xref: /dflybsd-src/sys/vfs/tmpfs/tmpfs_vnops.c (revision 06cdb70d3c2d4585f58c8f08e050a5d105d1bbb6)
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 
37 #include <sys/kernel.h>
38 #include <sys/kern_syscall.h>
39 #include <sys/param.h>
40 #include <sys/fcntl.h>
41 #include <sys/lockf.h>
42 #include <sys/priv.h>
43 #include <sys/proc.h>
44 #include <sys/resourcevar.h>
45 #include <sys/sched.h>
46 #include <sys/stat.h>
47 #include <sys/systm.h>
48 #include <sys/unistd.h>
49 #include <sys/vfsops.h>
50 #include <sys/vnode.h>
51 #include <sys/mountctl.h>
52 
53 #include <vm/vm.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pager.h>
57 #include <vm/swap_pager.h>
58 
59 #include <sys/buf2.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 void tmpfs_strategy_done(struct bio *bio);
68 
69 static __inline
70 void
71 tmpfs_knote(struct vnode *vp, int flags)
72 {
73 	if (flags)
74 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
75 }
76 
77 
78 /* --------------------------------------------------------------------- */
79 
80 static int
81 tmpfs_nresolve(struct vop_nresolve_args *v)
82 {
83 	struct vnode *dvp = v->a_dvp;
84 	struct vnode *vp = NULL;
85 	struct namecache *ncp = v->a_nch->ncp;
86 	struct tmpfs_node *tnode;
87 
88 	int error;
89 	struct tmpfs_dirent *de;
90 	struct tmpfs_node *dnode;
91 
92 	dnode = VP_TO_TMPFS_DIR(dvp);
93 
94 	de = tmpfs_dir_lookup(dnode, NULL, ncp);
95 	if (de == NULL) {
96 		error = ENOENT;
97 	} else {
98 		/*
99 		 * Allocate a vnode for the node we found.
100 		 */
101 		tnode = de->td_node;
102 		error = tmpfs_alloc_vp(dvp->v_mount, tnode,
103 				       LK_EXCLUSIVE | LK_RETRY, &vp);
104 		if (error)
105 			goto out;
106 		KKASSERT(vp);
107 	}
108 
109 out:
110 	/*
111 	 * Store the result of this lookup in the cache.  Avoid this if the
112 	 * request was for creation, as it does not improve timings on
113 	 * emprical tests.
114 	 */
115 	if (vp) {
116 		vn_unlock(vp);
117 		cache_setvp(v->a_nch, vp);
118 		vrele(vp);
119 	} else if (error == ENOENT) {
120 		cache_setvp(v->a_nch, NULL);
121 	}
122 	return error;
123 }
124 
125 static int
126 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
127 {
128 	struct vnode *dvp = v->a_dvp;
129 	struct vnode **vpp = v->a_vpp;
130 	struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
131 	struct ucred *cred = v->a_cred;
132 	int error;
133 
134 	*vpp = NULL;
135 	/* Check accessibility of requested node as a first step. */
136 	error = VOP_ACCESS(dvp, VEXEC, cred);
137 	if (error != 0)
138 		return error;
139 
140 	if (dnode->tn_dir.tn_parent != NULL) {
141 		/* Allocate a new vnode on the matching entry. */
142 		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
143 		    LK_EXCLUSIVE | LK_RETRY, vpp);
144 
145 		if (*vpp)
146 			vn_unlock(*vpp);
147 	}
148 
149 	return (*vpp == NULL) ? ENOENT : 0;
150 }
151 
152 /* --------------------------------------------------------------------- */
153 
154 static int
155 tmpfs_ncreate(struct vop_ncreate_args *v)
156 {
157 	struct vnode *dvp = v->a_dvp;
158 	struct vnode **vpp = v->a_vpp;
159 	struct namecache *ncp = v->a_nch->ncp;
160 	struct vattr *vap = v->a_vap;
161 	struct ucred *cred = v->a_cred;
162 	int error;
163 
164 	KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
165 
166 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
167 	if (error == 0) {
168 		cache_setunresolved(v->a_nch);
169 		cache_setvp(v->a_nch, *vpp);
170 		tmpfs_knote(dvp, NOTE_WRITE);
171 	}
172 
173 	return error;
174 }
175 /* --------------------------------------------------------------------- */
176 
