tmpfs_vnops.c revision 253927
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: head/sys/fs/tmpfs/tmpfs_vnops.c 253927 2013-08-04 15:56:19Z attilio $");
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/sf_buf.h>
49#include <sys/stat.h>
50#include <sys/systm.h>
51#include <sys/sysctl.h>
52#include <sys/unistd.h>
53#include <sys/vnode.h>
54
55#include <vm/vm.h>
56#include <vm/vm_param.h>
57#include <vm/vm_object.h>
58#include <vm/vm_page.h>
59#include <vm/vm_pager.h>
60
61#include <fs/tmpfs/tmpfs_vnops.h>
62#include <fs/tmpfs/tmpfs.h>
63
64SYSCTL_DECL(_vfs_tmpfs);
65
66static volatile int tmpfs_rename_restarts;
67SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD,
68    __DEVOLATILE(int *, &tmpfs_rename_restarts), 0,
69    "Times rename had to restart due to lock contention");
70
71/* --------------------------------------------------------------------- */
72
73static int
74tmpfs_lookup(struct vop_cachedlookup_args *v)
75{
76	struct vnode *dvp = v->a_dvp;
77	struct vnode **vpp = v->a_vpp;
78	struct componentname *cnp = v->a_cnp;
79
80	int error;
81	struct tmpfs_dirent *de;
82	struct tmpfs_node *dnode;
83
84	dnode = VP_TO_TMPFS_DIR(dvp);
85	*vpp = NULLVP;
86
87	/* Check accessibility of requested node as a first step. */
88	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
89	if (error != 0)
90		goto out;
91
92	/* We cannot be requesting the parent directory of the root node. */
93	MPASS(IMPLIES(dnode->tn_type == VDIR &&
94	    dnode->tn_dir.tn_parent == dnode,
95	    !(cnp->cn_flags & ISDOTDOT)));
96
97	TMPFS_ASSERT_LOCKED(dnode);
98	if (dnode->tn_dir.tn_parent == NULL) {
99		error = ENOENT;
100		goto out;
101	}
102	if (cnp->cn_flags & ISDOTDOT) {
103		int ltype = 0;
104
105		ltype = VOP_ISLOCKED(dvp);
106		vhold(dvp);
107		VOP_UNLOCK(dvp, 0);
108		/* Allocate a new vnode on the matching entry. */
109		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
110		    cnp->cn_lkflags, vpp);
111
112		vn_lock(dvp, ltype | LK_RETRY);
113		vdrop(dvp);
114	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
115		VREF(dvp);
116		*vpp = dvp;
117		error = 0;
118	} else {
119		de = tmpfs_dir_lookup(dnode, NULL, cnp);
120		if (de != NULL && de->td_node == NULL)
121			cnp->cn_flags |= ISWHITEOUT;
122		if (de == NULL || de->td_node == NULL) {
123			/* The entry was not found in the directory.
124			 * This is OK if we are creating or renaming an
125			 * entry and are working on the last component of
126			 * the path name. */
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				/* Keep the component name in the buffer for
139				 * future uses. */
140				cnp->cn_flags |= SAVENAME;
141
142				error = EJUSTRETURN;
143			} else
144				error = ENOENT;
145		} else {
146			struct tmpfs_node *tnode;
147
148			/* The entry was found, so get its associated
149			 * tmpfs_node. */
150			tnode = de->td_node;
151
152			/* If we are not at the last path component and
153			 * found a non-directory or non-link entry (which
154			 * may itself be pointing to a directory), raise
155			 * an error. */
156			if ((tnode->tn_type != VDIR &&
157			    tnode->tn_type != VLNK) &&
158			    !(cnp->cn_flags & ISLASTCN)) {
159				error = ENOTDIR;
160				goto out;
161			}
162
163			/* If we are deleting or renaming the entry, keep
164			 * track of its tmpfs_dirent so that it can be
165			 * easily deleted later. */
166			if ((cnp->cn_flags & ISLASTCN) &&
167			    (cnp->cn_nameiop == DELETE ||
168			    cnp->cn_nameiop == RENAME)) {
169				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
170				    cnp->cn_thread);
171				if (error != 0)
172					goto out;
173
174				/* Allocate a new vnode on the matching entry. */
175				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
176						cnp->cn_lkflags, vpp);
177				if (error != 0)
178					goto out;
179
180				if ((dnode->tn_mode & S_ISTXT) &&
181				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
182				  VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
183					error = EPERM;
184					vput(*vpp);
185					*vpp = NULL;
186					goto out;
187				}
188				cnp->cn_flags |= SAVENAME;
189			} else {
190				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
191						cnp->cn_lkflags, vpp);
192			}
193		}
194	}
195
196	/* Store the result of this lookup in the cache.  Avoid this if the
197	 * request was for creation, as it does not improve timings on
198	 * emprical tests. */
199	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
200		cache_enter(dvp, *vpp, cnp);
201
202out:
203	/* If there were no errors, *vpp cannot be null and it must be
204	 * locked. */
205	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
206
207	return error;
208}
209
210/* --------------------------------------------------------------------- */
211
212static int
213tmpfs_create(struct vop_create_args *v)
214{
215	struct vnode *dvp = v->a_dvp;
216	struct vnode **vpp = v->a_vpp;
217	struct componentname *cnp = v->a_cnp;
218	struct vattr *vap = v->a_vap;
219
220	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
221
222	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
223}
224/* --------------------------------------------------------------------- */
225
226static int
227tmpfs_mknod(struct vop_mknod_args *v)
228{
229	struct vnode *dvp = v->a_dvp;
230	struct vnode **vpp = v->a_vpp;
231	struct componentname *cnp = v->a_cnp;
232	struct vattr *vap = v->a_vap;
233
234	if (vap->va_type != VBLK && vap->va_type != VCHR &&
235	    vap->va_type != VFIFO)
236		return EINVAL;
237
238	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
239}
240
241/* --------------------------------------------------------------------- */
242
243static int
