Home | History | Annotate | Line # | Download | only in tmpfs
tmpfs_vnops.c revision 1.72
      1 /*	$NetBSD: tmpfs_vnops.c,v 1.72 2010/07/02 03:29:47 rmind 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 1.72 2010/07/02 03:29:47 rmind Exp $");
     39 
     40 #include <sys/param.h>
     41 #include <sys/dirent.h>
     42 #include <sys/fcntl.h>
     43 #include <sys/event.h>
     44 #include <sys/malloc.h>
     45 #include <sys/namei.h>
     46 #include <sys/proc.h>
     47 #include <sys/stat.h>
     48 #include <sys/uio.h>
     49 #include <sys/unistd.h>
     50 #include <sys/vnode.h>
     51 #include <sys/lockf.h>
     52 #include <sys/kauth.h>
     53 
     54 #include <uvm/uvm.h>
     55 
     56 #include <miscfs/fifofs/fifo.h>
     57 #include <miscfs/genfs/genfs.h>
     58 #include <fs/tmpfs/tmpfs_vnops.h>
     59 #include <fs/tmpfs/tmpfs.h>
     60 
     61 /* --------------------------------------------------------------------- */
     62 
     63 /*
     64  * vnode operations vector used for files stored in a tmpfs file system.
     65  */
     66 int (**tmpfs_vnodeop_p)(void *);
     67 const struct vnodeopv_entry_desc tmpfs_vnodeop_entries[] = {
     68 	{ &vop_default_desc,		vn_default_error },
     69 	{ &vop_lookup_desc,		tmpfs_lookup },
     70 	{ &vop_create_desc,		tmpfs_create },
     71 	{ &vop_mknod_desc,		tmpfs_mknod },
     72 	{ &vop_open_desc,		tmpfs_open },
     73 	{ &vop_close_desc,		tmpfs_close },
     74 	{ &vop_access_desc,		tmpfs_access },
     75 	{ &vop_getattr_desc,		tmpfs_getattr },
     76 	{ &vop_setattr_desc,		tmpfs_setattr },
     77 	{ &vop_read_desc,		tmpfs_read },
     78 	{ &vop_write_desc,		tmpfs_write },
     79 	{ &vop_ioctl_desc,		tmpfs_ioctl },
     80 	{ &vop_fcntl_desc,		tmpfs_fcntl },
     81 	{ &vop_poll_desc,		tmpfs_poll },
     82 	{ &vop_kqfilter_desc,		tmpfs_kqfilter },
     83 	{ &vop_revoke_desc,		tmpfs_revoke },
     84 	{ &vop_mmap_desc,		tmpfs_mmap },
     85 	{ &vop_fsync_desc,		tmpfs_fsync },
     86 	{ &vop_seek_desc,		tmpfs_seek },
     87 	{ &vop_remove_desc,		tmpfs_remove },
     88 	{ &vop_link_desc,		tmpfs_link },
     89 	{ &vop_rename_desc,		tmpfs_rename },
     90 	{ &vop_mkdir_desc,		tmpfs_mkdir },
     91 	{ &vop_rmdir_desc,		tmpfs_rmdir },
     92 	{ &vop_symlink_desc,		tmpfs_symlink },
     93 	{ &vop_readdir_desc,		tmpfs_readdir },
     94 	{ &vop_readlink_desc,		tmpfs_readlink },
     95 	{ &vop_abortop_desc,		tmpfs_abortop },
     96 	{ &vop_inactive_desc,		tmpfs_inactive },
     97 	{ &vop_reclaim_desc,		tmpfs_reclaim },
     98 	{ &vop_lock_desc,		tmpfs_lock },
     99 	{ &vop_unlock_desc,		tmpfs_unlock },
    100 	{ &vop_bmap_desc,		tmpfs_bmap },
    101 	{ &vop_strategy_desc,		tmpfs_strategy },
    102 	{ &vop_print_desc,		tmpfs_print },
    103 	{ &vop_pathconf_desc,		tmpfs_pathconf },
    104 	{ &vop_islocked_desc,		tmpfs_islocked },
    105 	{ &vop_advlock_desc,		tmpfs_advlock },
    106 	{ &vop_bwrite_desc,		tmpfs_bwrite },
    107 	{ &vop_getpages_desc,		tmpfs_getpages },
    108 	{ &vop_putpages_desc,		tmpfs_putpages },
    109 	{ NULL, NULL }
    110 };
    111 const struct vnodeopv_desc tmpfs_vnodeop_opv_desc =
    112 	{ &tmpfs_vnodeop_p, tmpfs_vnodeop_entries };
    113 
    114 /*
    115  * tmpfs_lookup: lookup routine.
    116  *
    117  * Arguments: dvp (directory being searched), vpp (result),
    118  * cnp (component name - path).
    119  *
    120  * => Caller holds a reference and lock on dvp.
    121  * => We return looked-up vnode (vpp) locked, with a reference held.
    122  */
    123 int
    124 tmpfs_lookup(void *v)
    125 {
    126 	struct vop_lookup_args /* {
    127 		struct vnode *a_dvp;
    128 		struct vnode **a_vpp;
    129 		struct componentname *a_cnp;
    130 	} */ *ap = v;
    131 	struct vnode *dvp = ap->a_dvp, **vpp = ap->a_vpp;
    132 	struct componentname *cnp = ap->a_cnp;
    133 	struct tmpfs_dirent *de;
    134 	struct tmpfs_node *dnode;
    135 	int error;
    136 
    137 	KASSERT(VOP_ISLOCKED(dvp));
    138 
    139 	dnode = VP_TO_TMPFS_DIR(dvp);
    140 	*vpp = NULL;
    141 
    142 	/* Check accessibility of requested node as a first step. */
    143 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
    144 	if (error != 0)
    145 		goto out;
    146 
    147 	/*
    148 	 * If requesting the last path component on a read-only file system
    149 	 * with a write operation, deny it.
    150 	 */
    151 	if ((cnp->cn_flags & ISLASTCN) &&
    152 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
    153 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    154 		error = EROFS;
    155 		goto out;
    156 	}
    157 
    158 	/*
    159 	 * Avoid doing a linear scan of the directory if the requested
    160 	 * directory/name couple is already in the cache.
    161 	 */
    162 	error = cache_lookup(dvp, vpp, cnp);
    163 	if (error >= 0)
    164 		goto out;
    165 
    166 	/* We cannot be requesting the parent directory of the root node. */
    167 	KASSERT(IMPLIES(dnode->tn_type == VDIR &&
    168 	    dnode->tn_spec.tn_dir.tn_parent == dnode,
    169 	    !(cnp->cn_flags & ISDOTDOT)));
    170 
    171 	if (cnp->cn_flags & ISDOTDOT) {
    172 		VOP_UNLOCK(dvp);
    173 
    174 		/* Allocate a new vnode on the matching entry. */
    175 		error = tmpfs_alloc_vp(dvp->v_mount,
    176 		    dnode->tn_spec.tn_dir.tn_parent, vpp);
    177 
    178 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
    179 		goto done;
    180 
    181 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
    182 		vref(dvp);
    183 		*vpp = dvp;
    184 		error = 0;
    185 		goto done;
    186 	}
    187 
    188 	de = tmpfs_dir_lookup(dnode, cnp);
    189 	if (de == NULL) {
    190 		/*
    191 		 * The entry was not found in the directory.  This is valid
    192 		 * if we are creating or renaming an entry and are working
    193 		 * on the last component of the path name.
