tmpfs_vnops.c revision 218949
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 218949 2011-02-22 14:47:10Z alc $");
38
39#include <sys/param.h>
40#include <sys/fcntl.h>
41#include <sys/lockf.h>
42#include <sys/namei.h>
43#include <sys/priv.h>
44#include <sys/proc.h>
45#include <sys/sched.h>
46#include <sys/sf_buf.h>
47#include <sys/stat.h>
48#include <sys/systm.h>
49#include <sys/unistd.h>
50#include <sys/vnode.h>
51
52#include <vm/vm.h>
53#include <vm/vm_object.h>
54#include <vm/vm_page.h>
55#include <vm/vm_pager.h>
56
57#include <machine/_inttypes.h>
58
59#include <fs/fifofs/fifo.h>
60#include <fs/tmpfs/tmpfs_vnops.h>
61#include <fs/tmpfs/tmpfs.h>
62
63/* --------------------------------------------------------------------- */
64
65static int
66tmpfs_lookup(struct vop_cachedlookup_args *v)
67{
68	struct vnode *dvp = v->a_dvp;
69	struct vnode **vpp = v->a_vpp;
70	struct componentname *cnp = v->a_cnp;
71
72	int error;
73	struct tmpfs_dirent *de;
74	struct tmpfs_node *dnode;
75
76	dnode = VP_TO_TMPFS_DIR(dvp);
77	*vpp = NULLVP;
78
79	/* Check accessibility of requested node as a first step. */
80	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
81	if (error != 0)
82		goto out;
83
84	/* We cannot be requesting the parent directory of the root node. */
85	MPASS(IMPLIES(dnode->tn_type == VDIR &&
86	    dnode->tn_dir.tn_parent == dnode,
87	    !(cnp->cn_flags & ISDOTDOT)));
88
89	TMPFS_ASSERT_LOCKED(dnode);
90	if (dnode->tn_dir.tn_parent == NULL) {
91		error = ENOENT;
92		goto out;
93	}
94	if (cnp->cn_flags & ISDOTDOT) {
95		int ltype = 0;
96
97		ltype = VOP_ISLOCKED(dvp);
98		vhold(dvp);
99		VOP_UNLOCK(dvp, 0);
100		/* Allocate a new vnode on the matching entry. */
101		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
102		    cnp->cn_lkflags, vpp);
103
104		vn_lock(dvp, ltype | LK_RETRY);
105		vdrop(dvp);
106	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
107		VREF(dvp);
108		*vpp = dvp;
109		error = 0;
110	} else {
111		de = tmpfs_dir_lookup(dnode, NULL, cnp);
112		if (de != NULL && de->td_node == NULL)
113			cnp->cn_flags |= ISWHITEOUT;
114		if (de == NULL || de->td_node == NULL) {
115			/* The entry was not found in the directory.
116			 * This is OK if we are creating or renaming an
117			 * entry and are working on the last component of
118			 * the path name. */
119			if ((cnp->cn_flags & ISLASTCN) &&
120			    (cnp->cn_nameiop == CREATE || \
121			    cnp->cn_nameiop == RENAME ||
122			    (cnp->cn_nameiop == DELETE &&
123			    cnp->cn_flags & DOWHITEOUT &&
124			    cnp->cn_flags & ISWHITEOUT))) {
125				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
126				    cnp->cn_thread);
127				if (error != 0)
128					goto out;
129
130				/* Keep the component name in the buffer for
131				 * future uses. */
132				cnp->cn_flags |= SAVENAME;
133
134				error = EJUSTRETURN;
135			} else
136				error = ENOENT;
137		} else {
138			struct tmpfs_node *tnode;
139
140			/* The entry was found, so get its associated
141			 * tmpfs_node. */
142			tnode = de->td_node;
143
144			/* If we are not at the last path component and
145			 * found a non-directory or non-link entry (which
146			 * may itself be pointing to a directory), raise
147			 * an error. */
148			if ((tnode->tn_type != VDIR &&
149			    tnode->tn_type != VLNK) &&
150			    !(cnp->cn_flags & ISLASTCN)) {
151				error = ENOTDIR;
152				goto out;
153			}
154
155			/* If we are deleting or renaming the entry, keep
156			 * track of its tmpfs_dirent so that it can be
157			 * easily deleted later. */
158			if ((cnp->cn_flags & ISLASTCN) &&
159			    (cnp->cn_nameiop == DELETE ||
160			    cnp->cn_nameiop == RENAME)) {
161				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
162				    cnp->cn_thread);
163				if (error != 0)
164					goto out;
165
166				/* Allocate a new vnode on the matching entry. */
167				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
168						cnp->cn_lkflags, vpp);
169				if (error != 0)
170					goto out;
171
172				if ((dnode->tn_mode & S_ISTXT) &&
173				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
174				  VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
175					error = EPERM;
176					vput(*vpp);
177					*vpp = NULL;
178					goto out;
179				}
180				cnp->cn_flags |= SAVENAME;
181			} else {
182				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
183						cnp->cn_lkflags, vpp);
184			}
185		}
186	}
187
188	/* Store the result of this lookup in the cache.  Avoid this if the
189	 * request was for creation, as it does not improve timings on
190	 * emprical tests. */
191	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
192		cache_enter(dvp, *vpp, cnp);
193
194out:
195	/* If there were no errors, *vpp cannot be null and it must be
196	 * locked. */
197	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
198
199	return error;
200}
201
202/* --------------------------------------------------------------------- */
203
204static int
205tmpfs_create(struct vop_create_args *v)
206{
207	struct vnode *dvp = v->a_dvp;
208	struct vnode **vpp = v->a_vpp;
209	struct componentname *cnp = v->a_cnp;
210	struct vattr *vap = v->a_vap;
211
212	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
213
214	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
215}
216/* --------------------------------------------------------------------- */
217
218static int
219tmpfs_mknod(struct vop_mknod_args *v)
220{
221	struct vnode *dvp = v->a_dvp;
222	struct vnode **vpp = v->a_vpp;
223	struct componentname *cnp = v->a_cnp;
