Home | History | Annotate | Line # | Download | only in tmpfs
tmpfs_vnops.c revision 1.49
      1 /*	$NetBSD: tmpfs_vnops.c,v 1.49 2008/02/06 11:22:12 jmmv 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  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *        This product includes software developed by the NetBSD
     22  *        Foundation, Inc. and its contributors.
     23  * 4. Neither the name of The NetBSD Foundation nor the names of its
     24  *    contributors may be used to endorse or promote products derived
     25  *    from this software without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     37  * POSSIBILITY OF SUCH DAMAGE.
     38  */
     39 
     40 /*
     41  * tmpfs vnode interface.
     42  */
     43 
     44 #include <sys/cdefs.h>
     45 __KERNEL_RCSID(0, "$NetBSD: tmpfs_vnops.c,v 1.49 2008/02/06 11:22:12 jmmv Exp $");
     46 
     47 #include <sys/param.h>
     48 #include <sys/dirent.h>
     49 #include <sys/fcntl.h>
     50 #include <sys/event.h>
     51 #include <sys/malloc.h>
     52 #include <sys/namei.h>
     53 #include <sys/proc.h>
     54 #include <sys/stat.h>
     55 #include <sys/uio.h>
     56 #include <sys/unistd.h>
     57 #include <sys/vnode.h>
     58 #include <sys/lockf.h>
     59 #include <sys/kauth.h>
     60 
     61 #include <uvm/uvm.h>
     62 
     63 #include <miscfs/fifofs/fifo.h>
     64 #include <fs/tmpfs/tmpfs_vnops.h>
     65 #include <fs/tmpfs/tmpfs.h>
     66 
     67 /* --------------------------------------------------------------------- */
     68 
     69 /*
     70  * vnode operations vector used for files stored in a tmpfs file system.
     71  */
     72 int (**tmpfs_vnodeop_p)(void *);
     73 const struct vnodeopv_entry_desc tmpfs_vnodeop_entries[] = {
     74 	{ &vop_default_desc,		vn_default_error },
     75 	{ &vop_lookup_desc,		tmpfs_lookup },
     76 	{ &vop_create_desc,		tmpfs_create },
     77 	{ &vop_mknod_desc,		tmpfs_mknod },
     78 	{ &vop_open_desc,		tmpfs_open },
     79 	{ &vop_close_desc,		tmpfs_close },
     80 	{ &vop_access_desc,		tmpfs_access },
     81 	{ &vop_getattr_desc,		tmpfs_getattr },
     82 	{ &vop_setattr_desc,		tmpfs_setattr },
     83 	{ &vop_read_desc,		tmpfs_read },
     84 	{ &vop_write_desc,		tmpfs_write },
     85 	{ &vop_ioctl_desc,		tmpfs_ioctl },
     86 	{ &vop_fcntl_desc,		tmpfs_fcntl },
     87 	{ &vop_poll_desc,		tmpfs_poll },
     88 	{ &vop_kqfilter_desc,		tmpfs_kqfilter },
     89 	{ &vop_revoke_desc,		tmpfs_revoke },
     90 	{ &vop_mmap_desc,		tmpfs_mmap },
     91 	{ &vop_fsync_desc,		tmpfs_fsync },
     92 	{ &vop_seek_desc,		tmpfs_seek },
     93 	{ &vop_remove_desc,		tmpfs_remove },
     94 	{ &vop_link_desc,		tmpfs_link },
     95 	{ &vop_rename_desc,		tmpfs_rename },
     96 	{ &vop_mkdir_desc,		tmpfs_mkdir },
     97 	{ &vop_rmdir_desc,		tmpfs_rmdir },
     98 	{ &vop_symlink_desc,		tmpfs_symlink },
     99 	{ &vop_readdir_desc,		tmpfs_readdir },
    100 	{ &vop_readlink_desc,		tmpfs_readlink },
    101 	{ &vop_abortop_desc,		tmpfs_abortop },
    102 	{ &vop_inactive_desc,		tmpfs_inactive },
    103 	{ &vop_reclaim_desc,		tmpfs_reclaim },
    104 	{ &vop_lock_desc,		tmpfs_lock },
    105 	{ &vop_unlock_desc,		tmpfs_unlock },
    106 	{ &vop_bmap_desc,		tmpfs_bmap },
    107 	{ &vop_strategy_desc,		tmpfs_strategy },
    108 	{ &vop_print_desc,		tmpfs_print },
    109 	{ &vop_pathconf_desc,		tmpfs_pathconf },
    110 	{ &vop_islocked_desc,		tmpfs_islocked },
    111 	{ &vop_advlock_desc,		tmpfs_advlock },
    112 	{ &vop_bwrite_desc,		tmpfs_bwrite },
    113 	{ &vop_getpages_desc,		tmpfs_getpages },
    114 	{ &vop_putpages_desc,		tmpfs_putpages },
    115 	{ NULL, NULL }
    116 };
    117 const struct vnodeopv_desc tmpfs_vnodeop_opv_desc =
    118 	{ &tmpfs_vnodeop_p, tmpfs_vnodeop_entries };
    119 
    120 /* --------------------------------------------------------------------- */
    121 
    122 int
    123 tmpfs_lookup(void *v)
    124 {
    125 	struct vnode *dvp = ((struct vop_lookup_args *)v)->a_dvp;
    126 	struct vnode **vpp = ((struct vop_lookup_args *)v)->a_vpp;
    127 	struct componentname *cnp = ((struct vop_lookup_args *)v)->a_cnp;
    128 
    129 	int error;
    130 	struct tmpfs_dirent *de;
    131 	struct tmpfs_node *dnode;
    132 
    133 	KASSERT(VOP_ISLOCKED(dvp));
    134 
    135 	dnode = VP_TO_TMPFS_DIR(dvp);
    136 	*vpp = NULL;
    137 
    138 	/* Check accessibility of requested node as a first step. */
    139 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
    140 	if (error != 0)
    141 		goto out;
    142 
    143 	/* If requesting the last path component on a read-only file system
    144 	 * with a write operation, deny it. */
    145 	if ((cnp->cn_flags & ISLASTCN) &&
    146 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
    147 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    148 		error = EROFS;
    149 		goto out;
    150 	}
    151 
    152 	/* Avoid doing a linear scan of the directory if the requested
    153 	 * directory/name couple is already in the cache. */
    154 	error = cache_lookup(dvp, vpp, cnp);
    155 	if (error >= 0)
    156 		goto out;
    157 
    158 	/* We cannot be requesting the parent directory of the root node. */
    159 	KASSERT(IMPLIES(dnode->tn_type == VDIR &&
    160 	    dnode->tn_spec.tn_dir.tn_parent == dnode,
    161 	    !(cnp->cn_flags & ISDOTDOT)));
    162 
    163 	if (cnp->cn_flags & ISDOTDOT) {
    164 		VOP_UNLOCK(dvp, 0);
    165 
    166 		/* Allocate a new vnode on the matching entry. */
    167 		error = tmpfs_alloc_vp(dvp->v_mount,
    168 		    dnode->tn_spec.tn_dir.tn_parent, vpp);
    169 
    170 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
    171 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
    172 		VREF(dvp);
    173 		*vpp = dvp;
    174 		error = 0;
    175 	} else {
    176 		de = tmpfs_dir_lookup(dnode, cnp);
    177 		if (de == NULL) {
    178 			/* The entry was not found in the directory.
