xref: /dflybsd-src/sys/vfs/tmpfs/tmpfs_vnops.c (revision 330d3c4b487f3fc5d0eb023645b0b2a569f7048e)
1 /*-
2  * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
3  * All rights reserved.
4  *
5  * This code is derived from software contributed to The NetBSD Foundation
6  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
7  * 2005 program.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28  * POSSIBILITY OF SUCH DAMAGE.
29  *
30  * $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $
31  */
32 
33 /*
34  * tmpfs vnode interface.
35  */
36 #include <sys/cdefs.h>
37 
38 #include <sys/kernel.h>
39 #include <sys/kern_syscall.h>
40 #include <sys/param.h>
41 #include <sys/fcntl.h>
42 #include <sys/lockf.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/resourcevar.h>
46 #include <sys/sched.h>
47 #include <sys/stat.h>
48 #include <sys/systm.h>
49 #include <sys/unistd.h>
50 #include <sys/vfsops.h>
51 #include <sys/vnode.h>
52 
53 #include <sys/mplock2.h>
54 
55 #include <vm/vm.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_pager.h>
59 #include <vm/swap_pager.h>
60 
61 #include <vfs/fifofs/fifo.h>
62 #include <vfs/tmpfs/tmpfs_vnops.h>
63 #include <vfs/tmpfs/tmpfs.h>
64 
65 MALLOC_DECLARE(M_TMPFS);
66 
67 static __inline
68 void
69 tmpfs_knote(struct vnode *vp, int flags)
70 {
71 	if (flags)
72 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
73 }
74 
75 
76 /* --------------------------------------------------------------------- */
77 
78 static int
79 tmpfs_nresolve(struct vop_nresolve_args *v)
80 {
81 	struct vnode *dvp = v->a_dvp;
82 	struct vnode *vp = NULL;
83 	struct namecache *ncp = v->a_nch->ncp;
84 	struct tmpfs_node *tnode;
85 
86 	int error;
87 	struct tmpfs_dirent *de;
88 	struct tmpfs_node *dnode;
89 
90 	dnode = VP_TO_TMPFS_DIR(dvp);
91 
92 	de = tmpfs_dir_lookup(dnode, NULL, ncp);
93 	if (de == NULL) {
94 		error = ENOENT;
95 	} else {
96 		/*
97 		 * Allocate a vnode for the node we found.
98 		 */
99 		tnode = de->td_node;
100 		error = tmpfs_alloc_vp(dvp->v_mount, tnode,
101 				       LK_EXCLUSIVE | LK_RETRY, &vp);
102 		if (error)
103 			goto out;
104 		KKASSERT(vp);
105 	}
106 
107 out:
108 	/*
109 	 * Store the result of this lookup in the cache.  Avoid this if the
110 	 * request was for creation, as it does not improve timings on
111 	 * emprical tests.
112 	 */
113 	if (vp) {
114 		vn_unlock(vp);
115 		cache_setvp(v->a_nch, vp);
116 		vrele(vp);
117 	} else if (error == ENOENT) {
118 		cache_setvp(v->a_nch, NULL);
119 	}
120 	return error;
121 }
122 
123 static int
124 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
125 {
126 	struct vnode *dvp = v->a_dvp;
127 	struct vnode **vpp = v->a_vpp;
128 	struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
129 	struct ucred *cred = v->a_cred;
130 	int error;
131 
132 	*vpp = NULL;
133 	/* Check accessibility of requested node as a first step. */
134 	error = VOP_ACCESS(dvp, VEXEC, cred);
135 	if (error != 0)
136 		return error;
137 
138 	if (dnode->tn_dir.tn_parent != NULL) {
139 		/* Allocate a new vnode on the matching entry. */
140 		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
141 		    LK_EXCLUSIVE | LK_RETRY, vpp);
142 
143 		if (*vpp)
144 			vn_unlock(*vpp);
145 	}
146 
147 	return (*vpp == NULL) ? ENOENT : 0;
148 }
149 
150 /* --------------------------------------------------------------------- */
151 
152 static int
153 tmpfs_ncreate(struct vop_ncreate_args *v)
154 {
155 	struct vnode *dvp = v->a_dvp;
156 	struct vnode **vpp = v->a_vpp;
157 	struct namecache *ncp = v->a_nch->ncp;
158 	struct vattr *vap = v->a_vap;
159 	struct ucred *cred = v->a_cred;
160 	int error;
161 
162 	KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
163 
164 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
165 	if (error == 0) {
166 		cache_setunresolved(v->a_nch);
167 		cache_setvp(v->a_nch, *vpp);
168 		tmpfs_knote(dvp, NOTE_WRITE);
169 	}
170 
171 	return error;
172 }
173 /* --------------------------------------------------------------------- */
174 
175 static int
176 tmpfs_nmknod(struct vop_nmknod_args *v)
177 {
178 	struct vnode *dvp = v->a_dvp;
179 	struct vnode **vpp = v->a_vpp;
180 	struct namecache *ncp = v->a_nch->ncp;
181 	struct vattr *vap = v->a_vap;
182 	struct ucred *cred = v->a_cred;
183 	int error;
184 
185 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
186 	    vap->va_type != VFIFO)
187 		return EINVAL;
188 
189 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
190 	if (error == 0) {
191 		cache_setunresolved(v->a_nch);
192 		cache_setvp(v->a_nch, *vpp);
193 		tmpfs_knote(dvp, NOTE_WRITE);
194 	}
195 
196 	return error;
197 }
198 
199 /* --------------------------------------------------------------------- */
200 
201 static int
202 tmpfs_open(struct vop_open_args *v)
203 {
204 	struct vnode *vp = v->a_vp;
205 	int mode = v->a_mode;
206 
207 	int error;
208 	struct tmpfs_node *node;
209 
210 	node = VP_TO_TMPFS_NODE(vp);
211 
212 	/* The file is still active but all its names have been removed
213 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
214 	 * it is about to die. */
215 	if (node->tn_links < 1)
216 		return (ENOENT);
217 
218 	/* If the file is marked append-only, deny write requests. */
219 	if ((node->tn_flags & APPEND) &&
220 	    (mode & (FWRITE | O_APPEND)) == FWRITE) {
221 		error = EPERM;
222 	} else {
223 		return (vop_stdopen(v));
224 	}
225 	return error;
226 }
227 
228 /* --------------------------------------------------------------------- */
229 
230 static int
231 tmpfs_close(struct vop_close_args *v)
232 {
233 	struct vnode *vp = v->a_vp;
234 	struct tmpfs_node *node;
235 
236 	node = VP_TO_TMPFS_NODE(vp);
237 
238 	if (node->tn_links > 0) {
239 		/* Update node times.  No need to do it if the node has
240 		 * been deleted, because it will vanish after we return. */
241 		tmpfs_update(vp);
242 	}
243 
244 	return vop_stdclose(v);
245 }
246 
247 /* --------------------------------------------------------------------- */
248 
