tmpfs_vnops.c revision 346286
1/*	$NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $	*/
2
3/*-
4 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9 * 2005 program.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 *    notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 *    notice, this list of conditions and the following disclaimer in the
18 *    documentation and/or other materials provided with the distribution.
19 *
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#include <sys/cdefs.h>
37__FBSDID("$FreeBSD: stable/11/sys/fs/tmpfs/tmpfs_vnops.c 346286 2019-04-16 17:43:14Z kib $");
38
39#include <sys/param.h>
40#include <sys/fcntl.h>
41#include <sys/lockf.h>
42#include <sys/lock.h>
43#include <sys/namei.h>
44#include <sys/priv.h>
45#include <sys/proc.h>
46#include <sys/rwlock.h>
47#include <sys/sched.h>
48#include <sys/stat.h>
49#include <sys/sysctl.h>
50#include <sys/unistd.h>
51#include <sys/vnode.h>
52
53#include <vm/vm.h>
54#include <vm/vm_param.h>
55#include <vm/vm_object.h>
56#include <vm/vm_page.h>
57#include <vm/vm_pager.h>
58
59#include <fs/tmpfs/tmpfs_vnops.h>
60#include <fs/tmpfs/tmpfs.h>
61
62SYSCTL_DECL(_vfs_tmpfs);
63
64static volatile int tmpfs_rename_restarts;
65SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD,
66    __DEVOLATILE(int *, &tmpfs_rename_restarts), 0,
67    "Times rename had to restart due to lock contention");
68
69static int
70tmpfs_vn_get_ino_alloc(struct mount *mp, void *arg, int lkflags,
71    struct vnode **rvp)
72{
73
74	return (tmpfs_alloc_vp(mp, arg, lkflags, rvp));
75}
76
77static int
78tmpfs_lookup1(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
79{
80	struct tmpfs_dirent *de;
81	struct tmpfs_node *dnode, *pnode;
82	struct tmpfs_mount *tm;
83	int error;
84
85	dnode = VP_TO_TMPFS_DIR(dvp);
86	*vpp = NULLVP;
87
88	/* Check accessibility of requested node as a first step. */
89	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
90	if (error != 0)
91		goto out;
92
93	/* We cannot be requesting the parent directory of the root node. */
94	MPASS(IMPLIES(dnode->tn_type == VDIR &&
95	    dnode->tn_dir.tn_parent == dnode,
96	    !(cnp->cn_flags & ISDOTDOT)));
97
98	TMPFS_ASSERT_LOCKED(dnode);
99	if (dnode->tn_dir.tn_parent == NULL) {
100		error = ENOENT;
101		goto out;
102	}
103	if (cnp->cn_flags & ISDOTDOT) {
104		tm = VFS_TO_TMPFS(dvp->v_mount);
105		pnode = dnode->tn_dir.tn_parent;
106		tmpfs_ref_node(pnode);
107		error = vn_vget_ino_gen(dvp, tmpfs_vn_get_ino_alloc,
108		    pnode, cnp->cn_lkflags, vpp);
109		tmpfs_free_node(tm, pnode);
110		if (error != 0)
111			goto out;
112	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
113		VREF(dvp);
114		*vpp = dvp;
115		error = 0;
116	} else {
117		de = tmpfs_dir_lookup(dnode, NULL, cnp);
118		if (de != NULL && de->td_node == NULL)
119			cnp->cn_flags |= ISWHITEOUT;
120		if (de == NULL || de->td_node == NULL) {
121			/*
122			 * The entry was not found in the directory.
123			 * This is OK if we are creating or renaming an
124			 * entry and are working on the last component of
125			 * the path name.
126			 */
127			if ((cnp->cn_flags & ISLASTCN) &&
128			    (cnp->cn_nameiop == CREATE || \
129			    cnp->cn_nameiop == RENAME ||
130			    (cnp->cn_nameiop == DELETE &&
131			    cnp->cn_flags & DOWHITEOUT &&
132			    cnp->cn_flags & ISWHITEOUT))) {
133				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
134				    cnp->cn_thread);
135				if (error != 0)
136					goto out;
137
138				/*
139				 * Keep the component name in the buffer for
140				 * future uses.
141				 */
142				cnp->cn_flags |= SAVENAME;
143
144				error = EJUSTRETURN;
145			} else
146				error = ENOENT;
147		} else {
148			struct tmpfs_node *tnode;
149
150			/*
151			 * The entry was found, so get its associated
152			 * tmpfs_node.
153			 */
154			tnode = de->td_node;
155
156			/*
157			 * If we are not at the last path component and
158			 * found a non-directory or non-link entry (which
159			 * may itself be pointing to a directory), raise
160			 * an error.
161			 */
162			if ((tnode->tn_type != VDIR &&
163			    tnode->tn_type != VLNK) &&
164			    !(cnp->cn_flags & ISLASTCN)) {
165				error = ENOTDIR;
166				goto out;
167			}
168
169			/*
170			 * If we are deleting or renaming the entry, keep
171			 * track of its tmpfs_dirent so that it can be
172			 * easily deleted later.
173			 */
174			if ((cnp->cn_flags & ISLASTCN) &&
175			    (cnp->cn_nameiop == DELETE ||
176			    cnp->cn_nameiop == RENAME)) {
177				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
178				    cnp->cn_thread);
179				if (error != 0)
180					goto out;
181
182				/* Allocate a new vnode on the matching entry. */
183				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
184				    cnp->cn_lkflags, vpp);
185				if (error != 0)
186					goto out;
187
188				if ((dnode->tn_mode & S_ISTXT) &&
189				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred,
190				  cnp->cn_thread) && VOP_ACCESS(*vpp, VADMIN,
191				  cnp->cn_cred, cnp->cn_thread)) {
192					error = EPERM;
193					vput(*vpp);
194					*vpp = NULL;
195					goto out;
196				}
197				cnp->cn_flags |= SAVENAME;
198			} else {
199				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
200				    cnp->cn_lkflags, vpp);
201				if (error != 0)
202					goto out;
203			}
204		}
205	}
206
207	/*
208	 * Store the result of this lookup in the cache.  Avoid this if the
209	 * request was for creation, as it does not improve timings on
210	 * emprical tests.
211	 */
212	if ((cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp))
213		cache_enter(dvp, *vpp, cnp);
214
215out:
216	/*
217	 * If there were no errors, *vpp cannot be null and it must be
218	 * locked.