    179 			 * This is OK iff we are creating or renaming an
    180 			 * entry and are working on the last component of
    181 			 * the path name. */
    182 			if ((cnp->cn_flags & ISLASTCN) &&
    183 			    (cnp->cn_nameiop == CREATE || \
    184 			    cnp->cn_nameiop == RENAME)) {
    185 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
    186 				if (error != 0)
    187 					goto out;
    188 
    189 				/* Keep the component name in the buffer for
    190 				 * future uses. */
    191 				cnp->cn_flags |= SAVENAME;
    192 
    193 				error = EJUSTRETURN;
    194 			} else
    195 				error = ENOENT;
    196 		} else {
    197 			struct tmpfs_node *tnode;
    198 
    199 			/* The entry was found, so get its associated
    200 			 * tmpfs_node. */
    201 			tnode = de->td_node;
    202 
    203 			/* If we are not at the last path component and
    204 			 * found a non-directory or non-link entry (which
    205 			 * may itself be pointing to a directory), raise
    206 			 * an error. */
    207 			if ((tnode->tn_type != VDIR &&
    208 			    tnode->tn_type != VLNK) &&
    209 			    !(cnp->cn_flags & ISLASTCN)) {
    210 				error = ENOTDIR;
    211 				goto out;
    212 			}
    213 
    214 			/* If we are deleting or renaming the entry, keep
    215 			 * track of its tmpfs_dirent so that it can be
    216 			 * easily deleted later. */
    217 			if ((cnp->cn_flags & ISLASTCN) &&
    218 			    (cnp->cn_nameiop == DELETE ||
    219 			    cnp->cn_nameiop == RENAME)) {
    220 				if ((dnode->tn_mode & S_ISTXT) != 0 &&
    221 				    kauth_authorize_generic(cnp->cn_cred,
    222 				     KAUTH_GENERIC_ISSUSER, NULL) != 0 &&
    223 				    kauth_cred_geteuid(cnp->cn_cred) != dnode->tn_uid &&
    224 				    kauth_cred_geteuid(cnp->cn_cred) != tnode->tn_uid)
    225 					return EPERM;
    226 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
    227 				if (error != 0)
    228 					goto out;
    229 			} else
    230 				de = NULL;
    231 
    232 			/* Allocate a new vnode on the matching entry. */
    233 			error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp);
    234 		}
    235 	}
    236 
    237 	/* Store the result of this lookup in the cache.  Avoid this if the
    238 	 * request was for creation, as it does not improve timings on
    239 	 * emprical tests. */
    240 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
    241 	    (cnp->cn_flags & ISDOTDOT) == 0)
    242 		cache_enter(dvp, *vpp, cnp);
    243 
    244 out:
    245 	/* If there were no errors, *vpp cannot be null and it must be
    246 	 * locked. */
    247 	KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
    248 
    249 	/* dvp must always be locked. */
    250 	KASSERT(VOP_ISLOCKED(dvp));
    251 
    252 	return error;
    253 }
    254 
    255 /* --------------------------------------------------------------------- */
    256 
    257 int
    258 tmpfs_create(void *v)
    259 {
    260 	struct vnode *dvp = ((struct vop_create_args *)v)->a_dvp;
    261 	struct vnode **vpp = ((struct vop_create_args *)v)->a_vpp;
    262 	struct componentname *cnp = ((struct vop_create_args *)v)->a_cnp;
    263 	struct vattr *vap = ((struct vop_create_args *)v)->a_vap;
    264 
    265 	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
    266 
    267 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    268 }
    269 /* --------------------------------------------------------------------- */
    270 
    271 int
    272 tmpfs_mknod(void *v)
    273 {
    274 	struct vnode *dvp = ((struct vop_mknod_args *)v)->a_dvp;
    275 	struct vnode **vpp = ((struct vop_mknod_args *)v)->a_vpp;
    276 	struct componentname *cnp = ((struct vop_mknod_args *)v)->a_cnp;
    277 	struct vattr *vap = ((struct vop_mknod_args *)v)->a_vap;
    278 
    279 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
    280 	    vap->va_type != VFIFO)
    281 		return EINVAL;
    282 
    283 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
    284 }
    285 
    286 /* --------------------------------------------------------------------- */
    287 
    288 int
    289 tmpfs_open(void *v)
    290 {
    291 	struct vnode *vp = ((struct vop_open_args *)v)->a_vp;
    292 	int mode = ((struct vop_open_args *)v)->a_mode;
    293 
    294 	int error;
    295 	struct tmpfs_node *node;
    296 
    297 	KASSERT(VOP_ISLOCKED(vp));
    298 
    299 	node = VP_TO_TMPFS_NODE(vp);
    300 
    301 	/* The file is still active but all its names have been removed
    302 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
    303 	 * it is about to die. */
    304 	if (node->tn_links < 1) {
    305 		error = ENOENT;
    306 		goto out;
    307 	}
    308 
    309 	/* If the file is marked append-only, deny write requests. */
    310 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
    311 		error = EPERM;
    312 	else
    313 		error = 0;
    314 
    315 out:
    316 	KASSERT(VOP_ISLOCKED(vp));
    317 
    318 	return error;
    319 }
    320 
    321 /* --------------------------------------------------------------------- */
    322 
    323 int
    324 tmpfs_close(void *v)
    325 {
    326 	struct vnode *vp = ((struct vop_close_args *)v)->a_vp;
    327 
    328 	struct tmpfs_node *node;
    329 
    330 	KASSERT(VOP_ISLOCKED(vp));
    331 
    332 	node = VP_TO_TMPFS_NODE(vp);
    333 
    334 	if (node->tn_links > 0) {
    335 		/* Update node times.  No need to do it if the node has
    336 		 * been deleted, because it will vanish after we return. */
    337 		tmpfs_update(vp, NULL, NULL, UPDATE_CLOSE);
    338 	}
    339 
    340 	return 0;
    341 }
    342 
    343 /* --------------------------------------------------------------------- */
    344 
    345 int
    346 tmpfs_access(void *v)
    347 {
    348 	struct vnode *vp = ((struct vop_access_args *)v)->a_vp;
    349 	int mode = ((struct vop_access_args *)v)->a_mode;
    350 	kauth_cred_t cred = ((struct vop_access_args *)v)->a_cred;
    351 
    352 	int error;
    353 	struct tmpfs_node *node;
    354 
    355 	KASSERT(VOP_ISLOCKED(vp));
    356 
    357 	node = VP_TO_TMPFS_NODE(vp);
    358 
    359 	switch (vp->v_type) {