249 int
250 tmpfs_access(struct vop_access_args *v)
251 {
252 	struct vnode *vp = v->a_vp;
253 	int error;
254 	struct tmpfs_node *node;
255 
256 	node = VP_TO_TMPFS_NODE(vp);
257 
258 	switch (vp->v_type) {
259 	case VDIR:
260 		/* FALLTHROUGH */
261 	case VLNK:
262 		/* FALLTHROUGH */
263 	case VREG:
264 		if (VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
265 			error = EROFS;
266 			goto out;
267 		}
268 		break;
269 
270 	case VBLK:
271 		/* FALLTHROUGH */
272 	case VCHR:
273 		/* FALLTHROUGH */
274 	case VSOCK:
275 		/* FALLTHROUGH */
276 	case VFIFO:
277 		break;
278 
279 	default:
280 		error = EINVAL;
281 		goto out;
282 	}
283 
284 	if (VWRITE && node->tn_flags & IMMUTABLE) {
285 		error = EPERM;
286 		goto out;
287 	}
288 
289 	error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
290 
291 out:
292 
293 	return error;
294 }
295 
296 /* --------------------------------------------------------------------- */
297 
298 int
299 tmpfs_getattr(struct vop_getattr_args *v)
300 {
301 	struct vnode *vp = v->a_vp;
302 	struct vattr *vap = v->a_vap;
303 	struct tmpfs_node *node;
304 
305 	node = VP_TO_TMPFS_NODE(vp);
306 
307 	tmpfs_update(vp);
308 
309 	vap->va_type = vp->v_type;
310 	vap->va_mode = node->tn_mode;
311 	vap->va_nlink = node->tn_links;
312 	vap->va_uid = node->tn_uid;
313 	vap->va_gid = node->tn_gid;
314 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
315 	vap->va_fileid = node->tn_id;
316 	vap->va_size = node->tn_size;
317 	vap->va_blocksize = PAGE_SIZE;
318 	vap->va_atime.tv_sec = node->tn_atime;
319 	vap->va_atime.tv_nsec = node->tn_atimensec;
320 	vap->va_mtime.tv_sec = node->tn_mtime;
321 	vap->va_mtime.tv_nsec = node->tn_mtimensec;
322 	vap->va_ctime.tv_sec = node->tn_ctime;
323 	vap->va_ctime.tv_nsec = node->tn_ctimensec;
324 	vap->va_gen = node->tn_gen;
325 	vap->va_flags = node->tn_flags;
326 	if (vp->v_type == VBLK || vp->v_type == VCHR)
327 	{
328 		vap->va_rmajor = umajor(node->tn_rdev);
329 		vap->va_rminor = uminor(node->tn_rdev);
330 	}
331 	vap->va_bytes = round_page(node->tn_size);
332 	vap->va_filerev = 0;
333 
334 	return 0;
335 }
336 
337 /* --------------------------------------------------------------------- */
338 
339 int
340 tmpfs_setattr(struct vop_setattr_args *v)
341 {
342 	struct vnode *vp = v->a_vp;
343 	struct vattr *vap = v->a_vap;
344 	struct ucred *cred = v->a_cred;
345 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
346 	int error = 0;
347 	int kflags = 0;
348 
349 	if (error == 0 && (vap->va_flags != VNOVAL)) {
350 		error = tmpfs_chflags(vp, vap->va_flags, cred);
351 		kflags |= NOTE_ATTRIB;
352 	}
353 
354 	if (error == 0 && (vap->va_size != VNOVAL)) {
355 		if (vap->va_size > node->tn_size)
356 			kflags |= NOTE_WRITE | NOTE_EXTEND;
357 		else
358 			kflags |= NOTE_WRITE;
359 		error = tmpfs_chsize(vp, vap->va_size, cred);
360 	}
361 
362 	if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
363 			   vap->va_gid != (gid_t)VNOVAL)) {
364 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
365 		kflags |= NOTE_ATTRIB;
366 	}
367 
368 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
369 		error = tmpfs_chmod(vp, vap->va_mode, cred);
370 		kflags |= NOTE_ATTRIB;
371 	}
372 
373 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
374 	    vap->va_atime.tv_nsec != VNOVAL) ||
375 	    (vap->va_mtime.tv_sec != VNOVAL &&
376 	    vap->va_mtime.tv_nsec != VNOVAL) )) {
377 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
378 				      vap->va_vaflags, cred);
379 		kflags |= NOTE_ATTRIB;
380 	}
381 
382 	/* Update the node times.  We give preference to the error codes
383 	 * generated by this function rather than the ones that may arise
384 	 * from tmpfs_update. */
385 	tmpfs_update(vp);
386 	tmpfs_knote(vp, kflags);
387 
388 	return error;
389 }
390 
391 /* --------------------------------------------------------------------- */
392 
393 /*
394  * fsync is usually a NOP, but we must take action when unmounting or
395  * when recycling.
396  */
397 static int
398 tmpfs_fsync(struct vop_fsync_args *v)
399 {
400 	struct tmpfs_mount *tmp;
401 	struct tmpfs_node *node;
402 	struct vnode *vp = v->a_vp;
403 
404 	tmp = VFS_TO_TMPFS(vp->v_mount);
405 	node = VP_TO_TMPFS_NODE(vp);
406 
407 	tmpfs_update(vp);
408 	if (vp->v_type == VREG) {
409 		if (vp->v_flag & VRECLAIMED) {
410 			if (node->tn_links == 0)
411 				tmpfs_truncate(vp, 0);
412 			else
413 				vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
414 		}
415 	}
416 	return 0;
417 }
418 
419 /* --------------------------------------------------------------------- */
420 
421 static int
422 tmpfs_read (struct vop_read_args *ap)
423 {
424 	struct buf *bp;
425 	struct vnode *vp = ap->a_vp;
426 	struct uio *uio = ap->a_uio;
427 	struct tmpfs_node *node;
428 	off_t base_offset;
429 	size_t offset;
430 	size_t len;
431 	int got_mplock;
432 	int error;
433 
434 	error = 0;
435 	if (uio->uio_resid == 0) {
436 		return error;
437 	}
438 
439 	node = VP_TO_TMPFS_NODE(vp);
440 
441 	if (uio->uio_offset < 0)
442 		return (EINVAL);
443 	if (vp->v_type != VREG)
444 		return (EINVAL);
445 
446 #ifdef SMP
447 	if(curthread->td_mpcount)
448 		got_mplock = -1;
449 	else
450 		got_mplock = 0;
451 #else
452 		got_mplock = -1;
453 #endif
454 
455 	while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
456 		/*
457 		 * Use buffer cache I/O (via tmpfs_strategy)
458 		 */
459 		offset = (size_t)uio->uio_offset & BMASK;
460 		base_offset = (off_t)uio->uio_offset - offset;
461 		bp = getcacheblk(vp, base_offset);
462 		if (bp == NULL)
463 		{
464 			if (got_mplock == 0) {
465 				got_mplock = 1;
466 				get_mplock();
467 			}
468 
469 			error = bread(vp, base_offset, BSIZE, &bp);
470 			if (error) {
471 				brelse(bp);
472 				kprintf("tmpfs_read bread error %d\n", error);
473 				break;
474 			}
475 		}
476 
477 		if (got_mplock == 0) {
478 			got_mplock = 1;
479 			get_mplock();
480 		}
481 
482 		/*
483 		 * Figure out how many bytes we can actually copy this loop.