224	struct vattr *vap = v->a_vap;
225
226	if (vap->va_type != VBLK && vap->va_type != VCHR &&
227	    vap->va_type != VFIFO)
228		return EINVAL;
229
230	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
231}
232
233/* --------------------------------------------------------------------- */
234
235static int
236tmpfs_open(struct vop_open_args *v)
237{
238	struct vnode *vp = v->a_vp;
239	int mode = v->a_mode;
240
241	int error;
242	struct tmpfs_node *node;
243
244	MPASS(VOP_ISLOCKED(vp));
245
246	node = VP_TO_TMPFS_NODE(vp);
247
248	/* The file is still active but all its names have been removed
249	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
250	 * it is about to die. */
251	if (node->tn_links < 1)
252		return (ENOENT);
253
254	/* If the file is marked append-only, deny write requests. */
255	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
256		error = EPERM;
257	else {
258		error = 0;
259		vnode_create_vobject(vp, node->tn_size, v->a_td);
260	}
261
262	MPASS(VOP_ISLOCKED(vp));
263	return error;
264}
265
266/* --------------------------------------------------------------------- */
267
268static int
269tmpfs_close(struct vop_close_args *v)
270{
271	struct vnode *vp = v->a_vp;
272
273	MPASS(VOP_ISLOCKED(vp));
274
275	/* Update node times. */
276	tmpfs_update(vp);
277
278	return (0);
279}
280
281/* --------------------------------------------------------------------- */
282
283int
284tmpfs_access(struct vop_access_args *v)
285{
286	struct vnode *vp = v->a_vp;
287	accmode_t accmode = v->a_accmode;
288	struct ucred *cred = v->a_cred;
289
290	int error;
291	struct tmpfs_node *node;
292
293	MPASS(VOP_ISLOCKED(vp));
294
295	node = VP_TO_TMPFS_NODE(vp);
296
297	switch (vp->v_type) {
298	case VDIR:
299		/* FALLTHROUGH */
300	case VLNK:
301		/* FALLTHROUGH */
302	case VREG:
303		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
304			error = EROFS;
305			goto out;
306		}
307		break;
308
309	case VBLK:
310		/* FALLTHROUGH */
311	case VCHR:
312		/* FALLTHROUGH */
313	case VSOCK:
314		/* FALLTHROUGH */
315	case VFIFO:
316		break;
317
318	default:
319		error = EINVAL;
320		goto out;
321	}
322
323	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
324		error = EPERM;
325		goto out;
326	}
327
328	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
329	    node->tn_gid, accmode, cred, NULL);
330
331out:
332	MPASS(VOP_ISLOCKED(vp));
333
334	return error;
335}
336
337/* --------------------------------------------------------------------- */
338
339int
340tmpfs_getattr(struct vop_getattr_args *v)
341{
342	struct vnode *vp = v->a_vp;
343	struct vattr *vap = v->a_vap;
344
345	struct tmpfs_node *node;
346
347	node = VP_TO_TMPFS_NODE(vp);
348
349	tmpfs_update(vp);
350
351	vap->va_type = vp->v_type;
352	vap->va_mode = node->tn_mode;
353	vap->va_nlink = node->tn_links;
354	vap->va_uid = node->tn_uid;
355	vap->va_gid = node->tn_gid;
356	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
357	vap->va_fileid = node->tn_id;
358	vap->va_size = node->tn_size;
359	vap->va_blocksize = PAGE_SIZE;
360	vap->va_atime = node->tn_atime;
361	vap->va_mtime = node->tn_mtime;
362	vap->va_ctime = node->tn_ctime;
363	vap->va_birthtime = node->tn_birthtime;
364	vap->va_gen = node->tn_gen;
365	vap->va_flags = node->tn_flags;
366	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
367		node->tn_rdev : NODEV;
368	vap->va_bytes = round_page(node->tn_size);
369	vap->va_filerev = 0;
370
371	return 0;
372}
373
374/* --------------------------------------------------------------------- */
375
376/* XXX Should this operation be atomic?  I think it should, but code in
377 * XXX other places (e.g., ufs) doesn't seem to be... */
378int
379tmpfs_setattr(struct vop_setattr_args *v)
380{
381	struct vnode *vp = v->a_vp;
382	struct vattr *vap = v->a_vap;
383	struct ucred *cred = v->a_cred;
384	struct thread *td = curthread;
385
386	int error;
387
388	MPASS(VOP_ISLOCKED(vp));
389
390	error = 0;
391
392	/* Abort if any unsettable attribute is given. */
393	if (vap->va_type != VNON ||
394	    vap->va_nlink != VNOVAL ||
395	    vap->va_fsid != VNOVAL ||
396	    vap->va_fileid != VNOVAL ||
397	    vap->va_blocksize != VNOVAL ||
398	    vap->va_gen != VNOVAL ||
399	    vap->va_rdev != VNOVAL ||
400	    vap->va_bytes != VNOVAL)
401		error = EINVAL;
402
403	if (error == 0 && (vap->va_flags != VNOVAL))
404		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
405
406	if (error == 0 && (vap->va_size != VNOVAL))
407		error = tmpfs_chsize(vp, vap->va_size, cred, td);
408
409	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
410		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
411
412	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
413		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
414
415	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
416	    vap->va_atime.tv_nsec != VNOVAL) ||
417	    (vap->va_mtime.tv_sec != VNOVAL &&
418	    vap->va_mtime.tv_nsec != VNOVAL) ||
419	    (vap->va_birthtime.tv_sec != VNOVAL &&
420	    vap->va_birthtime.tv_nsec != VNOVAL)))
421		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
422			&vap->va_birthtime, vap->va_vaflags, cred, td);
423
424	/* Update the node times.  We give preference to the error codes
425	 * generated by this function rather than the ones that may arise
426	 * from tmpfs_update. */
427	tmpfs_update(vp);
428
429	MPASS(VOP_ISLOCKED(vp));
430
431	return error;
432}
433
434/* --------------------------------------------------------------------- */
435static int
436tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