    194 		 */
    195 		if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == CREATE ||
    196 		    cnp->cn_nameiop == RENAME)) {
    197 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
    198 			if (error) {
    199 				goto out;
    200 			}
    201 			/* Keep the component name for future uses. */
    202 			cnp->cn_flags |= SAVENAME;
    203 			error = EJUSTRETURN;
    204 		} else {
    205 			error = ENOENT;
    206 		}
    207 	} else {
    208 		struct tmpfs_node *tnode = de->td_node;
    209 
    210 		/*
    211 		 * If we are not at the last path component and found a
    212 		 * non-directory or non-link entry (which may itself be
    213 		 * pointing to a directory), raise an error.
    214 		 */
    215 		if ((tnode->tn_type != VDIR && tnode->tn_type != VLNK) &&
    216 		    (cnp->cn_flags & ISLASTCN) == 0) {
    217 			error = ENOTDIR;
    218 			goto out;
    219 		}
    220 
    221 		/* Check permissions. */
    222 		if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == DELETE ||
    223 		    cnp->cn_nameiop == RENAME)) {
    224 			kauth_action_t action = 0;
    225 
    226 			/* This is the file-system's decision. */
    227 			if ((dnode->tn_mode & S_ISTXT) != 0 &&
    228 			    kauth_cred_geteuid(cnp->cn_cred) != dnode->tn_uid &&
    229 			    kauth_cred_geteuid(cnp->cn_cred) != tnode->tn_uid)
    230 				error = EPERM;
    231 			else
    232 				error = 0;
    233 
    234 			/* Only bother if we are not already failing it. */
    235 			if (!error) {
    236 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
    237 			}
    238 
    239 			if (cnp->cn_nameiop == DELETE) {
    240 				action |= KAUTH_VNODE_DELETE;
    241 			} else {
    242 				KASSERT(cnp->cn_nameiop == RENAME);
    243 				action |= KAUTH_VNODE_RENAME;
    244 			}
    245 			error = kauth_authorize_vnode(cnp->cn_cred,
    246 			    action, *vpp, dvp, error);
    247 			if (error) {
    248 				goto out;
    249 			}
    250 			cnp->cn_flags |= SAVENAME;
    251 		}
    252 		/* Allocate a new vnode on the matching entry. */
    253 		error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp);
    254 	}
    255 done:
    256 	/*
    257 	 * Store the result of this lookup in the cache.  Avoid this if the
    258 	 * request was for creation, as it does not improve timings on
    259 	 * emprical tests.
    260 	 */
    261 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
    262 	    (cnp->cn_flags & ISDOTDOT) == 0)
    263 		cache_enter(dvp, *vpp, cnp);
    264 
    265 out:
    266 	KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
    267 	KASSERT(VOP_ISLOCKED(dvp));
    268 	return error;
    269 }
    270 
    271 int
    272 tmpfs_create(void *v)
    273 {
    274 	struct vnode *dvp = ((struct vop_create_args *)v)->a_dvp;
    275 	struct vnode **vpp = ((struct vop_create_args *)v)->a_vpp;
    276 	struct componentname *cnp = ((struct vop_create_args *)v)->a_cnp;
    277 	struct vattr *vap = ((struct vop_create_args *)v)->a_vap;
    278 
    279 	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
    280 
    281 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    282 }
    283 /* --------------------------------------------------------------------- */
    284 
    285 int
    286 tmpfs_mknod(void *v)
    287 {
    288 	struct vnode *dvp = ((struct vop_mknod_args *)v)->a_dvp;
    289 	struct vnode **vpp = ((struct vop_mknod_args *)v)->a_vpp;
    290 	struct componentname *cnp = ((struct vop_mknod_args *)v)->a_cnp;
    291 	struct vattr *vap = ((struct vop_mknod_args *)v)->a_vap;
    292 
    293 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
    294 	    vap->va_type != VFIFO) {
    295 		vput(dvp);
    296 		return EINVAL;
    297 	}
    298 
    299 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    300 }
    301 
    302 /* --------------------------------------------------------------------- */
    303 
    304 int
    305 tmpfs_open(void *v)
    306 {
    307 	struct vnode *vp = ((struct vop_open_args *)v)->a_vp;
    308 	int mode = ((struct vop_open_args *)v)->a_mode;
    309 
    310 	int error;
    311 	struct tmpfs_node *node;
    312 
    313 	KASSERT(VOP_ISLOCKED(vp));
    314 
    315 	node = VP_TO_TMPFS_NODE(vp);
    316 
    317 	/* The file is still active but all its names have been removed
    318 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
    319 	 * it is about to die. */
    320 	if (node->tn_links < 1) {
    321 		error = ENOENT;
    322 		goto out;
    323 	}
    324 
    325 	/* If the file is marked append-only, deny write requests. */
    326 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
    327 		error = EPERM;
    328 	else
    329 		error = 0;
    330 
    331 out:
    332 	KASSERT(VOP_ISLOCKED(vp));
    333 
    334 	return error;
    335 }
    336 
    337 /* --------------------------------------------------------------------- */
    338 
    339 int
    340 tmpfs_close(void *v)
    341 {
    342 	struct vnode *vp = ((struct vop_close_args *)v)->a_vp;
    343 
    344 	struct tmpfs_node *node;
    345 
    346 	KASSERT(VOP_ISLOCKED(vp));
    347 
    348 	node = VP_TO_TMPFS_NODE(vp);
    349 
    350 	if (node->tn_links > 0) {
    351 		/* Update node times.  No need to do it if the node has
    352 		 * been deleted, because it will vanish after we return. */
    353 		tmpfs_update(vp, NULL, NULL, NULL, UPDATE_CLOSE);
    354 	}
    355 
    356 	return 0;
    357 }
    358 
    359 /* --------------------------------------------------------------------- */
    360 
    361 static int
    362 tmpfs_check_possible(struct vnode *vp, struct tmpfs_node *node, mode_t mode)
    363 {
    364 	int error = 0;
    365 
    366 	switch (vp->v_type) {
    367 	case VDIR:
    368 		/* FALLTHROUGH */
    369 	case VLNK:
    370 		/* FALLTHROUGH */
    371 	case VREG:
    372 		if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
    373 			error = EROFS;
    374 			goto out;
    375 		}
    376 		break;
    377 
    378 	case VBLK:
    379 		/* FALLTHROUGH */
    380 	case VCHR:
    381 		/* FALLTHROUGH */
    382 	case VSOCK:
    383 		/* FALLTHROUGH */
    384 	case VFIFO:
    385 		break;
    386 
    387 	default:
    388 		error = EINVAL;
    389 		goto out;
    390 	}
    391 
    392 	if (mode & VWRITE && node->tn_flags & IMMUTABLE) {
    393 		error = EPERM;
    394 		goto out;
    395 	}
    396 
    397  out:
    398 	return error;
    399 }
    400 