244tmpfs_open(struct vop_open_args *v)
245{
246	struct vnode *vp = v->a_vp;
247	int mode = v->a_mode;
248
249	int error;
250	struct tmpfs_node *node;
251
252	MPASS(VOP_ISLOCKED(vp));
253
254	node = VP_TO_TMPFS_NODE(vp);
255
256	/* The file is still active but all its names have been removed
257	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
258	 * it is about to die. */
259	if (node->tn_links < 1)
260		return (ENOENT);
261
262	/* If the file is marked append-only, deny write requests. */
263	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
264		error = EPERM;
265	else {
266		error = 0;
267		/* For regular files, the call below is nop. */
268		vnode_create_vobject(vp, node->tn_size, v->a_td);
269	}
270
271	MPASS(VOP_ISLOCKED(vp));
272	return error;
273}
274
275/* --------------------------------------------------------------------- */
276
277static int
278tmpfs_close(struct vop_close_args *v)
279{
280	struct vnode *vp = v->a_vp;
281
282	/* Update node times. */
283	tmpfs_update(vp);
284
285	return (0);
286}
287
288/* --------------------------------------------------------------------- */
289
290int
291tmpfs_access(struct vop_access_args *v)
292{
293	struct vnode *vp = v->a_vp;
294	accmode_t accmode = v->a_accmode;
295	struct ucred *cred = v->a_cred;
296
297	int error;
298	struct tmpfs_node *node;
299
300	MPASS(VOP_ISLOCKED(vp));
301
302	node = VP_TO_TMPFS_NODE(vp);
303
304	switch (vp->v_type) {
305	case VDIR:
306		/* FALLTHROUGH */
307	case VLNK:
308		/* FALLTHROUGH */
309	case VREG:
310		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
311			error = EROFS;
312			goto out;
313		}
314		break;
315
316	case VBLK:
317		/* FALLTHROUGH */
318	case VCHR:
319		/* FALLTHROUGH */
320	case VSOCK:
321		/* FALLTHROUGH */
322	case VFIFO:
323		break;
324
325	default:
326		error = EINVAL;
327		goto out;
328	}
329
330	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
331		error = EPERM;
332		goto out;
333	}
334
335	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
336	    node->tn_gid, accmode, cred, NULL);
337
338out:
339	MPASS(VOP_ISLOCKED(vp));
340
341	return error;
342}
343
344/* --------------------------------------------------------------------- */
345
346int
347tmpfs_getattr(struct vop_getattr_args *v)
348{
349	struct vnode *vp = v->a_vp;
350	struct vattr *vap = v->a_vap;
351
352	struct tmpfs_node *node;
353
354	node = VP_TO_TMPFS_NODE(vp);
355
356	tmpfs_update(vp);
357
358	vap->va_type = vp->v_type;
359	vap->va_mode = node->tn_mode;
360	vap->va_nlink = node->tn_links;
361	vap->va_uid = node->tn_uid;
362	vap->va_gid = node->tn_gid;
363	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
364	vap->va_fileid = node->tn_id;
365	vap->va_size = node->tn_size;
366	vap->va_blocksize = PAGE_SIZE;
367	vap->va_atime = node->tn_atime;
368	vap->va_mtime = node->tn_mtime;
369	vap->va_ctime = node->tn_ctime;
370	vap->va_birthtime = node->tn_birthtime;
371	vap->va_gen = node->tn_gen;
372	vap->va_flags = node->tn_flags;
373	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
374		node->tn_rdev : NODEV;
375	vap->va_bytes = round_page(node->tn_size);
376	vap->va_filerev = 0;
377
378	return 0;
379}
380
381/* --------------------------------------------------------------------- */
382
383/* XXX Should this operation be atomic?  I think it should, but code in
384 * XXX other places (e.g., ufs) doesn't seem to be... */
385int
386tmpfs_setattr(struct vop_setattr_args *v)
387{
388	struct vnode *vp = v->a_vp;
389	struct vattr *vap = v->a_vap;
390	struct ucred *cred = v->a_cred;
391	struct thread *td = curthread;
392
393	int error;
394
395	MPASS(VOP_ISLOCKED(vp));
396
397	error = 0;
398
399	/* Abort if any unsettable attribute is given. */
400	if (vap->va_type != VNON ||
401	    vap->va_nlink != VNOVAL ||
402	    vap->va_fsid != VNOVAL ||
403	    vap->va_fileid != VNOVAL ||
404	    vap->va_blocksize != VNOVAL ||
405	    vap->va_gen != VNOVAL ||
406	    vap->va_rdev != VNOVAL ||
407	    vap->va_bytes != VNOVAL)
408		error = EINVAL;
409
410	if (error == 0 && (vap->va_flags != VNOVAL))
411		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
412
413	if (error == 0 && (vap->va_size != VNOVAL))
414		error = tmpfs_chsize(vp, vap->va_size, cred, td);
415
416	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
417		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
418
419	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
420		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
421
422	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
423	    vap->va_atime.tv_nsec != VNOVAL) ||
424	    (vap->va_mtime.tv_sec != VNOVAL &&
425	    vap->va_mtime.tv_nsec != VNOVAL) ||
426	    (vap->va_birthtime.tv_sec != VNOVAL &&
427	    vap->va_birthtime.tv_nsec != VNOVAL)))
428		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
429			&vap->va_birthtime, vap->va_vaflags, cred, td);
430
431	/* Update the node times.  We give preference to the error codes
432	 * generated by this function rather than the ones that may arise
433	 * from tmpfs_update. */
434	tmpfs_update(vp);
435
436	MPASS(VOP_ISLOCKED(vp));
437
438	return error;
439}
440
441static int
442tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
443    vm_offset_t offset, size_t tlen, struct uio *uio)
444{
445	vm_page_t	m;
446	int		error, rv;
447
448	VM_OBJECT_WLOCK(tobj);
449
450	/*
451	 * Parallel reads of the page content from disk are prevented
452	 * by VPO_BUSY.
453	 *
454	 * Although the tmpfs vnode lock is held here, it is
455	 * nonetheless safe to sleep waiting for a free page.  The
456	 * pageout daemon does not need to acquire the tmpfs vnode
457	 * lock to page out tobj's pages because tobj is a OBJT_SWAP
458	 * type object.