437    vm_offset_t offset, size_t tlen, struct uio *uio)
438{
439	vm_page_t	m;
440	int		error;
441
442	VM_OBJECT_LOCK(tobj);
443	vm_object_pip_add(tobj, 1);
444	m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
445	    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
446	if (m->valid != VM_PAGE_BITS_ALL) {
447		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
448			error = vm_pager_get_pages(tobj, &m, 1, 0);
449			if (error != 0) {
450				printf("tmpfs get pages from pager error [read]\n");
451				goto out;
452			}
453		} else
454			vm_page_zero_invalid(m, TRUE);
455	}
456	VM_OBJECT_UNLOCK(tobj);
457	error = uiomove_fromphys(&m, offset, tlen, uio);
458	VM_OBJECT_LOCK(tobj);
459out:
460	vm_page_lock(m);
461	vm_page_unwire(m, TRUE);
462	vm_page_unlock(m);
463	vm_page_wakeup(m);
464	vm_object_pip_subtract(tobj, 1);
465	VM_OBJECT_UNLOCK(tobj);
466
467	return (error);
468}
469
470static __inline int
471tmpfs_nocacheread_buf(vm_object_t tobj, vm_pindex_t idx,
472    vm_offset_t offset, size_t tlen, void *buf)
473{
474	struct uio uio;
475	struct iovec iov;
476
477	uio.uio_iovcnt = 1;
478	uio.uio_iov = &iov;
479	iov.iov_base = buf;
480	iov.iov_len = tlen;
481
482	uio.uio_offset = 0;
483	uio.uio_resid = tlen;
484	uio.uio_rw = UIO_READ;
485	uio.uio_segflg = UIO_SYSSPACE;
486	uio.uio_td = curthread;
487
488	return (tmpfs_nocacheread(tobj, idx, offset, tlen, &uio));
489}
490
491static int
492tmpfs_mappedread(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
493{
494	struct sf_buf	*sf;
495	vm_pindex_t	idx;
496	vm_page_t	m;
497	vm_offset_t	offset;
498	off_t		addr;
499	size_t		tlen;
500	char		*ma;
501	int		error;
502
503	addr = uio->uio_offset;
504	idx = OFF_TO_IDX(addr);
505	offset = addr & PAGE_MASK;
506	tlen = MIN(PAGE_SIZE - offset, len);
507
508	if ((vobj == NULL) ||
509	    (vobj->resident_page_count == 0 && vobj->cache == NULL))
510		goto nocache;
511
512	VM_OBJECT_LOCK(vobj);
513lookupvpg:
514	if (((m = vm_page_lookup(vobj, idx)) != NULL) &&
515	    vm_page_is_valid(m, offset, tlen)) {
516		if ((m->oflags & VPO_BUSY) != 0) {
517			/*
518			 * Reference the page before unlocking and sleeping so
519			 * that the page daemon is less likely to reclaim it.
520			 */
521			vm_page_lock_queues();
522			vm_page_flag_set(m, PG_REFERENCED);
523			vm_page_sleep(m, "tmfsmr");
524			goto lookupvpg;
525		}
526		vm_page_busy(m);
527		VM_OBJECT_UNLOCK(vobj);
528		error = uiomove_fromphys(&m, offset, tlen, uio);
529		VM_OBJECT_LOCK(vobj);
530		vm_page_wakeup(m);
531		VM_OBJECT_UNLOCK(vobj);
532		return	(error);
533	} else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) {
534		KASSERT(offset == 0,
535		    ("unexpected offset in tmpfs_mappedread for sendfile"));
536		if ((m->oflags & VPO_BUSY) != 0) {
537			/*
538			 * Reference the page before unlocking and sleeping so
539			 * that the page daemon is less likely to reclaim it.
540			 */
541			vm_page_lock_queues();
542			vm_page_flag_set(m, PG_REFERENCED);
543			vm_page_sleep(m, "tmfsmr");
544			goto lookupvpg;
545		}
546		vm_page_busy(m);
547		VM_OBJECT_UNLOCK(vobj);
548		sched_pin();
549		sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
550		ma = (char *)sf_buf_kva(sf);
551		error = tmpfs_nocacheread_buf(tobj, idx, 0, tlen, ma);
552		if (error == 0) {
553			if (tlen != PAGE_SIZE)
554				bzero(ma + tlen, PAGE_SIZE - tlen);
555			uio->uio_offset += tlen;
556			uio->uio_resid -= tlen;
557		}
558		sf_buf_free(sf);
559		sched_unpin();
560		VM_OBJECT_LOCK(vobj);
561		if (error == 0)
562			m->valid = VM_PAGE_BITS_ALL;
563		vm_page_wakeup(m);
564		VM_OBJECT_UNLOCK(vobj);
565		return	(error);
566	}
567	VM_OBJECT_UNLOCK(vobj);
568nocache:
569	error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio);
570
571	return	(error);
572}
573
574static int
575tmpfs_read(struct vop_read_args *v)
576{
577	struct vnode *vp = v->a_vp;
578	struct uio *uio = v->a_uio;
579
580	struct tmpfs_node *node;
581	vm_object_t uobj;
582	size_t len;
583	int resid;
584
585	int error = 0;
586
587	node = VP_TO_TMPFS_NODE(vp);
588
589	if (vp->v_type != VREG) {
590		error = EISDIR;
591		goto out;
592	}
593
594	if (uio->uio_offset < 0) {
595		error = EINVAL;
596		goto out;
597	}
598
599	node->tn_status |= TMPFS_NODE_ACCESSED;
600
601	uobj = node->tn_reg.tn_aobj;
602	while ((resid = uio->uio_resid) > 0) {
603		error = 0;
604		if (node->tn_size <= uio->uio_offset)
605			break;
606		len = MIN(node->tn_size - uio->uio_offset, resid);
607		if (len == 0)
608			break;
609		error = tmpfs_mappedread(vp->v_object, uobj, len, uio);
610		if ((error != 0) || (resid == uio->uio_resid))
611			break;
612	}
613
614out:
615
616	return error;
617}
618
619/* --------------------------------------------------------------------- */
620
621static int
622tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
623{
624	vm_pindex_t	idx;
625	vm_page_t	vpg, tpg;
626	vm_offset_t	offset;
627	off_t		addr;
628	size_t		tlen;
629	int		error;
630
631	error = 0;
632
633	addr = uio->uio_offset;
634	idx = OFF_TO_IDX(addr);
635	offset = addr & PAGE_MASK;
636	tlen = MIN(PAGE_SIZE - offset, len);
637
638	if ((vobj == NULL) ||
639	    (vobj->resident_page_count == 0 && vobj->cache == NULL)) {
640		vpg = NULL;
641		goto nocache;
642	}
643
644	VM_OBJECT_LOCK(vobj);
645lookupvpg:
646	if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
647	    vm_page_is_valid(vpg, offset, tlen)) {
648		if ((vpg->oflags & VPO_BUSY) != 0) {
649			/*
650			 * Reference the page before unlocking and sleeping so
651			 * that the page daemon is less likely to reclaim it.