    401 static int
    402 tmpfs_check_permitted(struct vnode *vp, struct tmpfs_node *node, mode_t mode,
    403     kauth_cred_t cred)
    404 {
    405 
    406 	return genfs_can_access(vp->v_type, node->tn_mode, node->tn_uid,
    407 	    node->tn_gid, mode, cred);
    408 }
    409 
    410 int
    411 tmpfs_access(void *v)
    412 {
    413 	struct vnode *vp = ((struct vop_access_args *)v)->a_vp;
    414 	int mode = ((struct vop_access_args *)v)->a_mode;
    415 	kauth_cred_t cred = ((struct vop_access_args *)v)->a_cred;
    416 
    417 	int error;
    418 	struct tmpfs_node *node;
    419 
    420 	KASSERT(VOP_ISLOCKED(vp));
    421 
    422 	node = VP_TO_TMPFS_NODE(vp);
    423 
    424 	error = tmpfs_check_possible(vp, node, mode);
    425 	if (error)
    426 		goto out;
    427 
    428 	error = tmpfs_check_permitted(vp, node, mode, cred);
    429 
    430 	error = kauth_authorize_vnode(cred, kauth_mode_to_action(mode), vp,
    431 	    NULL, error);
    432 
    433 out:
    434 	KASSERT(VOP_ISLOCKED(vp));
    435 
    436 	return error;
    437 }
    438 
    439 /* --------------------------------------------------------------------- */
    440 
    441 int
    442 tmpfs_getattr(void *v)
    443 {
    444 	struct vnode *vp = ((struct vop_getattr_args *)v)->a_vp;
    445 	struct vattr *vap = ((struct vop_getattr_args *)v)->a_vap;
    446 
    447 	struct tmpfs_node *node;
    448 
    449 	node = VP_TO_TMPFS_NODE(vp);
    450 
    451 	vattr_null(vap);
    452 
    453 	tmpfs_itimes(vp, NULL, NULL, NULL);
    454 
    455 	vap->va_type = vp->v_type;
    456 	vap->va_mode = node->tn_mode;
    457 	vap->va_nlink = node->tn_links;
    458 	vap->va_uid = node->tn_uid;
    459 	vap->va_gid = node->tn_gid;
    460 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
    461 	vap->va_fileid = node->tn_id;
    462 	vap->va_size = node->tn_size;
    463 	vap->va_blocksize = PAGE_SIZE;
    464 	vap->va_atime = node->tn_atime;
    465 	vap->va_mtime = node->tn_mtime;
    466 	vap->va_ctime = node->tn_ctime;
    467 	vap->va_birthtime = node->tn_birthtime;
    468 	vap->va_gen = node->tn_gen;
    469 	vap->va_flags = node->tn_flags;
    470 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
    471 		node->tn_spec.tn_dev.tn_rdev : VNOVAL;
    472 	vap->va_bytes = round_page(node->tn_size);
    473 	vap->va_filerev = VNOVAL;
    474 	vap->va_vaflags = 0;
    475 	vap->va_spare = VNOVAL; /* XXX */
    476 
    477 	return 0;
    478 }
    479 
    480 /* --------------------------------------------------------------------- */
    481 
    482 #define GOODTIME(tv)	((tv)->tv_sec != VNOVAL || (tv)->tv_nsec != VNOVAL)
    483 /* XXX Should this operation be atomic?  I think it should, but code in
    484  * XXX other places (e.g., ufs) doesn't seem to be... */
    485 int
    486 tmpfs_setattr(void *v)
    487 {
    488 	struct vnode *vp = ((struct vop_setattr_args *)v)->a_vp;
    489 	struct vattr *vap = ((struct vop_setattr_args *)v)->a_vap;
    490 	kauth_cred_t cred = ((struct vop_setattr_args *)v)->a_cred;
    491 	struct lwp *l = curlwp;
    492 
    493 	int error;
    494 
    495 	KASSERT(VOP_ISLOCKED(vp));
    496 
    497 	error = 0;
    498 
    499 	/* Abort if any unsettable attribute is given. */
    500 	if (vap->va_type != VNON ||
    501 	    vap->va_nlink != VNOVAL ||
    502 	    vap->va_fsid != VNOVAL ||
    503 	    vap->va_fileid != VNOVAL ||
    504 	    vap->va_blocksize != VNOVAL ||
    505 	    GOODTIME(&vap->va_ctime) ||
    506 	    vap->va_gen != VNOVAL ||
    507 	    vap->va_rdev != VNOVAL ||
    508 	    vap->va_bytes != VNOVAL)
    509 		error = EINVAL;
    510 
    511 	if (error == 0 && (vap->va_flags != VNOVAL))
    512 		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
    513 
    514 	if (error == 0 && (vap->va_size != VNOVAL))
    515 		error = tmpfs_chsize(vp, vap->va_size, cred, l);
    516 
    517 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
    518 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
    519 
    520 	if (error == 0 && (vap->va_mode != VNOVAL))
    521 		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
    522 
    523 	if (error == 0 && (GOODTIME(&vap->va_atime) || GOODTIME(&vap->va_mtime)
    524 	    || GOODTIME(&vap->va_birthtime)))
    525 		if ((error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
    526 		    &vap->va_birthtime, vap->va_vaflags, cred, l)) == 0)
    527 			return 0;
    528 
    529 	/* Update the node times.  We give preference to the error codes
    530 	 * generated by this function rather than the ones that may arise
    531 	 * from tmpfs_update. */
    532 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    533 
    534 	KASSERT(VOP_ISLOCKED(vp));
    535 
    536 	return error;
    537 }
    538 
    539 /* --------------------------------------------------------------------- */
    540 
    541 int
    542 tmpfs_read(void *v)
    543 {
    544 	struct vnode *vp = ((struct vop_read_args *)v)->a_vp;
    545 	struct uio *uio = ((struct vop_read_args *)v)->a_uio;
    546 	int ioflag = ((struct vop_read_args *)v)->a_ioflag;
    547 
    548 	int error;
    549 	struct tmpfs_node *node;
    550 	struct uvm_object *uobj;
    551 
    552 	KASSERT(VOP_ISLOCKED(vp));
    553 
    554 	node = VP_TO_TMPFS_NODE(vp);
    555 
    556 	if (vp->v_type != VREG) {
    557 		error = EISDIR;
    558 		goto out;
    559 	}
    560 
    561 	if (uio->uio_offset < 0) {
    562 		error = EINVAL;
    563 		goto out;
    564 	}
    565 
    566 	node->tn_status |= TMPFS_NODE_ACCESSED;
    567 
    568 	uobj = node->tn_spec.tn_reg.tn_aobj;
    569 	error = 0;
    570 	while (error == 0 && uio->uio_resid > 0) {
    571 		vsize_t len;
    572 
    573 		if (node->tn_size <= uio->uio_offset)
    574 			break;
    575 
    576 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    577 		if (len == 0)
    578 			break;
    579 
    580 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
    581 		    UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
    582 	}
    583 
    584 out:
    585 	KASSERT(VOP_ISLOCKED(vp));
    586 
    587 	return error;
    588 }
    589 
    590 /* --------------------------------------------------------------------- */
    591 
    592 int
    593 tmpfs_write(void *v)
    594 {
    595 	struct vnode *vp = ((struct vop_write_args *)v)->a_vp;
    596 	struct uio *uio = ((struct vop_write_args *)v)->a_uio;
    597 	int ioflag = ((struct vop_write_args *)v)->a_ioflag;