    360 	case VDIR:
    361 		/* FALLTHROUGH */
    362 	case VLNK:
    363 		/* FALLTHROUGH */
    364 	case VREG:
    365 		if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
    366 			error = EROFS;
    367 			goto out;
    368 		}
    369 		break;
    370 
    371 	case VBLK:
    372 		/* FALLTHROUGH */
    373 	case VCHR:
    374 		/* FALLTHROUGH */
    375 	case VSOCK:
    376 		/* FALLTHROUGH */
    377 	case VFIFO:
    378 		break;
    379 
    380 	default:
    381 		error = EINVAL;
    382 		goto out;
    383 	}
    384 
    385 	if (mode & VWRITE && node->tn_flags & IMMUTABLE) {
    386 		error = EPERM;
    387 		goto out;
    388 	}
    389 
    390 	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
    391 	    node->tn_gid, mode, cred);
    392 
    393 out:
    394 	KASSERT(VOP_ISLOCKED(vp));
    395 
    396 	return error;
    397 }
    398 
    399 /* --------------------------------------------------------------------- */
    400 
    401 int
    402 tmpfs_getattr(void *v)
    403 {
    404 	struct vnode *vp = ((struct vop_getattr_args *)v)->a_vp;
    405 	struct vattr *vap = ((struct vop_getattr_args *)v)->a_vap;
    406 
    407 	struct tmpfs_node *node;
    408 
    409 	node = VP_TO_TMPFS_NODE(vp);
    410 
    411 	VATTR_NULL(vap);
    412 
    413 	tmpfs_itimes(vp, NULL, NULL);
    414 
    415 	vap->va_type = vp->v_type;
    416 	vap->va_mode = node->tn_mode;
    417 	vap->va_nlink = node->tn_links;
    418 	vap->va_uid = node->tn_uid;
    419 	vap->va_gid = node->tn_gid;
    420 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
    421 	vap->va_fileid = node->tn_id;
    422 	vap->va_size = node->tn_size;
    423 	vap->va_blocksize = PAGE_SIZE;
    424 	vap->va_atime = node->tn_atime;
    425 	vap->va_mtime = node->tn_mtime;
    426 	vap->va_ctime = node->tn_ctime;
    427 	vap->va_birthtime = node->tn_birthtime;
    428 	vap->va_gen = node->tn_gen;
    429 	vap->va_flags = node->tn_flags;
    430 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
    431 		node->tn_spec.tn_dev.tn_rdev : VNOVAL;
    432 	vap->va_bytes = round_page(node->tn_size);
    433 	vap->va_filerev = VNOVAL;
    434 	vap->va_vaflags = 0;
    435 	vap->va_spare = VNOVAL; /* XXX */
    436 
    437 	return 0;
    438 }
    439 
    440 /* --------------------------------------------------------------------- */
    441 
    442 /* XXX Should this operation be atomic?  I think it should, but code in
    443  * XXX other places (e.g., ufs) doesn't seem to be... */
    444 int
    445 tmpfs_setattr(void *v)
    446 {
    447 	struct vnode *vp = ((struct vop_setattr_args *)v)->a_vp;
    448 	struct vattr *vap = ((struct vop_setattr_args *)v)->a_vap;
    449 	kauth_cred_t cred = ((struct vop_setattr_args *)v)->a_cred;
    450 	struct lwp *l = curlwp;
    451 
    452 	int error;
    453 
    454 	KASSERT(VOP_ISLOCKED(vp));
    455 
    456 	error = 0;
    457 
    458 	/* Abort if any unsettable attribute is given. */
    459 	if (vap->va_type != VNON ||
    460 	    vap->va_nlink != VNOVAL ||
    461 	    vap->va_fsid != VNOVAL ||
    462 	    vap->va_fileid != VNOVAL ||
    463 	    vap->va_blocksize != VNOVAL ||
    464 	    vap->va_ctime.tv_sec != VNOVAL ||
    465 	    vap->va_ctime.tv_nsec != VNOVAL ||
    466 	    vap->va_birthtime.tv_sec != VNOVAL ||
    467 	    vap->va_birthtime.tv_nsec != VNOVAL ||
    468 	    vap->va_gen != VNOVAL ||
    469 	    vap->va_rdev != VNOVAL ||
    470 	    vap->va_bytes != VNOVAL)
    471 		error = EINVAL;
    472 
    473 	if (error == 0 && (vap->va_flags != VNOVAL))
    474 		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
    475 
    476 	if (error == 0 && (vap->va_size != VNOVAL))
    477 		error = tmpfs_chsize(vp, vap->va_size, cred, l);
    478 
    479 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
    480 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
    481 
    482 	if (error == 0 && (vap->va_mode != VNOVAL))
    483 		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
    484 
    485 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
    486 	    vap->va_atime.tv_nsec != VNOVAL) ||
    487 	    (vap->va_mtime.tv_sec != VNOVAL &&
    488 	    vap->va_mtime.tv_nsec != VNOVAL)))
    489 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
    490 		    vap->va_vaflags, cred, l);
    491 
    492 	/* Update the node times.  We give preference to the error codes
    493 	 * generated by this function rather than the ones that may arise
    494 	 * from tmpfs_update. */
    495 	tmpfs_update(vp, NULL, NULL, 0);
    496 
    497 	KASSERT(VOP_ISLOCKED(vp));
    498 
    499 	return error;
    500 }
    501 
    502 /* --------------------------------------------------------------------- */
    503 
    504 int
    505 tmpfs_read(void *v)
    506 {
    507 	struct vnode *vp = ((struct vop_read_args *)v)->a_vp;
    508 	struct uio *uio = ((struct vop_read_args *)v)->a_uio;
    509 
    510 	int error;
    511 	int flags;
    512 	struct tmpfs_node *node;
    513 	struct uvm_object *uobj;
    514 
    515 	KASSERT(VOP_ISLOCKED(vp));
    516 
    517 	node = VP_TO_TMPFS_NODE(vp);
    518 
    519 	if (vp->v_type != VREG) {
    520 		error = EISDIR;
    521 		goto out;
    522 	}
    523 
    524 	if (uio->uio_offset < 0) {
    525 		error = EINVAL;
    526 		goto out;
    527 	}
    528 
    529 	node->tn_status |= TMPFS_NODE_ACCESSED;
    530 
    531 	uobj = node->tn_spec.tn_reg.tn_aobj;
    532 	flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
    533 	error = 0;
    534 	while (error == 0 && uio->uio_resid > 0) {
    535 		vsize_t len;
    536 		void *win;
    537 
    538 		if (node->tn_size <= uio->uio_offset)
    539 			break;
    540 
    541 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    542 		if (len == 0)
    543 			break;
    544 
    545 		win = ubc_alloc(uobj, uio->uio_offset, &len, UVM_ADV_NORMAL,
    546 		    UBC_READ);
    547 		error = uiomove(win, len, uio);
    548 		ubc_release(win, flags);
    549 	}
    550 
    551 out:
    552 	KASSERT(VOP_ISLOCKED(vp));
    553 
    554 	return error;
    555 }
    556 
    557 /* --------------------------------------------------------------------- */