652			 */
653			vm_page_lock_queues();
654			vm_page_flag_set(vpg, PG_REFERENCED);
655			vm_page_sleep(vpg, "tmfsmw");
656			goto lookupvpg;
657		}
658		vm_page_busy(vpg);
659		vm_page_undirty(vpg);
660		VM_OBJECT_UNLOCK(vobj);
661		error = uiomove_fromphys(&vpg, offset, tlen, uio);
662	} else {
663		if (__predict_false(vobj->cache != NULL))
664			vm_page_cache_free(vobj, idx, idx + 1);
665		VM_OBJECT_UNLOCK(vobj);
666		vpg = NULL;
667	}
668nocache:
669	VM_OBJECT_LOCK(tobj);
670	vm_object_pip_add(tobj, 1);
671	tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
672	    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
673	if (tpg->valid != VM_PAGE_BITS_ALL) {
674		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
675			error = vm_pager_get_pages(tobj, &tpg, 1, 0);
676			if (error != 0) {
677				printf("tmpfs get pages from pager error [write]\n");
678				goto out;
679			}
680		} else
681			vm_page_zero_invalid(tpg, TRUE);
682	}
683	VM_OBJECT_UNLOCK(tobj);
684	if (vpg == NULL)
685		error = uiomove_fromphys(&tpg, offset, tlen, uio);
686	else {
687		KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
688		pmap_copy_page(vpg, tpg);
689	}
690	VM_OBJECT_LOCK(tobj);
691out:
692	if (vobj != NULL)
693		VM_OBJECT_LOCK(vobj);
694	if (error == 0) {
695		KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
696		    ("parts of tpg invalid"));
697		vm_page_dirty(tpg);
698	}
699	vm_page_lock(tpg);
700	vm_page_unwire(tpg, TRUE);
701	vm_page_unlock(tpg);
702	vm_page_wakeup(tpg);
703	if (vpg != NULL)
704		vm_page_wakeup(vpg);
705	if (vobj != NULL)
706		VM_OBJECT_UNLOCK(vobj);
707	vm_object_pip_subtract(tobj, 1);
708	VM_OBJECT_UNLOCK(tobj);
709
710	return	(error);
711}
712
713static int
714tmpfs_write(struct vop_write_args *v)
715{
716	struct vnode *vp = v->a_vp;
717	struct uio *uio = v->a_uio;
718	int ioflag = v->a_ioflag;
719
720	boolean_t extended;
721	int error = 0;
722	off_t oldsize;
723	struct tmpfs_node *node;
724	vm_object_t uobj;
725	size_t len;
726	int resid;
727
728	node = VP_TO_TMPFS_NODE(vp);
729	oldsize = node->tn_size;
730
731	if (uio->uio_offset < 0 || vp->v_type != VREG) {
732		error = EINVAL;
733		goto out;
734	}
735
736	if (uio->uio_resid == 0) {
737		error = 0;
738		goto out;
739	}
740
741	if (ioflag & IO_APPEND)
742		uio->uio_offset = node->tn_size;
743
744	if (uio->uio_offset + uio->uio_resid >
745	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
746		return (EFBIG);
747
748	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
749		return (EFBIG);
750
751	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
752	if (extended) {
753		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
754		if (error != 0)
755			goto out;
756	}
757
758	uobj = node->tn_reg.tn_aobj;
759	while ((resid = uio->uio_resid) > 0) {
760		if (node->tn_size <= uio->uio_offset)
761			break;
762		len = MIN(node->tn_size - uio->uio_offset, resid);
763		if (len == 0)
764			break;
765		error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio);
766		if ((error != 0) || (resid == uio->uio_resid))
767			break;
768	}
769
770	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
771	    (extended ? TMPFS_NODE_CHANGED : 0);
772
773	if (node->tn_mode & (S_ISUID | S_ISGID)) {
774		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
775			node->tn_mode &= ~(S_ISUID | S_ISGID);
776	}
777
778	if (error != 0)
779		(void)tmpfs_reg_resize(vp, oldsize);
780
781out:
782	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
783	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
784
785	return error;
786}
787
788/* --------------------------------------------------------------------- */
789
790static int
791tmpfs_fsync(struct vop_fsync_args *v)
792{
793	struct vnode *vp = v->a_vp;
794
795	MPASS(VOP_ISLOCKED(vp));
796
797	tmpfs_update(vp);
798
799	return 0;
800}
801
802/* --------------------------------------------------------------------- */
803
804static int
805tmpfs_remove(struct vop_remove_args *v)
806{
807	struct vnode *dvp = v->a_dvp;
808	struct vnode *vp = v->a_vp;
809
810	int error;
811	struct tmpfs_dirent *de;
812	struct tmpfs_mount *tmp;
813	struct tmpfs_node *dnode;
814	struct tmpfs_node *node;
815
816	MPASS(VOP_ISLOCKED(dvp));
817	MPASS(VOP_ISLOCKED(vp));
818
819	if (vp->v_type == VDIR) {
820		error = EISDIR;
821		goto out;
822	}
823
824	dnode = VP_TO_TMPFS_DIR(dvp);
825	node = VP_TO_TMPFS_NODE(vp);
826	tmp = VFS_TO_TMPFS(vp->v_mount);
827	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
828	MPASS(de != NULL);
829
830	/* Files marked as immutable or append-only cannot be deleted. */
831	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
832	    (dnode->tn_flags & APPEND)) {
833		error = EPERM;
834		goto out;
835	}
836
837	/* Remove the entry from the directory; as it is a file, we do not
838	 * have to change the number of hard links of the directory. */
839	tmpfs_dir_detach(dvp, de);
840	if (v->a_cnp->cn_flags & DOWHITEOUT)
841		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
842
843	/* Free the directory entry we just deleted.  Note that the node
844	 * referred by it will not be removed until the vnode is really
845	 * reclaimed. */
846	tmpfs_free_dirent(tmp, de, TRUE);
847
848	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
849	error = 0;
850
851out:
852
853	return error;
854}
855
856/* --------------------------------------------------------------------- */
857
858static int
859tmpfs_link(struct vop_link_args *v)
860{
861	struct vnode *dvp = v->a_tdvp;
862	struct vnode *vp = v->a_vp;
863	struct componentname *cnp = v->a_cnp;
864
865	int error;
866	struct tmpfs_dirent *de;
867	struct tmpfs_node *node;
868
869	MPASS(VOP_ISLOCKED(dvp));
870	MPASS(cnp->cn_flags & HASBUF);
871	MPASS(dvp != vp); /* XXX When can this be false? */
872
873	node = VP_TO_TMPFS_NODE(vp);
874
875	/* XXX: Why aren't the following two tests done by the caller? */
876
877	/* Hard links of directories are forbidden. */
878	if (vp->v_type == VDIR) {
879		error = EPERM;
880		goto out;
881	}
882
883	/* Cannot create cross-device links. */
884	if (dvp->v_mount != vp->v_mount) {
885		error = EXDEV;
886		goto out;
887	}
888
889	/* Ensure that we do not overflow the maximum number of links imposed