    598 
    599 	bool extended;
    600 	int error;
    601 	off_t oldsize;
    602 	struct tmpfs_node *node;
    603 	struct uvm_object *uobj;
    604 
    605 	KASSERT(VOP_ISLOCKED(vp));
    606 
    607 	node = VP_TO_TMPFS_NODE(vp);
    608 	oldsize = node->tn_size;
    609 
    610 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
    611 		error = EINVAL;
    612 		goto out;
    613 	}
    614 
    615 	if (uio->uio_resid == 0) {
    616 		error = 0;
    617 		goto out;
    618 	}
    619 
    620 	if (ioflag & IO_APPEND)
    621 		uio->uio_offset = node->tn_size;
    622 
    623 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
    624 	if (extended) {
    625 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
    626 		if (error != 0)
    627 			goto out;
    628 	}
    629 
    630 	uobj = node->tn_spec.tn_reg.tn_aobj;
    631 	error = 0;
    632 	while (error == 0 && uio->uio_resid > 0) {
    633 		vsize_t len;
    634 
    635 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    636 		if (len == 0)
    637 			break;
    638 
    639 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
    640 		    UBC_WRITE | UBC_UNMAP_FLAG(vp));
    641 	}
    642 
    643 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
    644 	    (extended ? TMPFS_NODE_CHANGED : 0);
    645 
    646 	if (error != 0)
    647 		(void)tmpfs_reg_resize(vp, oldsize);
    648 
    649 	VN_KNOTE(vp, NOTE_WRITE);
    650 
    651 out:
    652 	KASSERT(VOP_ISLOCKED(vp));
    653 	KASSERT(IMPLIES(error == 0, uio->uio_resid == 0));
    654 	KASSERT(IMPLIES(error != 0, oldsize == node->tn_size));
    655 
    656 	return error;
    657 }
    658 
    659 /* --------------------------------------------------------------------- */
    660 
    661 int
    662 tmpfs_fsync(void *v)
    663 {
    664 	struct vnode *vp = ((struct vop_fsync_args *)v)->a_vp;
    665 
    666 	KASSERT(VOP_ISLOCKED(vp));
    667 
    668 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    669 
    670 	return 0;
    671 }
    672 
    673 /* --------------------------------------------------------------------- */
    674 
    675 int
    676 tmpfs_remove(void *v)
    677 {
    678 	struct vnode *dvp = ((struct vop_remove_args *)v)->a_dvp;
    679 	struct vnode *vp = ((struct vop_remove_args *)v)->a_vp;
    680 	struct componentname *cnp = (((struct vop_remove_args *)v)->a_cnp);
    681 
    682 	int error;
    683 	struct tmpfs_dirent *de;
    684 	struct tmpfs_mount *tmp;
    685 	struct tmpfs_node *dnode;
    686 	struct tmpfs_node *node;
    687 
    688 	KASSERT(VOP_ISLOCKED(dvp));
    689 	KASSERT(VOP_ISLOCKED(vp));
    690 
    691 	if (vp->v_type == VDIR) {
    692 		error = EPERM;
    693 		goto out;
    694 	}
    695 
    696 	dnode = VP_TO_TMPFS_DIR(dvp);
    697 	node = VP_TO_TMPFS_NODE(vp);
    698 	tmp = VFS_TO_TMPFS(vp->v_mount);
    699 	de = tmpfs_dir_lookup(dnode, cnp);
    700 	KASSERT(de);
    701 	KASSERT(de->td_node == node);
    702 
    703 	/* Files marked as immutable or append-only cannot be deleted. */
    704 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    705 		error = EPERM;
    706 		goto out;
    707 	}
    708 
    709 	/* Remove the entry from the directory; as it is a file, we do not
    710 	 * have to change the number of hard links of the directory. */
    711 	tmpfs_dir_detach(dvp, de);
    712 
    713 	/* Free the directory entry we just deleted.  Note that the node
    714 	 * referred by it will not be removed until the vnode is really
    715 	 * reclaimed. */
    716 	tmpfs_free_dirent(tmp, de, true);
    717 
    718 	error = 0;
    719 
    720 out:
    721 	vput(vp);
    722 	if (dvp == vp)
    723 		vrele(dvp);
    724 	else
    725 		vput(dvp);
    726 	if (cnp->cn_flags & HASBUF) {
    727 		PNBUF_PUT(cnp->cn_pnbuf);
    728 		cnp->cn_flags &= ~HASBUF;
    729 	}
    730 
    731 	return error;
    732 }
    733 
    734 /* --------------------------------------------------------------------- */
    735 
    736 int
    737 tmpfs_link(void *v)
    738 {
    739 	struct vnode *dvp = ((struct vop_link_args *)v)->a_dvp;
    740 	struct vnode *vp = ((struct vop_link_args *)v)->a_vp;
    741 	struct componentname *cnp = ((struct vop_link_args *)v)->a_cnp;
    742 
    743 	int error;
    744 	struct tmpfs_dirent *de;
    745 	struct tmpfs_node *dnode;
    746 	struct tmpfs_node *node;
    747 
    748 	KASSERT(VOP_ISLOCKED(dvp));
    749 	KASSERT(cnp->cn_flags & HASBUF);
    750 	KASSERT(dvp != vp); /* XXX When can this be false? */
    751 
    752 	dnode = VP_TO_TMPFS_DIR(dvp);
    753 	node = VP_TO_TMPFS_NODE(vp);
    754 
    755 	/* Lock vp because we will need to run tmpfs_update over it, which
    756 	 * needs the vnode to be locked. */
    757 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    758 
    759 	/* XXX: Why aren't the following two tests done by the caller? */
    760 
    761 	/* Hard links of directories are forbidden. */
    762 	if (vp->v_type == VDIR) {
    763 		error = EPERM;
    764 		goto out;
    765 	}
    766 
    767 	/* Cannot create cross-device links. */
    768 	if (dvp->v_mount != vp->v_mount) {
    769 		error = EXDEV;
    770 		goto out;
    771 	}
    772 
    773 	/* Ensure that we do not overflow the maximum number of links imposed
    774 	 * by the system. */
    775 	KASSERT(node->tn_links <= LINK_MAX);
    776 	if (node->tn_links == LINK_MAX) {
    777 		error = EMLINK;
    778 		goto out;
    779 	}
    780 
    781 	/* We cannot create links of files marked immutable or append-only. */
    782 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    783 		error = EPERM;
    784 		goto out;
    785 	}
    786 
    787 	/* Allocate a new directory entry to represent the node. */
    788 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
    789 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
    790 	if (error != 0)
    791 		goto out;
    792 
    793 	/* Insert the new directory entry into the appropriate directory. */
    794 	tmpfs_dir_attach(dvp, de);
    795 
    796 	/* vp link count has changed, so update node times. */
    797 	node->tn_status |= TMPFS_NODE_CHANGED;
    798 	tmpfs_update(vp, NULL, NULL, NULL, 0);
    799 
    800 	error = 0;
    801 
    802 out:
    803 	VOP_UNLOCK(vp);
    804 	PNBUF_PUT(cnp->cn_pnbuf);
    805 	vput(dvp);
    806 
    807 	return error;
    808 }
    809 
    810 /*
    811  * tmpfs_rename: rename routine.