484 		 */
485 		len = BSIZE - offset;
486 		if (len > uio->uio_resid)
487 			len = uio->uio_resid;
488 		if (len > node->tn_size - uio->uio_offset)
489 			len = (size_t)(node->tn_size - uio->uio_offset);
490 
491 		error = uiomove((char *)bp->b_data + offset, len, uio);
492 		bqrelse(bp);
493 		if (error) {
494 			kprintf("tmpfs_read uiomove error %d\n", error);
495 			break;
496 		}
497 	}
498 
499 	if (got_mplock > 0)
500 		rel_mplock();
501 
502 	TMPFS_NODE_LOCK(node);
503 	node->tn_status |= TMPFS_NODE_ACCESSED;
504 	TMPFS_NODE_UNLOCK(node);
505 
506 	return(error);
507 }
508 
509 static int
510 tmpfs_write (struct vop_write_args *ap)
511 {
512 	struct buf *bp;
513 	struct vnode *vp = ap->a_vp;
514 	struct uio *uio = ap->a_uio;
515 	struct thread *td = uio->uio_td;
516 	struct tmpfs_node *node;
517 	boolean_t extended;
518 	off_t oldsize;
519 	int error;
520 	off_t base_offset;
521 	size_t offset;
522 	size_t len;
523 	struct rlimit limit;
524 	int got_mplock;
525 	int trivial = 0;
526 	int kflags = 0;
527 
528 	error = 0;
529 	if (uio->uio_resid == 0) {
530 		return error;
531 	}
532 
533 	node = VP_TO_TMPFS_NODE(vp);
534 
535 	if (vp->v_type != VREG)
536 		return (EINVAL);
537 
538 	oldsize = node->tn_size;
539 	if (ap->a_ioflag & IO_APPEND)
540 		uio->uio_offset = node->tn_size;
541 
542 	/*
543 	 * Check for illegal write offsets.
544 	 */
545 	if (uio->uio_offset + uio->uio_resid >
546 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
547 		return (EFBIG);
548 
549 	if (vp->v_type == VREG && td != NULL) {
550 		error = kern_getrlimit(RLIMIT_FSIZE, &limit);
551 		if (error != 0)
552 			return error;
553 		if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
554 			ksignal(td->td_proc, SIGXFSZ);
555 			return (EFBIG);
556 		}
557 	}
558 
559 
560 	/*
561 	 * Extend the file's size if necessary
562 	 */
563 	extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
564 
565 #ifdef SMP
566 	if (curthread->td_mpcount) {
567 		got_mplock = -1;
568 	} else {
569 		got_mplock = 1;
570 		get_mplock();
571 	}
572 #else
573 	got_mplock = -1;
574 #endif
575 	while (uio->uio_resid > 0) {
576 		/*
577 		 * Use buffer cache I/O (via tmpfs_strategy)
578 		 */
579 		offset = (size_t)uio->uio_offset & BMASK;
580 		base_offset = (off_t)uio->uio_offset - offset;
581 		len = BSIZE - offset;
582 		if (len > uio->uio_resid)
583 			len = uio->uio_resid;
584 
585 		if ((uio->uio_offset + len) > node->tn_size) {
586 			trivial = (uio->uio_offset <= node->tn_size);
587 			error = tmpfs_reg_resize(vp, uio->uio_offset + len,  trivial);
588 			if (error)
589 				break;
590 		}
591 
592 		/*
593 		 * Read to fill in any gaps.  Theoretically we could
594 		 * optimize this if the write covers the entire buffer
595 		 * and is not a UIO_NOCOPY write, however this can lead
596 		 * to a security violation exposing random kernel memory
597 		 * (whatever junk was in the backing VM pages before).
598 		 *
599 		 * So just use bread() to do the right thing.
600 		 */
601 		error = bread(vp, base_offset, BSIZE, &bp);
602 		error = uiomove((char *)bp->b_data + offset, len, uio);
603 		if (error) {
604 			kprintf("tmpfs_write uiomove error %d\n", error);
605 			brelse(bp);
606 			break;
607 		}
608 
609 		if (uio->uio_offset > node->tn_size) {
610 			node->tn_size = uio->uio_offset;
611 			kflags |= NOTE_EXTEND;
612 		}
613 		kflags |= NOTE_WRITE;
614 
615 		/*
616 		 * The data has been loaded into the buffer, write it out.
617 		 *
618 		 * We want tmpfs to be able to use all available ram, not
619 		 * just the buffer cache, so if not explicitly paging we
620 		 * use buwrite() to leave the buffer clean but mark all the
621 		 * VM pages valid+dirty.
622 		 *
623 		 * When the kernel is paging, either via normal pageout
624 		 * operation or when cleaning the object during a recycle,
625 		 * the underlying VM pages are going to get thrown away
626 		 * so we MUST write them to swap.
627 		 *
628 		 * XXX unfortunately this catches msync() system calls too
629 		 * for the moment.
630 		 */
631 		if (vm_swap_size == 0) {
632 			/*
633 			 * if swap isn't configured yet, force a buwrite() to
634 			 * avoid problems further down the line, due to flushing
635 			 * to swap.