    558 
    559 int
    560 tmpfs_write(void *v)
    561 {
    562 	struct vnode *vp = ((struct vop_write_args *)v)->a_vp;
    563 	struct uio *uio = ((struct vop_write_args *)v)->a_uio;
    564 	int ioflag = ((struct vop_write_args *)v)->a_ioflag;
    565 
    566 	bool extended;
    567 	int error;
    568 	int flags;
    569 	off_t oldsize;
    570 	struct tmpfs_node *node;
    571 	struct uvm_object *uobj;
    572 
    573 	KASSERT(VOP_ISLOCKED(vp));
    574 
    575 	node = VP_TO_TMPFS_NODE(vp);
    576 	oldsize = node->tn_size;
    577 
    578 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
    579 		error = EINVAL;
    580 		goto out;
    581 	}
    582 
    583 	if (uio->uio_resid == 0) {
    584 		error = 0;
    585 		goto out;
    586 	}
    587 
    588 	if (ioflag & IO_APPEND)
    589 		uio->uio_offset = node->tn_size;
    590 
    591 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
    592 	if (extended) {
    593 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
    594 		if (error != 0)
    595 			goto out;
    596 	}
    597 
    598 	uobj = node->tn_spec.tn_reg.tn_aobj;
    599 	flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
    600 	error = 0;
    601 	while (error == 0 && uio->uio_resid > 0) {
    602 		vsize_t len;
    603 		void *win;
    604 
    605 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
    606 		if (len == 0)
    607 			break;
    608 
    609 		win = ubc_alloc(uobj, uio->uio_offset, &len, UVM_ADV_NORMAL,
    610 		    UBC_WRITE);
    611 		error = uiomove(win, len, uio);
    612 		ubc_release(win, flags);
    613 	}
    614 
    615 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
    616 	    (extended ? TMPFS_NODE_CHANGED : 0);
    617 
    618 	if (error != 0)
    619 		(void)tmpfs_reg_resize(vp, oldsize);
    620 
    621 	VN_KNOTE(vp, NOTE_WRITE);
    622 
    623 out:
    624 	KASSERT(VOP_ISLOCKED(vp));
    625 	KASSERT(IMPLIES(error == 0, uio->uio_resid == 0));
    626 	KASSERT(IMPLIES(error != 0, oldsize == node->tn_size));
    627 
    628 	return error;
    629 }
    630 
    631 /* --------------------------------------------------------------------- */
    632 
    633 int
    634 tmpfs_fsync(void *v)
    635 {
    636 	struct vnode *vp = ((struct vop_fsync_args *)v)->a_vp;
    637 
    638 	KASSERT(VOP_ISLOCKED(vp));
    639 
    640 	tmpfs_update(vp, NULL, NULL, 0);
    641 
    642 	return 0;
    643 }
    644 
    645 /* --------------------------------------------------------------------- */
    646 
    647 int
    648 tmpfs_remove(void *v)
    649 {
    650 	struct vnode *dvp = ((struct vop_remove_args *)v)->a_dvp;
    651 	struct vnode *vp = ((struct vop_remove_args *)v)->a_vp;
    652 	struct componentname *cnp = (((struct vop_remove_args *)v)->a_cnp);
    653 
    654 	int error;
    655 	struct tmpfs_dirent *de;
    656 	struct tmpfs_mount *tmp;
    657 	struct tmpfs_node *dnode;
    658 	struct tmpfs_node *node;
    659 
    660 	KASSERT(VOP_ISLOCKED(dvp));
    661 	KASSERT(VOP_ISLOCKED(vp));
    662 
    663 	if (vp->v_type == VDIR) {
    664 		error = EPERM;
    665 		goto out;
    666 	}
    667 
    668 	dnode = VP_TO_TMPFS_DIR(dvp);
    669 	node = VP_TO_TMPFS_NODE(vp);
    670 	tmp = VFS_TO_TMPFS(vp->v_mount);
    671 	de = tmpfs_dir_lookup(dnode, cnp);
    672 	if (de == NULL) {
    673 		error = ENOENT;
    674 		goto out;
    675 	}
    676 	KASSERT(de->td_node == node);
    677 
    678 	/* Files marked as immutable or append-only cannot be deleted. */
    679 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    680 		error = EPERM;
    681 		goto out;
    682 	}
    683 
    684 	/* Remove the entry from the directory; as it is a file, we do not
    685 	 * have to change the number of hard links of the directory. */
    686 	tmpfs_dir_detach(dvp, de);
    687 
    688 	/* Free the directory entry we just deleted.  Note that the node
    689 	 * referred by it will not be removed until the vnode is really
    690 	 * reclaimed. */
    691 	tmpfs_free_dirent(tmp, de, true);
    692 
    693 	error = 0;
    694 
    695 out:
    696 	vput(vp);
    697 	if (dvp == vp)
    698 		vrele(dvp);
    699 	else
    700 		vput(dvp);
    701 
    702 	return error;
    703 }
    704 
    705 /* --------------------------------------------------------------------- */
    706 
    707 int
    708 tmpfs_link(void *v)
    709 {
    710 	struct vnode *dvp = ((struct vop_link_args *)v)->a_dvp;
    711 	struct vnode *vp = ((struct vop_link_args *)v)->a_vp;
    712 	struct componentname *cnp = ((struct vop_link_args *)v)->a_cnp;
    713 
    714 	int error;
    715 	struct tmpfs_dirent *de;
    716 	struct tmpfs_node *dnode;
    717 	struct tmpfs_node *node;
    718 
    719 	KASSERT(VOP_ISLOCKED(dvp));
    720 	KASSERT(cnp->cn_flags & HASBUF);
    721 	KASSERT(dvp != vp); /* XXX When can this be false? */
    722 
    723 	dnode = VP_TO_TMPFS_DIR(dvp);
    724 	node = VP_TO_TMPFS_NODE(vp);
    725 
    726 	/* Lock vp because we will need to run tmpfs_update over it, which
    727 	 * needs the vnode to be locked. */
    728 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    729 	if (error != 0)
    730 		goto out1;
    731 
    732 	/* XXX: Why aren't the following two tests done by the caller? */
    733 
    734 	/* Hard links of directories are forbidden. */
    735 	if (vp->v_type == VDIR) {
    736 		error = EPERM;
    737 		goto out;
    738 	}
    739 
    740 	/* Cannot create cross-device links. */
    741 	if (dvp->v_mount != vp->v_mount) {
    742 		error = EXDEV;
    743 		goto out;
    744 	}
    745 
    746 	/* Ensure that we do not overflow the maximum number of links imposed
    747 	 * by the system. */
    748 	KASSERT(node->tn_links <= LINK_MAX);
    749 	if (node->tn_links == LINK_MAX) {
    750 		error = EMLINK;
    751 		goto out;
    752 	}
    753 
    754 	/* We cannot create links of files marked immutable or append-only. */
    755 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
    756 		error = EPERM;
    757 		goto out;
    758 	}
    759 
    760 	/* Allocate a new directory entry to represent the node. */