890	 * by the system. */
891	MPASS(node->tn_links <= LINK_MAX);
892	if (node->tn_links == LINK_MAX) {
893		error = EMLINK;
894		goto out;
895	}
896
897	/* We cannot create links of files marked immutable or append-only. */
898	if (node->tn_flags & (IMMUTABLE | APPEND)) {
899		error = EPERM;
900		goto out;
901	}
902
903	/* Allocate a new directory entry to represent the node. */
904	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
905	    cnp->cn_nameptr, cnp->cn_namelen, &de);
906	if (error != 0)
907		goto out;
908
909	/* Insert the new directory entry into the appropriate directory. */
910	if (cnp->cn_flags & ISWHITEOUT)
911		tmpfs_dir_whiteout_remove(dvp, cnp);
912	tmpfs_dir_attach(dvp, de);
913
914	/* vp link count has changed, so update node times. */
915	node->tn_status |= TMPFS_NODE_CHANGED;
916	tmpfs_update(vp);
917
918	error = 0;
919
920out:
921	return error;
922}
923
924/* --------------------------------------------------------------------- */
925
926static int
927tmpfs_rename(struct vop_rename_args *v)
928{
929	struct vnode *fdvp = v->a_fdvp;
930	struct vnode *fvp = v->a_fvp;
931	struct componentname *fcnp = v->a_fcnp;
932	struct vnode *tdvp = v->a_tdvp;
933	struct vnode *tvp = v->a_tvp;
934	struct componentname *tcnp = v->a_tcnp;
935
936	char *newname;
937	int error;
938	struct tmpfs_dirent *de;
939	struct tmpfs_mount *tmp;
940	struct tmpfs_node *fdnode;
941	struct tmpfs_node *fnode;
942	struct tmpfs_node *tnode;
943	struct tmpfs_node *tdnode;
944
945	MPASS(VOP_ISLOCKED(tdvp));
946	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
947	MPASS(fcnp->cn_flags & HASBUF);
948	MPASS(tcnp->cn_flags & HASBUF);
949
950  	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
951
952	/* Disallow cross-device renames.
953	 * XXX Why isn't this done by the caller? */
954	if (fvp->v_mount != tdvp->v_mount ||
955	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
956		error = EXDEV;
957		goto out;
958	}
959
960	tmp = VFS_TO_TMPFS(tdvp->v_mount);
961	tdnode = VP_TO_TMPFS_DIR(tdvp);
962
963	/* If source and target are the same file, there is nothing to do. */
964	if (fvp == tvp) {
965		error = 0;
966		goto out;
967	}
968
969	/* If we need to move the directory between entries, lock the
970	 * source so that we can safely operate on it. */
971	if (tdvp != fdvp) {
972		error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
973		if (error != 0)
974			goto out;
975	}
976	fdnode = VP_TO_TMPFS_DIR(fdvp);
977	fnode = VP_TO_TMPFS_NODE(fvp);
978	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
979
980	/* Entry can disappear before we lock fdvp,
981	 * also avoid manipulating '.' and '..' entries. */
982	if (de == NULL) {
983		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
984		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
985			error = EINVAL;
986		else
987			error = ENOENT;
988		goto out_locked;
989	}
990	MPASS(de->td_node == fnode);
991
992	/* If re-naming a directory to another preexisting directory
993	 * ensure that the target directory is empty so that its
994	 * removal causes no side effects.
995	 * Kern_rename gurantees the destination to be a directory
996	 * if the source is one. */
997	if (tvp != NULL) {
998		MPASS(tnode != NULL);
999
1000		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1001		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1002			error = EPERM;
1003			goto out_locked;
1004		}
1005
1006		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1007			if (tnode->tn_size > 0) {
1008				error = ENOTEMPTY;
1009				goto out_locked;
1010			}
1011		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1012			error = ENOTDIR;
1013			goto out_locked;
1014		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1015			error = EISDIR;
1016			goto out_locked;
1017		} else {
1018			MPASS(fnode->tn_type != VDIR &&
1019				tnode->tn_type != VDIR);
1020		}
1021	}
1022
1023	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1024	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1025		error = EPERM;
1026		goto out_locked;
1027	}
1028
1029	/* Ensure that we have enough memory to hold the new name, if it
1030	 * has to be changed. */
1031	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1032	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1033		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1034	} else
1035		newname = NULL;
1036
1037	/* If the node is being moved to another directory, we have to do
1038	 * the move. */
1039	if (fdnode != tdnode) {
1040		/* In case we are moving a directory, we have to adjust its
1041		 * parent to point to the new parent. */
1042		if (de->td_node->tn_type == VDIR) {
1043			struct tmpfs_node *n;
1044
1045			/* Ensure the target directory is not a child of the
1046			 * directory being moved.  Otherwise, we'd end up
1047			 * with stale nodes. */
1048			n = tdnode;
1049			/* TMPFS_LOCK garanties that no nodes are freed while
1050			 * traversing the list. Nodes can only be marked as
1051			 * removed: tn_parent == NULL. */
1052			TMPFS_LOCK(tmp);
1053			TMPFS_NODE_LOCK(n);
1054			while (n != n->tn_dir.tn_parent) {
1055				struct tmpfs_node *parent;
1056
1057				if (n == fnode) {
1058					TMPFS_NODE_UNLOCK(n);
1059					TMPFS_UNLOCK(tmp);
1060					error = EINVAL;
1061					if (newname != NULL)
1062						    free(newname, M_TMPFSNAME);
1063					goto out_locked;
1064				}
1065				parent = n->tn_dir.tn_parent;
1066				TMPFS_NODE_UNLOCK(n);
1067				if (parent == NULL) {
1068					n = NULL;
1069					break;
1070				}
1071				TMPFS_NODE_LOCK(parent);
1072				if (parent->tn_dir.tn_parent == NULL) {
1073					TMPFS_NODE_UNLOCK(parent);
1074					n = NULL;
1075					break;
1076				}
1077				n = parent;
1078			}
1079			TMPFS_UNLOCK(tmp);
1080			if (n == NULL) {
1081				error = EINVAL;
1082				if (newname != NULL)
1083					    free(newname, M_TMPFSNAME);
1084				goto out_locked;
1085			}
1086			TMPFS_NODE_UNLOCK(n);
1087
1088			/* Adjust the parent pointer. */
1089			TMPFS_VALIDATE_DIR(fnode);
1090			TMPFS_NODE_LOCK(de->td_node);
1091			de->td_node->tn_dir.tn_parent = tdnode;
1092			TMPFS_NODE_UNLOCK(de->td_node);
1093
1094			/* As a result of changing the target of the '..'