459	 */
460	m = vm_page_grab(tobj, idx, VM_ALLOC_NORMAL | VM_ALLOC_RETRY |
461	    VM_ALLOC_NOBUSY);
462	if (m->valid != VM_PAGE_BITS_ALL) {
463		vm_page_busy(m);
464		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
465			rv = vm_pager_get_pages(tobj, &m, 1, 0);
466			m = vm_page_lookup(tobj, idx);
467			if (m == NULL) {
468				printf(
469		    "tmpfs: vm_obj %p idx %jd null lookup rv %d\n",
470				    tobj, idx, rv);
471				VM_OBJECT_WUNLOCK(tobj);
472				return (EIO);
473			}
474			if (rv != VM_PAGER_OK) {
475				printf(
476		    "tmpfs: vm_obj %p idx %jd valid %x pager error %d\n",
477				    tobj, idx, m->valid, rv);
478				vm_page_lock(m);
479				vm_page_free(m);
480				vm_page_unlock(m);
481				VM_OBJECT_WUNLOCK(tobj);
482				return (EIO);
483			}
484		} else
485			vm_page_zero_invalid(m, TRUE);
486		vm_page_wakeup(m);
487	}
488	vm_page_lock(m);
489	vm_page_hold(m);
490	vm_page_unlock(m);
491	VM_OBJECT_WUNLOCK(tobj);
492	error = uiomove_fromphys(&m, offset, tlen, uio);
493	vm_page_lock(m);
494	vm_page_unhold(m);
495	if (m->queue == PQ_NONE) {
496		vm_page_deactivate(m);
497	} else {
498		/* Requeue to maintain LRU ordering. */
499		vm_page_requeue(m);
500	}
501	vm_page_unlock(m);
502
503	return (error);
504}
505
506static int
507tmpfs_read(struct vop_read_args *v)
508{
509	struct vnode *vp = v->a_vp;
510	struct uio *uio = v->a_uio;
511	struct tmpfs_node *node;
512	vm_object_t uobj;
513	size_t len;
514	int resid;
515	int error = 0;
516	vm_pindex_t	idx;
517	vm_offset_t	offset;
518	off_t		addr;
519	size_t		tlen;
520
521	node = VP_TO_TMPFS_NODE(vp);
522
523	if (vp->v_type != VREG) {
524		error = EISDIR;
525		goto out;
526	}
527
528	if (uio->uio_offset < 0) {
529		error = EINVAL;
530		goto out;
531	}
532
533	node->tn_status |= TMPFS_NODE_ACCESSED;
534
535	uobj = node->tn_reg.tn_aobj;
536	while ((resid = uio->uio_resid) > 0) {
537		error = 0;
538		if (node->tn_size <= uio->uio_offset)
539			break;
540		len = MIN(node->tn_size - uio->uio_offset, resid);
541		if (len == 0)
542			break;
543		addr = uio->uio_offset;
544		idx = OFF_TO_IDX(addr);
545		offset = addr & PAGE_MASK;
546		tlen = MIN(PAGE_SIZE - offset, len);
547		error = tmpfs_nocacheread(uobj, idx, offset, tlen, uio);
548		if ((error != 0) || (resid == uio->uio_resid))
549			break;
550	}
551
552out:
553
554	return error;
555}
556
557/* --------------------------------------------------------------------- */
558
559static int
560tmpfs_mappedwrite(vm_object_t tobj, size_t len, struct uio *uio)
561{
562	vm_pindex_t	idx;
563	vm_page_t	tpg;
564	vm_offset_t	offset;
565	off_t		addr;
566	size_t		tlen;
567	int		error, rv;
568
569	error = 0;
570
571	addr = uio->uio_offset;
572	idx = OFF_TO_IDX(addr);
573	offset = addr & PAGE_MASK;
574	tlen = MIN(PAGE_SIZE - offset, len);
575
576	VM_OBJECT_WLOCK(tobj);
577	tpg = vm_page_grab(tobj, idx, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
578	    VM_ALLOC_RETRY);
579	if (tpg->valid != VM_PAGE_BITS_ALL) {
580		vm_page_busy(tpg);
581		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
582			rv = vm_pager_get_pages(tobj, &tpg, 1, 0);
583			tpg = vm_page_lookup(tobj, idx);
584			if (tpg == NULL) {
585				printf(
586		    "tmpfs: vm_obj %p idx %jd null lookup rv %d\n",
587				    tobj, idx, rv);
588				VM_OBJECT_WUNLOCK(tobj);
589				return (EIO);
590			}
591			if (rv != VM_PAGER_OK) {
592				printf(
593		    "tmpfs: vm_obj %p idx %jd valid %x pager error %d\n",
594				    tobj, idx, tpg->valid, rv);
595				vm_page_lock(tpg);
596				vm_page_free(tpg);
597				vm_page_unlock(tpg);
598				VM_OBJECT_WUNLOCK(tobj);
599				return (EIO);
600			}
601		} else
602			vm_page_zero_invalid(tpg, TRUE);
603		vm_page_wakeup(tpg);
604	}
605	vm_page_lock(tpg);
606	vm_page_hold(tpg);
607	vm_page_unlock(tpg);
608	VM_OBJECT_WUNLOCK(tobj);
609	error = uiomove_fromphys(&tpg, offset, tlen, uio);
610	VM_OBJECT_WLOCK(tobj);
611	if (error == 0)
612		vm_page_dirty(tpg);
613	vm_page_lock(tpg);
614	vm_page_unhold(tpg);
615	if (tpg->queue == PQ_NONE) {
616		vm_page_deactivate(tpg);
617	} else {
618		/* Requeue to maintain LRU ordering. */
619		vm_page_requeue(tpg);
620	}
621	vm_page_unlock(tpg);
622	VM_OBJECT_WUNLOCK(tobj);
623
624	return	(error);
625}
626
627static int
628tmpfs_write(struct vop_write_args *v)
629{
630	struct vnode *vp = v->a_vp;
631	struct uio *uio = v->a_uio;
632	int ioflag = v->a_ioflag;
633
634	boolean_t extended;
635	int error = 0;
636	off_t oldsize;
637	struct tmpfs_node *node;
638	vm_object_t uobj;
639	size_t len;
640	int resid;
641
642	node = VP_TO_TMPFS_NODE(vp);
643	oldsize = node->tn_size;
644
645	if (uio->uio_offset < 0 || vp->v_type != VREG) {
646		error = EINVAL;
647		goto out;
648	}
649
650	if (uio->uio_resid == 0) {
651		error = 0;
652		goto out;
653	}
654
655	if (ioflag & IO_APPEND)
656		uio->uio_offset = node->tn_size;
657
658	if (uio->uio_offset + uio->uio_resid >
659	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
660		return (EFBIG);
661
662	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
663		return (EFBIG);
664
665	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
666	if (extended) {
667		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
668		    FALSE);
669		if (error != 0)
670			goto out;
671	}
672
673	uobj = node->tn_reg.tn_aobj;
674	while ((resid = uio->uio_resid) > 0) {
675		if (node->tn_size <= uio->uio_offset)
676			break;
677		len = MIN(node->tn_size - uio->uio_offset, resid);
678		if (len == 0)
679			break;
680		error = tmpfs_mappedwrite(uobj, len, uio);