636 			 */
637 			buwrite(bp);
638 		} else {
639 			if (ap->a_ioflag & IO_SYNC) {
640 				bwrite(bp);
641 			} else if ((ap->a_ioflag & IO_ASYNC) ||
642 				 (uio->uio_segflg == UIO_NOCOPY)) {
643 				bawrite(bp);
644 			} else {
645 				buwrite(bp);
646 			}
647 		}
648 
649 		if (bp->b_error) {
650 			kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
651 			break;
652 		}
653 	}
654 
655 	if (got_mplock > 0)
656 		rel_mplock();
657 
658 	if (error) {
659 		if (extended) {
660 			(void)tmpfs_reg_resize(vp, oldsize, trivial);
661 			kflags &= ~NOTE_EXTEND;
662 		}
663 		goto done;
664 	}
665 
666 	TMPFS_NODE_LOCK(node);
667 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
668 	    (extended? TMPFS_NODE_CHANGED : 0);
669 
670 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
671 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
672 			node->tn_mode &= ~(S_ISUID | S_ISGID);
673 	}
674 	TMPFS_NODE_UNLOCK(node);
675 done:
676 
677 	tmpfs_knote(vp, kflags);
678 	return(error);
679 }
680 
681 static int
682 tmpfs_advlock (struct vop_advlock_args *ap)
683 {
684 	struct tmpfs_node *node;
685 	struct vnode *vp = ap->a_vp;
686 
687 	node = VP_TO_TMPFS_NODE(vp);
688 
689 	return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
690 }
691 
692 static int
693 tmpfs_strategy(struct vop_strategy_args *ap)
694 {
695 	struct bio *bio = ap->a_bio;
696 	struct buf *bp = bio->bio_buf;
697 	struct vnode *vp = ap->a_vp;
698 	struct tmpfs_node *node;
699 	vm_object_t uobj;
700 
701 	if (vp->v_type != VREG) {
702 		bp->b_resid = bp->b_bcount;
703 		bp->b_flags |= B_ERROR | B_INVAL;
704 		bp->b_error = EINVAL;
705 		biodone(bio);
706 		return(0);
707 	}
708 
709 	node = VP_TO_TMPFS_NODE(vp);
710 
711 	uobj = node->tn_reg.tn_aobj;
712 
713 	/*
714 	 * Call swap_pager_strategy to read or write between the VM
715 	 * object and the buffer cache.
716 	 */
717 	swap_pager_strategy(uobj, bio);
718 
719 	return 0;
720 }
721 
722 static int
723 tmpfs_bmap(struct vop_bmap_args *ap)
724 {
725 	if (ap->a_doffsetp != NULL)
726 		*ap->a_doffsetp = ap->a_loffset;
727 	if (ap->a_runp != NULL)
728 		*ap->a_runp = 0;
729 	if (ap->a_runb != NULL)
730 		*ap->a_runb = 0;
731 
732 	return 0;
733 }
734 
735 /* --------------------------------------------------------------------- */
736 
737 static int
738 tmpfs_nremove(struct vop_nremove_args *v)
739 {
740 	struct vnode *dvp = v->a_dvp;
741 	struct namecache *ncp = v->a_nch->ncp;
742 	struct vnode *vp;
743 	int error;
744 	struct tmpfs_dirent *de;
745 	struct tmpfs_mount *tmp;
746 	struct tmpfs_node *dnode;
747 	struct tmpfs_node *node;
748 
749 	/*
750 	 * We have to acquire the vp from v->a_nch because
751 	 * we will likely unresolve the namecache entry, and
752 	 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
753 	 * sequence to recover space from the file.
754 	 */
755 	error = cache_vref(v->a_nch, v->a_cred, &vp);
756 	KKASSERT(error == 0);
757 
758 	if (vp->v_type == VDIR) {
759 		error = EISDIR;
760 		goto out;
761 	}
762 
763 	dnode = VP_TO_TMPFS_DIR(dvp);
764 	node = VP_TO_TMPFS_NODE(vp);
765 	tmp = VFS_TO_TMPFS(vp->v_mount);
766 	de = tmpfs_dir_lookup(dnode, node, ncp);
767 	if (de == NULL) {
768 		error = ENOENT;
769 		goto out;
770 	}
771 
772 	/* Files marked as immutable or append-only cannot be deleted. */
773 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
774 	    (dnode->tn_flags & APPEND)) {
775 		error = EPERM;
776 		goto out;
777 	}
778 
779 	/* Remove the entry from the directory; as it is a file, we do not
780 	 * have to change the number of hard links of the directory. */
781 	tmpfs_dir_detach(dnode, de);
782 
783 	/* Free the directory entry we just deleted.  Note that the node
784 	 * referred by it will not be removed until the vnode is really
785 	 * reclaimed. */
786 	tmpfs_free_dirent(tmp, de);
787 
788 	if (node->tn_links > 0) {
789 	        TMPFS_NODE_LOCK(node);
790 		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
791 	                TMPFS_NODE_MODIFIED;
792 	        TMPFS_NODE_UNLOCK(node);
793 	}
794 
795 	cache_setunresolved(v->a_nch);
796 	cache_setvp(v->a_nch, NULL);
797 	tmpfs_knote(vp, NOTE_DELETE);
798 	/*cache_inval_vp(vp, CINV_DESTROY);*/
799 	tmpfs_knote(dvp, NOTE_WRITE);
800 	error = 0;
801 
802 out:
803 	vrele(vp);
804 
805 	return error;
806 }
807 
808 /* --------------------------------------------------------------------- */
809 
810 static int
811 tmpfs_nlink(struct vop_nlink_args *v)
812 {
813 	struct vnode *dvp = v->a_dvp;
814 	struct vnode *vp = v->a_vp;
815 	struct namecache *ncp = v->a_nch->ncp;
816 	struct tmpfs_dirent *de;
817 	struct tmpfs_node *node;
818 	struct tmpfs_node *dnode;
819 	int error;
820 
821 	KKASSERT(dvp != vp); /* XXX When can this be false? */
822 
823 	node = VP_TO_TMPFS_NODE(vp);
824 	dnode = VP_TO_TMPFS_NODE(dvp);
825 
826 	/* XXX: Why aren't the following two tests done by the caller? */
827 
828 	/* Hard links of directories are forbidden. */
829 	if (vp->v_type == VDIR) {
830 		error = EPERM;
831 		goto out;
832 	}
833 
834 	/* Cannot create cross-device links. */
835 	if (dvp->v_mount != vp->v_mount) {
836 		error = EXDEV;
837 		goto out;
838 	}
839 
840 	/* Ensure that we do not overflow the maximum number of links imposed
841 	 * by the system. */
842 	KKASSERT(node->tn_links <= LINK_MAX);
843 	if (node->tn_links == LINK_MAX) {
844 		error = EMLINK;
845 		goto out;
846 	}
847 
848 	/* We cannot create links of files marked immutable or append-only. */
849 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
850 		error = EPERM;
851 		goto out;
852 	}
853 
854 	/* Allocate a new directory entry to represent the node. */
855 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
856 	    ncp->nc_name, ncp->nc_nlen, &de);
857 	if (error != 0)
858 		goto out;
859 
860 	/* Insert the new directory entry into the appropriate directory. */
861 	tmpfs_dir_attach(dnode, de);
862 
863 	/* vp link count has changed, so update node times. */
864 
865 	TMPFS_NODE_LOCK(node);
866 	node->tn_status |= TMPFS_NODE_CHANGED;
867 	TMPFS_NODE_UNLOCK(node);
868 	tmpfs_update(vp);
869 
870 	tmpfs_knote(vp, NOTE_LINK);
871 	cache_setunresolved(v->a_nch);
872 	cache_setvp(v->a_nch, vp);
873 	tmpfs_knote(dvp, NOTE_WRITE);
874 	error = 0;
875 
876 out:
877 	return error;
878 }
879 
880 /* --------------------------------------------------------------------- */
881 
882 static int
883 tmpfs_nrename(struct vop_nrename_args *v)
884 {
885 	struct vnode *fdvp = v->a_fdvp;
886 	struct namecache *fncp = v->a_fnch->ncp;
887 	struct vnode *fvp = fncp->nc_vp;
888 	struct vnode *tdvp = v->a_tdvp;
889 	struct namecache *tncp = v->a_tnch->ncp;
890 	struct vnode *tvp = tncp->nc_vp;
891 	struct tmpfs_dirent *de;
892 	struct tmpfs_mount *tmp;
893 	struct tmpfs_node *fdnode;
894 	struct tmpfs_node *fnode;
895 	struct tmpfs_node *tnode;
896 	struct tmpfs_node *tdnode;
897 	char *newname;
898 	char *oldname;
899 	int error;
900 
901 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
902 
903 	/* Disallow cross-device renames.
