Home | History | Annotate | Line # | Download | only in tmpfs
tmpfs_vnops.c revision 1.51.6.5
      1 /*	$NetBSD: tmpfs_vnops.c,v 1.51.6.5 2009/04/19 15:27:32 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 1.51.6.5 2009/04/19 15:27:32 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 				cnp->cn_flags |= SAVENAME;
    223 			} else
    224 				de = NULL;
    225 
    226 			/* Allocate a new vnode on the matching entry. */
    227 			error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp);
    228 		}
    229 	}
    230 
    231 	/* Store the result of this lookup in the cache.  Avoid this if the
    232 	 * request was for creation, as it does not improve timings on
    233 	 * emprical tests. */
    234 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
    235 	    (cnp->cn_flags & ISDOTDOT) == 0)
    236 		cache_enter(dvp, *vpp, cnp);
    237 
    238 out:
    239 	/* If there were no errors, *vpp cannot be null and it must be
    240 	 * locked. */
    241 	KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
    242 
    243 	/* dvp must always be locked. */
    244 	KASSERT(VOP_ISLOCKED(dvp));
    245 
    246 	return error;
    247 }
    248 
    249 /* --------------------------------------------------------------------- */
    250 
    251 int
    252 tmpfs_create(void *v)
    253 {
    254 	struct vnode *dvp = ((struct vop_create_args *)v)->a_dvp;
    255 	struct vnode **vpp = ((struct vop_create_args *)v)->a_vpp;
    256 	struct componentname *cnp = ((struct vop_create_args *)v)->a_cnp;
    257 	struct vattr *vap = ((struct vop_create_args *)v)->a_vap;
    258 
    259 	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
    260 
    261 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    262 }
    263 /* --------------------------------------------------------------------- */
    264 
    265 int
    266 tmpfs_mknod(void *v)
    267 {
    268 	struct vnode *dvp = ((struct vop_mknod_args *)v)->a_dvp;
    269 	struct vnode **vpp = ((struct vop_mknod_args *)v)->a_vpp;
    270 	struct componentname *cnp = ((struct vop_mknod_args *)v)->a_cnp;
    271 	struct vattr *vap = ((struct vop_mknod_args *)v)->a_vap;
    272 
    273 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
    274 	    vap->va_type != VFIFO) {
    275 		vput(dvp);
    276 		return EINVAL;
    277 	}
    278 
    279 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    280 }
    281 
    282 /* --------------------------------------------------------------------- */
    283 
    284 int
    285 tmpfs_open(void *v)
    286 {
    287 	struct vnode *vp = ((struct vop_open_args *)v)->a_vp;
    288 	int mode = ((struct vop_open_args *)v)->a_mode;
    289 
    290 	int error;
    291 	struct tmpfs_node *node;
    292 
    293 	KASSERT(VOP_ISLOCKED(vp));
    294 
    295 	node = VP_TO_TMPFS_NODE(vp);
    296 
    297 	/* The file is still active but all its names have been removed
    298 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
    299 	 * it is about to die. */
    300 	if (node->tn_links < 1) {
    301 		error = ENOENT;
    302 		goto out;
    303 	}
    304 
    305 	/* If the file is marked append-only, deny write requests. */
    306 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
    307 		error = EPERM;
    308 	else
    309 		error = 0;
    310 
    311 out:
    312 	KASSERT(VOP_ISLOCKED(vp));
    313 
    314 	return error;
    315 }
    316 
    317 /* --------------------------------------------------------------------- */
    318 
    319 int
    320 tmpfs_close(void *v)
    321 {
    322 	struct vnode *vp = ((struct vop_close_args *)v)->a_vp;
    323 
    324 	struct tmpfs_node *node;
    325 
    326 	KASSERT(VOP_ISLOCKED(vp));
    327 
    328 	node = VP_TO_TMPFS_NODE(vp);
    329 
    330 	if (node->tn_links > 0) {
    331 		/* Update node times.  No need to do it if the node has
    332 		 * been deleted, because it will vanish after we return. */
    333 		tmpfs_update(vp, NULL, NULL, NULL, UPDATE_CLOSE);
    334 	}
    335 
    336 	return 0;
    337 }
    338 
    339 /* --------------------------------------------------------------------- */
    340 
    341 int
    342 tmpfs_access(void *v)
    343 {
    344 	struct vnode *vp = ((struct vop_access_args *)v)->a_vp;
    345 	int mode = ((struct vop_access_args *)v)->a_mode;
    346 	kauth_cred_t cred = ((struct vop_access_args *)v)->a_cred;
    347 
    348 	int error;
    349 	struct tmpfs_node *node;
    350 
    351 	KASSERT(VOP_ISLOCKED(vp));
    352 
    353 	node = VP_TO_TMPFS_NODE(vp);
    354 
    355 	switch (vp->v_type) {
    356 	case VDIR:
    357 		/* FALLTHROUGH */
    358 	case VLNK:
    359 		/* FALLTHROUGH */
    360 	case VREG:
    361 		if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
    362 			error = EROFS;
    363 			goto out;
    364 		}
    365 		break;
    366 
    367 	case VBLK:
    368 		/* FALLTHROUGH */
    369 	case VCHR:
    370 		/* FALLTHROUGH */
    371 	case VSOCK:
    372 		/* FALLTHROUGH */
    373 	case VFIFO:
    374 		break;
    375 
    376 	default:
    377 		error = EINVAL;
    378 		goto out;
    379 	}
    380 
    381 	if (mode & VWRITE && node->tn_flags & IMMUTABLE) {
    382 		error = EPERM;
    383 		goto out;
    384 	}
    385 
    386 	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
    387 	    node->tn_gid, mode, cred);
    388 
    389 out:
    390 	KASSERT(VOP_ISLOCKED(vp));
    391 
    392 	return error;
    393 }
    394 
    395 /* --------------------------------------------------------------------- */