681		if ((error != 0) || (resid == uio->uio_resid))
682			break;
683	}
684
685	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
686	    (extended ? TMPFS_NODE_CHANGED : 0);
687
688	if (node->tn_mode & (S_ISUID | S_ISGID)) {
689		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
690			node->tn_mode &= ~(S_ISUID | S_ISGID);
691	}
692
693	if (error != 0)
694		(void)tmpfs_reg_resize(vp, oldsize, TRUE);
695
696out:
697	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
698	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
699
700	return error;
701}
702
703/* --------------------------------------------------------------------- */
704
705static int
706tmpfs_fsync(struct vop_fsync_args *v)
707{
708	struct vnode *vp = v->a_vp;
709
710	MPASS(VOP_ISLOCKED(vp));
711
712	tmpfs_update(vp);
713
714	return 0;
715}
716
717/* --------------------------------------------------------------------- */
718
719static int
720tmpfs_remove(struct vop_remove_args *v)
721{
722	struct vnode *dvp = v->a_dvp;
723	struct vnode *vp = v->a_vp;
724
725	int error;
726	struct tmpfs_dirent *de;
727	struct tmpfs_mount *tmp;
728	struct tmpfs_node *dnode;
729	struct tmpfs_node *node;
730
731	MPASS(VOP_ISLOCKED(dvp));
732	MPASS(VOP_ISLOCKED(vp));
733
734	if (vp->v_type == VDIR) {
735		error = EISDIR;
736		goto out;
737	}
738
739	dnode = VP_TO_TMPFS_DIR(dvp);
740	node = VP_TO_TMPFS_NODE(vp);
741	tmp = VFS_TO_TMPFS(vp->v_mount);
742	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
743	MPASS(de != NULL);
744
745	/* Files marked as immutable or append-only cannot be deleted. */
746	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
747	    (dnode->tn_flags & APPEND)) {
748		error = EPERM;
749		goto out;
750	}
751
752	/* Remove the entry from the directory; as it is a file, we do not
753	 * have to change the number of hard links of the directory. */
754	tmpfs_dir_detach(dvp, de);
755	if (v->a_cnp->cn_flags & DOWHITEOUT)
756		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
757
758	/* Free the directory entry we just deleted.  Note that the node
759	 * referred by it will not be removed until the vnode is really
760	 * reclaimed. */
761	tmpfs_free_dirent(tmp, de);
762
763	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
764	error = 0;
765
766out:
767
768	return error;
769}
770
771/* --------------------------------------------------------------------- */
772
773static int
774tmpfs_link(struct vop_link_args *v)
775{
776	struct vnode *dvp = v->a_tdvp;
777	struct vnode *vp = v->a_vp;
778	struct componentname *cnp = v->a_cnp;
779
780	int error;
781	struct tmpfs_dirent *de;
782	struct tmpfs_node *node;
783
784	MPASS(VOP_ISLOCKED(dvp));
785	MPASS(cnp->cn_flags & HASBUF);
786	MPASS(dvp != vp); /* XXX When can this be false? */
787
788	node = VP_TO_TMPFS_NODE(vp);
789
790	/* XXX: Why aren't the following two tests done by the caller? */
791
792	/* Hard links of directories are forbidden. */
793	if (vp->v_type == VDIR) {
794		error = EPERM;
795		goto out;
796	}
797
798	/* Cannot create cross-device links. */
799	if (dvp->v_mount != vp->v_mount) {
800		error = EXDEV;
801		goto out;
802	}
803
804	/* Ensure that we do not overflow the maximum number of links imposed
805	 * by the system. */
806	MPASS(node->tn_links <= LINK_MAX);
807	if (node->tn_links == LINK_MAX) {
808		error = EMLINK;
809		goto out;
810	}
811
812	/* We cannot create links of files marked immutable or append-only. */
813	if (node->tn_flags & (IMMUTABLE | APPEND)) {
814		error = EPERM;
815		goto out;
816	}
817
818	/* Allocate a new directory entry to represent the node. */
819	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
820	    cnp->cn_nameptr, cnp->cn_namelen, &de);
821	if (error != 0)
822		goto out;
823
824	/* Insert the new directory entry into the appropriate directory. */
825	if (cnp->cn_flags & ISWHITEOUT)
826		tmpfs_dir_whiteout_remove(dvp, cnp);
827	tmpfs_dir_attach(dvp, de);
828
829	/* vp link count has changed, so update node times. */
830	node->tn_status |= TMPFS_NODE_CHANGED;
831	tmpfs_update(vp);
832
833	error = 0;
834
835out:
836	return error;
837}
838
839/* --------------------------------------------------------------------- */
840
841/*
842 * We acquire all but fdvp locks using non-blocking acquisitions.  If we
843 * fail to acquire any lock in the path we will drop all held locks,
844 * acquire the new lock in a blocking fashion, and then release it and
845 * restart the rename.  This acquire/release step ensures that we do not
846 * spin on a lock waiting for release.  On error release all vnode locks
847 * and decrement references the way tmpfs_rename() would do.
848 */
849static int
850tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp,
851    struct vnode *tdvp, struct vnode **tvpp,
852    struct componentname *fcnp, struct componentname *tcnp)
853{
854	struct vnode *nvp;
855	struct mount *mp;
856	struct tmpfs_dirent *de;
857	int error, restarts = 0;
858
859	VOP_UNLOCK(tdvp, 0);
860	if (*tvpp != NULL && *tvpp != tdvp)
861		VOP_UNLOCK(*tvpp, 0);
862	mp = fdvp->v_mount;
863
864relock:
865	restarts += 1;
866	error = vn_lock(fdvp, LK_EXCLUSIVE);
867	if (error)
868		goto releout;
869	if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
870		VOP_UNLOCK(fdvp, 0);
871		error = vn_lock(tdvp, LK_EXCLUSIVE);
872		if (error)
873			goto releout;
874		VOP_UNLOCK(tdvp, 0);
875		goto relock;
876	}
877	/*
878	 * Re-resolve fvp to be certain it still exists and fetch the
879	 * correct vnode.