904 	 * XXX Why isn't this done by the caller? */
905 	if (fvp->v_mount != tdvp->v_mount ||
906 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
907 		error = EXDEV;
908 		goto out;
909 	}
910 
911 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
912 	tdnode = VP_TO_TMPFS_DIR(tdvp);
913 
914 	/* If source and target are the same file, there is nothing to do. */
915 	if (fvp == tvp) {
916 		error = 0;
917 		goto out;
918 	}
919 
920 	fdnode = VP_TO_TMPFS_DIR(fdvp);
921 	fnode = VP_TO_TMPFS_NODE(fvp);
922 	de = tmpfs_dir_lookup(fdnode, fnode, fncp);
923 
924 	/* Avoid manipulating '.' and '..' entries. */
925 	if (de == NULL) {
926 		error = ENOENT;
927 		goto out_locked;
928 	}
929 	KKASSERT(de->td_node == fnode);
930 
931 	/*
932 	 * If replacing an entry in the target directory and that entry
933 	 * is a directory, it must be empty.
934 	 *
935 	 * Kern_rename gurantees the destination to be a directory
936 	 * if the source is one (it does?).
937 	 */
938 	if (tvp != NULL) {
939 		KKASSERT(tnode != NULL);
940 
941 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
942 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
943 			error = EPERM;
944 			goto out_locked;
945 		}
946 
947 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
948 			if (tnode->tn_size > 0) {
949 				error = ENOTEMPTY;
950 				goto out_locked;
951 			}
952 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
953 			error = ENOTDIR;
954 			goto out_locked;
955 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
956 			error = EISDIR;
957 			goto out_locked;
958 		} else {
959 			KKASSERT(fnode->tn_type != VDIR &&
960 				tnode->tn_type != VDIR);
961 		}
962 	}
963 
964 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
965 	    (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
966 		error = EPERM;
967 		goto out_locked;
968 	}
969 
970 	/*
971 	 * Ensure that we have enough memory to hold the new name, if it
972 	 * has to be changed.
973 	 */
974 	if (fncp->nc_nlen != tncp->nc_nlen ||
975 	    bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
976 		newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
977 				  M_WAITOK | M_NULLOK);
978 		if (newname == NULL) {
979 			error = ENOSPC;
980 			goto out_locked;
981 		}
982 		bcopy(tncp->nc_name, newname, tncp->nc_nlen);
983 		newname[tncp->nc_nlen] = '\0';
984 	} else {
985 		newname = NULL;
986 	}
987 
988 	/*
989 	 * Unlink entry from source directory.  Note that the kernel has
990 	 * already checked for illegal recursion cases (renaming a directory
991 	 * into a subdirectory of itself).
992 	 */
993 	if (fdnode != tdnode)
994 		tmpfs_dir_detach(fdnode, de);
995 
996 	/*
997 	 * Handle any name change.  Swap with newname, we will
998 	 * deallocate it at the end.
999 	 */
1000 	if (newname != NULL) {
1001 #if 0
1002 		TMPFS_NODE_LOCK(fnode);
1003 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1004 		TMPFS_NODE_UNLOCK(fnode);
1005 #endif
1006 		oldname = de->td_name;
1007 		de->td_name = newname;
1008 		de->td_namelen = (uint16_t)tncp->nc_nlen;
1009 		newname = oldname;
1010 	}
1011 
1012 	/*
1013 	 * Link entry to target directory.  If the entry
1014 	 * represents a directory move the parent linkage
1015 	 * as well.
1016 	 */
1017 	if (fdnode != tdnode) {
1018 		if (de->td_node->tn_type == VDIR) {
1019 			TMPFS_VALIDATE_DIR(fnode);
1020 
1021 			TMPFS_NODE_LOCK(tdnode);
1022 			tdnode->tn_links++;
1023 			tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1024 			TMPFS_NODE_UNLOCK(tdnode);
1025 
1026 			TMPFS_NODE_LOCK(fnode);
1027 			fnode->tn_dir.tn_parent = tdnode;
1028 			fnode->tn_status |= TMPFS_NODE_CHANGED;
1029 			TMPFS_NODE_UNLOCK(fnode);
1030 
1031 			TMPFS_NODE_LOCK(fdnode);
1032 			fdnode->tn_links--;
1033 			fdnode->tn_status |= TMPFS_NODE_MODIFIED;
1034 			TMPFS_NODE_UNLOCK(fdnode);
1035 		}
1036 		tmpfs_dir_attach(tdnode, de);
1037 	} else {
1038 		TMPFS_NODE_LOCK(tdnode);
1039 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1040 		TMPFS_NODE_UNLOCK(tdnode);
1041 	}
1042 
1043 	/*
1044 	 * If we are overwriting an entry, we have to remove the old one
1045 	 * from the target directory.
1046 	 */
1047 	if (tvp != NULL) {
1048 		/* Remove the old entry from the target directory. */
1049 		de = tmpfs_dir_lookup(tdnode, tnode, tncp);
1050 		tmpfs_dir_detach(tdnode, de);
1051 		tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1052 
1053 		/*
1054 		 * Free the directory entry we just deleted.  Note that the
1055 		 * node referred by it will not be removed until the vnode is
1056 		 * really reclaimed.
