if (newname != NULL)
    991 		tmpfs_str_pool_put(&tmp->tm_str_pool, newname, namelen);
    992 
    993 	return error;
    994 }
    995 
    996 /* --------------------------------------------------------------------- */
    997 
    998 int
    999 tmpfs_mkdir(void *v)
   1000 {
   1001 	struct vnode *dvp = ((struct vop_mkdir_args *)v)->a_dvp;
   1002 	struct vnode **vpp = ((struct vop_mkdir_args *)v)->a_vpp;
   1003 	struct componentname *cnp = ((struct vop_mkdir_args *)v)->a_cnp;
   1004 	struct vattr *vap = ((struct vop_mkdir_args *)v)->a_vap;
   1005 
   1006 	KASSERT(vap->va_type == VDIR);
   1007 
   1008 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
   1009 }
   1010 
   1011 /* --------------------------------------------------------------------- */
   1012 
   1013 int
   1014 tmpfs_rmdir(void *v)
   1015 {
   1016 	struct vnode *dvp = ((struct vop_rmdir_args *)v)->a_dvp;
   1017 	struct vnode *vp = ((struct vop_rmdir_args *)v)->a_vp;
   1018 	struct componentname *cnp = ((struct vop_rmdir_args *)v)->a_cnp;
   1019 
   1020 	int error;
   1021 	struct tmpfs_dirent *de;
   1022 	struct tmpfs_mount *tmp;
   1023 	struct tmpfs_node *dnode;
   1024 	struct tmpfs_node *node;
   1025 
   1026 	KASSERT(VOP_ISLOCKED(dvp));
   1027 	KASSERT(VOP_ISLOCKED(vp));
   1028 
   1029 	tmp = VFS_TO_TMPFS(dvp->v_mount);
   1030 	dnode = VP_TO_TMPFS_DIR(dvp);
   1031 	node = VP_TO_TMPFS_DIR(vp);
   1032 	error = 0;
   1033 
   1034 	/* Directories with more than two entries ('.' and '..') cannot be
   1035 	 * removed. */
   1036 	if (node->tn_size > 0) {
   1037 		error = ENOTEMPTY;
   1038 		goto out;
   1039 	}
   1040 
   1041 	/* This invariant holds only if we are not trying to remove "..".
   1042 	 * We checked for that above so this is safe now. */
   1043 	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
   1044 
   1045 	/* Get the directory entry associated with node (vp). */
   1046 	de = tmpfs_dir_lookup(dnode, cnp);
   1047 	if (de == NULL) {
   1048 		error = ENOENT;
   1049 		goto out;
   1050 	}
   1051 	KASSERT(de->td_node == node);
   1052 
   1053 	/* Check flags to see if we are allowed to remove the directory. */
   1054 	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
   1055 		error = EPERM;
   1056 		goto out;
   1057 	}
   1058 
   1059 	/* Detach the directory entry from the directory (dnode). */
   1060 	tmpfs_dir_detach(dvp, de);
   1061 
   1062 	node->tn_links--;
   1063 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
   1064 	    TMPFS_NODE_MODIFIED;
   1065 	node->tn_spec.tn_dir.tn_parent->tn_links--;
   1066 	node->tn_spec.tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \
   1067 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
   1068 
   1069 	/* Release the parent. */
   1070 	cache_purge(dvp); /* XXX Is this needed? */
   1071 
   1072 	/* Free the directory entry we just deleted.  Note that the node
   1073 	 * referred by it will not be removed until the vnode is really
   1074 	 * reclaimed. */
   1075 	tmpfs_free_dirent(tmp, de, true);
   1076 
   1077 	KASSERT(node->tn_links == 0);
   1078  out:
   1079 	/* Release the nodes. */
   1080 	vput(dvp);
   1081 	vput(vp);
   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 
   1241 	struct tmpfs_node *node;
   1242 
   1243 	KASSERT(VOP_ISLOCKED(vp));
   1244 
   1245 	node = VP_TO_TMPFS_NODE(vp);
   1246 	*((struct vop_inactive_args *)v)->a_recycle = (node->tn_links == 0);
   1247 	VOP_UNLOCK(vp, 0);
   1248 
   1249 	return 0;
   1250 }
   1251 