    761 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
    762 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
    763 	if (error != 0)
    764 		goto out;
    765 
    766 	/* Insert the new directory entry into the appropriate directory. */
    767 	tmpfs_dir_attach(dvp, de);
    768 
    769 	/* vp link count has changed, so update node times. */
    770 	node->tn_status |= TMPFS_NODE_CHANGED;
    771 	tmpfs_update(vp, NULL, NULL, 0);
    772 
    773 	error = 0;
    774 
    775 out:
    776 	VOP_UNLOCK(vp, 0);
    777 out1:
    778 	PNBUF_PUT(cnp->cn_pnbuf);
    779 
    780 	vput(dvp);
    781 
    782 	return error;
    783 }
    784 
    785 /* --------------------------------------------------------------------- */
    786 
    787 int
    788 tmpfs_rename(void *v)
    789 {
    790 	struct vnode *fdvp = ((struct vop_rename_args *)v)->a_fdvp;
    791 	struct vnode *fvp = ((struct vop_rename_args *)v)->a_fvp;
    792 	struct componentname *fcnp = ((struct vop_rename_args *)v)->a_fcnp;
    793 	struct vnode *tdvp = ((struct vop_rename_args *)v)->a_tdvp;
    794 	struct vnode *tvp = ((struct vop_rename_args *)v)->a_tvp;
    795 	struct componentname *tcnp = ((struct vop_rename_args *)v)->a_tcnp;
    796 
    797 	char *newname;
    798 	int error;
    799 	struct tmpfs_dirent *de, *de2;
    800 	struct tmpfs_mount *tmp;
    801 	struct tmpfs_node *fdnode;
    802 	struct tmpfs_node *fnode;
    803 	struct tmpfs_node *tnode;
    804 	struct tmpfs_node *tdnode;
    805 	size_t namelen;
    806 
    807 	KASSERT(VOP_ISLOCKED(tdvp));
    808 	KASSERT(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
    809 	KASSERT(fcnp->cn_flags & HASBUF);
    810 	KASSERT(tcnp->cn_flags & HASBUF);
    811 
    812 	newname = NULL;
    813 	namelen = 0;
    814 	tmp = NULL;
    815 
    816 	/* Disallow cross-device renames. */
    817 	if (fvp->v_mount != tdvp->v_mount ||
    818 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
    819 		error = EXDEV;
    820 		goto out_unlocked;
    821 	}
    822 
    823 	fnode = VP_TO_TMPFS_NODE(fvp);
    824 	fdnode = VP_TO_TMPFS_DIR(fdvp);
    825 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
    826 	tdnode = VP_TO_TMPFS_DIR(tdvp);
    827 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
    828 
    829 	/* If we need to move the directory between entries, lock the
    830 	 * source so that we can safely operate on it. */
    831 
    832 	/* XXX: this is a potential locking order violation! */
    833 	if (fdnode != tdnode) {
    834 		error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
    835 		if (error != 0)
    836 			goto out_unlocked;
    837 	}
    838 
    839 	de = tmpfs_dir_lookup(fdnode, fcnp);
    840 	if (de == NULL) {
    841 		error = ENOENT;
    842 		goto out;
    843 	}
    844 	KASSERT(de->td_node == fnode);
    845 
    846 	/* If source and target are the same file, there is nothing to do. */
    847 	if (fvp == tvp) {
    848 		error = 0;
    849 		goto out;
    850 	}
    851 
    852 	/* Avoid manipulating '.' and '..' entries. */
    853 	if (de == NULL) {
    854 		KASSERT(fvp->v_type == VDIR);
    855 		error = EINVAL;
    856 		goto out;
    857 	}
    858 	KASSERT(de->td_node == fnode);
    859 
    860 	/* If replacing an existing entry, ensure we can do the operation. */
    861 	if (tvp != NULL) {
    862 		KASSERT(tnode != NULL);
    863 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
    864 			if (tnode->tn_size > 0) {
    865 				error = ENOTEMPTY;
    866 				goto out;
    867 			}
    868 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
    869 			error = ENOTDIR;
    870 			goto out;
    871 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
    872 			error = EISDIR;
    873 			goto out;
    874 		} else {
    875 			KASSERT(fnode->tn_type != VDIR &&
    876 			        tnode->tn_type != VDIR);
    877 		}
    878 	}
    879 
    880 	/* Ensure that we have enough memory to hold the new name, if it
    881 	 * has to be changed. */
    882 	namelen = tcnp->cn_namelen;
    883 	if (fcnp->cn_namelen != tcnp->cn_namelen ||
    884 	    memcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
    885 		newname = tmpfs_str_pool_get(&tmp->tm_str_pool, namelen, 0);
    886 		if (newname == NULL) {
    887 			error = ENOSPC;
    888 			goto out;
    889 		}
    890 	}
    891 
    892 	/* If the node is being moved to another directory, we have to do
    893 	 * the move. */
    894 	if (fdnode != tdnode) {
    895 		/* In case we are moving a directory, we have to adjust its
    896 		 * parent to point to the new parent. */
    897 		if (de->td_node->tn_type == VDIR) {
    898 			struct tmpfs_node *n;
    899 
    900 			/* Ensure the target directory is not a child of the
    901 			 * directory being moved.  Otherwise, we'd end up
    902 			 * with stale nodes. */
    903 			n = tdnode;
    904 			while (n != n->tn_spec.tn_dir.tn_parent) {
    905 				if (n == fnode) {
    906 					error = EINVAL;
    907 					goto out;
    908 				}
    909 				n = n->tn_spec.tn_dir.tn_parent;
    910 			}
    911 
    912 			/* Adjust the parent pointer. */
    913 			TMPFS_VALIDATE_DIR(fnode);
    914 			de->td_node->tn_spec.tn_dir.tn_parent = tdnode;
    915 
    916 			/* As a result of changing the target of the '..'
    917 			 * entry, the link count of the source and target
    918 			 * directories has to be adjusted. */
    919 			fdnode->tn_links--;
    920 			tdnode->tn_links++;
    921 		}
    922 
    923 		/* Do the move: just remove the entry from the source directory
    924 		 * and insert it into the target one. */
    925 		tmpfs_dir_detach(fdvp, de);
    926 		tmpfs_dir_attach(tdvp, de);
    927 
    928 		/* Notify listeners of fdvp about the change in the directory.
    929 		 * We can do it at this point because we aren't touching fdvp
    930 		 * any more below. */
    931 		VN_KNOTE(fdvp, NOTE_WRITE);
    932 	}
    933 
    934 	/* If we are overwriting an entry, we have to remove the old one
    935 	 * from the target directory. */
    936 	if (tvp != NULL) {
    937 		KASSERT(tnode != NULL);
    938 
    939 		/* Remove the old entry from the target directory.