1057 		 */
1058 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de);
1059 		/*cache_inval_vp(tvp, CINV_DESTROY);*/
1060 	}
1061 
1062 	/*
1063 	 * Finish up
1064 	 */
1065 	if (newname) {
1066 		kfree(newname, tmp->tm_name_zone);
1067 		newname = NULL;
1068 	}
1069 	cache_rename(v->a_fnch, v->a_tnch);
1070 	tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1071 	tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1072 	if (fnode->tn_vnode)
1073 		tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1074 	error = 0;
1075 
1076 out_locked:
1077 	;
1078 
1079 out:
1080 	/* Release target nodes. */
1081 	/* XXX: I don't understand when tdvp can be the same as tvp, but
1082 	 * other code takes care of this... */
1083 	if (tdvp == tvp)
1084 		vrele(tdvp);
1085 
1086 	return error;
1087 }
1088 
1089 /* --------------------------------------------------------------------- */
1090 
1091 static int
1092 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1093 {
1094 	struct vnode *dvp = v->a_dvp;
1095 	struct vnode **vpp = v->a_vpp;
1096 	struct namecache *ncp = v->a_nch->ncp;
1097 	struct vattr *vap = v->a_vap;
1098 	struct ucred *cred = v->a_cred;
1099 	int error;
1100 
1101 	KKASSERT(vap->va_type == VDIR);
1102 
1103 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1104 	if (error == 0) {
1105 		cache_setunresolved(v->a_nch);
1106 		cache_setvp(v->a_nch, *vpp);
1107 		tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1108 	}
1109 
1110 	return error;
1111 }
1112 
1113 /* --------------------------------------------------------------------- */
1114 
1115 static int
1116 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1117 {
1118 	struct vnode *dvp = v->a_dvp;
1119 	struct namecache *ncp = v->a_nch->ncp;
1120 	struct vnode *vp;
1121 	struct tmpfs_dirent *de;
1122 	struct tmpfs_mount *tmp;
1123 	struct tmpfs_node *dnode;
1124 	struct tmpfs_node *node;
1125 	int error;
1126 
1127 	/*
1128 	 * We have to acquire the vp from v->a_nch because
1129 	 * we will likely unresolve the namecache entry, and
1130 	 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
1131 	 * sequence.
1132 	 */
1133 	error = cache_vref(v->a_nch, v->a_cred, &vp);
1134 	KKASSERT(error == 0);
1135 
1136 	/*
1137 	 * Prevalidate so we don't hit an assertion later
1138 	 */
1139 	if (vp->v_type != VDIR) {
1140 		error = ENOTDIR;
1141 		goto out;
1142 	}
1143 
1144 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1145 	dnode = VP_TO_TMPFS_DIR(dvp);
1146 	node = VP_TO_TMPFS_DIR(vp);
1147 
1148 	/* Directories with more than two entries ('.' and '..') cannot be
1149 	 * removed. */
1150 	 if (node->tn_size > 0) {
1151 		 error = ENOTEMPTY;
1152 		 goto out;
1153 	 }
1154 
1155 	if ((dnode->tn_flags & APPEND)
1156 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1157 		error = EPERM;
1158 		goto out;
1159 	}
1160 
1161 	/* This invariant holds only if we are not trying to remove "..".
1162 	  * We checked for that above so this is safe now. */
1163 	KKASSERT(node->tn_dir.tn_parent == dnode);
1164 
1165 	/* Get the directory entry associated with node (vp).  This was
1166 	 * filled by tmpfs_lookup while looking up the entry. */
1167 	de = tmpfs_dir_lookup(dnode, node, ncp);
1168 	KKASSERT(TMPFS_DIRENT_MATCHES(de,
1169 	    ncp->nc_name,
1170 	    ncp->nc_nlen));
1171 
1172 	/* Check flags to see if we are allowed to remove the directory. */
1173 	if ((dnode->tn_flags & APPEND) ||
1174 	    node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1175 		error = EPERM;
1176 		goto out;
1177 	}
1178 
1179 
1180 	/* Detach the directory entry from the directory (dnode). */
1181 	tmpfs_dir_detach(dnode, de);
1182 
1183 	/* No vnode should be allocated for this entry from this point */
1184 	TMPFS_NODE_LOCK(node);
1185 	TMPFS_ASSERT_ELOCKED(node);
1186 	TMPFS_NODE_LOCK(dnode);
1187 	TMPFS_ASSERT_ELOCKED(dnode);
1188 
1189 #if 0
1190 	/* handled by tmpfs_free_node */
1191 	KKASSERT(node->tn_links > 0);
1192 	node->tn_links--;
1193 	node->tn_dir.tn_parent = NULL;
1194 #endif
1195 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1196 	    TMPFS_NODE_MODIFIED;
1197 
1198 #if 0
1199 	/* handled by tmpfs_free_node */
1200 	KKASSERT(dnode->tn_links > 0);
1201 	dnode->tn_links--;
1202 #endif
1203 	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1204 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1205 
1206 	TMPFS_NODE_UNLOCK(dnode);
1207 	TMPFS_NODE_UNLOCK(node);
1208 
1209 	/* Free the directory entry we just deleted.  Note that the node
1210 	 * referred by it will not be removed until the vnode is really
1211 	 * reclaimed. */
1212 	tmpfs_free_dirent(tmp, de);
1213 
1214 	/* Release the deleted vnode (will destroy the node, notify
1215 	 * interested parties and clean it from the cache). */
1216 
1217 	TMPFS_NODE_LOCK(dnode);
1218 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1219 	TMPFS_NODE_UNLOCK(dnode);
1220 	tmpfs_update(dvp);
1221 
1222 	cache_setunresolved(v->a_nch);
1223 	cache_setvp(v->a_nch, NULL);
1224 	/*cache_inval_vp(vp, CINV_DESTROY);*/
1225 	tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1226 	error = 0;
1227 
1228 out:
1229 	vrele(vp);
1230 
1231 	return error;
1232 }
1233 
1234 /* --------------------------------------------------------------------- */
1235 
1236 static int