   1252 /* --------------------------------------------------------------------- */
   1253 
   1254 int
   1255 tmpfs_reclaim(void *v)
   1256 {
   1257 	struct vnode *vp = ((struct vop_reclaim_args *)v)->a_vp;
   1258 
   1259 	struct tmpfs_mount *tmp;
   1260 	struct tmpfs_node *node;
   1261 
   1262 	node = VP_TO_TMPFS_NODE(vp);
   1263 	tmp = VFS_TO_TMPFS(vp->v_mount);
   1264 
   1265 	cache_purge(vp);
   1266 	tmpfs_free_vp(vp);
   1267 
   1268 	/* If the node referenced by this vnode was deleted by the user,
   1269 	 * we must free its associated data structures (now that the vnode
   1270 	 * is being reclaimed). */
   1271 	if (node->tn_links == 0)
   1272 		tmpfs_free_node(tmp, node);
   1273 
   1274 	KASSERT(vp->v_data == NULL);
   1275 
   1276 	return 0;
   1277 }
   1278 
   1279 /* --------------------------------------------------------------------- */
   1280 
   1281 int
   1282 tmpfs_print(void *v)
   1283 {
   1284 	struct vnode *vp = ((struct vop_print_args *)v)->a_vp;
   1285 
   1286 	struct tmpfs_node *node;
   1287 
   1288 	node = VP_TO_TMPFS_NODE(vp);
   1289 
   1290 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
   1291 	    node, node->tn_flags, node->tn_links);
   1292 	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
   1293 	    ", status 0x%x\n",
   1294 	    node->tn_mode, node->tn_uid, node->tn_gid,
   1295 	    (uintmax_t)node->tn_size, node->tn_status);
   1296 	if (vp->v_type == VFIFO)
   1297 		fifo_printinfo(vp);
   1298 	printf("\n");
   1299 
   1300 	return 0;
   1301 }
   1302 
   1303 /* --------------------------------------------------------------------- */
   1304 
   1305 int
   1306 tmpfs_pathconf(void *v)
   1307 {
   1308 	int name = ((struct vop_pathconf_args *)v)->a_name;
   1309 	register_t *retval = ((struct vop_pathconf_args *)v)->a_retval;
   1310 
   1311 	int error;
   1312 
   1313 	error = 0;
   1314 
   1315 	switch (name) {
   1316 	case _PC_LINK_MAX:
   1317 		*retval = LINK_MAX;
   1318 		break;
   1319 
   1320 	case _PC_NAME_MAX:
   1321 		*retval = NAME_MAX;
   1322 		break;
   1323 
   1324 	case _PC_PATH_MAX:
   1325 		*retval = PATH_MAX;
   1326 		break;
   1327 
   1328 	case _PC_PIPE_BUF:
   1329 		*retval = PIPE_BUF;
   1330 		break;
   1331 
   1332 	case _PC_CHOWN_RESTRICTED:
   1333 		*retval = 1;
   1334 		break;
   1335 
   1336 	case _PC_NO_TRUNC:
   1337 		*retval = 1;
   1338 		break;
   1339 
   1340 	case _PC_SYNC_IO:
   1341 		*retval = 1;
   1342 		break;
   1343 
   1344 	case _PC_FILESIZEBITS:
   1345 		*retval = 0; /* XXX Don't know which value should I return. */
   1346 		break;
   1347 
   1348 	default:
   1349 		error = EINVAL;
   1350 	}
   1351 
   1352 	return error;
   1353 }
   1354 
   1355 /* --------------------------------------------------------------------- */
   1356 
   1357 int
   1358 tmpfs_advlock(void *v)
   1359 {
   1360 	struct vnode *vp = ((struct vop_advlock_args *)v)->a_vp;
   1361 
   1362 	struct tmpfs_node *node;
   1363 
   1364 	node = VP_TO_TMPFS_NODE(vp);
   1365 
   1366 	return lf_advlock(v, &node->tn_lockf, node->tn_size);
   1367 }
   1368 
   1369 /* --------------------------------------------------------------------- */
   1370 
   1371 int
   1372 tmpfs_getpages(void *v)
   1373 {
   1374 	struct vnode *vp = ((struct vop_getpages_args *)v)->a_vp;
   1375 	voff_t offset = ((struct vop_getpages_args *)v)->a_offset;
   1376 	struct vm_page **m = ((struct vop_getpages_args *)v)->a_m;
   1377 	int *count = ((struct vop_getpages_args *)v)->a_count;