    812  *
    813  * Arguments: fdvp (from-parent vnode), fvp (from-leaf), tdvp (to-parent)
    814  * and tvp (to-leaf), if exists (NULL if not).
    815  *
    816  * => Caller holds a reference on fdvp and fvp, they are unlocked.
    817  *    Note: fdvp and fvp can refer to the same object (i.e. when it is root).
    818  *
    819  * => Both tdvp and tvp are referenced and locked.  It is our responsibility
    820  *    to release the references and unlock them (or destroy).
    821  */
    822 int
    823 tmpfs_rename(void *v)
    824 {
    825 	struct vnode *fdvp = ((struct vop_rename_args *)v)->a_fdvp;
    826 	struct vnode *fvp = ((struct vop_rename_args *)v)->a_fvp;
    827 	struct componentname *fcnp = ((struct vop_rename_args *)v)->a_fcnp;
    828 	struct vnode *tdvp = ((struct vop_rename_args *)v)->a_tdvp;
    829 	struct vnode *tvp = ((struct vop_rename_args *)v)->a_tvp;
    830 	struct componentname *tcnp = ((struct vop_rename_args *)v)->a_tcnp;
    831 
    832 	char *newname;
    833 	int error;
    834 	struct tmpfs_dirent *de, *de2;
    835 	struct tmpfs_mount *tmp;
    836 	struct tmpfs_node *fdnode;
    837 	struct tmpfs_node *fnode;
    838 	struct tmpfs_node *tnode;
    839 	struct tmpfs_node *tdnode;
    840 	size_t namelen;
    841 
    842 	KASSERT(VOP_ISLOCKED(tdvp));
    843 	KASSERT(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
    844 	KASSERT(fcnp->cn_flags & HASBUF);
    845 	KASSERT(tcnp->cn_flags & HASBUF);
    846 
    847 	newname = NULL;
    848 	namelen = 0;
    849 	tmp = NULL;
    850 
    851 	/* Disallow cross-device renames. */
    852 	if (fvp->v_mount != tdvp->v_mount ||
    853 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
    854 		error = EXDEV;
    855 		goto out_unlocked;
    856 	}
    857 
    858 	fnode = VP_TO_TMPFS_NODE(fvp);
    859 	fdnode = VP_TO_TMPFS_DIR(fdvp);
    860 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
    861 	tdnode = VP_TO_TMPFS_DIR(tdvp);
    862 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
    863 
    864 	if (fdvp == tvp) {
    865 		error = 0;
    866 		goto out_unlocked;
    867 	}
    868 
    869 	/* Allocate memory, if necessary, for a new name. */
    870 	namelen = tcnp->cn_namelen;
    871 	if (tmpfs_strname_neqlen(fcnp, tcnp)) {
    872 		newname = tmpfs_strname_alloc(tmp, namelen);
    873 		if (newname == NULL) {
    874 			error = ENOSPC;
    875 			goto out_unlocked;
    876 		}
    877 	}
    878 
    879 	/* If we need to move the directory between entries, lock the
    880 	 * source so that we can safely operate on it. */
    881 
    882 	/* XXX: this is a potential locking order violation! */
    883 	if (fdnode != tdnode) {
    884 		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
    885 	}
    886 
    887 	/*
    888 	 * If the node we were renaming has scarpered, just give up.
    889 	 */
    890 	de = tmpfs_dir_lookup(fdnode, fcnp);
    891 	if (de == NULL || de->td_node != fnode) {
    892 		error = ENOENT;
    893 		goto out;
    894 	}
    895 
    896 	/* If source and target is the same vnode, remove the source link. */
    897 	if (fvp == tvp) {
    898 		/*
    899 		 * Detach and free the directory entry.  Drops the link
    900 		 * count on the node.
    901 		 */
    902 		tmpfs_dir_detach(fdvp, de);
    903 		tmpfs_free_dirent(VFS_TO_TMPFS(fvp->v_mount), de, true);
    904 		VN_KNOTE(fdvp, NOTE_WRITE);
    905 		goto out_ok;
    906 	}
    907 
    908 	/* If replacing an existing entry, ensure we can do the operation. */
    909 	if (tvp != NULL) {
    910 		KASSERT(tnode != NULL);
    911 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
    912 			if (tnode->tn_size > 0) {
    913 				error = ENOTEMPTY;
    914 				goto out;
    915 			}
    916 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
    917 			error = ENOTDIR;
    918 			goto out;
    919 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
    920 			error = EISDIR;
    921 			goto out;
    922 		} else {
    923 			KASSERT(fnode->tn_type != VDIR &&
    924 			        tnode->tn_type != VDIR);
    925 		}
    926 	}
    927 
    928 	/* If the node is being moved to another directory, we have to do
    929 	 * the move. */
    930 	if (fdnode != tdnode) {
    931 		/* In case we are moving a directory, we have to adjust its
    932 		 * parent to point to the new parent. */
    933 		if (de->td_node->tn_type == VDIR) {
    934 			struct tmpfs_node *n;
    935 
    936 			/* Ensure the target directory is not a child of the
    937 			 * directory being moved.  Otherwise, we'd end up
    938 			 * with stale nodes. */
    939 			n = tdnode;
    940 			while (n != n->tn_spec.tn_dir.tn_parent) {
    941 				if (n == fnode) {
    942 					error = EINVAL;
    943 					goto out;
    944 				}
    945 				n = n->tn_spec.tn_dir.tn_parent;
    946 			}
    947 
    948 			/* Adjust the parent pointer. */
    949 			TMPFS_VALIDATE_DIR(fnode);
    950 			de->td_node->tn_spec.tn_dir.tn_parent = tdnode;
    951 
    952 			/* As a result of changing the target of the '..'
    953 			 * entry, the link count of the source and target
    954 			 * directories has to be adjusted. */
    955 			fdnode->tn_links--;
    956 			tdnode->tn_links++;
    957 		}
    958 
    959 		/* Do the move: just remove the entry from the source directory
    960 		 * and insert it into the target one. */
    961 		tmpfs_dir_detach(fdvp, de);
    962 		tmpfs_dir_attach(tdvp, de);
    963 
    964 		/* Notify listeners of fdvp about the change in the directory.