    396 
    397 int
    398 tmpfs_getattr(void *v)
    399 {
    400 	struct vnode *vp = ((struct vop_getattr_args *)v)->a_vp;
    401 	struct vattr *vap = ((struct vop_getattr_args *)v)->a_vap;
    402 
    403 	struct tmpfs_node *node;
    404 
    405 	node = VP_TO_TMPFS_NODE(vp);
    406 
    407 	VATTR_NULL(vap);
    408 
    409 	tmpfs_itimes(vp, NULL, NULL, NULL);
    410 
    411 	vap->va_type = vp->v_type;
    412 	vap->va_mode = node->tn_mode;
    413 	vap->va_nlink = node->tn_links;
    414 	vap->va_uid = node->tn_uid;
    415 	vap->va_gid = node->tn_gid;
    416 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
    417 	vap->va_fileid = node->tn_id;
    418 	vap->va_size = node->tn_size;
    419 	vap->va_blocksize = PAGE_SIZE;
    420 	vap->va_atime = node->tn_atime;
    421 	vap->va_mtime = node->tn_mtime;
    422 	vap->va_ctime = node->tn_ctime;
    423 	vap->va_birthtime = node->tn_birthtime;
    424 	vap->va_gen = node->tn_gen;
    425 	vap->va_flags = node->tn_flags;
    426 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
    427 		node->tn_spec.tn_dev.tn_rdev : VNOVAL;
    428 	vap->va_bytes = round_page(node->tn_size);
    429 	vap->va_filerev = VNOVAL;
    430 	vap->va_vaflags = 0;
    431 	vap->va_spare = VNOVAL; /* XXX */
    432 
    433 	return 0;
    434 }
    435 
    436 /* --------------------------------------------------------------------- */
    437 
    438 #define GOODTIME(tv)	((tv)->tv_sec != VNOVAL || (tv)->tv_nsec != VNOVAL)
    439 /* XXX Should this operation be atomic?  I think it should, but code in
    440  * XXX other places (e.g., ufs) doesn't seem to be... */
    441 int
    442 tmpfs_setattr(void *v)
    443 {
    444 	struct vnode *vp = ((struct vop_setattr_args *)v)->a_vp;
    445 	struct vattr *vap = ((struct vop_setattr_args *)v)->a_vap;
    446 	kauth_cred_t cred = ((struct vop_setattr_args *)v)->a_cred;
    447 	struct lwp *l = curlwp;
    448 
    449 	int error;
    450 
    451 	KASSERT(VOP_ISLOCKED(vp));
    452 
    453 	error = 0;
    454 
    455 	/* Abort if any unsettable attribute is given. */
    456 	if (vap->va_type != VNON ||
    457 	    vap->va_nlink != VNOVAL ||
    458 	    vap->va_fsid != VNOVAL ||
    459 	    vap->va_fileid != VNOVAL ||
    460 	    vap->va_blocksize != VNOVAL ||
    461 	    GOODTIME(&vap->va_ctime) ||
    462 	    vap->va_gen != VNOVAL ||
    463 	    vap->va_rdev != VNOVAL ||
    464 	    vap->va_bytes != VNOVAL)
    465 		error = EINVAL;
    466 
    467 	if (error == 0 && (vap->va_flags != VNOVAL))
    468 		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
    469 
    470 	if (error == 0 && (vap->va_size != VNOVAL))
    471 		error = tmpfs_chsize(vp, vap->va_size, cred, l);
    472 
    473 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
    474 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
    475 
    476 	if (error == 0 && (vap->va_mode != VNOVAL))
    477 		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
    478 
    479 	if (error == 0 && (GOODTIME(&vap->va_atime) || GOODTIME(&vap->va_mtime)
    480 	    || GOODTIME(&vap->va_birthtime)))
    481 		if ((error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
    482 		    &vap->va_birthtime, vap->va_vaflags, cred, l)) == 0)
    483 			return 0;
    484 
    485 	/* Update the node times.  We give preference to the error codes
    486 	 * generated by this function rather than the ones that may arise
    487 	 * from tmpfs_update. */
    488 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    489 
    490 	KASSERT(VOP_ISLOCKED(vp));
    491 
    492 	return error;
    493 }
    494 
    495 /* --------------------------------------------------------------------- */
    496 
    497 int
    498 tmpfs_read(void *v)
    499 {
    500 	struct vnode *vp = ((struct vop_read_args *)v)->a_vp;
    501 	struct uio *uio = ((struct vop_read_args *)v)->a_uio;
    502 
    503 	int error;
    504 	int flags;
    505 	struct tmpfs_node *node;
    506 	struct uvm_object *uobj;
    507 
    508 	KASSERT(VOP_ISLOCKED(vp));
    509 
    510 	node = VP_TO_TMPFS_NODE(vp);
    511 
    512 	if (vp->v_type != VREG) {
    513 		error = EISDIR;
    514 		goto out;
    515 	}
    516 
    517 	if (uio->uio_offset < 0) {
    518 		error = EINVAL;
    519 		goto out;
    520 	}
    521 
    522 	node->tn_status |= TMPFS_NODE_ACCESSED;
    523 
    524 	uobj = node->tn_spec.tn_reg.tn_aobj;
    525 	flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
    526 	error = 0;
    527 	while (error == 0 && uio->uio_resid > 0) {
    528 		vsize_t len;
    529 		void *win;
    530 
    531 		if (node->tn_size <= uio->uio_offset)
    532 			break;
    533 
    534 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    535 		if (len == 0)
    536 			break;
    537 
    538 		win = ubc_alloc(uobj, uio->uio_offset, &len, UVM_ADV_NORMAL,
    539 		    UBC_READ);
    540 		error = uiomove(win, len, uio);
    541 		ubc_release(win, flags);
    542 	}
    543 
    544 out:
    545 	KASSERT(VOP_ISLOCKED(vp));
    546 
    547 	return error;
    548 }
    549 
    550 /* --------------------------------------------------------------------- */
    551 
    552 int
    553 tmpfs_write(void *v)
    554 {
    555 	struct vnode *vp = ((struct vop_write_args *)v)->a_vp;
    556 	struct uio *uio = ((struct vop_write_args *)v)->a_uio;
    557 	int ioflag = ((struct vop_write_args *)v)->a_ioflag;
    558 
    559 	bool extended;
    560 	int error;
    561 	int flags;
    562 	off_t oldsize;
    563 	struct tmpfs_node *node;