   1378 	int centeridx = ((struct vop_getpages_args *)v)->a_centeridx;
   1379 	vm_prot_t access_type = ((struct vop_getpages_args *)v)->a_access_type;
   1380 	int advice = ((struct vop_getpages_args *)v)->a_advice;
   1381 	int flags = ((struct vop_getpages_args *)v)->a_flags;
   1382 
   1383 	int error;
   1384 	int i;
   1385 	struct tmpfs_node *node;
   1386 	struct uvm_object *uobj;
   1387 	int npages = *count;
   1388 
   1389 	KASSERT(vp->v_type == VREG);
   1390 	KASSERT(mutex_owned(&vp->v_interlock));
   1391 
   1392 	node = VP_TO_TMPFS_NODE(vp);
   1393 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1394 
   1395 	/* We currently don't rely on PGO_PASTEOF. */
   1396 
   1397 	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
   1398 		if ((flags & PGO_LOCKED) == 0)
   1399 			mutex_exit(&vp->v_interlock);
   1400 		return EINVAL;
   1401 	}
   1402 
   1403 	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
   1404 		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
   1405 	}
   1406 
   1407 	if ((flags & PGO_LOCKED) != 0)
   1408 		return EBUSY;
   1409 
   1410 	if ((flags & PGO_NOTIMESTAMP) == 0) {
   1411 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
   1412 			node->tn_status |= TMPFS_NODE_ACCESSED;
   1413 
   1414 		if ((access_type & VM_PROT_WRITE) != 0)
   1415 			node->tn_status |= TMPFS_NODE_MODIFIED;
   1416 	}
   1417 
   1418 	mutex_exit(&vp->v_interlock);
   1419 
   1420 	/*
   1421 	 * Make sure that the array on which we will store the
   1422 	 * gotten pages is clean.  Otherwise uao_get (pointed to by
   1423 	 * the pgo_get below) gets confused and does not return the
   1424 	 * appropriate pages.
   1425 	 *
   1426 	 * XXX This shall be revisited when kern/32166 is addressed
   1427 	 * because the loop to clean m[i] will most likely be redundant
   1428 	 * as well as the PGO_ALLPAGES flag.
   1429 	 */
   1430 	if (m != NULL)
   1431 		for (i = 0; i < npages; i++)
   1432 			m[i] = NULL;
   1433 	mutex_enter(&uobj->vmobjlock);
   1434 	error = (*uobj->pgops->pgo_get)(uobj, offset, m, &npages, centeridx,
   1435 	    access_type, advice, flags | PGO_ALLPAGES);
   1436 #if defined(DEBUG)
   1437 	{
   1438 		/* Make sure that all the pages we return are valid. */
   1439 		int dbgi;
   1440 		if (error == 0 && m != NULL)
   1441 			for (dbgi = 0; dbgi < npages; dbgi++)
   1442 				KASSERT(m[dbgi] != NULL);
   1443 	}
   1444 #endif
   1445 
   1446 	return error;
   1447 }
   1448 
   1449 /* --------------------------------------------------------------------- */
   1450 
   1451 int
   1452 tmpfs_putpages(void *v)
   1453 {
   1454 	struct vnode *vp = ((struct vop_putpages_args *)v)->a_vp;
   1455 	voff_t offlo = ((struct vop_putpages_args *)v)->a_offlo;
   1456 	voff_t offhi = ((struct vop_putpages_args *)v)->a_offhi;
   1457 	int flags = ((struct vop_putpages_args *)v)->a_flags;
   1458 
   1459 	int error;
   1460 	struct tmpfs_node *node;
   1461 	struct uvm_object *uobj;
   1462 
   1463 	KASSERT(mutex_owned(&vp->v_interlock));
   1464 
   1465 	node = VP_TO_TMPFS_NODE(vp);
   1466 
   1467 	if (vp->v_type != VREG) {
   1468 		mutex_exit(&vp->v_interlock);
   1469 		return 0;
   1470 	}
   1471 
   1472 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1473 	mutex_exit(&vp->v_interlock);
   1474 
   1475 	mutex_enter(&uobj->vmobjlock);
   1476 	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
   1477 
   1478 	/* XXX mtime */
   1479 
   1480 	return error;
   1481 }
   1482