1237 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1238 {
1239 	struct vnode *dvp = v->a_dvp;
1240 	struct vnode **vpp = v->a_vpp;
1241 	struct namecache *ncp = v->a_nch->ncp;
1242 	struct vattr *vap = v->a_vap;
1243 	struct ucred *cred = v->a_cred;
1244 	char *target = v->a_target;
1245 	int error;
1246 
1247 	vap->va_type = VLNK;
1248 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1249 	if (error == 0) {
1250 		tmpfs_knote(*vpp, NOTE_WRITE);
1251 		cache_setunresolved(v->a_nch);
1252 		cache_setvp(v->a_nch, *vpp);
1253 	}
1254 
1255 	return error;
1256 }
1257 
1258 /* --------------------------------------------------------------------- */
1259 
1260 static int
1261 tmpfs_readdir(struct vop_readdir_args *v)
1262 {
1263 	struct vnode *vp = v->a_vp;
1264 	struct uio *uio = v->a_uio;
1265 	int *eofflag = v->a_eofflag;
1266 	off_t **cookies = v->a_cookies;
1267 	int *ncookies = v->a_ncookies;
1268 	struct tmpfs_mount *tmp;
1269 	int error;
1270 	off_t startoff;
1271 	off_t cnt = 0;
1272 	struct tmpfs_node *node;
1273 
1274 	/* This operation only makes sense on directory nodes. */
1275 	if (vp->v_type != VDIR)
1276 		return ENOTDIR;
1277 
1278 	tmp = VFS_TO_TMPFS(vp->v_mount);
1279 	node = VP_TO_TMPFS_DIR(vp);
1280 	startoff = uio->uio_offset;
1281 
1282 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1283 		error = tmpfs_dir_getdotdent(node, uio);
1284 		if (error != 0)
1285 			goto outok;
1286 		cnt++;
1287 	}
1288 
1289 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1290 		error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1291 		if (error != 0)
1292 			goto outok;
1293 		cnt++;
1294 	}
1295 
1296 	error = tmpfs_dir_getdents(node, uio, &cnt);
1297 
1298 outok:
1299 	KKASSERT(error >= -1);
1300 
1301 	if (error == -1)
1302 		error = 0;
1303 
1304 	if (eofflag != NULL)
1305 		*eofflag =
1306 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1307 
1308 	/* Update NFS-related variables. */
1309 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1310 		off_t i;
1311 		off_t off = startoff;
1312 		struct tmpfs_dirent *de = NULL;
1313 
1314 		*ncookies = cnt;
1315 		*cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1316 
1317 		for (i = 0; i < cnt; i++) {
1318 			KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1319 			if (off == TMPFS_DIRCOOKIE_DOT) {
1320 				off = TMPFS_DIRCOOKIE_DOTDOT;
1321 			} else {
1322 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1323 					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1324 				} else if (de != NULL) {
1325 					de = TAILQ_NEXT(de, td_entries);
1326 				} else {
1327 					de = tmpfs_dir_lookupbycookie(node,
1328 					    off);
1329 					KKASSERT(de != NULL);
1330 					de = TAILQ_NEXT(de, td_entries);
1331 				}
1332 				if (de == NULL)
1333 					off = TMPFS_DIRCOOKIE_EOF;
1334 				else
1335 					off = tmpfs_dircookie(de);
1336 			}
1337 
1338 			(*cookies)[i] = off;
1339 		}
1340 		KKASSERT(uio->uio_offset == off);
1341 	}
1342 
1343 	return error;
1344 }
1345 
1346 /* --------------------------------------------------------------------- */
1347 
1348 static int
1349 tmpfs_readlink(struct vop_readlink_args *v)
1350 {
1351 	struct vnode *vp = v->a_vp;
1352 	struct uio *uio = v->a_uio;
1353 
1354 	int error;
1355 	struct tmpfs_node *node;
1356 
1357 	KKASSERT(uio->uio_offset == 0);
1358 	KKASSERT(vp->v_type == VLNK);
1359 
1360 	node = VP_TO_TMPFS_NODE(vp);
1361 
1362 	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1363 	    uio);
1364 	TMPFS_NODE_LOCK(node);
1365 	node->tn_status |= TMPFS_NODE_ACCESSED;
1366 	TMPFS_NODE_UNLOCK(node);
1367 
1368 	return error;
1369 }
1370 
1371 /* --------------------------------------------------------------------- */
1372 
1373 static int
1374 tmpfs_inactive(struct vop_inactive_args *v)
1375 {
1376 	struct vnode *vp = v->a_vp;
1377 
1378 	struct tmpfs_node *node;
1379 
1380 	node = VP_TO_TMPFS_NODE(vp);
1381 
1382 	/*
1383 	 * Get rid of unreferenced deleted vnodes sooner rather than
1384 	 * later so the data memory can be recovered immediately.
1385 	 *
1386 	 * We must truncate the vnode to prevent the normal reclamation
1387 	 * path from flushing the data for the removed file to disk.
1388 	 */
1389 	TMPFS_NODE_LOCK(node);
1390 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1391 	    (node->tn_links == 0 ||
1392 	     (node->tn_links == 1 && node->tn_type == VDIR &&
1393 	      node->tn_dir.tn_parent)))
1394 	{
1395 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1396 		TMPFS_NODE_UNLOCK(node);
1397 		if (node->tn_type == VREG)
1398 			tmpfs_truncate(vp, 0);
1399 		vrecycle(vp);
1400 	} else {
1401 		TMPFS_NODE_UNLOCK(node);
1402 	}
1403 
1404 	return 0;
1405 }
1406 
1407 /* --------------------------------------------------------------------- */
1408 
1409 int
1410 tmpfs_reclaim(struct vop_reclaim_args *v)
1411 {
1412 	struct vnode *vp = v->a_vp;
1413 	struct tmpfs_mount *tmp;
1414 	struct tmpfs_node *node;
1415 
1416 	node = VP_TO_TMPFS_NODE(vp);
1417 	tmp = VFS_TO_TMPFS(vp->v_mount);
1418 
1419 	tmpfs_free_vp(vp);
1420 
1421 	/*
1422 	 * If the node referenced by this vnode was deleted by the
1423 	 * user, we must free its associated data structures now that
1424 	 * the vnode is being reclaimed.
1425 	 *
1426 	 * Directories have an extra link ref.