    965 		 * We can do it at this point because we aren't touching fdvp
    966 		 * any more below. */
    967 		VN_KNOTE(fdvp, NOTE_WRITE);
    968 	}
    969 
    970 	/* If we are overwriting an entry, we have to remove the old one
    971 	 * from the target directory. */
    972 	if (tvp != NULL) {
    973 		KASSERT(tnode != NULL);
    974 
    975 		/* Remove the old entry from the target directory.
    976 		 * Note! This relies on tmpfs_dir_attach() putting the new
    977 		 * node on the end of the target's node list. */
    978 		de2 = tmpfs_dir_lookup(tdnode, tcnp);
    979 		KASSERT(de2 != NULL);
    980 		KASSERT(de2->td_node == tnode);
    981 		tmpfs_dir_detach(tdvp, de2);
    982 
    983 		/* Free the directory entry we just deleted.  Note that the
    984 		 * node referred by it will not be removed until the vnode is
    985 		 * really reclaimed. */
    986 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de2, true);
    987 	}
    988 
    989 	/* If the name has changed, we need to make it effective by changing
    990 	 * it in the directory entry. */
    991 	if (newname != NULL) {
    992 		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
    993 		KASSERT(tcnp->cn_namelen < 0xffff);
    994 
    995 		tmpfs_strname_free(tmp, de->td_name, de->td_namelen);
    996 		de->td_namelen = (uint16_t)namelen;
    997 		memcpy(newname, tcnp->cn_nameptr, namelen);
    998 		de->td_name = newname;
    999 		newname = NULL;
   1000 
   1001 		fnode->tn_status |= TMPFS_NODE_CHANGED;
   1002 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
   1003 	}
   1004  out_ok:
   1005 	/* Notify listeners of tdvp about the change in the directory (either
   1006 	 * because a new entry was added or because one was removed) and
   1007 	 * listeners of fvp about the rename. */
   1008 	VN_KNOTE(tdvp, NOTE_WRITE);
   1009 	VN_KNOTE(fvp, NOTE_RENAME);
   1010 
   1011 	error = 0;
   1012 
   1013  out:
   1014 	if (fdnode != tdnode)
   1015 		VOP_UNLOCK(fdvp);
   1016 
   1017  out_unlocked:
   1018 	/* Release target nodes. */
   1019 	if (tdvp == tvp)
   1020 		vrele(tdvp);
   1021 	else
   1022 		vput(tdvp);
   1023 	if (tvp != NULL)
   1024 		vput(tvp);
   1025 
   1026 	/* Release source nodes. */
   1027 	vrele(fdvp);
   1028 	vrele(fvp);
   1029 
   1030 	if (newname != NULL) {
   1031 		tmpfs_strname_free(tmp, newname, namelen);
   1032 	}
   1033 	return error;
   1034 }
   1035 
   1036 /* --------------------------------------------------------------------- */
   1037 
   1038 int
   1039 tmpfs_mkdir(void *v)
   1040 {
   1041 	struct vnode *dvp = ((struct vop_mkdir_args *)v)->a_dvp;
   1042 	struct vnode **vpp = ((struct vop_mkdir_args *)v)->a_vpp;
   1043 	struct componentname *cnp = ((struct vop_mkdir_args *)v)->a_cnp;
   1044 	struct vattr *vap = ((struct vop_mkdir_args *)v)->a_vap;
   1045 
   1046 	KASSERT(vap->va_type == VDIR);
   1047 
   1048 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
   1049 }
   1050 
   1051 /* --------------------------------------------------------------------- */
   1052 
   1053 int
   1054 tmpfs_rmdir(void *v)
   1055 {
   1056 	struct vnode *dvp = ((struct vop_rmdir_args *)v)->a_dvp;
   1057 	struct vnode *vp = ((struct vop_rmdir_args *)v)->a_vp;
   1058 	struct componentname *cnp = ((struct vop_rmdir_args *)v)->a_cnp;
   1059 
   1060 	int error;
   1061 	struct tmpfs_dirent *de;
   1062 	struct tmpfs_mount *tmp;
   1063 	struct tmpfs_node *dnode;
   1064 	struct tmpfs_node *node;
   1065 
   1066 	KASSERT(VOP_ISLOCKED(dvp));
   1067 	KASSERT(VOP_ISLOCKED(vp));
   1068 
   1069 	tmp = VFS_TO_TMPFS(dvp->v_mount);
   1070 	dnode = VP_TO_TMPFS_DIR(dvp);
   1071 	node = VP_TO_TMPFS_DIR(vp);
   1072 	error = 0;
   1073 
   1074 	/* Directories with more than two entries ('.' and '..') cannot be
   1075 	 * removed. */
   1076 	if (node->tn_size > 0) {
   1077 		error = ENOTEMPTY;
   1078 		goto out;
   1079 	}
   1080 
   1081 	/* This invariant holds only if we are not trying to remove "..".
   1082 	 * We checked for that above so this is safe now. */
   1083 	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
   1084 
   1085 	/* Get the directory entry associated with node (vp). */
   1086 	de = tmpfs_dir_lookup(dnode, cnp);
   1087 	KASSERT(de);
   1088 	KASSERT(de->td_node == node);
   1089 
   1090 	/* Check flags to see if we are allowed to remove the directory. */
   1091 	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
   1092 		error = EPERM;
   1093 		goto out;
   1094 	}
   1095 
   1096 	/* Detach the directory entry from the directory (dnode). */
   1097 	tmpfs_dir_detach(dvp, de);
   1098 
   1099 	node->tn_links--;
   1100 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
   1101 	    TMPFS_NODE_MODIFIED;
   1102 	node->tn_spec.tn_dir.tn_parent->tn_links--;
   1103 	node->tn_spec.tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \
   1104 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
   1105 
   1106 	/* Release the parent. */
   1107 	cache_purge(dvp); /* XXX Is this needed? */
   1108 
   1109 	/* Free the directory entry we just deleted.  Note that the node
   1110 	 * referred by it will not be removed until the vnode is really
   1111 	 * reclaimed. */
   1112 	tmpfs_free_dirent(tmp, de, true);
   1113 
   1114 	KASSERT(node->tn_links == 0);
   1115  out:
   1116 	/* Release the nodes. */
   1117 	vput(dvp);
   1118 	vput(vp);
   1119 	PNBUF_PUT(cnp->cn_pnbuf);
   1120 
   1121 	return error;
   1122 }
   1123 
   1124 /* --------------------------------------------------------------------- */
   1125 
   1126 int
   1127 tmpfs_symlink(void *v)
   1128 {
   1129 	struct vnode *dvp = ((struct vop_symlink_args *)v)->a_dvp;
   1130 	struct vnode **vpp = ((struct vop_symlink_args *)v)->a_vpp;
   1131 	struct componentname *cnp = ((struct vop_symlink_args *)v)->a_cnp;
   1132 	struct vattr *vap = ((struct vop_symlink_args *)v)->a_vap;