219	 */
220	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
221
222	return (error);
223}
224
225static int
226tmpfs_cached_lookup(struct vop_cachedlookup_args *v)
227{
228
229	return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp));
230}
231
232static int
233tmpfs_lookup(struct vop_lookup_args *v)
234{
235
236	return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp));
237}
238
239static int
240tmpfs_create(struct vop_create_args *v)
241{
242	struct vnode *dvp = v->a_dvp;
243	struct vnode **vpp = v->a_vpp;
244	struct componentname *cnp = v->a_cnp;
245	struct vattr *vap = v->a_vap;
246	int error;
247
248	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
249
250	error = tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
251	if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp))
252		cache_enter(dvp, *vpp, cnp);
253	return (error);
254}
255
256static int
257tmpfs_mknod(struct vop_mknod_args *v)
258{
259	struct vnode *dvp = v->a_dvp;
260	struct vnode **vpp = v->a_vpp;
261	struct componentname *cnp = v->a_cnp;
262	struct vattr *vap = v->a_vap;
263
264	if (vap->va_type != VBLK && vap->va_type != VCHR &&
265	    vap->va_type != VFIFO)
266		return EINVAL;
267
268	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
269}
270
271static int
272tmpfs_open(struct vop_open_args *v)
273{
274	struct vnode *vp = v->a_vp;
275	int mode = v->a_mode;
276
277	int error;
278	struct tmpfs_node *node;
279
280	MPASS(VOP_ISLOCKED(vp));
281
282	node = VP_TO_TMPFS_NODE(vp);
283
284	/* The file is still active but all its names have been removed
285	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
286	 * it is about to die. */
287	if (node->tn_links < 1)
288		return (ENOENT);
289
290	/* If the file is marked append-only, deny write requests. */
291	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
292		error = EPERM;
293	else {
294		error = 0;
295		/* For regular files, the call below is nop. */
296		KASSERT(vp->v_type != VREG || (node->tn_reg.tn_aobj->flags &
297		    OBJ_DEAD) == 0, ("dead object"));
298		vnode_create_vobject(vp, node->tn_size, v->a_td);
299	}
300
301	MPASS(VOP_ISLOCKED(vp));
302	return error;
303}
304
305static int
306tmpfs_close(struct vop_close_args *v)
307{
308	struct vnode *vp = v->a_vp;
309
310	/* Update node times. */
311	tmpfs_update(vp);
312
313	return (0);
314}
315
316int
317tmpfs_access(struct vop_access_args *v)
318{
319	struct vnode *vp = v->a_vp;
320	accmode_t accmode = v->a_accmode;
321	struct ucred *cred = v->a_cred;
322
323	int error;
324	struct tmpfs_node *node;
325
326	MPASS(VOP_ISLOCKED(vp));
327
328	node = VP_TO_TMPFS_NODE(vp);
329
330	switch (vp->v_type) {
331	case VDIR:
332		/* FALLTHROUGH */
333	case VLNK:
334		/* FALLTHROUGH */
335	case VREG:
336		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
337			error = EROFS;
338			goto out;
339		}
340		break;
341
342	case VBLK:
343		/* FALLTHROUGH */
344	case VCHR:
345		/* FALLTHROUGH */
346	case VSOCK:
347		/* FALLTHROUGH */
348	case VFIFO:
349		break;
350
351	default:
352		error = EINVAL;
353		goto out;
354	}
355
356	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
357		error = EPERM;
358		goto out;
359	}
360
361	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
362	    node->tn_gid, accmode, cred, NULL);
363
364out:
365	MPASS(VOP_ISLOCKED(vp));
366
367	return error;
368}
369
370int
371tmpfs_getattr(struct vop_getattr_args *v)
372{
373	struct vnode *vp = v->a_vp;
374	struct vattr *vap = v->a_vap;
375	vm_object_t obj;
376	struct tmpfs_node *node;
377
378	node = VP_TO_TMPFS_NODE(vp);
379
380	tmpfs_update(vp);
381
382	vap->va_type = vp->v_type;
383	vap->va_mode = node->tn_mode;
384	vap->va_nlink = node->tn_links;
385	vap->va_uid = node->tn_uid;
386	vap->va_gid = node->tn_gid;
387	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
388	vap->va_fileid = node->tn_id;
389	vap->va_size = node->tn_size;
390	vap->va_blocksize = PAGE_SIZE;
391	vap->va_atime = node->tn_atime;
392	vap->va_mtime = node->tn_mtime;
393	vap->va_ctime = node->tn_ctime;
394	vap->va_birthtime = node->tn_birthtime;
395	vap->va_gen = node->tn_gen;
396	vap->va_flags = node->tn_flags;
397	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
398		node->tn_rdev : NODEV;
399	if (vp->v_type == VREG) {
400		obj = node->tn_reg.tn_aobj;
401		vap->va_bytes = (u_quad_t)obj->resident_page_count * PAGE_SIZE;
402	} else
403		vap->va_bytes = node->tn_size;
404	vap->va_filerev = 0;
405
406	return 0;
407}
408
409int
410tmpfs_setattr(struct vop_setattr_args *v)
411{
412	struct vnode *vp = v->a_vp;
413	struct vattr *vap = v->a_vap;
414	struct ucred *cred = v->a_cred;
415	struct thread *td = curthread;
416
417	int error;
418
419	MPASS(VOP_ISLOCKED(vp));
420
421	error = 0;
422
423	/* Abort if any unsettable attribute is given. */
424	if (vap->va_type != VNON ||
425	    vap->va_nlink != VNOVAL ||
426	    vap->va_fsid != VNOVAL ||
427	    vap->va_fileid != VNOVAL ||
428	    vap->va_blocksize != VNOVAL ||
429	    vap->va_gen != VNOVAL ||
430	    vap->va_rdev != VNOVAL ||
431	    vap->va_bytes != VNOVAL)
432		error = EINVAL;
433
434	if (error == 0 && (vap->va_flags != VNOVAL))
435		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
436
437	if (error == 0 && (vap->va_size != VNOVAL))
438		error = tmpfs_chsize(vp, vap->va_size, cred, td);
439
440	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
441		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
442
443	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
444		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
445
446	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
447	    vap->va_atime.tv_nsec != VNOVAL) ||
448	    (vap->va_mtime.tv_sec != VNOVAL &&
449	    vap->va_mtime.tv_nsec != VNOVAL) ||