    564 	struct uvm_object *uobj;
    565 
    566 	KASSERT(VOP_ISLOCKED(vp));
    567 
    568 	node = VP_TO_TMPFS_NODE(vp);
    569 	oldsize = node->tn_size;
    570 
    571 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
    572 		error = EINVAL;
    573 		goto out;
    574 	}
    575 
    576 	if (uio->uio_resid == 0) {
    577 		error = 0;
    578 		goto out;
    579 	}
    580 
    581 	if (ioflag & IO_APPEND)
    582 		uio->uio_offset = node->tn_size;
    583 
    584 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
    585 	if (extended) {
    586 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
    587 		if (error != 0)
    588 			goto out;
    589 	}
    590 
    591 	uobj = node->tn_spec.tn_reg.tn_aobj;
    592 	flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
    593 	error = 0;
    594 	while (error == 0 && uio->uio_resid > 0) {
    595 		vsize_t len;
    596 		void *win;
    597 
    598 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    599 		if (len == 0)
    600 			break;
    601 
    602 		win = ubc_alloc(uobj, uio->uio_offset, &len, UVM_ADV_NORMAL,
    603 		    UBC_WRITE);
    604 		error = uiomove(win, len, uio);
    605 		ubc_release(win, flags);
    606 	}
    607 
    608 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
    609 	    (extended ? TMPFS_NODE_CHANGED : 0);
    610 
    611 	if (error != 0)
    612 		(void)tmpfs_reg_resize(vp, oldsize);
    613 
    614 	VN_KNOTE(vp, NOTE_WRITE);
    615 
    616 out:
    617 	KASSERT(VOP_ISLOCKED(vp));
    618 	KASSERT(IMPLIES(error == 0, uio->uio_resid == 0));
    619 	KASSERT(IMPLIES(error != 0, oldsize == node->tn_size));
    620 
    621 	return error;
    622 }
    623 
    624 /* --------------------------------------------------------------------- */
    625 
    626 int
    627 tmpfs_fsync(void *v)
    628 {
    629 	struct vnode *vp = ((struct vop_fsync_args *)v)->a_vp;
    630 
    631 	KASSERT(VOP_ISLOCKED(vp));
    632 
    633 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    634 
    635 	return 0;
    636 }
    637 
    638 /* --------------------------------------------------------------------- */
    639 
    640 int
    641 tmpfs_remove(void *v)
    642 {
    643 	struct vnode *dvp = ((struct vop_remove_args *)v)->a_dvp;
    644 	struct vnode *vp = ((struct vop_remove_args *)v)->a_vp;
    645 	struct componentname *cnp = (((struct vop_remove_args *)v)->a_cnp);
    646 
    647 	int error;
    648 	struct tmpfs_dirent *de;
    649 	struct tmpfs_mount *tmp;
    650 	struct tmpfs_node *dnode;
    651 	struct tmpfs_node *node;
    652 
    653 	KASSERT(VOP_ISLOCKED(dvp));
    654 	KASSERT(VOP_ISLOCKED(vp));
    655 
    656 	if (vp->v_type == VDIR) {
    657 		error = EPERM;
    658 		goto out;
    659 	}
    660 
    661 	dnode = VP_TO_TMPFS_DIR(dvp);
    662 	node = VP_TO_TMPFS_NODE(vp);
    663 	tmp = VFS_TO_TMPFS(vp->v_mount);
    664 	de = tmpfs_dir_lookup(dnode, cnp);
    665 	if (de == NULL) {
    666 		error = ENOENT;
    667 		goto out;
    668 	}
    669 	KASSERT(de->td_node == node);
    670 
    671 	/* Files marked as immutable or append-only cannot be deleted. */
    672 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    673 		error = EPERM;
    674 		goto out;
    675 	}
    676 
    677 	/* Remove the entry from the directory; as it is a file, we do not
    678 	 * have to change the number of hard links of the directory. */
    679 	tmpfs_dir_detach(dvp, de);
    680 
    681 	/* Free the directory entry we just deleted.  Note that the node
    682 	 * referred by it will not be removed until the vnode is really
    683 	 * reclaimed. */
    684 	tmpfs_free_dirent(tmp, de, true);
    685 
    686 	error = 0;
    687 
    688 out:
    689 	vput(vp);
    690 	if (dvp == vp)
    691 		vrele(dvp);
    692 	else
    693 		vput(dvp);
    694 	PNBUF_PUT(cnp->cn_pnbuf);
    695 
    696 	return error;
    697 }
    698 
    699 /* --------------------------------------------------------------------- */
    700 
    701 int
    702 tmpfs_link(void *v)
    703 {
    704 	struct vnode *dvp = ((struct vop_link_args *)v)->a_dvp;
    705 	struct vnode *vp = ((struct vop_link_args *)v)->a_vp;
    706 	struct componentname *cnp = ((struct vop_link_args *)v)->a_cnp;
    707 
    708 	int error;
    709 	struct tmpfs_dirent *de;
    710 	struct tmpfs_node *dnode;
    711 	struct tmpfs_node *node;
    712 
    713 	KASSERT(VOP_ISLOCKED(dvp));
    714 	KASSERT(cnp->cn_flags & HASBUF);
    715 	KASSERT(dvp != vp); /* XXX When can this be false? */
    716 
    717 	dnode = VP_TO_TMPFS_DIR(dvp);
    718 	node = VP_TO_TMPFS_NODE(vp);
    719 
    720 	/* Lock vp because we will need to run tmpfs_update over it, which
    721 	 * needs the vnode to be locked. */
    722 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    723 	if (error != 0)
    724 		goto out1;
    725 
    726 	/* XXX: Why aren't the following two tests done by the caller? */
    727 
    728 	/* Hard links of directories are forbidden. */
    729 	if (vp->v_type == VDIR) {
    730 		error = EPERM;
    731 		goto out;
    732 	}
    733 
    734 	/* Cannot create cross-device links. */
    735 	if (dvp->v_mount != vp->v_mount) {
    736 		error = EXDEV;
    737 		goto out;
    738 	}
    739 
    740 	/* Ensure that we do not overflow the maximum number of links imposed
    741 	 * by the system. */
    742 	KASSERT(node->tn_links <= LINK_MAX);
    743 	if (node->tn_links == LINK_MAX) {
    744 		error = EMLINK;
    745 		goto out;
    746 	}
    747 