880	 */
881	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp);
882	if (de == NULL) {
883		VOP_UNLOCK(fdvp, 0);
884		VOP_UNLOCK(tdvp, 0);
885		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
886		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
887			error = EINVAL;
888		else
889			error = ENOENT;
890		goto releout;
891	}
892	error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
893	if (error != 0) {
894		VOP_UNLOCK(fdvp, 0);
895		VOP_UNLOCK(tdvp, 0);
896		if (error != EBUSY)
897			goto releout;
898		error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp);
899		if (error != 0)
900			goto releout;
901		VOP_UNLOCK(nvp, 0);
902		/*
903		 * Concurrent rename race.
904		 */
905		if (nvp == tdvp) {
906			vrele(nvp);
907			error = EINVAL;
908			goto releout;
909		}
910		vrele(*fvpp);
911		*fvpp = nvp;
912		goto relock;
913	}
914	vrele(*fvpp);
915	*fvpp = nvp;
916	VOP_UNLOCK(*fvpp, 0);
917	/*
918	 * Re-resolve tvp and acquire the vnode lock if present.
919	 */
920	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp);
921	/*
922	 * If tvp disappeared we just carry on.
923	 */
924	if (de == NULL && *tvpp != NULL) {
925		vrele(*tvpp);
926		*tvpp = NULL;
927	}
928	/*
929	 * Get the tvp ino if the lookup succeeded.  We may have to restart
930	 * if the non-blocking acquire fails.
931	 */
932	if (de != NULL) {
933		nvp = NULL;
934		error = tmpfs_alloc_vp(mp, de->td_node,
935		    LK_EXCLUSIVE | LK_NOWAIT, &nvp);
936		if (*tvpp != NULL)
937			vrele(*tvpp);
938		*tvpp = nvp;
939		if (error != 0) {
940			VOP_UNLOCK(fdvp, 0);
941			VOP_UNLOCK(tdvp, 0);
942			if (error != EBUSY)
943				goto releout;
944			error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE,
945			    &nvp);
946			if (error != 0)
947				goto releout;
948			VOP_UNLOCK(nvp, 0);
949			/*
950			 * fdvp contains fvp, thus tvp (=fdvp) is not empty.
951			 */
952			if (nvp == fdvp) {
953				error = ENOTEMPTY;
954				goto releout;
955			}
956			goto relock;
957		}
958	}
959	tmpfs_rename_restarts += restarts;
960
961	return (0);
962
963releout:
964	vrele(fdvp);
965	vrele(*fvpp);
966	vrele(tdvp);
967	if (*tvpp != NULL)
968		vrele(*tvpp);
969	tmpfs_rename_restarts += restarts;
970
971	return (error);
972}
973
974static int
975tmpfs_rename(struct vop_rename_args *v)
976{
977	struct vnode *fdvp = v->a_fdvp;
978	struct vnode *fvp = v->a_fvp;
979	struct componentname *fcnp = v->a_fcnp;
980	struct vnode *tdvp = v->a_tdvp;
981	struct vnode *tvp = v->a_tvp;
982	struct componentname *tcnp = v->a_tcnp;
983	struct mount *mp = NULL;
984
985	char *newname;
986	int error;
987	struct tmpfs_dirent *de;
988	struct tmpfs_mount *tmp;
989	struct tmpfs_node *fdnode;
990	struct tmpfs_node *fnode;
991	struct tmpfs_node *tnode;
992	struct tmpfs_node *tdnode;
993
994	MPASS(VOP_ISLOCKED(tdvp));
995	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
996	MPASS(fcnp->cn_flags & HASBUF);
997	MPASS(tcnp->cn_flags & HASBUF);
998
999	/* Disallow cross-device renames.
1000	 * XXX Why isn't this done by the caller? */
1001	if (fvp->v_mount != tdvp->v_mount ||
1002	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
1003		error = EXDEV;
1004		goto out;
1005	}
1006
1007	/* If source and target are the same file, there is nothing to do. */
1008	if (fvp == tvp) {
1009		error = 0;
1010		goto out;
1011	}
1012
1013	/* If we need to move the directory between entries, lock the
1014	 * source so that we can safely operate on it. */
1015	if (fdvp != tdvp && fdvp != tvp) {
1016		if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1017			mp = tdvp->v_mount;
1018			error = vfs_busy(mp, 0);
1019			if (error != 0) {
1020				mp = NULL;
1021				goto out;
1022			}
1023			error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp,
1024			    fcnp, tcnp);
1025			if (error != 0) {
1026				vfs_unbusy(mp);
1027				return (error);
1028			}
1029			ASSERT_VOP_ELOCKED(fdvp,
1030			    "tmpfs_rename: fdvp not locked");
1031			ASSERT_VOP_ELOCKED(tdvp,
1032			    "tmpfs_rename: tdvp not locked");
1033			if (tvp != NULL)
1034				ASSERT_VOP_ELOCKED(tvp,
1035				    "tmpfs_rename: tvp not locked");
1036			if (fvp == tvp) {
1037				error = 0;
1038				goto out_locked;
1039			}
1040		}
1041	}
1042
1043	tmp = VFS_TO_TMPFS(tdvp->v_mount);
1044	tdnode = VP_TO_TMPFS_DIR(tdvp);
1045	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
1046	fdnode = VP_TO_TMPFS_DIR(fdvp);
1047	fnode = VP_TO_TMPFS_NODE(fvp);
1048	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
1049
1050	/* Entry can disappear before we lock fdvp,
1051	 * also avoid manipulating '.' and '..' entries. */
1052	if (de == NULL) {
1053		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
1054		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
1055			error = EINVAL;
1056		else
1057			error = ENOENT;
1058		goto out_locked;
1059	}
1060	MPASS(de->td_node == fnode);
1061
1062	/* If re-naming a directory to another preexisting directory
1063	 * ensure that the target directory is empty so that its
1064	 * removal causes no side effects.