450	    (vap->va_birthtime.tv_sec != VNOVAL &&
451	    vap->va_birthtime.tv_nsec != VNOVAL)))
452		error = tmpfs_chtimes(vp, vap, cred, td);
453
454	/* Update the node times.  We give preference to the error codes
455	 * generated by this function rather than the ones that may arise
456	 * from tmpfs_update. */
457	tmpfs_update(vp);
458
459	MPASS(VOP_ISLOCKED(vp));
460
461	return error;
462}
463
464static int
465tmpfs_read(struct vop_read_args *v)
466{
467	struct vnode *vp;
468	struct uio *uio;
469	struct tmpfs_node *node;
470
471	vp = v->a_vp;
472	if (vp->v_type != VREG)
473		return (EISDIR);
474	uio = v->a_uio;
475	if (uio->uio_offset < 0)
476		return (EINVAL);
477	node = VP_TO_TMPFS_NODE(vp);
478	tmpfs_set_status(VFS_TO_TMPFS(vp->v_mount), node, TMPFS_NODE_ACCESSED);
479	return (uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio));
480}
481
482static int
483tmpfs_write(struct vop_write_args *v)
484{
485	struct vnode *vp;
486	struct uio *uio;
487	struct tmpfs_node *node;
488	off_t oldsize;
489	int error, ioflag;
490
491	vp = v->a_vp;
492	uio = v->a_uio;
493	ioflag = v->a_ioflag;
494	error = 0;
495	node = VP_TO_TMPFS_NODE(vp);
496	oldsize = node->tn_size;
497
498	if (uio->uio_offset < 0 || vp->v_type != VREG)
499		return (EINVAL);
500	if (uio->uio_resid == 0)
501		return (0);
502	if (ioflag & IO_APPEND)
503		uio->uio_offset = node->tn_size;
504	if (uio->uio_offset + uio->uio_resid >
505	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
506		return (EFBIG);
507	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
508		return (EFBIG);
509	if (uio->uio_offset + uio->uio_resid > node->tn_size) {
510		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
511		    FALSE);
512		if (error != 0)
513			goto out;
514	}
515
516	error = uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio);
517	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
518	    TMPFS_NODE_CHANGED;
519	if (node->tn_mode & (S_ISUID | S_ISGID)) {
520		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
521			node->tn_mode &= ~(S_ISUID | S_ISGID);
522	}
523	if (error != 0)
524		(void)tmpfs_reg_resize(vp, oldsize, TRUE);
525
526out:
527	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
528	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
529
530	return (error);
531}
532
533static int
534tmpfs_fsync(struct vop_fsync_args *v)
535{
536	struct vnode *vp = v->a_vp;
537
538	MPASS(VOP_ISLOCKED(vp));
539
540	tmpfs_check_mtime(vp);
541	tmpfs_update(vp);
542
543	return 0;
544}
545
546static int
547tmpfs_remove(struct vop_remove_args *v)
548{
549	struct vnode *dvp = v->a_dvp;
550	struct vnode *vp = v->a_vp;
551
552	int error;
553	struct tmpfs_dirent *de;
554	struct tmpfs_mount *tmp;
555	struct tmpfs_node *dnode;
556	struct tmpfs_node *node;
557
558	MPASS(VOP_ISLOCKED(dvp));
559	MPASS(VOP_ISLOCKED(vp));
560
561	if (vp->v_type == VDIR) {
562		error = EISDIR;
563		goto out;
564	}
565
566	dnode = VP_TO_TMPFS_DIR(dvp);
567	node = VP_TO_TMPFS_NODE(vp);
568	tmp = VFS_TO_TMPFS(vp->v_mount);
569	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
570	MPASS(de != NULL);
571
572	/* Files marked as immutable or append-only cannot be deleted. */
573	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
574	    (dnode->tn_flags & APPEND)) {
575		error = EPERM;
576		goto out;
577	}
578
579	/* Remove the entry from the directory; as it is a file, we do not
580	 * have to change the number of hard links of the directory. */
581	tmpfs_dir_detach(dvp, de);
582	if (v->a_cnp->cn_flags & DOWHITEOUT)
583		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
584
585	/* Free the directory entry we just deleted.  Note that the node
586	 * referred by it will not be removed until the vnode is really
587	 * reclaimed. */
588	tmpfs_free_dirent(tmp, de);
589
590	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
591	error = 0;
592
593out:
594
595	return error;
596}
597
598static int
599tmpfs_link(struct vop_link_args *v)
600{
601	struct vnode *dvp = v->a_tdvp;
602	struct vnode *vp = v->a_vp;
603	struct componentname *cnp = v->a_cnp;
604
605	int error;
606	struct tmpfs_dirent *de;
607	struct tmpfs_node *node;
608
609	MPASS(VOP_ISLOCKED(dvp));
610	MPASS(cnp->cn_flags & HASBUF);
611	MPASS(dvp != vp); /* XXX When can this be false? */
612	node = VP_TO_TMPFS_NODE(vp);
613
614	/* Ensure that we do not overflow the maximum number of links imposed
615	 * by the system. */
616	MPASS(node->tn_links <= LINK_MAX);
617	if (node->tn_links == LINK_MAX) {
618		error = EMLINK;
619		goto out;
620	}
621
622	/* We cannot create links of files marked immutable or append-only. */
623	if (node->tn_flags & (IMMUTABLE | APPEND)) {
624		error = EPERM;
625		goto out;
626	}
627
628	/* Allocate a new directory entry to represent the node. */
629	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
630	    cnp->cn_nameptr, cnp->cn_namelen, &de);
631	if (error != 0)
632		goto out;
633
634	/* Insert the new directory entry into the appropriate directory. */
635	if (cnp->cn_flags & ISWHITEOUT)
636		tmpfs_dir_whiteout_remove(dvp, cnp);
637	tmpfs_dir_attach(dvp, de);
638
639	/* vp link count has changed, so update node times. */
640	node->tn_status |= TMPFS_NODE_CHANGED;
641	tmpfs_update(vp);
642
643	error = 0;
644
645out:
646	return error;
647}
648
649/*
650 * We acquire all but fdvp locks using non-blocking acquisitions.  If we
651 * fail to acquire any lock in the path we will drop all held locks,
652 * acquire the new lock in a blocking fashion, and then release it and
653 * restart the rename.  This acquire/release step ensures that we do not
654 * spin on a lock waiting for release.  On error release all vnode locks
655 * and decrement references the way tmpfs_rename() would do.