    748 	/* We cannot create links of files marked immutable or append-only. */
    749 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    750 		error = EPERM;
    751 		goto out;
    752 	}
    753 
    754 	/* Allocate a new directory entry to represent the node. */
    755 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
    756 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
    757 	if (error != 0)
    758 		goto out;
    759 
    760 	/* Insert the new directory entry into the appropriate directory. */
    761 	tmpfs_dir_attach(dvp, de);
    762 
    763 	/* vp link count has changed, so update node times. */
    764 	node->tn_status |= TMPFS_NODE_CHANGED;
    765 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    766 
    767 	error = 0;
    768 
    769 out:
    770 	VOP_UNLOCK(vp, 0);
    771 out1:
    772 	PNBUF_PUT(cnp->cn_pnbuf);
    773 
    774 	vput(dvp);
    775 
    776 	return error;
    777 }
    778 
    779 /* --------------------------------------------------------------------- */
    780 
    781 int
    782 tmpfs_rename(void *v)
    783 {
    784 	struct vnode *fdvp = ((struct vop_rename_args *)v)->a_fdvp;
    785 	struct vnode *fvp = ((struct vop_rename_args *)v)->a_fvp;
    786 	struct componentname *fcnp = ((struct vop_rename_args *)v)->a_fcnp;
    787 	struct vnode *tdvp = ((struct vop_rename_args *)v)->a_tdvp;
    788 	struct vnode *tvp = ((struct vop_rename_args *)v)->a_tvp;
    789 	struct componentname *tcnp = ((struct vop_rename_args *)v)->a_tcnp;
    790 
    791 	char *newname;
    792 	int error;
    793 	struct tmpfs_dirent *de, *de2;
    794 	struct tmpfs_mount *tmp;
    795 	struct tmpfs_node *fdnode;
    796 	struct tmpfs_node *fnode;
    797 	struct tmpfs_node *tnode;
    798 	struct tmpfs_node *tdnode;
    799 	size_t namelen;
    800 
    801 	KASSERT(VOP_ISLOCKED(tdvp));
    802 	KASSERT(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
    803 	KASSERT(fcnp->cn_flags & HASBUF);
    804 	KASSERT(tcnp->cn_flags & HASBUF);
    805 
    806 	newname = NULL;
    807 	namelen = 0;
    808 	tmp = NULL;
    809 
    810 	/* Disallow cross-device renames. */
    811 	if (fvp->v_mount != tdvp->v_mount ||
    812 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
    813 		error = EXDEV;
    814 		goto out_unlocked;
    815 	}
    816 
    817 	fnode = VP_TO_TMPFS_NODE(fvp);
    818 	fdnode = VP_TO_TMPFS_DIR(fdvp);
    819 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
    820 	tdnode = VP_TO_TMPFS_DIR(tdvp);
    821 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
    822 
    823 	if (fdvp == tvp) {
    824 		error = 0;
    825 		goto out_unlocked;
    826 	}
    827 
    828 	/* If we need to move the directory between entries, lock the
    829 	 * source so that we can safely operate on it. */
    830 
    831 	/* XXX: this is a potential locking order violation! */
    832 	if (fdnode != tdnode) {
    833 		error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
    834 		if (error != 0)
    835 			goto out_unlocked;
    836 	}
    837 
    838 	/*
    839 	 * If the node we were renaming has scarpered, just give up.
    840 	 */
    841 	de = tmpfs_dir_lookup(fdnode, fcnp);
    842 	if (de == NULL || de->td_node != fnode) {
    843 		error = ENOENT;
    844 		goto out;
    845 	}
    846 
    847 	/* If source and target are the same file, there is nothing to do. */
    848 	if (fvp == tvp) {
    849 		error = 0;
    850 		goto out;
    851 	}
    852 
    853 	/* If replacing an existing entry, ensure we can do the operation. */
    854 	if (tvp != NULL) {
    855 		KASSERT(tnode != NULL);
    856 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
    857 			if (tnode->tn_size > 0) {
    858 				error = ENOTEMPTY;
    859 				goto out;
    860 			}
    861 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
    862 			error = ENOTDIR;
    863 			goto out;
    864 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
    865 			error = EISDIR;
    866 			goto out;
    867 		} else {
    868 			KASSERT(fnode->tn_type != VDIR &&
    869 			        tnode->tn_type != VDIR);
    870 		}
    871 	}
    872 
    873 	/* Ensure that we have enough memory to hold the new name, if it
    874 	 * has to be changed. */
    875 	namelen = tcnp->cn_namelen;
    876 	if (fcnp->cn_namelen != tcnp->cn_namelen ||
    877 	    memcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
    878 		newname = tmpfs_str_pool_get(&tmp->tm_str_pool, namelen, 0);
    879 		if (newname == NULL) {
    880 			error = ENOSPC;
    881 			goto out;
    882 		}
    883 	}
    884 
    885 	/* If the node is being moved to another directory, we have to do
    886 	 * the move. */
    887 	if (fdnode != tdnode) {
    888 		/* In case we are moving a directory, we have to adjust its
    889 		 * parent to point to the new parent. */
    890 		if (de->td_node->tn_type == VDIR) {
    891 			struct tmpfs_node *n;
    892 
    893 			/* Ensure the target directory is not a child of the
    894 			 * directory being moved.  Otherwise, we'd end up
    895 			 * with stale nodes. */
    896 			n = tdnode;
    897 			while (n != n->tn_spec.tn_dir.tn_parent) {
    898 				if (n == fnode) {
    899 					error = EINVAL;
    900 					goto out;
    901 				}
    902 				n = n->tn_spec.tn_dir.tn_parent;
    903 			}
    904 
    905 			/* Adjust the parent pointer. */
    906 			TMPFS_VALIDATE_DIR(fnode);
    907 			de->td_node->tn_spec.tn_dir.tn_parent = tdnode;
    908 
    909 			/* As a result of changing the target of the '..'