1095			 * entry, the link count of the source and target
1096			 * directories has to be adjusted. */
1097			TMPFS_NODE_LOCK(tdnode);
1098			TMPFS_ASSERT_LOCKED(tdnode);
1099			tdnode->tn_links++;
1100			TMPFS_NODE_UNLOCK(tdnode);
1101
1102			TMPFS_NODE_LOCK(fdnode);
1103			TMPFS_ASSERT_LOCKED(fdnode);
1104			fdnode->tn_links--;
1105			TMPFS_NODE_UNLOCK(fdnode);
1106		}
1107
1108		/* Do the move: just remove the entry from the source directory
1109		 * and insert it into the target one. */
1110		tmpfs_dir_detach(fdvp, de);
1111		if (fcnp->cn_flags & DOWHITEOUT)
1112			tmpfs_dir_whiteout_add(fdvp, fcnp);
1113		if (tcnp->cn_flags & ISWHITEOUT)
1114			tmpfs_dir_whiteout_remove(tdvp, tcnp);
1115		tmpfs_dir_attach(tdvp, de);
1116	}
1117
1118	/* If the name has changed, we need to make it effective by changing
1119	 * it in the directory entry. */
1120	if (newname != NULL) {
1121		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1122
1123		free(de->td_name, M_TMPFSNAME);
1124		de->td_namelen = (uint16_t)tcnp->cn_namelen;
1125		memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
1126		de->td_name = newname;
1127
1128		fnode->tn_status |= TMPFS_NODE_CHANGED;
1129		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1130	}
1131
1132	/* If we are overwriting an entry, we have to remove the old one
1133	 * from the target directory. */
1134	if (tvp != NULL) {
1135		/* Remove the old entry from the target directory. */
1136		de = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1137		tmpfs_dir_detach(tdvp, de);
1138
1139		/* Free the directory entry we just deleted.  Note that the
1140		 * node referred by it will not be removed until the vnode is
1141		 * really reclaimed. */
1142		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE);
1143	}
1144
1145	error = 0;
1146
1147out_locked:
1148	if (fdnode != tdnode)
1149		VOP_UNLOCK(fdvp, 0);
1150
1151out:
1152	/* Release target nodes. */
1153	/* XXX: I don't understand when tdvp can be the same as tvp, but
1154	 * other code takes care of this... */
1155	if (tdvp == tvp)
1156		vrele(tdvp);
1157	else
1158		vput(tdvp);
1159	if (tvp != NULL)
1160		vput(tvp);
1161
1162	/* Release source nodes. */
1163	vrele(fdvp);
1164	vrele(fvp);
1165
1166	return error;
1167}
1168
1169/* --------------------------------------------------------------------- */
1170
1171static int
1172tmpfs_mkdir(struct vop_mkdir_args *v)
1173{
1174	struct vnode *dvp = v->a_dvp;
1175	struct vnode **vpp = v->a_vpp;
1176	struct componentname *cnp = v->a_cnp;
1177	struct vattr *vap = v->a_vap;
1178
1179	MPASS(vap->va_type == VDIR);
1180
1181	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1182}
1183
1184/* --------------------------------------------------------------------- */
1185
1186static int
1187tmpfs_rmdir(struct vop_rmdir_args *v)
1188{
1189	struct vnode *dvp = v->a_dvp;
1190	struct vnode *vp = v->a_vp;
1191
1192	int error;
1193	struct tmpfs_dirent *de;
1194	struct tmpfs_mount *tmp;
1195	struct tmpfs_node *dnode;
1196	struct tmpfs_node *node;
1197
1198	MPASS(VOP_ISLOCKED(dvp));
1199	MPASS(VOP_ISLOCKED(vp));
1200
1201	tmp = VFS_TO_TMPFS(dvp->v_mount);
1202	dnode = VP_TO_TMPFS_DIR(dvp);
1203	node = VP_TO_TMPFS_DIR(vp);
1204
1205	/* Directories with more than two entries ('.' and '..') cannot be
1206	 * removed. */
1207	 if (node->tn_size > 0) {
1208		 error = ENOTEMPTY;
1209		 goto out;
1210	 }
1211
1212	if ((dnode->tn_flags & APPEND)
1213	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1214		error = EPERM;
1215		goto out;
1216	}
1217
1218	/* This invariant holds only if we are not trying to remove "..".