656 */
657static int
658tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp,
659    struct vnode *tdvp, struct vnode **tvpp,
660    struct componentname *fcnp, struct componentname *tcnp)
661{
662	struct vnode *nvp;
663	struct mount *mp;
664	struct tmpfs_dirent *de;
665	int error, restarts = 0;
666
667	VOP_UNLOCK(tdvp, 0);
668	if (*tvpp != NULL && *tvpp != tdvp)
669		VOP_UNLOCK(*tvpp, 0);
670	mp = fdvp->v_mount;
671
672relock:
673	restarts += 1;
674	error = vn_lock(fdvp, LK_EXCLUSIVE);
675	if (error)
676		goto releout;
677	if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
678		VOP_UNLOCK(fdvp, 0);
679		error = vn_lock(tdvp, LK_EXCLUSIVE);
680		if (error)
681			goto releout;
682		VOP_UNLOCK(tdvp, 0);
683		goto relock;
684	}
685	/*
686	 * Re-resolve fvp to be certain it still exists and fetch the
687	 * correct vnode.
688	 */
689	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp);
690	if (de == NULL) {
691		VOP_UNLOCK(fdvp, 0);
692		VOP_UNLOCK(tdvp, 0);
693		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
694		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
695			error = EINVAL;
696		else
697			error = ENOENT;
698		goto releout;
699	}
700	error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
701	if (error != 0) {
702		VOP_UNLOCK(fdvp, 0);
703		VOP_UNLOCK(tdvp, 0);
704		if (error != EBUSY)
705			goto releout;
706		error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp);
707		if (error != 0)
708			goto releout;
709		VOP_UNLOCK(nvp, 0);
710		/*
711		 * Concurrent rename race.
712		 */
713		if (nvp == tdvp) {
714			vrele(nvp);
715			error = EINVAL;
716			goto releout;
717		}
718		vrele(*fvpp);
719		*fvpp = nvp;
720		goto relock;
721	}
722	vrele(*fvpp);
723	*fvpp = nvp;
724	VOP_UNLOCK(*fvpp, 0);
725	/*
726	 * Re-resolve tvp and acquire the vnode lock if present.
727	 */
728	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp);
729	/*
730	 * If tvp disappeared we just carry on.
731	 */
732	if (de == NULL && *tvpp != NULL) {
733		vrele(*tvpp);
734		*tvpp = NULL;
735	}
736	/*
737	 * Get the tvp ino if the lookup succeeded.  We may have to restart
738	 * if the non-blocking acquire fails.
739	 */
740	if (de != NULL) {
741		nvp = NULL;
742		error = tmpfs_alloc_vp(mp, de->td_node,
743		    LK_EXCLUSIVE | LK_NOWAIT, &nvp);
744		if (*tvpp != NULL)
745			vrele(*tvpp);
746		*tvpp = nvp;
747		if (error != 0) {
748			VOP_UNLOCK(fdvp, 0);
749			VOP_UNLOCK(tdvp, 0);
750			if (error != EBUSY)
751				goto releout;
752			error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE,
753			    &nvp);
754			if (error != 0)
755				goto releout;
756			VOP_UNLOCK(nvp, 0);
757			/*
758			 * fdvp contains fvp, thus tvp (=fdvp) is not empty.
759			 */
760			if (nvp == fdvp) {
761				error = ENOTEMPTY;
762				goto releout;
763			}
764			goto relock;
765		}
766	}
767	tmpfs_rename_restarts += restarts;
768
769	return (0);
770
771releout:
772	vrele(fdvp);
773	vrele(*fvpp);
774	vrele(tdvp);
775	if (*tvpp != NULL)
776		vrele(*tvpp);
777	tmpfs_rename_restarts += restarts;
778
779	return (error);
780}
781
782static int
783tmpfs_rename(struct vop_rename_args *v)
784{
785	struct vnode *fdvp = v->a_fdvp;
786	struct vnode *fvp = v->a_fvp;
787	struct componentname *fcnp = v->a_fcnp;
788	struct vnode *tdvp = v->a_tdvp;
789	struct vnode *tvp = v->a_tvp;
790	struct componentname *tcnp = v->a_tcnp;
791	struct mount *mp = NULL;
792
793	char *newname;
794	int error;
795	struct tmpfs_dirent *de;
796	struct tmpfs_mount *tmp;
797	struct tmpfs_node *fdnode;
798	struct tmpfs_node *fnode;
799	struct tmpfs_node *tnode;
800	struct tmpfs_node *tdnode;
801
802	MPASS(VOP_ISLOCKED(tdvp));
803	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
804	MPASS(fcnp->cn_flags & HASBUF);
805	MPASS(tcnp->cn_flags & HASBUF);
806
807	/* Disallow cross-device renames.
808	 * XXX Why isn't this done by the caller? */
809	if (fvp->v_mount != tdvp->v_mount ||
810	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
811		error = EXDEV;
812		goto out;
813	}
814
815	/* If source and target are the same file, there is nothing to do. */
816	if (fvp == tvp) {
817		error = 0;
818		goto out;
819	}
820
821	/* If we need to move the directory between entries, lock the
822	 * source so that we can safely operate on it. */
823	if (fdvp != tdvp && fdvp != tvp) {
824		if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
825			mp = tdvp->v_mount;
826			error = vfs_busy(mp, 0);
827			if (error != 0) {
828				mp = NULL;
829				goto out;
830			}
831			error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp,
832			    fcnp, tcnp);
833			if (error != 0) {
834				vfs_unbusy(mp);
835				return (error);
836			}
837			ASSERT_VOP_ELOCKED(fdvp,
838			    "tmpfs_rename: fdvp not locked");
839			ASSERT_VOP_ELOCKED(tdvp,
840			    "tmpfs_rename: tdvp not locked");
841			if (tvp != NULL)
842				ASSERT_VOP_ELOCKED(tvp,
843				    "tmpfs_rename: tvp not locked");
844			if (fvp == tvp) {
845				error = 0;
846				goto out_locked;
847			}
848		}
849	}
850
851	tmp = VFS_TO_TMPFS(tdvp->v_mount);
852	tdnode = VP_TO_TMPFS_DIR(tdvp);
853	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
854	fdnode = VP_TO_TMPFS_DIR(fdvp);
855	fnode = VP_TO_TMPFS_NODE(fvp);
856	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
857
858	/* Entry can disappear before we lock fdvp,
859	 * also avoid manipulating '.' and '..' entries. */
860	if (de == NULL) {
861		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
862		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
863			error = EINVAL;
864		else
865			error = ENOENT;
866		goto out_locked;
867	}
868	MPASS(de->td_node == fnode);
869
870	/* If re-naming a directory to another preexisting directory
871	 * ensure that the target directory is empty so that its
872	 * removal causes no side effects.