    940 		 * Note! This relies on tmpfs_dir_attach() putting the new
    941 		 * node on the end of the target's node list. */
    942 		de2 = tmpfs_dir_lookup(tdnode, tcnp);
    943 		KASSERT(de2 != NULL);
    944 		KASSERT(de2->td_node == tnode);
    945 		tmpfs_dir_detach(tdvp, de2);
    946 
    947 		/* Free the directory entry we just deleted.  Note that the
    948 		 * node referred by it will not be removed until the vnode is
    949 		 * really reclaimed. */
    950 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de2, true);
    951 	}
    952 
    953 	/* If the name has changed, we need to make it effective by changing
    954 	 * it in the directory entry. */
    955 	if (newname != NULL) {
    956 		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
    957 		KASSERT(tcnp->cn_namelen < 0xffff);
    958 
    959 		tmpfs_str_pool_put(&tmp->tm_str_pool, de->td_name,
    960 		    de->td_namelen);
    961 		de->td_namelen = (uint16_t)namelen;
    962 		memcpy(newname, tcnp->cn_nameptr, namelen);
    963 		de->td_name = newname;
    964 		newname = NULL;
    965 
    966 		fnode->tn_status |= TMPFS_NODE_CHANGED;
    967 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
    968 	}
    969 
    970 	/* Notify listeners of tdvp about the change in the directory (either
    971 	 * because a new entry was added or because one was removed) and
    972 	 * listeners of fvp about the rename. */
    973 	VN_KNOTE(tdvp, NOTE_WRITE);
    974 	VN_KNOTE(fvp, NOTE_RENAME);
    975 
    976 	error = 0;
    977 
    978  out:
    979 	if (fdnode != tdnode)
    980 		VOP_UNLOCK(fdvp, 0);
    981 
    982  out_unlocked:
    983 	/* Release target nodes. */
    984 	if (tdvp == tvp)
    985 		vrele(tdvp);
    986 	else
    987 		vput(tdvp);
    988 	if (tvp != NULL)
    989 		vput(tvp);
    990 
    991 	/* Release source nodes. */
    992 	vrele(fdvp);
    993 	vrele(fvp);
    994 
    995 	if (newname != NULL)
    996 		tmpfs_str_pool_put(&tmp->tm_str_pool, newname, namelen);
    997 
    998 	return error;
    999 }
   1000 
   1001 /* --------------------------------------------------------------------- */
   1002 
   1003 int
   1004 tmpfs_mkdir(void *v)
   1005 {
   1006 	struct vnode *dvp = ((struct vop_mkdir_args *)v)->a_dvp;
   1007 	struct vnode **vpp = ((struct vop_mkdir_args *)v)->a_vpp;
   1008 	struct componentname *cnp = ((struct vop_mkdir_args *)v)->a_cnp;
   1009 	struct vattr *vap = ((struct vop_mkdir_args *)v)->a_vap;
   1010 
   1011 	KASSERT(vap->va_type == VDIR);
   1012 
   1013 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
   1014 }
   1015 
   1016 /* --------------------------------------------------------------------- */
   1017 
   1018 int
   1019 tmpfs_rmdir(void *v)
   1020 {
   1021 	struct vnode *dvp = ((struct vop_rmdir_args *)v)->a_dvp;
   1022 	struct vnode *vp = ((struct vop_rmdir_args *)v)->a_vp;
   1023 	struct componentname *cnp = ((struct vop_rmdir_args *)v)->a_cnp;
   1024 
   1025 	int error;
   1026 	struct tmpfs_dirent *de;
   1027 	struct tmpfs_mount *tmp;
   1028 	struct tmpfs_node *dnode;
   1029 	struct tmpfs_node *node;
   1030 
   1031 	KASSERT(VOP_ISLOCKED(dvp));
   1032 	KASSERT(VOP_ISLOCKED(vp));
   1033 
   1034 	tmp = VFS_TO_TMPFS(dvp->v_mount);
   1035 	dnode = VP_TO_TMPFS_DIR(dvp);
   1036 	node = VP_TO_TMPFS_DIR(vp);
   1037 	error = 0;
   1038 
   1039 	/* Directories with more than two entries ('.' and '..') cannot be
   1040 	 * removed. */
   1041 	if (node->tn_size > 0) {
   1042 		error = ENOTEMPTY;
   1043 		goto out;
   1044 	}
   1045 
   1046 	/* This invariant holds only if we are not trying to remove "..".
   1047 	 * We checked for that above so this is safe now. */
   1048 	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
   1049 
   1050 	/* Get the directory entry associated with node (vp). */
   1051 	de = tmpfs_dir_lookup(dnode, cnp);
   1052 	if (de == NULL) {
   1053 		error = ENOENT;
   1054 		goto out;
   1055 	}
   1056 	KASSERT(de->td_node == node);
   1057 
   1058 	/* Check flags to see if we are allowed to remove the directory. */
   1059 	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
   1060 		error = EPERM;
   1061 		goto out;
   1062 	}
   1063 
   1064 	/* Detach the directory entry from the directory (dnode). */
   1065 	tmpfs_dir_detach(dvp, de);
   1066 
   1067 	node->tn_links--;
   1068 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
   1069 	    TMPFS_NODE_MODIFIED;
   1070 	node->tn_spec.tn_dir.tn_parent->tn_links--;
   1071 	node->tn_spec.tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \
   1072 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
   1073 
   1074 	/* Release the parent. */
   1075 	cache_purge(dvp); /* XXX Is this needed? */
   1076 
   1077 	/* Free the directory entry we just deleted.  Note that the node
   1078 	 * referred by it will not be removed until the vnode is really
   1079 	 * reclaimed. */
   1080 	tmpfs_free_dirent(tmp, de, true);
   1081 
   1082 	KASSERT(node->tn_links == 0);
   1083  out:
   1084 	/* Release the nodes. */
   1085 	vput(dvp);
   1086 	vput(vp);
   1087 
   1088 	return error;
   1089 }
   1090 
   1091 /* --------------------------------------------------------------------- */
   1092 
   1093 int
   1094 tmpfs_symlink(void *v)
   1095 {
   1096 	struct vnode *dvp = ((struct vop_symlink_args *)v)->a_dvp;
   1097 	struct vnode **vpp = ((struct vop_symlink_args *)v)->a_vpp;
   1098 	struct componentname *cnp = ((struct vop_symlink_args *)v)->a_cnp;
   1099 	struct vattr *vap = ((struct vop_symlink_args *)v)->a_vap;
   1100 	char *target = ((struct vop_symlink_args *)v)->a_target;
   1101 
   1102 	KASSERT(vap->va_type == VLNK);
   1103 
   1104 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
   1105 }