   1133 	char *target = ((struct vop_symlink_args *)v)->a_target;
   1134 
   1135 	KASSERT(vap->va_type == VLNK);
   1136 
   1137 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
   1138 }
   1139 
   1140 /* --------------------------------------------------------------------- */
   1141 
   1142 int
   1143 tmpfs_readdir(void *v)
   1144 {
   1145 	struct vnode *vp = ((struct vop_readdir_args *)v)->a_vp;
   1146 	struct uio *uio = ((struct vop_readdir_args *)v)->a_uio;
   1147 	int *eofflag = ((struct vop_readdir_args *)v)->a_eofflag;
   1148 	off_t **cookies = ((struct vop_readdir_args *)v)->a_cookies;
   1149 	int *ncookies = ((struct vop_readdir_args *)v)->a_ncookies;
   1150 
   1151 	int error;
   1152 	off_t startoff;
   1153 	off_t cnt;
   1154 	struct tmpfs_node *node;
   1155 
   1156 	KASSERT(VOP_ISLOCKED(vp));
   1157 
   1158 	/* This operation only makes sense on directory nodes. */
   1159 	if (vp->v_type != VDIR) {
   1160 		error = ENOTDIR;
   1161 		goto out;
   1162 	}
   1163 
   1164 	node = VP_TO_TMPFS_DIR(vp);
   1165 
   1166 	startoff = uio->uio_offset;
   1167 
   1168 	cnt = 0;
   1169 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
   1170 		error = tmpfs_dir_getdotdent(node, uio);
   1171 		if (error == -1) {
   1172 			error = 0;
   1173 			goto outok;
   1174 		} else if (error != 0)
   1175 			goto outok;
   1176 		cnt++;
   1177 	}
   1178 
   1179 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
   1180 		error = tmpfs_dir_getdotdotdent(node, uio);
   1181 		if (error == -1) {
   1182 			error = 0;
   1183 			goto outok;
   1184 		} else if (error != 0)
   1185 			goto outok;
   1186 		cnt++;
   1187 	}
   1188 
   1189 	error = tmpfs_dir_getdents(node, uio, &cnt);
   1190 	if (error == -1)
   1191 		error = 0;
   1192 	KASSERT(error >= 0);
   1193 
   1194 outok:
   1195 	/* This label assumes that startoff has been
   1196 	 * initialized.  If the compiler didn't spit out warnings, we'd
   1197 	 * simply make this one be 'out' and drop 'outok'. */
   1198 
   1199 	if (eofflag != NULL)
   1200 		*eofflag =
   1201 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
   1202 
   1203 	/* Update NFS-related variables. */
   1204 	if (error == 0 && cookies != NULL && ncookies != NULL) {
   1205 		off_t i;
   1206 		off_t off = startoff;
   1207 		struct tmpfs_dirent *de = NULL;
   1208 
   1209 		*ncookies = cnt;
   1210 		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
   1211 
   1212 		for (i = 0; i < cnt; i++) {
   1213 			KASSERT(off != TMPFS_DIRCOOKIE_EOF);
   1214 			if (off == TMPFS_DIRCOOKIE_DOT) {
   1215 				off = TMPFS_DIRCOOKIE_DOTDOT;
   1216 			} else {
   1217 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
   1218 					de = TAILQ_FIRST(&node->tn_spec.
   1219 					    tn_dir.tn_dir);
   1220 				} else if (de != NULL) {
   1221 					de = TAILQ_NEXT(de, td_entries);
   1222 				} else {
   1223 					de = tmpfs_dir_lookupbycookie(node,
   1224 					    off);
   1225 					KASSERT(de != NULL);
   1226 					de = TAILQ_NEXT(de, td_entries);
   1227 				}
   1228 				if (de == NULL) {
   1229 					off = TMPFS_DIRCOOKIE_EOF;
   1230 				} else {
   1231 					off = tmpfs_dircookie(de);
   1232 				}
   1233 			}
   1234 
   1235 			(*cookies)[i] = off;
   1236 		}
   1237 		KASSERT(uio->uio_offset == off);
   1238 	}
   1239 
   1240 out:
   1241 	KASSERT(VOP_ISLOCKED(vp));
   1242 
   1243 	return error;
   1244 }
   1245 
   1246 /* --------------------------------------------------------------------- */
   1247 
   1248 int
   1249 tmpfs_readlink(void *v)
   1250 {
   1251 	struct vnode *vp = ((struct vop_readlink_args *)v)->a_vp;
   1252 	struct uio *uio = ((struct vop_readlink_args *)v)->a_uio;
   1253 
   1254 	int error;
   1255 	struct tmpfs_node *node;
   1256 
   1257 	KASSERT(VOP_ISLOCKED(vp));
   1258 	KASSERT(uio->uio_offset == 0);
   1259 	KASSERT(vp->v_type == VLNK);
   1260 
   1261 	node = VP_TO_TMPFS_NODE(vp);
   1262 
   1263 	error = uiomove(node->tn_spec.tn_lnk.tn_link,
   1264 	    MIN(node->tn_size, uio->uio_resid), uio);
   1265 	node->tn_status |= TMPFS_NODE_ACCESSED;
   1266 
   1267 	KASSERT(VOP_ISLOCKED(vp));
   1268 
   1269 	return error;
   1270 }
   1271 
   1272 /* --------------------------------------------------------------------- */
   1273 
   1274 int
   1275 tmpfs_inactive(void *v)
   1276 {
   1277 	struct vnode *vp = ((struct vop_inactive_args *)v)->a_vp;
   1278 
   1279 	struct tmpfs_node *node;
   1280 
   1281 	KASSERT(VOP_ISLOCKED(vp));
   1282 
   1283 	node = VP_TO_TMPFS_NODE(vp);
   1284 	*((struct vop_inactive_args *)v)->a_recycle = (node->tn_links == 0);
   1285 	VOP_UNLOCK(vp);
   1286 
   1287 	return 0;
   1288 }
   1289 
   1290 /* --------------------------------------------------------------------- */
   1291 
   1292 int
   1293 tmpfs_reclaim(void *v)
   1294 {
   1295 	struct vnode *vp = ((struct vop_reclaim_args *)v)->a_vp;
   1296 
   1297 	struct tmpfs_mount *tmp;
   1298 	struct tmpfs_node *node;
   1299 
   1300 	node = VP_TO_TMPFS_NODE(vp);
   1301 	tmp = VFS_TO_TMPFS(vp->v_mount);
   1302 
   1303 	cache_purge(vp);
   1304 	tmpfs_free_vp(vp);
   1305 
   1306 	/* If the node referenced by this vnode was deleted by the user,
   1307 	 * we must free its associated data structures (now that the vnode
   1308 	 * is being reclaimed). */
   1309 	if (node->tn_links == 0)
   1310 		tmpfs_free_node(tmp, node);
   1311 
   1312 	KASSERT(vp->v_data == NULL);
   1313 
   1314 	return 0;
   1315 }
   1316 
   1317 /* --------------------------------------------------------------------- */
   1318 
   1319 int
   1320 tmpfs_print(void *v)
   1321 {
   1322 	struct vnode *vp = ((struct vop_print_args *)v)->a_vp;
   1323 
   1324 	struct tmpfs_node *node;
   1325 
   1326 	node = VP_TO_TMPFS_NODE(vp);
   1327 
   1328 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