873	 * Kern_rename guarantees the destination to be a directory
874	 * if the source is one. */
875	if (tvp != NULL) {
876		MPASS(tnode != NULL);
877
878		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
879		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
880			error = EPERM;
881			goto out_locked;
882		}
883
884		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
885			if (tnode->tn_size > 0) {
886				error = ENOTEMPTY;
887				goto out_locked;
888			}
889		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
890			error = ENOTDIR;
891			goto out_locked;
892		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
893			error = EISDIR;
894			goto out_locked;
895		} else {
896			MPASS(fnode->tn_type != VDIR &&
897				tnode->tn_type != VDIR);
898		}
899	}
900
901	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
902	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
903		error = EPERM;
904		goto out_locked;
905	}
906
907	/* Ensure that we have enough memory to hold the new name, if it
908	 * has to be changed. */
909	if (fcnp->cn_namelen != tcnp->cn_namelen ||
910	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
911		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
912	} else
913		newname = NULL;
914
915	/* If the node is being moved to another directory, we have to do
916	 * the move. */
917	if (fdnode != tdnode) {
918		/* In case we are moving a directory, we have to adjust its
919		 * parent to point to the new parent. */
920		if (de->td_node->tn_type == VDIR) {
921			struct tmpfs_node *n;
922
923			/* Ensure the target directory is not a child of the
924			 * directory being moved.  Otherwise, we'd end up
925			 * with stale nodes. */
926			n = tdnode;
927			/* TMPFS_LOCK garanties that no nodes are freed while
928			 * traversing the list. Nodes can only be marked as
929			 * removed: tn_parent == NULL. */
930			TMPFS_LOCK(tmp);
931			TMPFS_NODE_LOCK(n);
932			while (n != n->tn_dir.tn_parent) {
933				struct tmpfs_node *parent;
934
935				if (n == fnode) {
936					TMPFS_NODE_UNLOCK(n);
937					TMPFS_UNLOCK(tmp);
938					error = EINVAL;
939					if (newname != NULL)
940						    free(newname, M_TMPFSNAME);
941					goto out_locked;
942				}
943				parent = n->tn_dir.tn_parent;
944				TMPFS_NODE_UNLOCK(n);
945				if (parent == NULL) {
946					n = NULL;
947					break;
948				}
949				TMPFS_NODE_LOCK(parent);
950				if (parent->tn_dir.tn_parent == NULL) {
951					TMPFS_NODE_UNLOCK(parent);
952					n = NULL;
953					break;
954				}
955				n = parent;
956			}
957			TMPFS_UNLOCK(tmp);
958			if (n == NULL) {
959				error = EINVAL;
960				if (newname != NULL)
961					    free(newname, M_TMPFSNAME);
962				goto out_locked;
963			}
964			TMPFS_NODE_UNLOCK(n);
965
966			/* Adjust the parent pointer. */
967			TMPFS_VALIDATE_DIR(fnode);
968			TMPFS_NODE_LOCK(de->td_node);
969			de->td_node->tn_dir.tn_parent = tdnode;
970			TMPFS_NODE_UNLOCK(de->td_node);
971
972			/* As a result of changing the target of the '..'
973			 * entry, the link count of the source and target
974			 * directories has to be adjusted. */
975			TMPFS_NODE_LOCK(tdnode);
976			TMPFS_ASSERT_LOCKED(tdnode);
977			tdnode->tn_links++;
978			TMPFS_NODE_UNLOCK(tdnode);
979
980			TMPFS_NODE_LOCK(fdnode);
981			TMPFS_ASSERT_LOCKED(fdnode);
982			fdnode->tn_links--;
983			TMPFS_NODE_UNLOCK(fdnode);
984		}
985	}
986
987	/* Do the move: just remove the entry from the source directory
988	 * and insert it into the target one. */
989	tmpfs_dir_detach(fdvp, de);
990
991	if (fcnp->cn_flags & DOWHITEOUT)
992		tmpfs_dir_whiteout_add(fdvp, fcnp);
993	if (tcnp->cn_flags & ISWHITEOUT)
994		tmpfs_dir_whiteout_remove(tdvp, tcnp);
995
996	/* If the name has changed, we need to make it effective by changing
997	 * it in the directory entry. */
998	if (newname != NULL) {
999		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1000
1001		free(de->ud.td_name, M_TMPFSNAME);
1002		de->ud.td_name = newname;
1003		tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen);
1004
1005		fnode->tn_status |= TMPFS_NODE_CHANGED;
1006		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1007	}
1008
1009	/* If we are overwriting an entry, we have to remove the old one
1010	 * from the target directory. */
1011	if (tvp != NULL) {
1012		struct tmpfs_dirent *tde;
1013
1014		/* Remove the old entry from the target directory. */
1015		tde = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1016		tmpfs_dir_detach(tdvp, tde);
1017
1018		/* Free the directory entry we just deleted.  Note that the
1019		 * node referred by it will not be removed until the vnode is
1020		 * really reclaimed. */
1021		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1022	}
1023
1024	tmpfs_dir_attach(tdvp, de);
1025
1026	if (tmpfs_use_nc(fvp)) {
1027		cache_purge(fvp);
1028		if (tvp != NULL)
1029			cache_purge(tvp);
1030		cache_purge_negative(tdvp);
1031	}
1032
1033	error = 0;
1034
1035out_locked:
1036	if (fdvp != tdvp && fdvp != tvp)
1037		VOP_UNLOCK(fdvp, 0);
1038
1039out:
1040	/* Release target nodes. */
1041	/* XXX: I don't understand when tdvp can be the same as tvp, but
1042	 * other code takes care of this... */
1043	if (tdvp == tvp)
1044		vrele(tdvp);
1045	else
1046		vput(tdvp);
1047	if (tvp != NULL)
1048		vput(tvp);
1049
1050	/* Release source nodes. */
1051	vrele(fdvp);
1052	vrele(fvp);
1053
1054	if (mp != NULL)
1055		vfs_unbusy(mp);
1056
1057	return error;
1058}
1059
1060static int
1061tmpfs_mkdir(struct vop_mkdir_args *v)
1062{
1063	struct vnode *dvp = v->a_dvp;
1064	struct vnode **vpp = v->a_vpp;
1065	struct componentname *cnp = v->a_cnp;
1066	struct vattr *vap = v->a_vap;
1067
1068	MPASS(vap->va_type == VDIR);
1069
1070	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1071}
1072
1073static int
1074tmpfs_rmdir(struct vop_rmdir_args *v)
1075{
1076	struct vnode *dvp = v->a_dvp;
1077	struct vnode *vp = v->a_vp;
1078
1079	int error;
1080	struct tmpfs_dirent *de;
1081	struct tmpfs_mount *tmp;
1082	struct tmpfs_node *dnode;
1083	struct tmpfs_node *node;
1084
1085	MPASS(VOP_ISLOCKED(dvp));
1086	MPASS(VOP_ISLOCKED(vp));
1087
1088	tmp = VFS_TO_TMPFS(dvp->v_mount);
1089	dnode = VP_TO_TMPFS_DIR(dvp);
1090	node = VP_TO_TMPFS_DIR(vp);
1091
1092	/* Directories with more than two entries ('.' and '..') cannot be
1093	 * removed. */
1094	 if (node->tn_size > 0) {
1095		 error = ENOTEMPTY;
1096		 goto out;
1097	 }
1098
1099	if ((dnode->tn_flags & APPEND)
1100	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1101		error = EPERM;
1102		goto out;
1103	}
1104
1105	/* This invariant holds only if we are not trying to remove "..".