1219	  * We checked for that above so this is safe now. */
1220	MPASS(node->tn_dir.tn_parent == dnode);
1221
1222	/* Get the directory entry associated with node (vp).  This was
1223	 * filled by tmpfs_lookup while looking up the entry. */
1224	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1225	MPASS(TMPFS_DIRENT_MATCHES(de,
1226	    v->a_cnp->cn_nameptr,
1227	    v->a_cnp->cn_namelen));
1228
1229	/* Check flags to see if we are allowed to remove the directory. */
1230	if (dnode->tn_flags & APPEND
1231		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1232		error = EPERM;
1233		goto out;
1234	}
1235
1236
1237	/* Detach the directory entry from the directory (dnode). */
1238	tmpfs_dir_detach(dvp, de);
1239	if (v->a_cnp->cn_flags & DOWHITEOUT)
1240		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1241
1242	/* No vnode should be allocated for this entry from this point */
1243	TMPFS_NODE_LOCK(node);
1244	TMPFS_ASSERT_ELOCKED(node);
1245	node->tn_links--;
1246	node->tn_dir.tn_parent = NULL;
1247	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1248	    TMPFS_NODE_MODIFIED;
1249
1250	TMPFS_NODE_UNLOCK(node);
1251
1252	TMPFS_NODE_LOCK(dnode);
1253	TMPFS_ASSERT_ELOCKED(dnode);
1254	dnode->tn_links--;
1255	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1256	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1257	TMPFS_NODE_UNLOCK(dnode);
1258
1259	cache_purge(dvp);
1260	cache_purge(vp);
1261
1262	/* Free the directory entry we just deleted.  Note that the node
1263	 * referred by it will not be removed until the vnode is really
1264	 * reclaimed. */
1265	tmpfs_free_dirent(tmp, de, TRUE);
1266
1267	/* Release the deleted vnode (will destroy the node, notify
1268	 * interested parties and clean it from the cache). */
1269
1270	dnode->tn_status |= TMPFS_NODE_CHANGED;
1271	tmpfs_update(dvp);
1272
1273	error = 0;
1274
1275out:
1276	return error;
1277}
1278
1279/* --------------------------------------------------------------------- */
1280
1281static int
1282tmpfs_symlink(struct vop_symlink_args *v)
1283{
1284	struct vnode *dvp = v->a_dvp;
1285	struct vnode **vpp = v->a_vpp;
1286	struct componentname *cnp = v->a_cnp;
1287	struct vattr *vap = v->a_vap;
1288	char *target = v->a_target;
1289
1290#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1291	MPASS(vap->va_type == VLNK);
1292#else
1293	vap->va_type = VLNK;
1294#endif
1295
1296	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1297}
1298
1299/* --------------------------------------------------------------------- */
1300
1301static int
1302tmpfs_readdir(struct vop_readdir_args *v)
1303{
1304	struct vnode *vp = v->a_vp;
1305	struct uio *uio = v->a_uio;
1306	int *eofflag = v->a_eofflag;
1307	u_long **cookies = v->a_cookies;
1308	int *ncookies = v->a_ncookies;
1309
1310	int error;
1311	off_t startoff;
1312	off_t cnt = 0;
1313	struct tmpfs_node *node;
1314
1315	/* This operation only makes sense on directory nodes. */
1316	if (vp->v_type != VDIR)
1317		return ENOTDIR;
1318
1319	node = VP_TO_TMPFS_DIR(vp);
1320
1321	startoff = uio->uio_offset;
1322
1323	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1324		error = tmpfs_dir_getdotdent(node, uio);
1325		if (error != 0)
1326			goto outok;
1327		cnt++;
1328	}
1329
1330	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1331		error = tmpfs_dir_getdotdotdent(node, uio);
1332		if (error != 0)
1333			goto outok;
1334		cnt++;
1335	}
1336
1337	error = tmpfs_dir_getdents(node, uio, &cnt);
1338
1339outok:
1340	MPASS(error >= -1);
1341
1342	if (error == -1)
1343		error = (cnt != 0) ? 0 : EINVAL;
1344
1345	if (eofflag != NULL)
1346		*eofflag =
1347		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1348
1349	/* Update NFS-related variables. */
1350	if (error == 0 && cookies != NULL && ncookies != NULL) {
1351		off_t i;
1352		off_t off = startoff;
1353		struct tmpfs_dirent *de = NULL;
1354
1355		*ncookies = cnt;
1356		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1357
1358		for (i = 0; i < cnt; i++) {
1359			MPASS(off != TMPFS_DIRCOOKIE_EOF);
1360			if (off == TMPFS_DIRCOOKIE_DOT) {
1361				off = TMPFS_DIRCOOKIE_DOTDOT;
1362			} else {
1363				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1364					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1365				} else if (de != NULL) {
1366					de = TAILQ_NEXT(de, td_entries);
1367				} else {
1368					de = tmpfs_dir_lookupbycookie(node,
1369					    off);
1370					MPASS(de != NULL);
1371					de = TAILQ_NEXT(de, td_entries);
1372				}
1373				if (de == NULL)
1374					off = TMPFS_DIRCOOKIE_EOF;
1375				else
1376					off = tmpfs_dircookie(de);
1377			}
1378
1379			(*cookies)[i] = off;
1380		}
1381		MPASS(uio->uio_offset == off);
1382	}
1383
1384	return error;
1385}
1386
1387/* --------------------------------------------------------------------- */
1388
1389static int
1390tmpfs_readlink(struct vop_readlink_args *v)
1391{
1392	struct vnode *vp = v->a_vp;
1393	struct uio *uio = v->a_uio;
1394
1395	int error;
1396	struct tmpfs_node *node;
1397
1398	MPASS(uio->uio_offset == 0);
1399	MPASS(vp->v_type == VLNK);
1400
1401	node = VP_TO_TMPFS_NODE(vp);
1402
1403	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1404	    uio);
1405	node->tn_status |= TMPFS_NODE_ACCESSED;
1406