   1329 	    node, node->tn_flags, node->tn_links);
   1330 	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
   1331 	    ", status 0x%x",
   1332 	    node->tn_mode, node->tn_uid, node->tn_gid,
   1333 	    (uintmax_t)node->tn_size, node->tn_status);
   1334 	if (vp->v_type == VFIFO)
   1335 		VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
   1336 	printf("\n");
   1337 
   1338 	return 0;
   1339 }
   1340 
   1341 /* --------------------------------------------------------------------- */
   1342 
   1343 int
   1344 tmpfs_pathconf(void *v)
   1345 {
   1346 	int name = ((struct vop_pathconf_args *)v)->a_name;
   1347 	register_t *retval = ((struct vop_pathconf_args *)v)->a_retval;
   1348 
   1349 	int error;
   1350 
   1351 	error = 0;
   1352 
   1353 	switch (name) {
   1354 	case _PC_LINK_MAX:
   1355 		*retval = LINK_MAX;
   1356 		break;
   1357 
   1358 	case _PC_NAME_MAX:
   1359 		*retval = NAME_MAX;
   1360 		break;
   1361 
   1362 	case _PC_PATH_MAX:
   1363 		*retval = PATH_MAX;
   1364 		break;
   1365 
   1366 	case _PC_PIPE_BUF:
   1367 		*retval = PIPE_BUF;
   1368 		break;
   1369 
   1370 	case _PC_CHOWN_RESTRICTED:
   1371 		*retval = 1;
   1372 		break;
   1373 
   1374 	case _PC_NO_TRUNC:
   1375 		*retval = 1;
   1376 		break;
   1377 
   1378 	case _PC_SYNC_IO:
   1379 		*retval = 1;
   1380 		break;
   1381 
   1382 	case _PC_FILESIZEBITS:
   1383 		*retval = 0; /* XXX Don't know which value should I return. */
   1384 		break;
   1385 
   1386 	default:
   1387 		error = EINVAL;
   1388 	}
   1389 
   1390 	return error;
   1391 }
   1392 
   1393 /* --------------------------------------------------------------------- */
   1394 
   1395 int
   1396 tmpfs_advlock(void *v)
   1397 {
   1398 	struct vnode *vp = ((struct vop_advlock_args *)v)->a_vp;
   1399 
   1400 	struct tmpfs_node *node;
   1401 
   1402 	node = VP_TO_TMPFS_NODE(vp);
   1403 
   1404 	return lf_advlock(v, &node->tn_lockf, node->tn_size);
   1405 }
   1406 
   1407 /* --------------------------------------------------------------------- */
   1408 
   1409 int
   1410 tmpfs_getpages(void *v)
   1411 {
   1412 	struct vnode *vp = ((struct vop_getpages_args *)v)->a_vp;
   1413 	voff_t offset = ((struct vop_getpages_args *)v)->a_offset;
   1414 	struct vm_page **m = ((struct vop_getpages_args *)v)->a_m;
   1415 	int *count = ((struct vop_getpages_args *)v)->a_count;
   1416 	int centeridx = ((struct vop_getpages_args *)v)->a_centeridx;
   1417 	vm_prot_t access_type = ((struct vop_getpages_args *)v)->a_access_type;
   1418 	int advice = ((struct vop_getpages_args *)v)->a_advice;
   1419 	int flags = ((struct vop_getpages_args *)v)->a_flags;
   1420 
   1421 	int error;
   1422 	int i;
   1423 	struct tmpfs_node *node;
   1424 	struct uvm_object *uobj;
   1425 	int npages = *count;
   1426 
   1427 	KASSERT(vp->v_type == VREG);
   1428 	KASSERT(mutex_owned(&vp->v_interlock));
   1429 
   1430 	node = VP_TO_TMPFS_NODE(vp);
   1431 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1432 
   1433 	/* We currently don't rely on PGO_PASTEOF. */
   1434 
   1435 	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
   1436 		if ((flags & PGO_LOCKED) == 0)
   1437 			mutex_exit(&vp->v_interlock);
   1438 		return EINVAL;
   1439 	}
   1440 
   1441 	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
   1442 		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
   1443 	}
   1444 
   1445 	if ((flags & PGO_LOCKED) != 0)
   1446 		return EBUSY;
   1447 
   1448 	if ((flags & PGO_NOTIMESTAMP) == 0) {
   1449 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
   1450 			node->tn_status |= TMPFS_NODE_ACCESSED;
   1451 
   1452 		if ((access_type & VM_PROT_WRITE) != 0)
   1453 			node->tn_status |= TMPFS_NODE_MODIFIED;
   1454 	}
   1455 
   1456 	mutex_exit(&vp->v_interlock);
   1457 
   1458 	/*
   1459 	 * Make sure that the array on which we will store the
   1460 	 * gotten pages is clean.  Otherwise uao_get (pointed to by
   1461 	 * the pgo_get below) gets confused and does not return the
   1462 	 * appropriate pages.
   1463 	 *
   1464 	 * XXX This shall be revisited when kern/32166 is addressed
   1465 	 * because the loop to clean m[i] will most likely be redundant
   1466 	 * as well as the PGO_ALLPAGES flag.
   1467 	 */
   1468 	if (m != NULL)
   1469 		for (i = 0; i < npages; i++)
   1470 			m[i] = NULL;
   1471 	mutex_enter(&uobj->vmobjlock);
   1472 	error = (*uobj->pgops->pgo_get)(uobj, offset, m, &npages, centeridx,
   1473 	    access_type, advice, flags | PGO_ALLPAGES);
   1474 #if defined(DEBUG)
   1475 	{
   1476 		/* Make sure that all the pages we return are valid. */
   1477 		int dbgi;
   1478 		if (error == 0 && m != NULL)
   1479 			for (dbgi = 0; dbgi < npages; dbgi++)
   1480 				KASSERT(m[dbgi] != NULL);
   1481 	}
   1482 #endif
   1483 
   1484 	return error;
   1485 }
   1486 
   1487 /* --------------------------------------------------------------------- */
   1488 
   1489 int
   1490 tmpfs_putpages(void *v)
   1491 {
   1492 	struct vnode *vp = ((struct vop_putpages_args *)v)->a_vp;
   1493 	voff_t offlo = ((struct vop_putpages_args *)v)->a_offlo;
   1494 	voff_t offhi = ((struct vop_putpages_args *)v)->a_offhi;
   1495 	int flags = ((struct vop_putpages_args *)v)->a_flags;
   1496 
   1497 	int error;
   1498 	struct tmpfs_node *node;
   1499 	struct uvm_object *uobj;
   1500 
   1501 	KASSERT(mutex_owned(&vp->v_interlock));
   1502 
   1503 	node = VP_TO_TMPFS_NODE(vp);
   1504 
   1505 	if (vp->v_type != VREG) {
   1506 		mutex_exit(&vp->v_interlock);
   1507 		return 0;
   1508 	}
   1509 
   1510 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1511 	mutex_exit(&vp->v_interlock);
   1512 
   1513 	mutex_enter(&uobj->vmobjlock);
   1514 	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
   1515 
   1516 	/* XXX mtime */
   1517 
   1518 	return error;
   1519 }
   1520