   1106 
   1107 /* --------------------------------------------------------------------- */
   1108 
   1109 int
   1110 tmpfs_readdir(void *v)
   1111 {
   1112 	struct vnode *vp = ((struct vop_readdir_args *)v)->a_vp;
   1113 	struct uio *uio = ((struct vop_readdir_args *)v)->a_uio;
   1114 	int *eofflag = ((struct vop_readdir_args *)v)->a_eofflag;
   1115 	off_t **cookies = ((struct vop_readdir_args *)v)->a_cookies;
   1116 	int *ncookies = ((struct vop_readdir_args *)v)->a_ncookies;
   1117 
   1118 	int error;
   1119 	off_t startoff;
   1120 	off_t cnt;
   1121 	struct tmpfs_node *node;
   1122 
   1123 	KASSERT(VOP_ISLOCKED(vp));
   1124 
   1125 	/* This operation only makes sense on directory nodes. */
   1126 	if (vp->v_type != VDIR) {
   1127 		error = ENOTDIR;
   1128 		goto out;
   1129 	}
   1130 
   1131 	node = VP_TO_TMPFS_DIR(vp);
   1132 
   1133 	startoff = uio->uio_offset;
   1134 
   1135 	cnt = 0;
   1136 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
   1137 		error = tmpfs_dir_getdotdent(node, uio);
   1138 		if (error == -1) {
   1139 			error = 0;
   1140 			goto outok;
   1141 		} else if (error != 0)
   1142 			goto outok;
   1143 		cnt++;
   1144 	}
   1145 
   1146 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
   1147 		error = tmpfs_dir_getdotdotdent(node, uio);
   1148 		if (error == -1) {
   1149 			error = 0;
   1150 			goto outok;
   1151 		} else if (error != 0)
   1152 			goto outok;
   1153 		cnt++;
   1154 	}
   1155 
   1156 	error = tmpfs_dir_getdents(node, uio, &cnt);
   1157 	if (error == -1)
   1158 		error = 0;
   1159 	KASSERT(error >= 0);
   1160 
   1161 outok:
   1162 	/* This label assumes that startoff has been
   1163 	 * initialized.  If the compiler didn't spit out warnings, we'd
   1164 	 * simply make this one be 'out' and drop 'outok'. */
   1165 
   1166 	if (eofflag != NULL)
   1167 		*eofflag =
   1168 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
   1169 
   1170 	/* Update NFS-related variables. */
   1171 	if (error == 0 && cookies != NULL && ncookies != NULL) {
   1172 		off_t i;
   1173 		off_t off = startoff;
   1174 		struct tmpfs_dirent *de = NULL;
   1175 
   1176 		*ncookies = cnt;
   1177 		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
   1178 
   1179 		for (i = 0; i < cnt; i++) {
   1180 			KASSERT(off != TMPFS_DIRCOOKIE_EOF);
   1181 			if (off == TMPFS_DIRCOOKIE_DOT) {
   1182 				off = TMPFS_DIRCOOKIE_DOTDOT;
   1183 			} else {
   1184 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
   1185 					de = TAILQ_FIRST(&node->tn_spec.
   1186 					    tn_dir.tn_dir);
   1187 				} else if (de != NULL) {
   1188 					de = TAILQ_NEXT(de, td_entries);
   1189 				} else {
   1190 					de = tmpfs_dir_lookupbycookie(node,
   1191 					    off);
   1192 					KASSERT(de != NULL);
   1193 					de = TAILQ_NEXT(de, td_entries);
   1194 				}
   1195 				if (de == NULL) {
   1196 					off = TMPFS_DIRCOOKIE_EOF;
   1197 				} else {
   1198 					off = tmpfs_dircookie(de);
   1199 				}
   1200 			}
   1201 
   1202 			(*cookies)[i] = off;
   1203 		}
   1204 		KASSERT(uio->uio_offset == off);
   1205 	}
   1206 
   1207 out:
   1208 	KASSERT(VOP_ISLOCKED(vp));
   1209 
   1210 	return error;
   1211 }
   1212 
   1213 /* --------------------------------------------------------------------- */
   1214 
   1215 int
   1216 tmpfs_readlink(void *v)
   1217 {
   1218 	struct vnode *vp = ((struct vop_readlink_args *)v)->a_vp;
   1219 	struct uio *uio = ((struct vop_readlink_args *)v)->a_uio;
   1220 
   1221 	int error;
   1222 	struct tmpfs_node *node;
   1223 
   1224 	KASSERT(VOP_ISLOCKED(vp));
   1225 	KASSERT(uio->uio_offset == 0);
   1226 	KASSERT(vp->v_type == VLNK);
   1227 
   1228 	node = VP_TO_TMPFS_NODE(vp);
   1229 
   1230 	error = uiomove(node->tn_spec.tn_lnk.tn_link,
   1231 	    MIN(node->tn_size, uio->uio_resid), uio);
   1232 	node->tn_status |= TMPFS_NODE_ACCESSED;
   1233 
   1234 	KASSERT(VOP_ISLOCKED(vp));
   1235 
   1236 	return error;
   1237 }
   1238 
   1239 /* --------------------------------------------------------------------- */
   1240 
   1241 int
   1242 tmpfs_inactive(void *v)
   1243 {
   1244 	struct vnode *vp = ((struct vop_inactive_args *)v)->a_vp;
   1245 
   1246 	struct tmpfs_node *node;
   1247 
   1248 	KASSERT(VOP_ISLOCKED(vp));
   1249 
   1250 	node = VP_TO_TMPFS_NODE(vp);
   1251 	*((struct vop_inactive_args *)v)->a_recycle = (node->tn_links == 0);
   1252 	VOP_UNLOCK(vp, 0);
   1253 
   1254 	return 0;
   1255 }
   1256 
   1257 /* --------------------------------------------------------------------- */
   1258 
   1259 int
   1260 tmpfs_reclaim(void *v)
   1261 {
   1262 	struct vnode *vp = ((struct vop_reclaim_args *)v)->a_vp;
   1263 
   1264 	struct tmpfs_mount *tmp;
   1265 	struct tmpfs_node *node;
   1266 
   1267 	node = VP_TO_TMPFS_NODE(vp);
   1268 	tmp = VFS_TO_TMPFS(vp->v_mount);
   1269 
   1270 	cache_purge(vp);
   1271 	tmpfs_free_vp(vp);
   1272 
   1273 	/* If the node referenced by this vnode was deleted by the user,
   1274 	 * we must free its associated data structures (now that the vnode
   1275 	 * is being reclaimed). */
   1276 	if (node->tn_links == 0)
   1277 		tmpfs_free_node(tmp, node);
   1278 
   1279 	KASSERT(vp->v_data == NULL);
   1280 
   1281 	return 0;
   1282 }
   1283 
   1284 /* --------------------------------------------------------------------- */
   1285 
   1286 int
   1287 tmpfs_print(void *v)
   1288 {
   1289 	struct vnode *vp = ((struct vop_print_args *)v)->a_vp;
   1290 
   1291 	struct tmpfs_node *node;
   1292 
   1293 	node = VP_TO_TMPFS_NODE(vp);
   1294 
   1295 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
   1296 	    node, node->tn_flags, node->tn_links);
   1297 	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