1407	return error;
1408}
1409
1410/* --------------------------------------------------------------------- */
1411
1412static int
1413tmpfs_inactive(struct vop_inactive_args *v)
1414{
1415	struct vnode *vp = v->a_vp;
1416	struct thread *l = v->a_td;
1417
1418	struct tmpfs_node *node;
1419
1420	MPASS(VOP_ISLOCKED(vp));
1421
1422	node = VP_TO_TMPFS_NODE(vp);
1423
1424	if (node->tn_links == 0)
1425		vrecycle(vp, l);
1426
1427	return 0;
1428}
1429
1430/* --------------------------------------------------------------------- */
1431
1432int
1433tmpfs_reclaim(struct vop_reclaim_args *v)
1434{
1435	struct vnode *vp = v->a_vp;
1436
1437	struct tmpfs_mount *tmp;
1438	struct tmpfs_node *node;
1439
1440	node = VP_TO_TMPFS_NODE(vp);
1441	tmp = VFS_TO_TMPFS(vp->v_mount);
1442
1443	vnode_destroy_vobject(vp);
1444	cache_purge(vp);
1445
1446	TMPFS_NODE_LOCK(node);
1447	TMPFS_ASSERT_ELOCKED(node);
1448	tmpfs_free_vp(vp);
1449
1450	/* If the node referenced by this vnode was deleted by the user,
1451	 * we must free its associated data structures (now that the vnode
1452	 * is being reclaimed). */
1453	if (node->tn_links == 0 &&
1454	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1455		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1456		TMPFS_NODE_UNLOCK(node);
1457		tmpfs_free_node(tmp, node);
1458	} else
1459		TMPFS_NODE_UNLOCK(node);
1460
1461	MPASS(vp->v_data == NULL);
1462	return 0;
1463}
1464
1465/* --------------------------------------------------------------------- */
1466
1467static int
1468tmpfs_print(struct vop_print_args *v)
1469{
1470	struct vnode *vp = v->a_vp;
1471
1472	struct tmpfs_node *node;
1473
1474	node = VP_TO_TMPFS_NODE(vp);
1475
1476	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1477	    node, node->tn_flags, node->tn_links);
1478	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1479	    ", status 0x%x\n",
1480	    node->tn_mode, node->tn_uid, node->tn_gid,
1481	    (uintmax_t)node->tn_size, node->tn_status);
1482
1483	if (vp->v_type == VFIFO)
1484		fifo_printinfo(vp);
1485
1486	printf("\n");
1487
1488	return 0;
1489}
1490
1491/* --------------------------------------------------------------------- */
1492
1493static int
1494tmpfs_pathconf(struct vop_pathconf_args *v)
1495{
1496	int name = v->a_name;
1497	register_t *retval = v->a_retval;
1498
1499	int error;
1500
1501	error = 0;
1502
1503	switch (name) {
1504	case _PC_LINK_MAX:
1505		*retval = LINK_MAX;
1506		break;
1507
1508	case _PC_NAME_MAX:
1509		*retval = NAME_MAX;
1510		break;
1511
1512	case _PC_PATH_MAX:
1513		*retval = PATH_MAX;
1514		break;
1515
1516	case _PC_PIPE_BUF:
1517		*retval = PIPE_BUF;
1518		break;
1519
1520	case _PC_CHOWN_RESTRICTED:
1521		*retval = 1;
1522		break;
1523
1524	case _PC_NO_TRUNC:
1525		*retval = 1;
1526		break;
1527
1528	case _PC_SYNC_IO:
1529		*retval = 1;
1530		break;
1531
1532	case _PC_FILESIZEBITS:
1533		*retval = 0; /* XXX Don't know which value should I return. */
1534		break;
1535
1536	default:
1537		error = EINVAL;
1538	}
1539
1540	return error;
1541}
1542
1543static int
1544tmpfs_vptofh(struct vop_vptofh_args *ap)
1545{
1546	struct tmpfs_fid *tfhp;
1547	struct tmpfs_node *node;
1548
1549	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1550	node = VP_TO_TMPFS_NODE(ap->a_vp);
1551
1552	tfhp->tf_len = sizeof(struct tmpfs_fid);
1553	tfhp->tf_id = node->tn_id;
1554	tfhp->tf_gen = node->tn_gen;
1555
1556	return (0);
1557}
1558
1559static int
1560tmpfs_whiteout(struct vop_whiteout_args *ap)
1561{
1562	struct vnode *dvp = ap->a_dvp;
1563	struct componentname *cnp = ap->a_cnp;
1564	struct tmpfs_dirent *de;
1565
1566	switch (ap->a_flags) {
1567	case LOOKUP:
1568		return (0);
1569	case CREATE:
1570		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1571		if (de != NULL)
1572			return (de->td_node == NULL ? 0 : EEXIST);
1573		return (tmpfs_dir_whiteout_add(dvp, cnp));
1574	case DELETE:
1575		tmpfs_dir_whiteout_remove(dvp, cnp);
1576		return (0);
1577	default:
1578		panic("tmpfs_whiteout: unknown op");
1579	}
1580}
1581
1582/* --------------------------------------------------------------------- */
1583
1584/*
1585 * vnode operations vector used for files stored in a tmpfs file system.
1586 */
1587struct vop_vector tmpfs_vnodeop_entries = {
1588	.vop_default =			&default_vnodeops,
1589	.vop_lookup =			vfs_cache_lookup,
1590	.vop_cachedlookup =		tmpfs_lookup,
1591	.vop_create =			tmpfs_create,
1592	.vop_mknod =			tmpfs_mknod,
1593	.vop_open =			tmpfs_open,
1594	.vop_close =			tmpfs_close,
1595	.vop_access =			tmpfs_access,
1596	.vop_getattr =			tmpfs_getattr,
1597	.vop_setattr =			tmpfs_setattr,
1598	.vop_read =			tmpfs_read,
1599	.vop_write =			tmpfs_write,
1600	.vop_fsync =			tmpfs_fsync,
1601	.vop_remove =			tmpfs_remove,
1602	.vop_link =			tmpfs_link,
1603	.vop_rename =			tmpfs_rename,
1604	.vop_mkdir =			tmpfs_mkdir,
1605	.vop_rmdir =			tmpfs_rmdir,
1606	.vop_symlink =			tmpfs_symlink,
1607	.vop_readdir =			tmpfs_readdir,
1608	.vop_readlink =			tmpfs_readlink,
1609	.vop_inactive =			tmpfs_inactive,
1610	.vop_reclaim =			tmpfs_reclaim,
1611	.vop_print =			tmpfs_print,
1612	.vop_pathconf =			tmpfs_pathconf,
1613	.vop_vptofh =			tmpfs_vptofh,
1614	.vop_whiteout =			tmpfs_whiteout,
1615	.vop_bmap =			VOP_EOPNOTSUPP,
1616};
1617
1618