1106	  * We checked for that above so this is safe now. */
1107	MPASS(node->tn_dir.tn_parent == dnode);
1108
1109	/* Get the directory entry associated with node (vp).  This was
1110	 * filled by tmpfs_lookup while looking up the entry. */
1111	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1112	MPASS(TMPFS_DIRENT_MATCHES(de,
1113	    v->a_cnp->cn_nameptr,
1114	    v->a_cnp->cn_namelen));
1115
1116	/* Check flags to see if we are allowed to remove the directory. */
1117	if ((dnode->tn_flags & APPEND) != 0 ||
1118	    (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) != 0) {
1119		error = EPERM;
1120		goto out;
1121	}
1122
1123
1124	/* Detach the directory entry from the directory (dnode). */
1125	tmpfs_dir_detach(dvp, de);
1126	if (v->a_cnp->cn_flags & DOWHITEOUT)
1127		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1128
1129	/* No vnode should be allocated for this entry from this point */
1130	TMPFS_NODE_LOCK(node);
1131	node->tn_links--;
1132	node->tn_dir.tn_parent = NULL;
1133	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1134	    TMPFS_NODE_MODIFIED;
1135
1136	TMPFS_NODE_UNLOCK(node);
1137
1138	TMPFS_NODE_LOCK(dnode);
1139	dnode->tn_links--;
1140	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1141	    TMPFS_NODE_MODIFIED;
1142	TMPFS_NODE_UNLOCK(dnode);
1143
1144	if (tmpfs_use_nc(dvp)) {
1145		cache_purge(dvp);
1146		cache_purge(vp);
1147	}
1148
1149	/* Free the directory entry we just deleted.  Note that the node
1150	 * referred by it will not be removed until the vnode is really
1151	 * reclaimed. */
1152	tmpfs_free_dirent(tmp, de);
1153
1154	/* Release the deleted vnode (will destroy the node, notify
1155	 * interested parties and clean it from the cache). */
1156
1157	dnode->tn_status |= TMPFS_NODE_CHANGED;
1158	tmpfs_update(dvp);
1159
1160	error = 0;
1161
1162out:
1163	return error;
1164}
1165
1166static int
1167tmpfs_symlink(struct vop_symlink_args *v)
1168{
1169	struct vnode *dvp = v->a_dvp;
1170	struct vnode **vpp = v->a_vpp;
1171	struct componentname *cnp = v->a_cnp;
1172	struct vattr *vap = v->a_vap;
1173	char *target = v->a_target;
1174
1175#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1176	MPASS(vap->va_type == VLNK);
1177#else
1178	vap->va_type = VLNK;
1179#endif
1180
1181	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1182}
1183
1184static int
1185tmpfs_readdir(struct vop_readdir_args *va)
1186{
1187	struct vnode *vp;
1188	struct uio *uio;
1189	struct tmpfs_mount *tm;
1190	struct tmpfs_node *node;
1191	u_long **cookies;
1192	int *eofflag, *ncookies;
1193	ssize_t startresid;
1194	int error, maxcookies;
1195
1196	vp = va->a_vp;
1197	uio = va->a_uio;
1198	eofflag = va->a_eofflag;
1199	cookies = va->a_cookies;
1200	ncookies = va->a_ncookies;
1201
1202	/* This operation only makes sense on directory nodes. */
1203	if (vp->v_type != VDIR)
1204		return ENOTDIR;
1205
1206	maxcookies = 0;
1207	node = VP_TO_TMPFS_DIR(vp);
1208	tm = VFS_TO_TMPFS(vp->v_mount);
1209
1210	startresid = uio->uio_resid;
1211
1212	/* Allocate cookies for NFS and compat modules. */
1213	if (cookies != NULL && ncookies != NULL) {
1214		maxcookies = howmany(node->tn_size,
1215		    sizeof(struct tmpfs_dirent)) + 2;
1216		*cookies = malloc(maxcookies * sizeof(**cookies), M_TEMP,
1217		    M_WAITOK);
1218		*ncookies = 0;
1219	}
1220
1221	if (cookies == NULL)
1222		error = tmpfs_dir_getdents(tm, node, uio, 0, NULL, NULL);
1223	else
1224		error = tmpfs_dir_getdents(tm, node, uio, maxcookies, *cookies,
1225		    ncookies);
1226
1227	/* Buffer was filled without hitting EOF. */
1228	if (error == EJUSTRETURN)
1229		error = (uio->uio_resid != startresid) ? 0 : EINVAL;
1230
1231	if (error != 0 && cookies != NULL && ncookies != NULL) {
1232		free(*cookies, M_TEMP);
1233		*cookies = NULL;
1234		*ncookies = 0;
1235	}
1236
1237	if (eofflag != NULL)
1238		*eofflag =
1239		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1240
1241	return error;
1242}
1243
1244static int
1245tmpfs_readlink(struct vop_readlink_args *v)
1246{
1247	struct vnode *vp = v->a_vp;
1248	struct uio *uio = v->a_uio;
1249
1250	int error;
1251	struct tmpfs_node *node;
1252
1253	MPASS(uio->uio_offset == 0);
1254	MPASS(vp->v_type == VLNK);
1255
1256	node = VP_TO_TMPFS_NODE(vp);
1257