    910 			 * entry, the link count of the source and target
    911 			 * directories has to be adjusted. */
    912 			fdnode->tn_links--;
    913 			tdnode->tn_links++;
    914 		}
    915 
    916 		/* Do the move: just remove the entry from the source directory
    917 		 * and insert it into the target one. */
    918 		tmpfs_dir_detach(fdvp, de);
    919 		tmpfs_dir_attach(tdvp, de);
    920 
    921 		/* Notify listeners of fdvp about the change in the directory.
    922 		 * We can do it at this point because we aren't touching fdvp
    923 		 * any more below. */
    924 		VN_KNOTE(fdvp, NOTE_WRITE);
    925 	}
    926 
    927 	/* If we are overwriting an entry, we have to remove the old one
    928 	 * from the target directory. */
    929 	if (tvp != NULL) {
    930 		KASSERT(tnode != NULL);
    931 
    932 		/* Remove the old entry from the target directory.
    933 		 * Note! This relies on tmpfs_dir_attach() putting the new
    934 		 * node on the end of the target's node list. */
    935 		de2 = tmpfs_dir_lookup(tdnode, tcnp);
    936 		KASSERT(de2 != NULL);
    937 		KASSERT(de2->td_node == tnode);
    938 		tmpfs_dir_detach(tdvp, de2);
    939 
    940 		/* Free the directory entry we just deleted.  Note that the
    941 		 * node referred by it will not be removed until the vnode is
    942 		 * really reclaimed. */
    943 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de2, true);
    944 	}
    945 
    946 	/* If the name has changed, we need to make it effective by changing
    947 	 * it in the directory entry. */
    948 	if (newname != NULL) {
    949 		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
    950 		KASSERT(tcnp->cn_namelen < 0xffff);
    951 
    952 		tmpfs_str_pool_put(&tmp->tm_str_pool, de->td_name,
    953 		    de->td_namelen);
    954 		de->td_namelen = (uint16_t)namelen;
    955 		memcpy(newname, tcnp->cn_nameptr, namelen);
    956 		de->td_name = newname;
    957 		newname = NULL;
    958 
    959 		fnode->tn_status |= TMPFS_NODE_CHANGED;
    960 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
    961 	}
    962 
    963 	/* Notify listeners of tdvp about the change in the directory (either
    964 	 * because a new entry was added or because one was removed) and
    965 	 * listeners of fvp about the rename. */
    966 	VN_KNOTE(tdvp, NOTE_WRITE);
    967 	VN_KNOTE(fvp, NOTE_RENAME);
    968 
    969 	error = 0;
    970 
    971  out:
    972 	if (fdnode != tdnode)
    973 		VOP_UNLOCK(fdvp, 0);
    974 
    975  out_unlocked:
    976 	/* Release target nodes. */
    977 	if (tdvp == tvp)
    978 		vrele(tdvp);
    979 	else
    980 		vput(tdvp);
    981 	if (tvp != NULL)
    982 		vput(tvp);
    983 
    984 	/* Release source nodes. */
    985 	vrele(fdvp);
    986 	vrele(fvp);
    987 
    988 	if (newname != NULL)
    989 		tmpfs_str_pool_put(&tmp->tm_str_pool, newname, namelen);
    990 
    991 	return error;
    992 }
    993 
    994 /* --------------------------------------------------------------------- */
    995 
    996 int
    997 tmpfs_mkdir(void *v)
    998 {
    999 	struct vnode *dvp = ((struct vop_mkdir_args *)v)->a_dvp;
   1000 	struct vnode **vpp = ((struct vop_mkdir_args *)v)->a_vpp;
   1001 	struct componentname *cnp = ((struct vop_mkdir_args *)v)->a_cnp;
   1002 	struct vattr *vap = ((struct vop_mkdir_args *)v)->a_vap;
   1003 
   1004 	KASSERT(vap->va_type == VDIR);
   1005 
   1006 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
   1007 }
   1008 
   1009 /* --------------------------------------------------------------------- */
   1010 
   1011 int
   1012 tmpfs_rmdir(void *v)
   1013 {
   1014 	struct vnode *dvp = ((struct vop_rmdir_args *)v)->a_dvp;
   1015 	struct vnode *vp = ((struct vop_rmdir_args *)v)->a_vp;
   1016 	struct componentname *cnp = ((struct vop_rmdir_args *)v)->a_cnp;
   1017 
   1018 	int error;
   1019 	struct tmpfs_dirent *de;
   1020 	struct tmpfs_mount *tmp;
   1021 	struct tmpfs_node *dnode;
   1022 	struct tmpfs_node *node;
   1023 
   1024 	KASSERT(VOP_ISLOCKED(dvp));
   1025 	KASSERT(VOP_ISLOCKED(vp));
   1026 
   1027 	tmp = VFS_TO_TMPFS(dvp->v_mount);
   1028 	dnode = VP_TO_TMPFS_DIR(dvp);
   1029 	node = VP_TO_TMPFS_DIR(vp);
   1030 	error = 0;
   1031 
   1032 	/* Directories with more than two entries ('.' and '..') cannot be
   1033 	 * removed. */
   1034 	if (node->tn_size > 0) {
   1035 		error = ENOTEMPTY;
   1036 		goto out;
   1037 	}
   1038 
   1039 	/* This invariant holds only if we are not trying to remove "..".