1065	 * Kern_rename gurantees the destination to be a directory
1066	 * if the source is one. */
1067	if (tvp != NULL) {
1068		MPASS(tnode != NULL);
1069
1070		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1071		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1072			error = EPERM;
1073			goto out_locked;
1074		}
1075
1076		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1077			if (tnode->tn_size > 0) {
1078				error = ENOTEMPTY;
1079				goto out_locked;
1080			}
1081		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1082			error = ENOTDIR;
1083			goto out_locked;
1084		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1085			error = EISDIR;
1086			goto out_locked;
1087		} else {
1088			MPASS(fnode->tn_type != VDIR &&
1089				tnode->tn_type != VDIR);
1090		}
1091	}
1092
1093	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1094	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1095		error = EPERM;
1096		goto out_locked;
1097	}
1098
1099	/* Ensure that we have enough memory to hold the new name, if it
1100	 * has to be changed. */
1101	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1102	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1103		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1104	} else
1105		newname = NULL;
1106
1107	/* If the node is being moved to another directory, we have to do
1108	 * the move. */
1109	if (fdnode != tdnode) {
1110		/* In case we are moving a directory, we have to adjust its
1111		 * parent to point to the new parent. */
1112		if (de->td_node->tn_type == VDIR) {
1113			struct tmpfs_node *n;
1114
1115			/* Ensure the target directory is not a child of the
1116			 * directory being moved.  Otherwise, we'd end up
1117			 * with stale nodes. */
1118			n = tdnode;
1119			/* TMPFS_LOCK garanties that no nodes are freed while
1120			 * traversing the list. Nodes can only be marked as
1121			 * removed: tn_parent == NULL. */
1122			TMPFS_LOCK(tmp);
1123			TMPFS_NODE_LOCK(n);
1124			while (n != n->tn_dir.tn_parent) {
1125				struct tmpfs_node *parent;
1126
1127				if (n == fnode) {
1128					TMPFS_NODE_UNLOCK(n);
1129					TMPFS_UNLOCK(tmp);
1130					error = EINVAL;
1131					if (newname != NULL)
1132						    free(newname, M_TMPFSNAME);
1133					goto out_locked;
1134				}
1135				parent = n->tn_dir.tn_parent;
1136				TMPFS_NODE_UNLOCK(n);
1137				if (parent == NULL) {
1138					n = NULL;
1139					break;
1140				}
1141				TMPFS_NODE_LOCK(parent);
1142				if (parent->tn_dir.tn_parent == NULL) {
1143					TMPFS_NODE_UNLOCK(parent);
1144					n = NULL;
1145					break;
1146				}
1147				n = parent;
1148			}
1149			TMPFS_UNLOCK(tmp);
1150			if (n == NULL) {
1151				error = EINVAL;
1152				if (newname != NULL)
1153					    free(newname, M_TMPFSNAME);
1154				goto out_locked;
1155			}
1156			TMPFS_NODE_UNLOCK(n);
1157
1158			/* Adjust the parent pointer. */
1159			TMPFS_VALIDATE_DIR(fnode);
1160			TMPFS_NODE_LOCK(de->td_node);
1161			de->td_node->tn_dir.tn_parent = tdnode;
1162			TMPFS_NODE_UNLOCK(de->td_node);
1163
1164			/* As a result of changing the target of the '..'
1165			 * entry, the link count of the source and target
1166			 * directories has to be adjusted. */
1167			TMPFS_NODE_LOCK(tdnode);
1168			TMPFS_ASSERT_LOCKED(tdnode);
1169			tdnode->tn_links++;
1170			TMPFS_NODE_UNLOCK(tdnode);
1171
1172			TMPFS_NODE_LOCK(fdnode);
1173			TMPFS_ASSERT_LOCKED(fdnode);
1174			fdnode->tn_links--;
1175			TMPFS_NODE_UNLOCK(fdnode);
1176		}
1177	}
1178
1179	/* Do the move: just remove the entry from the source directory
1180	 * and insert it into the target one. */
1181	tmpfs_dir_detach(fdvp, de);
1182
1183	if (fcnp->cn_flags & DOWHITEOUT)
1184		tmpfs_dir_whiteout_add(fdvp, fcnp);
1185	if (tcnp->cn_flags & ISWHITEOUT)
1186		tmpfs_dir_whiteout_remove(tdvp, tcnp);
1187
1188	/* If the name has changed, we need to make it effective by changing
1189	 * it in the directory entry. */
1190	if (newname != NULL) {
1191		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1192
1193		free(de->ud.td_name, M_TMPFSNAME);
1194		de->ud.td_name = newname;
1195		tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen);
1196
1197		fnode->tn_status |= TMPFS_NODE_CHANGED;
1198		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1199	}
1200
1201	/* If we are overwriting an entry, we have to remove the old one
1202	 * from the target directory. */
1203	if (tvp != NULL) {
1204		struct tmpfs_dirent *tde;
1205
1206		/* Remove the old entry from the target directory. */
1207		tde = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1208		tmpfs_dir_detach(tdvp, tde);
1209
1210		/* Free the directory entry we just deleted.  Note that the
1211		 * node referred by it will not be removed until the vnode is
1212		 * really reclaimed. */
1213		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1214	}
1215
1216	tmpfs_dir_attach(tdvp, de);
1217
1218	cache_purge(fvp);
1219	if (tvp != NULL)
1220		cache_purge(tvp);
1221	cache_purge_negative(tdvp);
1222
1223	error = 0;
1224
1225out_locked:
1226	if (fdvp != tdvp && fdvp != tvp)
1227		VOP_UNLOCK(fdvp, 0);
1228
1229out:
1230	/* Release target nodes. */
1231	/* XXX: I don't understand when tdvp can be the same as tvp, but
1232	 * other code takes care of this... */
1233	if (tdvp == tvp)
1234		vrele(tdvp);
1235	else
1236		vput(tdvp);
1237	if (tvp != NULL)
1238		vput(tvp);
1239
1240	/* Release source nodes. */
1241	vrele(fdvp);
1242	vrele(fvp);
1243
1244	if (mp != NULL)
1245		vfs_unbusy(mp);
1246
1247	return error;
1248}
1249
1250/* --------------------------------------------------------------------- */
1251
1252static int
1253tmpfs_mkdir(struct vop_mkdir_args *v)
1254{
1255	struct vnode *dvp = v->a_dvp;
1256	struct vnode **vpp = v->a_vpp;
1257	struct componentname *cnp = v->a_cnp;
1258	struct vattr *vap = v->a_vap;
1259
1260	MPASS(vap->va_type == VDIR);
1261
1262	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1263}
1264
1265/* --------------------------------------------------------------------- */
1266
1267static int
1268tmpfs_rmdir(struct vop_rmdir_args *v)
1269{
1270	struct vnode *dvp = v->a_dvp;
1271	struct vnode *vp = v->a_vp;
1272
1273	int error;
1274	struct tmpfs_dirent *de;
1275	struct tmpfs_mount *tmp;
1276	struct tmpfs_node *dnode;
1277	struct tmpfs_node *node;
1278
1279	MPASS(VOP_ISLOCKED(dvp));
1280	MPASS(VOP_ISLOCKED(vp));
1281
1282	tmp = VFS_TO_TMPFS(dvp->v_mount);
1283	dnode = VP_TO_TMPFS_DIR(dvp);
1284	node = VP_TO_TMPFS_DIR(vp);
1285
1286	/* Directories with more than two entries ('.' and '..') cannot be
1287	 * removed. */
1288	 if (node->tn_size > 0) {
1289		 error = ENOTEMPTY;
1290		 goto out;
1291	 }
1292
1293	if ((dnode->tn_flags & APPEND)
1294	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1295		error = EPERM;
1296		goto out;
1297	}
1298
1299	/* This invariant holds only if we are not trying to remove "..".