1427 	 */
1428 	TMPFS_NODE_LOCK(node);
1429 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1430 	    (node->tn_links == 0 ||
1431 	     (node->tn_links == 1 && node->tn_type == VDIR &&
1432 	      node->tn_dir.tn_parent)))
1433 	{
1434 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1435 		tmpfs_free_node(tmp, node);
1436 		/* eats the lock */
1437 	} else {
1438 		TMPFS_NODE_UNLOCK(node);
1439 	}
1440 
1441 	KKASSERT(vp->v_data == NULL);
1442 	return 0;
1443 }
1444 
1445 /* --------------------------------------------------------------------- */
1446 
1447 static int
1448 tmpfs_print(struct vop_print_args *v)
1449 {
1450 	struct vnode *vp = v->a_vp;
1451 
1452 	struct tmpfs_node *node;
1453 
1454 	node = VP_TO_TMPFS_NODE(vp);
1455 
1456 	kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1457 	    node, node->tn_flags, node->tn_links);
1458 	kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1459 	    node->tn_mode, node->tn_uid, node->tn_gid,
1460 	    (uintmax_t)node->tn_size, node->tn_status);
1461 
1462 	if (vp->v_type == VFIFO)
1463 		fifo_printinfo(vp);
1464 
1465 	kprintf("\n");
1466 
1467 	return 0;
1468 }
1469 
1470 /* --------------------------------------------------------------------- */
1471 
1472 static int
1473 tmpfs_pathconf(struct vop_pathconf_args *v)
1474 {
1475 	int name = v->a_name;
1476 	register_t *retval = v->a_retval;
1477 
1478 	int error;
1479 
1480 	error = 0;
1481 
1482 	switch (name) {
1483 	case _PC_LINK_MAX:
1484 		*retval = LINK_MAX;
1485 		break;
1486 
1487 	case _PC_NAME_MAX:
1488 		*retval = NAME_MAX;
1489 		break;
1490 
1491 	case _PC_PATH_MAX:
1492 		*retval = PATH_MAX;
1493 		break;
1494 
1495 	case _PC_PIPE_BUF:
1496 		*retval = PIPE_BUF;
1497 		break;
1498 
1499 	case _PC_CHOWN_RESTRICTED:
1500 		*retval = 1;
1501 		break;
1502 
1503 	case _PC_NO_TRUNC:
1504 		*retval = 1;
1505 		break;
1506 
1507 	case _PC_SYNC_IO:
1508 		*retval = 1;
1509 		break;
1510 
1511 	case _PC_FILESIZEBITS:
1512 		*retval = 0; /* XXX Don't know which value should I return. */
1513 		break;
1514 
1515 	default:
1516 		error = EINVAL;
1517 	}
1518 
1519 	return error;
1520 }
1521 
1522 /************************************************************************
1523  *                          KQFILTER OPS                                *
1524  ************************************************************************/
1525 
1526 static void filt_tmpfsdetach(struct knote *kn);
1527 static int filt_tmpfsread(struct knote *kn, long hint);
1528 static int filt_tmpfswrite(struct knote *kn, long hint);
1529 static int filt_tmpfsvnode(struct knote *kn, long hint);
1530 
1531 static struct filterops tmpfsread_filtops =
1532 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1533 static struct filterops tmpfswrite_filtops =
1534 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1535 static struct filterops tmpfsvnode_filtops =
1536 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1537 
1538 static int
1539 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1540 {
1541 	struct vnode *vp = ap->a_vp;
1542 	struct knote *kn = ap->a_kn;
1543 
1544 	switch (kn->kn_filter) {
1545 	case EVFILT_READ:
1546 		kn->kn_fop = &tmpfsread_filtops;
1547 		break;
1548 	case EVFILT_WRITE:
1549 		kn->kn_fop = &tmpfswrite_filtops;
1550 		break;
1551 	case EVFILT_VNODE:
1552 		kn->kn_fop = &tmpfsvnode_filtops;
1553 		break;
1554 	default:
1555 		return (EOPNOTSUPP);
1556 	}
1557 
1558 	kn->kn_hook = (caddr_t)vp;
1559 
1560 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1561 
1562 	return(0);
1563 }
1564 
1565 static void
1566 filt_tmpfsdetach(struct knote *kn)
1567 {
1568 	struct vnode *vp = (void *)kn->kn_hook;
1569 
1570 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1571 }
1572 
1573 static int
1574 filt_tmpfsread(struct knote *kn, long hint)
1575 {
1576 	struct vnode *vp = (void *)kn->kn_hook;
1577 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1578 	off_t off;
1579 
1580 	if (hint == NOTE_REVOKE) {
1581 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1582 		return(1);
1583 	}
1584 	off = node->tn_size - kn->kn_fp->f_offset;
1585 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1586 	if (kn->kn_sfflags & NOTE_OLDAPI)
1587 		return(1);
1588 
1589 	/*
1590 	 * Handle possible MP race interlock on filter check/write
1591 	 */
1592 	if (kn->kn_data == 0) {
1593 		get_mplock();
1594 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1595 		rel_mplock();
1596 	}
1597 	return (kn->kn_data != 0);
1598 }
1599 
1600 static int
1601 filt_tmpfswrite(struct knote *kn, long hint)
1602 {
1603 	if (hint == NOTE_REVOKE)
1604 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1605 	kn->kn_data = 0;
1606 	return (1);
1607 }
1608 
1609 static int
1610 filt_tmpfsvnode(struct knote *kn, long hint)
1611 {
1612 	if (kn->kn_sfflags & hint)
1613 		kn->kn_fflags |= hint;
1614 	if (hint == NOTE_REVOKE) {
1615 		kn->kn_flags |= EV_EOF;
1616 		return (1);
1617 	}
1618 	return (kn->kn_fflags != 0);
1619 }
1620 
1621 
1622 /* --------------------------------------------------------------------- */
1623 
1624 /*
1625  * vnode operations vector used for files stored in a tmpfs file system.
1626  */
1627 struct vop_ops tmpfs_vnode_vops = {
1628 	.vop_default =			vop_defaultop,
1629 	.vop_getpages = 		vop_stdgetpages,
1630 	.vop_putpages = 		vop_stdputpages,
1631 	.vop_ncreate =			tmpfs_ncreate,
1632 	.vop_nresolve =			tmpfs_nresolve,
1633 	.vop_nlookupdotdot =		tmpfs_nlookupdotdot,
1634 	.vop_nmknod =			tmpfs_nmknod,
1635 	.vop_open =			tmpfs_open,
1636 	.vop_close =			tmpfs_close,
1637 	.vop_access =			tmpfs_access,
1638 	.vop_getattr =			tmpfs_getattr,
1639 	.vop_setattr =			tmpfs_setattr,
1640 	.vop_read =			tmpfs_read,
1641 	.vop_write =			tmpfs_write,
1642 	.vop_fsync =			tmpfs_fsync,
1643 	.vop_nremove =			tmpfs_nremove,
1644 	.vop_nlink =			tmpfs_nlink,
1645 	.vop_nrename =			tmpfs_nrename,
1646 	.vop_nmkdir =			tmpfs_nmkdir,
1647 	.vop_nrmdir =			tmpfs_nrmdir,
1648 	.vop_nsymlink =			tmpfs_nsymlink,
1649 	.vop_readdir =			tmpfs_readdir,
1650 	.vop_readlink =			tmpfs_readlink,
1651 	.vop_inactive =			tmpfs_inactive,
1652 	.vop_reclaim =			tmpfs_reclaim,
1653 	.vop_print =			tmpfs_print,
1654 	.vop_pathconf =			tmpfs_pathconf,
1655 	.vop_bmap =			tmpfs_bmap,
1656 	.vop_strategy =			tmpfs_strategy,
1657 	.vop_advlock =			tmpfs_advlock,
1658 	.vop_kqfilter =			tmpfs_kqfilter
1659 };
1660