   1040 	 * We checked for that above so this is safe now. */
   1041 	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
   1042 
   1043 	/* Get the directory entry associated with node (vp). */
   1044 	de = tmpfs_dir_lookup(dnode, cnp);
   1045 	if (de == NULL) {
   1046 		error = ENOENT;
   1047 		goto out;
   1048 	}
   1049 	KASSERT(de->td_node == node);
   1050 
   1051 	/* Check flags to see if we are allowed to remove the directory. */
   1052 	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
   1053 		error = EPERM;
   1054 		goto out;
   1055 	}
   1056 
   1057 	/* Detach the directory entry from the directory (dnode). */
   1058 	tmpfs_dir_detach(dvp, de);
   1059 
   1060 	node->tn_links--;
   1061 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
   1062 	    TMPFS_NODE_MODIFIED;
   1063 	node->tn_spec.tn_dir.tn_parent->tn_links--;
   1064 	node->tn_spec.tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \
   1065 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
   1066 
   1067 	/* Release the parent. */
   1068 	cache_purge(dvp); /* XXX Is this needed? */
   1069 
   1070 	/* Free the directory entry we just deleted.  Note that the node
   1071 	 * referred by it will not be removed until the vnode is really
   1072 	 * reclaimed. */
   1073 	tmpfs_free_dirent(tmp, de, true);
   1074 
   1075 	KASSERT(node->tn_links == 0);
   1076  out:
   1077 	/* Release the nodes. */
   1078 	vput(dvp);
   1079 	vput(vp);
   1080 	PNBUF_PUT(cnp->cn_pnbuf);
   1081 
   1082 	return error;
   1083 }
   1084 
   1085 /* --------------------------------------------------------------------- */
   1086 
   1087 int
   1088 tmpfs_symlink(void *v)
   1089 {
   1090 	struct vnode *dvp = ((struct vop_symlink_args *)v)->a_dvp;
   1091 	struct vnode **vpp = ((struct vop_symlink_args *)v)->a_vpp;
   1092 	struct componentname *cnp = ((struct vop_symlink_args *)v)->a_cnp;
   1093 	struct vattr *vap = ((struct vop_symlink_args *)v)->a_vap;
   1094 	char *target = ((struct vop_symlink_args *)v)->a_target;
   1095 
   1096 	KASSERT(vap->va_type == VLNK);
   1097 
   1098 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
   1099 }
   1100 
   1101 /* --------------------------------------------------------------------- */
   1102 
   1103 int
   1104 tmpfs_readdir(void *v)
   1105 {
   1106 	struct vnode *vp = ((struct vop_readdir_args *)v)->a_vp;
   1107 	struct uio *uio = ((struct vop_readdir_args *)v)->a_uio;
   1108 	int *eofflag = ((struct vop_readdir_args *)v)->a_eofflag;
   1109 	off_t **cookies = ((struct vop_readdir_args *)v)->a_cookies;
   1110 	int *ncookies = ((struct vop_readdir_args *)v)->a_ncookies;
   1111 
   1112 	int error;
   1113 	off_t startoff;
   1114 	off_t cnt;
   1115 	struct tmpfs_node *node;
   1116 
   1117 	KASSERT(VOP_ISLOCKED(vp));
   1118 
   1119 	/* This operation only makes sense on directory nodes. */
   1120 	if (vp->v_type != VDIR) {
   1121 		error = ENOTDIR;
   1122 		goto out;
   1123 	}
   1124 
   1125 	node = VP_TO_TMPFS_DIR(vp);
   1126 
   1127 	startoff = uio->uio_offset;
   1128 
   1129 	cnt = 0;
   1130 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
   1131 		error = tmpfs_dir_getdotdent(node, uio);
   1132 		if (error == -1) {
   1133 			error = 0;
   1134 			goto outok;
   1135 		} else if (error != 0)
   1136 			goto outok;
   1137 		cnt++;
   1138 	}
   1139 
   1140 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
   1141 		error = tmpfs_dir_getdotdotdent(node, uio);
   1142 		if (error == -1) {
   1143 			error = 0;
   1144 			goto outok;
   1145 		} else if (error != 0)
   1146 			goto outok;
   1147 		cnt++;
   1148 	}
   1149 
   1150 	error = tmpfs_dir_getdents(node, uio, &cnt);
   1151 	if (error == -1)
   1152 		error = 0;
   1153 	KASSERT(error >= 0);
   1154 
   1155 outok:
   1156 	/* This label assumes that startoff has been
   1157 	 * initialized.  If the compiler didn't spit out warnings, we'd
   1158 	 * simply make this one be 'out' and drop 'outok'. */
   1159 
   1160 	if (eofflag != NULL)
   1161 		*eofflag =
   1162 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
   1163 
   1164 	/* Update NFS-related variables. */
   1165 	if (error == 0 && cookies != NULL && ncookies != NULL) {
   1166 		off_t i;
   1167 		off_t off = startoff;
   1168 		struct tmpfs_dirent *de = NULL;
   1169 
   1170 		*ncookies = cnt;
   1171 		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
   1172 
   1173 		for (i = 0; i < cnt; i++) {
   1174 			KASSERT(off != TMPFS_DIRCOOKIE_EOF);
   1175 			if (off == TMPFS_DIRCOOKIE_DOT) {
   1176 				off = TMPFS_DIRCOOKIE_DOTDOT;
   1177 			} else {
   1178 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
   1179 					de = TAILQ_FIRST(&node->tn_spec.