   1298 	    ", status 0x%x\n",
   1299 	    node->tn_mode, node->tn_uid, node->tn_gid,
   1300 	    (uintmax_t)node->tn_size, node->tn_status);
   1301 	if (vp->v_type == VFIFO)
   1302 		fifo_printinfo(vp);
   1303 	printf("\n");
   1304 
   1305 	return 0;
   1306 }
   1307 
   1308 /* --------------------------------------------------------------------- */
   1309 
   1310 int
   1311 tmpfs_pathconf(void *v)
   1312 {
   1313 	int name = ((struct vop_pathconf_args *)v)->a_name;
   1314 	register_t *retval = ((struct vop_pathconf_args *)v)->a_retval;
   1315 
   1316 	int error;
   1317 
   1318 	error = 0;
   1319 
   1320 	switch (name) {
   1321 	case _PC_LINK_MAX:
   1322 		*retval = LINK_MAX;
   1323 		break;
   1324 
   1325 	case _PC_NAME_MAX:
   1326 		*retval = NAME_MAX;
   1327 		break;
   1328 
   1329 	case _PC_PATH_MAX:
   1330 		*retval = PATH_MAX;
   1331 		break;
   1332 
   1333 	case _PC_PIPE_BUF:
   1334 		*retval = PIPE_BUF;
   1335 		break;
   1336 
   1337 	case _PC_CHOWN_RESTRICTED:
   1338 		*retval = 1;
   1339 		break;
   1340 
   1341 	case _PC_NO_TRUNC:
   1342 		*retval = 1;
   1343 		break;
   1344 
   1345 	case _PC_SYNC_IO:
   1346 		*retval = 1;
   1347 		break;
   1348 
   1349 	case _PC_FILESIZEBITS:
   1350 		*retval = 0; /* XXX Don't know which value should I return. */
   1351 		break;
   1352 
   1353 	default:
   1354 		error = EINVAL;
   1355 	}
   1356 
   1357 	return error;
   1358 }
   1359 
   1360 /* --------------------------------------------------------------------- */
   1361 
   1362 int
   1363 tmpfs_advlock(void *v)
   1364 {
   1365 	struct vnode *vp = ((struct vop_advlock_args *)v)->a_vp;
   1366 
   1367 	struct tmpfs_node *node;
   1368 
   1369 	node = VP_TO_TMPFS_NODE(vp);
   1370 
   1371 	return lf_advlock(v, &node->tn_lockf, node->tn_size);
   1372 }
   1373 
   1374 /* --------------------------------------------------------------------- */
   1375 
   1376 int
   1377 tmpfs_getpages(void *v)
   1378 {
   1379 	struct vnode *vp = ((struct vop_getpages_args *)v)->a_vp;
   1380 	voff_t offset = ((struct vop_getpages_args *)v)->a_offset;
   1381 	struct vm_page **m = ((struct vop_getpages_args *)v)->a_m;
   1382 	int *count = ((struct vop_getpages_args *)v)->a_count;
   1383 	int centeridx = ((struct vop_getpages_args *)v)->a_centeridx;
   1384 	vm_prot_t access_type = ((struct vop_getpages_args *)v)->a_access_type;
   1385 	int advice = ((struct vop_getpages_args *)v)->a_advice;
   1386 	int flags = ((struct vop_getpages_args *)v)->a_flags;
   1387 
   1388 	int error;
   1389 	int i;
   1390 	struct tmpfs_node *node;
   1391 	struct uvm_object *uobj;
   1392 	int npages = *count;
   1393 
   1394 	KASSERT(vp->v_type == VREG);
   1395 	KASSERT(mutex_owned(&vp->v_interlock));
   1396 
   1397 	node = VP_TO_TMPFS_NODE(vp);
   1398 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1399 
   1400 	/* We currently don't rely on PGO_PASTEOF. */
   1401 
   1402 	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
   1403 		if ((flags & PGO_LOCKED) == 0)
   1404 			mutex_exit(&vp->v_interlock);
   1405 		return EINVAL;
   1406 	}
   1407 
   1408 	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
   1409 		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
   1410 	}
   1411 
   1412 	if ((flags & PGO_LOCKED) != 0)
   1413 		return EBUSY;
   1414 
   1415 	if ((flags & PGO_NOTIMESTAMP) == 0) {
   1416 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
   1417 			node->tn_status |= TMPFS_NODE_ACCESSED;
   1418 
   1419 		if ((access_type & VM_PROT_WRITE) != 0)
   1420 			node->tn_status |= TMPFS_NODE_MODIFIED;
   1421 	}
   1422 
   1423 	mutex_exit(&vp->v_interlock);
   1424 
   1425 	/*
   1426 	 * Make sure that the array on which we will store the
   1427 	 * gotten pages is clean.  Otherwise uao_get (pointed to by
   1428 	 * the pgo_get below) gets confused and does not return the
   1429 	 * appropriate pages.
   1430 	 *
   1431 	 * XXX This shall be revisited when kern/32166 is addressed
   1432 	 * because the loop to clean m[i] will most likely be redundant
   1433 	 * as well as the PGO_ALLPAGES flag.
   1434 	 */
   1435 	if (m != NULL)
   1436 		for (i = 0; i < npages; i++)
   1437 			m[i] = NULL;
   1438 	mutex_enter(&uobj->vmobjlock);
   1439 	error = (*uobj->pgops->pgo_get)(uobj, offset, m, &npages, centeridx,
   1440 	    access_type, advice, flags | PGO_ALLPAGES);
   1441 #if defined(DEBUG)
   1442 	{
   1443 		/* Make sure that all the pages we return are valid. */
   1444 		int dbgi;
   1445 		if (error == 0 && m != NULL)
   1446 			for (dbgi = 0; dbgi < npages; dbgi++)
   1447 				KASSERT(m[dbgi] != NULL);
   1448 	}
   1449 #endif
   1450 
   1451 	return error;
   1452 }
   1453 
   1454 /* --------------------------------------------------------------------- */
   1455 
   1456 int
   1457 tmpfs_putpages(void *v)
   1458 {
   1459 	struct vnode *vp = ((struct vop_putpages_args *)v)->a_vp;
   1460 	voff_t offlo = ((struct vop_putpages_args *)v)->a_offlo;
   1461 	voff_t offhi = ((struct vop_putpages_args *)v)->a_offhi;
   1462 	int flags = ((struct vop_putpages_args *)v)->a_flags;
   1463 
   1464 	int error;
   1465 	struct tmpfs_node *node;
   1466 	struct uvm_object *uobj;
   1467 
   1468 	KASSERT(mutex_owned(&vp->v_interlock));
   1469 
   1470 	node = VP_TO_TMPFS_NODE(vp);
   1471 
   1472 	if (vp->v_type != VREG) {
   1473 		mutex_exit(&vp->v_interlock);
   1474 		return 0;
   1475 	}
   1476 
   1477 	uobj = node->tn_spec.tn_reg.tn_aobj;
   1478 	mutex_exit(&vp->v_interlock);
   1479 
   1480 	mutex_enter(&uobj->vmobjlock);
   1481 	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
   1482 
   1483 	/* XXX mtime */
   1484 
   1485 	return error;
   1486 }
   1487