1300	  * We checked for that above so this is safe now. */
1301	MPASS(node->tn_dir.tn_parent == dnode);
1302
1303	/* Get the directory entry associated with node (vp).  This was
1304	 * filled by tmpfs_lookup while looking up the entry. */
1305	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1306	MPASS(TMPFS_DIRENT_MATCHES(de,
1307	    v->a_cnp->cn_nameptr,
1308	    v->a_cnp->cn_namelen));
1309
1310	/* Check flags to see if we are allowed to remove the directory. */
1311	if (dnode->tn_flags & APPEND
1312		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1313		error = EPERM;
1314		goto out;
1315	}
1316
1317
1318	/* Detach the directory entry from the directory (dnode). */
1319	tmpfs_dir_detach(dvp, de);
1320	if (v->a_cnp->cn_flags & DOWHITEOUT)
1321		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1322
1323	/* No vnode should be allocated for this entry from this point */
1324	TMPFS_NODE_LOCK(node);
1325	TMPFS_ASSERT_ELOCKED(node);
1326	node->tn_links--;
1327	node->tn_dir.tn_parent = NULL;
1328	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1329	    TMPFS_NODE_MODIFIED;
1330
1331	TMPFS_NODE_UNLOCK(node);
1332
1333	TMPFS_NODE_LOCK(dnode);
1334	TMPFS_ASSERT_ELOCKED(dnode);
1335	dnode->tn_links--;
1336	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1337	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1338	TMPFS_NODE_UNLOCK(dnode);
1339
1340	cache_purge(dvp);
1341	cache_purge(vp);
1342
1343	/* Free the directory entry we just deleted.  Note that the node
1344	 * referred by it will not be removed until the vnode is really
1345	 * reclaimed. */
1346	tmpfs_free_dirent(tmp, de);
1347
1348	/* Release the deleted vnode (will destroy the node, notify
1349	 * interested parties and clean it from the cache). */
1350
1351	dnode->tn_status |= TMPFS_NODE_CHANGED;
1352	tmpfs_update(dvp);
1353
1354	error = 0;
1355
1356out:
1357	return error;
1358}
1359
1360/* --------------------------------------------------------------------- */
1361
1362static int
1363tmpfs_symlink(struct vop_symlink_args *v)
1364{
1365	struct vnode *dvp = v->a_dvp;
1366	struct vnode **vpp = v->a_vpp;
1367	struct componentname *cnp = v->a_cnp;
1368	struct vattr *vap = v->a_vap;
1369	char *target = v->a_target;
1370
1371#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1372	MPASS(vap->va_type == VLNK);
1373#else
1374	vap->va_type = VLNK;
1375#endif
1376
1377	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1378}
1379
1380/* --------------------------------------------------------------------- */
1381
1382static int
1383tmpfs_readdir(struct vop_readdir_args *v)
1384{
1385	struct vnode *vp = v->a_vp;
1386	struct uio *uio = v->a_uio;
1387	int *eofflag = v->a_eofflag;
1388	u_long **cookies = v->a_cookies;
1389	int *ncookies = v->a_ncookies;
1390
1391	int error;
1392	ssize_t startresid;
1393	int cnt = 0;
1394	struct tmpfs_node *node;
1395
1396	/* This operation only makes sense on directory nodes. */
1397	if (vp->v_type != VDIR)
1398		return ENOTDIR;
1399
1400	node = VP_TO_TMPFS_DIR(vp);
1401
1402	startresid = uio->uio_resid;
1403
1404	if (cookies != NULL && ncookies != NULL) {
1405		cnt = howmany(node->tn_size, sizeof(struct tmpfs_dirent)) + 2;
1406		*cookies = malloc(cnt * sizeof(**cookies), M_TEMP, M_WAITOK);
1407		*ncookies = 0;
1408	}
1409
1410	if (cnt == 0)
1411		error = tmpfs_dir_getdents(node, uio, 0, NULL, NULL);
1412	else
1413		error = tmpfs_dir_getdents(node, uio, cnt, *cookies, ncookies);
1414
1415	if (error == EJUSTRETURN)
1416		error = (uio->uio_resid != startresid) ? 0 : EINVAL;
1417
1418	if (error != 0 && cnt != 0)
1419		free(*cookies, M_TEMP);
1420
1421	if (eofflag != NULL)
1422		*eofflag =
1423		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1424
1425	return error;
1426}
1427
1428/* --------------------------------------------------------------------- */
1429
1430static int
1431tmpfs_readlink(struct vop_readlink_args *v)
1432{
1433	struct vnode *vp = v->a_vp;
1434	struct uio *uio = v->a_uio;
1435
1436	int error;
1437	struct tmpfs_node *node;
1438
1439	MPASS(uio->uio_offset == 0);
1440	MPASS(vp->v_type == VLNK);
1441
1442	node = VP_TO_TMPFS_NODE(vp);
1443
1444	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1445	    uio);
1446	node->tn_status |= TMPFS_NODE_ACCESSED;
1447
1448	return error;
1449}
1450
1451/* --------------------------------------------------------------------- */
1452
1453static int
1454tmpfs_inactive(struct vop_inactive_args *v)
1455{
1456	struct vnode *vp = v->a_vp;
1457
1458	struct tmpfs_node *node;
1459
1460	node = VP_TO_TMPFS_NODE(vp);
1461
1462	if (node->tn_links == 0)
1463		vrecycle(vp);
1464
1465	return 0;
1466}
1467
1468/* --------------------------------------------------------------------- */
1469
1470int
1471tmpfs_reclaim(struct vop_reclaim_args *v)