1258	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1259	    uio);
1260	tmpfs_set_status(VFS_TO_TMPFS(vp->v_mount), node, TMPFS_NODE_ACCESSED);
1261
1262	return (error);
1263}
1264
1265static int
1266tmpfs_inactive(struct vop_inactive_args *v)
1267{
1268	struct vnode *vp;
1269	struct tmpfs_node *node;
1270
1271	vp = v->a_vp;
1272	node = VP_TO_TMPFS_NODE(vp);
1273	if (node->tn_links == 0)
1274		vrecycle(vp);
1275	else
1276		tmpfs_check_mtime(vp);
1277	return (0);
1278}
1279
1280int
1281tmpfs_reclaim(struct vop_reclaim_args *v)
1282{
1283	struct vnode *vp = v->a_vp;
1284
1285	struct tmpfs_mount *tmp;
1286	struct tmpfs_node *node;
1287
1288	node = VP_TO_TMPFS_NODE(vp);
1289	tmp = VFS_TO_TMPFS(vp->v_mount);
1290
1291	if (vp->v_type == VREG)
1292		tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj);
1293	else
1294		vnode_destroy_vobject(vp);
1295	vp->v_object = NULL;
1296	if (tmpfs_use_nc(vp))
1297		cache_purge(vp);
1298
1299	TMPFS_NODE_LOCK(node);
1300	tmpfs_free_vp(vp);
1301
1302	/* If the node referenced by this vnode was deleted by the user,
1303	 * we must free its associated data structures (now that the vnode
1304	 * is being reclaimed). */
1305	if (node->tn_links == 0 &&
1306	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1307		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1308		TMPFS_NODE_UNLOCK(node);
1309		tmpfs_free_node(tmp, node);
1310	} else
1311		TMPFS_NODE_UNLOCK(node);
1312
1313	MPASS(vp->v_data == NULL);
1314	return 0;
1315}
1316
1317int
1318tmpfs_print(struct vop_print_args *v)
1319{
1320	struct vnode *vp = v->a_vp;
1321
1322	struct tmpfs_node *node;
1323
1324	node = VP_TO_TMPFS_NODE(vp);
1325
1326	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %jd\n",
1327	    node, node->tn_flags, (uintmax_t)node->tn_links);
1328	printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n",
1329	    node->tn_mode, node->tn_uid, node->tn_gid,
1330	    (intmax_t)node->tn_size, node->tn_status);
1331
1332	if (vp->v_type == VFIFO)
1333		fifo_printinfo(vp);
1334
1335	printf("\n");
1336
1337	return 0;
1338}
1339
1340int
1341tmpfs_pathconf(struct vop_pathconf_args *v)
1342{
1343	struct vnode *vp = v->a_vp;
1344	int name = v->a_name;
1345	register_t *retval = v->a_retval;
1346
1347	int error;
1348
1349	error = 0;
1350
1351	switch (name) {
1352	case _PC_LINK_MAX:
1353		*retval = LINK_MAX;
1354		break;
1355
1356	case _PC_NAME_MAX:
1357		*retval = NAME_MAX;
1358		break;
1359
1360	case _PC_PIPE_BUF:
1361		if (vp->v_type == VDIR || vp->v_type == VFIFO)
1362			*retval = PIPE_BUF;
1363		else
1364			error = EINVAL;
1365		break;
1366
1367	case _PC_CHOWN_RESTRICTED:
1368		*retval = 1;
1369		break;
1370
1371	case _PC_NO_TRUNC:
1372		*retval = 1;
1373		break;
1374
1375	case _PC_SYNC_IO:
1376		*retval = 1;
1377		break;
1378
1379	case _PC_FILESIZEBITS:
1380		*retval = 64;
1381		break;
1382
1383	default:
1384		error = vop_stdpathconf(v);
1385	}
1386
1387	return error;
1388}
1389
1390static int
1391tmpfs_vptofh(struct vop_vptofh_args *ap)
1392{
1393	struct tmpfs_fid *tfhp;
1394	struct tmpfs_node *node;
1395
1396	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1397	node = VP_TO_TMPFS_NODE(ap->a_vp);
1398
1399	tfhp->tf_len = sizeof(struct tmpfs_fid);
1400	tfhp->tf_id = node->tn_id;
1401	tfhp->tf_gen = node->tn_gen;
1402
1403	return (0);
1404}
1405
1406static int
1407tmpfs_whiteout(struct vop_whiteout_args *ap)
1408{
1409	struct vnode *dvp = ap->a_dvp;
1410	struct componentname *cnp = ap->a_cnp;
1411	struct tmpfs_dirent *de;
1412
1413	switch (ap->a_flags) {
1414	case LOOKUP:
1415		return (0);
1416	case CREATE:
1417		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1418		if (de != NULL)
1419			return (de->td_node == NULL ? 0 : EEXIST);
1420		return (tmpfs_dir_whiteout_add(dvp, cnp));
1421	case DELETE:
1422		tmpfs_dir_whiteout_remove(dvp, cnp);
1423		return (0);
1424	default:
1425		panic("tmpfs_whiteout: unknown op");
1426	}
1427}
1428
1429static int
1430tmpfs_vptocnp_dir(struct tmpfs_node *tn, struct tmpfs_node *tnp,
1431    struct tmpfs_dirent **pde)
1432{
1433	struct tmpfs_dir_cursor dc;
1434	struct tmpfs_dirent *de;
1435
1436	for (de = tmpfs_dir_first(tnp, &dc); de != NULL;
1437	     de = tmpfs_dir_next(tnp, &dc)) {
1438		if (de->td_node == tn) {
1439			*pde = de;
1440			return (0);
1441		}
1442	}