   1180 					    tn_dir.tn_dir);
   1181 				} else if (de != NULL) {
   1182 					de = TAILQ_NEXT(de, td_entries);
   1183 				} else {
   1184 					de = tmpfs_dir_lookupbycookie(node,
   1185 					    off);
   1186 					KASSERT(de != NULL);
   1187 					de = TAILQ_NEXT(de, td_entries);
   1188 				}
   1189 				if (de == NULL) {
   1190 					off = TMPFS_DIRCOOKIE_EOF;
   1191 				} else {
   1192 					off = tmpfs_dircookie(de);
   1193 				}
   1194 			}
   1195 
   1196 			(*cookies)[i] = off;
   1197 		}
   1198 		KASSERT(uio->uio_offset == off);
   1199 	}
   1200 
   1201 out:
   1202 	KASSERT(VOP_ISLOCKED(vp));
   1203 
   1204 	return error;
   1205 }
   1206 
   1207 /* --------------------------------------------------------------------- */
   1208 
   1209 int
   1210 tmpfs_readlink(void *v)
   1211 {
   1212 	struct vnode *vp = ((struct vop_readlink_args *)v)->a_vp;
   1213 	struct uio *uio = ((struct vop_readlink_args *)v)->a_uio;
   1214 
   1215 	int error;
   1216 	struct tmpfs_node *node;
   1217 
   1218 	KASSERT(VOP_ISLOCKED(vp));
   1219 	KASSERT(uio->uio_offset == 0);
   1220 	KASSERT(vp->v_type == VLNK);
   1221 
   1222 	node = VP_TO_TMPFS_NODE(vp);
   1223 
   1224 	error = uiomove(node->tn_spec.tn_lnk.tn_link,
   1225 	    MIN(node->tn_size, uio->uio_resid), uio);
   1226 	node->tn_status |= TMPFS_NODE_ACCESSED;
   1227 
   1228 	KASSERT(VOP_ISLOCKED(vp));
   1229 
   1230 	return error;
   1231 }
   1232 
   1233 /* --------------------------------------------------------------------- */
   1234 
   1235 int
   1236 tmpfs_inactive(void *v)
   1237 {
   1238 	struct vnode *vp = ((struct vop_inactive_args *)v)->a_vp;
   1239 
   1240 	struct tmpfs_node *node;
   1241 
   1242 	KASSERT(VOP_ISLOCKED(vp));
   1243 
   1244 	node = VP_TO_TMPFS_NODE(vp);
   1245 	*((struct vop_inactive_args *)v)->a_recycle = (node->tn_links == 0);
   1246 	VOP_UNLOCK(vp, 0);
   1247 
   1248 	return 0;
   1249 }
   1250 
   1251 /* --------------------------------------------------------------------- */
   1252 
   1253 int
   1254 tmpfs_reclaim(void *v)
   1255 {
   1256 	struct vnode *vp = ((struct vop_reclaim_args *)v)->a_vp;
   1257 
   1258 	struct tmpfs_mount *tmp;
   1259 	struct tmpfs_node *node;
   1260 
   1261 	node = VP_TO_TMPFS_NODE(vp);
   1262 	tmp = VFS_TO_TMPFS(vp->v_mount);
   1263 
   1264 	cache_purge(vp);
   1265 	tmpfs_free_vp(vp);
   1266 
   1267 	/* If the node referenced by this vnode was deleted by the user,
   1268 	 * we must free its associated data structures (now that the vnode
   1269 	 * is being reclaimed). */
   1270 	if (node->tn_links == 0)
   1271 		tmpfs_free_node(tmp, node);
   1272 
   1273 	KASSERT(vp->v_data == NULL);
   1274 
   1275 	return 0;
   1276 }
   1277 
   1278 /* --------------------------------------------------------------------- */
   1279 
   1280 int
   1281 tmpfs_print(void *v)
   1282 {
   1283 	struct vnode *vp = ((struct vop_print_args *)v)->a_vp;
   1284 
   1285 	struct tmpfs_node *node;
   1286 
   1287 	node = VP_TO_TMPFS_NODE(vp);
   1288 
   1289 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
   1290 	    node, node->tn_flags, node->tn_links);
   1291 	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
   1292 	    ", status 0x%x\n",
   1293 	    node->tn_mode, node->tn_uid, node->tn_gid,
   1294 	    (uintmax_t)node->tn_size, node->tn_status);
   1295 	if (vp->v_type == VFIFO)
   1296 		fifo_printinfo(vp);
   1297 	printf("\n");
   1298 
   1299 	return 0;
   1300 }
   1301 
   1302 /* --------------------------------------------------------------------- */
   1303 
   1304 int
   1305 tmpfs_pathconf(void *v)
   1306 {
   1307 	int name = ((struct vop_pathconf_args *)v)->a_name;
   1308 	register_t *retval = ((struct vop_pathconf_args *)v)->a_retval;
   1309 
   1310 	int error;
   1311 
   1312 	error = 0;
   1313 
   1314 	switch (name) {
   1315 	case _PC_LINK_MAX:
   1316 		*retval = LINK_MAX;
   1317 		break;
   1318 
   1319 	case _PC_NAME_MAX:
   1320 		*retval = NAME_MAX;
   1321 		break;
   1322 
   1323 	case _PC_PATH_MAX:
   1324 		*retval = PATH_MAX;
   1325 		break;
   1326 
   1327 	case _PC_PIPE_BUF:
   1328 		*retval = PIPE_BUF;
   1329 		break;
   1330 
   1331 	case _PC_CHOWN_RESTRICTED:
   1332 		*retval = 1;
   1333 		break;
   1334 
   1335 	case _PC_NO_TRUNC:
   1336 		*retval = 1;
   1337 		break;
   1338 
   1339 	case _PC_SYNC_IO:
   1340 		*retval = 1;
   1341 		break;
   1342 
   1343 	case _PC_FILESIZEBITS:
   1344 		*retval = 0; /* XXX Don't know which value should I return. */
   1345 		break;
   1346 
   1347 	default:
   1348 		error = EINVAL;
   1349 	}
   1350 
   1351 	return error;
   1352 }
   1353 
   1354 /* --------------------------------------------------------------------- */
   1355 
   1356 int
   1357 tmpfs_advlock(void *v)
   1358 {
   1359 	struct vnode *vp = ((struct vop_advlock_args *)v)->a_vp;
   1360 
   1361 	struct tmpfs_node *node;
   1362 
   1363 	node = VP_TO_TMPFS_NODE(vp);
   1364 
   1365 	return lf_advlock(v, &node->tn_lockf, node->tn_size);
   1366 }
   1367 
   1368 /* --------------------------------------------------------------------- */
   1369 
   1370 int
   1371 tmpfs_getpages(void *v)
   1372 {
   1373 	struct vnode *vp = ((struct vop_getpages_args *)v)->a_vp;
   1374 	voff_t offset = ((struct vop_getpages_args *)v)->a_offset;
   1375 	struct vm_page **m = ((struct vop_getpages_args *)v)->a_m;
   1376 	int *count = ((struct vop_getpages_args *)v)->a_count;
   1377 	int centeridx = ((struct vop_getpages_args *)v)->a_centeridx;
   1378 	vm_prot_t access_type = ((struct vop_getpages_args *)v)->a_access_type;
   1379 	int advice = ((struct vop_getpages_args *)v)->a_advice;
   1380 	int flags = ((struct vop_getpages_args *)v)->a_flags;
   1381 
   1382 	int error;
   1383 	int i;
   1384 	struct tmpfs_node *node;
   1385 	struct uvm_object *uobj;
   1386 	int npages = *count;
   1387 
   1388 	KASSERT(vp->v_type == VREG);
   1389 	KASSERT(mutex_owned(&vp->v_interlock));
   1390 
   1391 	node = VP_TO_TMPFS_NODE(vp);
   1392 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1393 
   1394 	/* We currently don't rely on PGO_PASTEOF. */
   1395 
   1396 	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
   1397 		if ((flags & PGO_LOCKED) == 0)
   1398 			mutex_exit(&vp->v_interlock);
   1399 		return EINVAL;
   1400 	}
   1401 
   1402 	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
   1403 		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
   1404 	}
   1405 
   1406 	if ((flags & PGO_LOCKED) != 0)
   1407 		return EBUSY;
   1408 
   1409 	if ((flags & PGO_NOTIMESTAMP) == 0) {
   1410 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
   1411 			node->tn_status |= TMPFS_NODE_ACCESSED;
   1412 
   1413 		if ((access_type & VM_PROT_WRITE) != 0)
   1414 			node->tn_status |= TMPFS_NODE_MODIFIED;
   1415 	}
   1416 
   1417 	mutex_exit(&vp->v_interlock);
   1418 
   1419 	/*
   1420 	 * Make sure that the array on which we will store the
   1421 	 * gotten pages is clean.  Otherwise uao_get (pointed to by
   1422 	 * the pgo_get below) gets confused and does not return the
   1423 	 * appropriate pages.
   1424 	 *
   1425 	 * XXX This shall be revisited when kern/32166 is addressed
   1426 	 * because the loop to clean m[i] will most likely be redundant
   1427 	 * as well as the PGO_ALLPAGES flag.
   1428 	 */
   1429 	if (m != NULL)
   1430 		for (i = 0; i < npages; i++)
   1431 			m[i] = NULL;
   1432 	mutex_enter(&uobj->vmobjlock);
   1433 	error = (*uobj->pgops->pgo_get)(uobj, offset, m, &npages, centeridx,
   1434 	    access_type, advice, flags | PGO_ALLPAGES);
   1435 #if defined(DEBUG)
   1436 	{
   1437 		/* Make sure that all the pages we return are valid. */
   1438 		int dbgi;
   1439 		if (error == 0 && m != NULL)
   1440 			for (dbgi = 0; dbgi < npages; dbgi++)
   1441 				KASSERT(m[dbgi] != NULL);
   1442 	}
   1443 #endif
   1444 
   1445 	return error;
   1446 }
   1447 
   1448 /* --------------------------------------------------------------------- */
   1449 
   1450 int
   1451 tmpfs_putpages(void *v)
   1452 {
   1453 	struct vnode *vp = ((struct vop_putpages_args *)v)->a_vp;
   1454 	voff_t offlo = ((struct vop_putpages_args *)v)->a_offlo;
   1455 	voff_t offhi = ((struct vop_putpages_args *)v)->a_offhi;
   1456 	int flags = ((struct vop_putpages_args *)v)->a_flags;
   1457 
   1458 	int error;
   1459 	struct tmpfs_node *node;
   1460 	struct uvm_object *uobj;
   1461 
   1462 	KASSERT(mutex_owned(&vp->v_interlock));
   1463 
   1464 	node = VP_TO_TMPFS_NODE(vp);
   1465 
   1466 	if (vp->v_type != VREG) {
   1467 		mutex_exit(&vp->v_interlock);
   1468 		return 0;
   1469 	}
   1470 
   1471 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1472 	mutex_exit(&vp->v_interlock);
   1473 
   1474 	mutex_enter(&uobj->vmobjlock);
   1475 	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
   1476 
   1477 	/* XXX mtime */
   1478 
   1479 	return error;
   1480 }
   1481