1472{
1473	struct vnode *vp = v->a_vp;
1474
1475	struct tmpfs_mount *tmp;
1476	struct tmpfs_node *node;
1477
1478	node = VP_TO_TMPFS_NODE(vp);
1479	tmp = VFS_TO_TMPFS(vp->v_mount);
1480
1481	if (vp->v_type == VREG)
1482		tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj);
1483	else
1484		vnode_destroy_vobject(vp);
1485	vp->v_object = NULL;
1486	cache_purge(vp);
1487
1488	TMPFS_NODE_LOCK(node);
1489	TMPFS_ASSERT_ELOCKED(node);
1490	tmpfs_free_vp(vp);
1491
1492	/* If the node referenced by this vnode was deleted by the user,
1493	 * we must free its associated data structures (now that the vnode
1494	 * is being reclaimed). */
1495	if (node->tn_links == 0 &&
1496	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1497		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1498		TMPFS_NODE_UNLOCK(node);
1499		tmpfs_free_node(tmp, node);
1500	} else
1501		TMPFS_NODE_UNLOCK(node);
1502
1503	MPASS(vp->v_data == NULL);
1504	return 0;
1505}
1506
1507/* --------------------------------------------------------------------- */
1508
1509static int
1510tmpfs_print(struct vop_print_args *v)
1511{
1512	struct vnode *vp = v->a_vp;
1513
1514	struct tmpfs_node *node;
1515
1516	node = VP_TO_TMPFS_NODE(vp);
1517
1518	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %d\n",
1519	    node, node->tn_flags, node->tn_links);
1520	printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n",
1521	    node->tn_mode, node->tn_uid, node->tn_gid,
1522	    (intmax_t)node->tn_size, node->tn_status);
1523
1524	if (vp->v_type == VFIFO)
1525		fifo_printinfo(vp);
1526
1527	printf("\n");
1528
1529	return 0;
1530}
1531
1532/* --------------------------------------------------------------------- */
1533
1534static int
1535tmpfs_pathconf(struct vop_pathconf_args *v)
1536{
1537	int name = v->a_name;
1538	register_t *retval = v->a_retval;
1539
1540	int error;
1541
1542	error = 0;
1543
1544	switch (name) {
1545	case _PC_LINK_MAX:
1546		*retval = LINK_MAX;
1547		break;
1548
1549	case _PC_NAME_MAX:
1550		*retval = NAME_MAX;
1551		break;
1552
1553	case _PC_PATH_MAX:
1554		*retval = PATH_MAX;
1555		break;
1556
1557	case _PC_PIPE_BUF:
1558		*retval = PIPE_BUF;
1559		break;
1560
1561	case _PC_CHOWN_RESTRICTED:
1562		*retval = 1;
1563		break;
1564
1565	case _PC_NO_TRUNC:
1566		*retval = 1;
1567		break;
1568
1569	case _PC_SYNC_IO:
1570		*retval = 1;
1571		break;
1572
1573	case _PC_FILESIZEBITS:
1574		*retval = 0; /* XXX Don't know which value should I return. */
1575		break;
1576
1577	default:
1578		error = EINVAL;
1579	}
1580
1581	return error;
1582}
1583
1584static int
1585tmpfs_vptofh(struct vop_vptofh_args *ap)
1586{
1587	struct tmpfs_fid *tfhp;
1588	struct tmpfs_node *node;
1589
1590	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1591	node = VP_TO_TMPFS_NODE(ap->a_vp);
1592
1593	tfhp->tf_len = sizeof(struct tmpfs_fid);
1594	tfhp->tf_id = node->tn_id;
1595	tfhp->tf_gen = node->tn_gen;
1596
1597	return (0);
1598}
1599
1600static int
1601tmpfs_whiteout(struct vop_whiteout_args *ap)
1602{
1603	struct vnode *dvp = ap->a_dvp;
1604	struct componentname *cnp = ap->a_cnp;
1605	struct tmpfs_dirent *de;
1606
1607	switch (ap->a_flags) {
1608	case LOOKUP:
1609		return (0);
1610	case CREATE:
1611		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1612		if (de != NULL)
1613			return (de->td_node == NULL ? 0 : EEXIST);
1614		return (tmpfs_dir_whiteout_add(dvp, cnp));
1615	case DELETE:
1616		tmpfs_dir_whiteout_remove(dvp, cnp);
1617		return (0);
1618	default:
1619		panic("tmpfs_whiteout: unknown op");
1620	}
1621}
1622
1623/* --------------------------------------------------------------------- */
1624
1625/*
1626 * vnode operations vector used for files stored in a tmpfs file system.
1627 */
1628struct vop_vector tmpfs_vnodeop_entries = {
1629	.vop_default =			&default_vnodeops,
1630	.vop_lookup =			vfs_cache_lookup,
1631	.vop_cachedlookup =		tmpfs_lookup,
1632	.vop_create =			tmpfs_create,
1633	.vop_mknod =			tmpfs_mknod,
1634	.vop_open =			tmpfs_open,
1635	.vop_close =			tmpfs_close,
1636	.vop_access =			tmpfs_access,
1637	.vop_getattr =			tmpfs_getattr,
1638	.vop_setattr =			tmpfs_setattr,
1639	.vop_read =			tmpfs_read,
1640	.vop_write =			tmpfs_write,
1641	.vop_fsync =			tmpfs_fsync,
1642	.vop_remove =			tmpfs_remove,
1643	.vop_link =			tmpfs_link,
1644	.vop_rename =			tmpfs_rename,
1645	.vop_mkdir =			tmpfs_mkdir,
1646	.vop_rmdir =			tmpfs_rmdir,
1647	.vop_symlink =			tmpfs_symlink,
1648	.vop_readdir =			tmpfs_readdir,
1649	.vop_readlink =			tmpfs_readlink,
1650	.vop_inactive =			tmpfs_inactive,
1651	.vop_reclaim =			tmpfs_reclaim,
1652	.vop_print =			tmpfs_print,
1653	.vop_pathconf =			tmpfs_pathconf,
1654	.vop_vptofh =			tmpfs_vptofh,
1655	.vop_whiteout =			tmpfs_whiteout,
1656	.vop_bmap =			VOP_EOPNOTSUPP,
1657};
1658
1659