1443	return (ENOENT);
1444}
1445
1446static int
1447tmpfs_vptocnp_fill(struct vnode *vp, struct tmpfs_node *tn,
1448    struct tmpfs_node *tnp, char *buf, int *buflen, struct vnode **dvp)
1449{
1450	struct tmpfs_dirent *de;
1451	int error, i;
1452
1453	error = vn_vget_ino_gen(vp, tmpfs_vn_get_ino_alloc, tnp, LK_SHARED,
1454	    dvp);
1455	if (error != 0)
1456		return (error);
1457	error = tmpfs_vptocnp_dir(tn, tnp, &de);
1458	if (error == 0) {
1459		i = *buflen;
1460		i -= de->td_namelen;
1461		if (i < 0) {
1462			error = ENOMEM;
1463		} else {
1464			bcopy(de->ud.td_name, buf + i, de->td_namelen);
1465			*buflen = i;
1466		}
1467	}
1468	if (error == 0) {
1469		if (vp != *dvp)
1470			VOP_UNLOCK(*dvp, 0);
1471	} else {
1472		if (vp != *dvp)
1473			vput(*dvp);
1474		else
1475			vrele(vp);
1476	}
1477	return (error);
1478}
1479
1480static int
1481tmpfs_vptocnp(struct vop_vptocnp_args *ap)
1482{
1483	struct vnode *vp, **dvp;
1484	struct tmpfs_node *tn, *tnp, *tnp1;
1485	struct tmpfs_dirent *de;
1486	struct tmpfs_mount *tm;
1487	char *buf;
1488	int *buflen;
1489	int error;
1490
1491	vp = ap->a_vp;
1492	dvp = ap->a_vpp;
1493	buf = ap->a_buf;
1494	buflen = ap->a_buflen;
1495
1496	tm = VFS_TO_TMPFS(vp->v_mount);
1497	tn = VP_TO_TMPFS_NODE(vp);
1498	if (tn->tn_type == VDIR) {
1499		tnp = tn->tn_dir.tn_parent;
1500		if (tnp == NULL)
1501			return (ENOENT);
1502		tmpfs_ref_node(tnp);
1503		error = tmpfs_vptocnp_fill(vp, tn, tn->tn_dir.tn_parent, buf,
1504		    buflen, dvp);
1505		tmpfs_free_node(tm, tnp);
1506		return (error);
1507	}
1508restart:
1509	TMPFS_LOCK(tm);
1510	LIST_FOREACH_SAFE(tnp, &tm->tm_nodes_used, tn_entries, tnp1) {
1511		if (tnp->tn_type != VDIR)
1512			continue;
1513		TMPFS_NODE_LOCK(tnp);
1514		tmpfs_ref_node_locked(tnp);
1515
1516		/*
1517		 * tn_vnode cannot be instantiated while we hold the
1518		 * node lock, so the directory cannot be changed while
1519		 * we iterate over it.  Do this to avoid instantiating
1520		 * vnode for directories which cannot point to our
1521		 * node.
1522		 */
1523		error = tnp->tn_vnode == NULL ? tmpfs_vptocnp_dir(tn, tnp,
1524		    &de) : 0;
1525
1526		if (error == 0) {
1527			TMPFS_NODE_UNLOCK(tnp);
1528			TMPFS_UNLOCK(tm);
1529			error = tmpfs_vptocnp_fill(vp, tn, tnp, buf, buflen,
1530			    dvp);
1531			if (error == 0) {
1532				tmpfs_free_node(tm, tnp);
1533				return (0);
1534			}
1535			if ((vp->v_iflag & VI_DOOMED) != 0) {
1536				tmpfs_free_node(tm, tnp);
1537				return (ENOENT);
1538			}
1539			TMPFS_LOCK(tm);
1540			TMPFS_NODE_LOCK(tnp);
1541		}
1542		if (tmpfs_free_node_locked(tm, tnp, false)) {
1543			goto restart;
1544		} else {
1545			KASSERT(tnp->tn_refcount > 0,
1546			    ("node %p refcount zero", tnp));
1547			tnp1 = LIST_NEXT(tnp, tn_entries);
1548			TMPFS_NODE_UNLOCK(tnp);
1549		}
1550	}
1551	TMPFS_UNLOCK(tm);
1552	return (ENOENT);
1553}
1554
1555/*
1556 * Vnode operations vector used for files stored in a tmpfs file system.
1557 */
1558struct vop_vector tmpfs_vnodeop_entries = {
1559	.vop_default =			&default_vnodeops,
1560	.vop_lookup =			vfs_cache_lookup,
1561	.vop_cachedlookup =		tmpfs_cached_lookup,
1562	.vop_create =			tmpfs_create,
1563	.vop_mknod =			tmpfs_mknod,
1564	.vop_open =			tmpfs_open,
1565	.vop_close =			tmpfs_close,
1566	.vop_access =			tmpfs_access,
1567	.vop_getattr =			tmpfs_getattr,
1568	.vop_setattr =			tmpfs_setattr,
1569	.vop_read =			tmpfs_read,
1570	.vop_write =			tmpfs_write,
1571	.vop_fsync =			tmpfs_fsync,
1572	.vop_remove =			tmpfs_remove,
1573	.vop_link =			tmpfs_link,
1574	.vop_rename =			tmpfs_rename,
1575	.vop_mkdir =			tmpfs_mkdir,
1576	.vop_rmdir =			tmpfs_rmdir,
1577	.vop_symlink =			tmpfs_symlink,
1578	.vop_readdir =			tmpfs_readdir,
1579	.vop_readlink =			tmpfs_readlink,
1580	.vop_inactive =			tmpfs_inactive,
1581	.vop_reclaim =			tmpfs_reclaim,
1582	.vop_print =			tmpfs_print,
1583	.vop_pathconf =			tmpfs_pathconf,
1584	.vop_vptofh =			tmpfs_vptofh,
1585	.vop_whiteout =			tmpfs_whiteout,
1586	.vop_bmap =			VOP_EOPNOTSUPP,
1587	.vop_vptocnp =			tmpfs_vptocnp,
1588};
1589
1590/*
1591 * Same vector for mounts which do not use namecache.
1592 */
1593struct vop_vector tmpfs_vnodeop_nonc_entries = {
1594	.vop_default =			&tmpfs_vnodeop_entries,
1595	.vop_lookup =			tmpfs_lookup,
1596};
1597