177 static int
178 tmpfs_nmknod(struct vop_nmknod_args *v)
179 {
180 	struct vnode *dvp = v->a_dvp;
181 	struct vnode **vpp = v->a_vpp;
182 	struct namecache *ncp = v->a_nch->ncp;
183 	struct vattr *vap = v->a_vap;
184 	struct ucred *cred = v->a_cred;
185 	int error;
186 
187 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
188 	    vap->va_type != VFIFO)
189 		return EINVAL;
190 
191 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
192 	if (error == 0) {
193 		cache_setunresolved(v->a_nch);
194 		cache_setvp(v->a_nch, *vpp);
195 		tmpfs_knote(dvp, NOTE_WRITE);
196 	}
197 
198 	return error;
199 }
200 
201 /* --------------------------------------------------------------------- */
202 
203 static int
204 tmpfs_open(struct vop_open_args *v)
205 {
206 	struct vnode *vp = v->a_vp;
207 	int mode = v->a_mode;
208 
209 	int error;
210 	struct tmpfs_node *node;
211 
212 	node = VP_TO_TMPFS_NODE(vp);
213 
214 	/* The file is still active but all its names have been removed
215 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
216 	 * it is about to die. */
217 	if (node->tn_links < 1)
218 		return (ENOENT);
219 
220 	/* If the file is marked append-only, deny write requests. */
221 	if ((node->tn_flags & APPEND) &&
222 	    (mode & (FWRITE | O_APPEND)) == FWRITE) {
223 		error = EPERM;
224 	} else {
225 		return (vop_stdopen(v));
226 	}
227 	return error;
228 }
229 
230 /* --------------------------------------------------------------------- */
231 
232 static int
233 tmpfs_close(struct vop_close_args *v)
234 {
235 	struct vnode *vp = v->a_vp;
236 	struct tmpfs_node *node;
237 
238 	node = VP_TO_TMPFS_NODE(vp);
239 
240 	if (node->tn_links > 0) {
241 		/* Update node times.  No need to do it if the node has
242 		 * been deleted, because it will vanish after we return. */
243 		tmpfs_update(vp);
244 	}
245 
246 	return vop_stdclose(v);
247 }
248 
249 /* --------------------------------------------------------------------- */
250 
251 int
252 tmpfs_access(struct vop_access_args *v)
253 {
254 	struct vnode *vp = v->a_vp;
255 	int error;
256 	struct tmpfs_node *node;
257 
258 	node = VP_TO_TMPFS_NODE(vp);
259 
260 	switch (vp->v_type) {
261 	case VDIR:
262 		/* FALLTHROUGH */
263 	case VLNK:
264 		/* FALLTHROUGH */
265 	case VREG:
266 		if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
267 			error = EROFS;
268 			goto out;
269 		}
270 		break;
271 
272 	case VBLK:
273 		/* FALLTHROUGH */
274 	case VCHR:
275 		/* FALLTHROUGH */
276 	case VSOCK:
277 		/* FALLTHROUGH */
278 	case VFIFO:
279 		break;
280 
281 	default:
282 		error = EINVAL;
283 		goto out;
284 	}
285 
286 	if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
287 		error = EPERM;
288 		goto out;
289 	}
290 
291 	error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
292 
293 out:
294 
295 	return error;
296 }
297 
298 /* --------------------------------------------------------------------- */
299 
300 int
301 tmpfs_getattr(struct vop_getattr_args *v)
302 {
303 	struct vnode *vp = v->a_vp;
304 	struct vattr *vap = v->a_vap;
305 	struct tmpfs_node *node;
306 
307 	node = VP_TO_TMPFS_NODE(vp);
308 
309 	lwkt_gettoken(&vp->v_mount->mnt_token);
310 	tmpfs_update(vp);
311 
312 	vap->va_type = vp->v_type;
313 	vap->va_mode = node->tn_mode;
314 	vap->va_nlink = node->tn_links;
315 	vap->va_uid = node->tn_uid;
316 	vap->va_gid = node->tn_gid;
317 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
318 	vap->va_fileid = node->tn_id;
319 	vap->va_size = node->tn_size;
320 	vap->va_blocksize = PAGE_SIZE;
321 	vap->va_atime.tv_sec = node->tn_atime;
322 	vap->va_atime.tv_nsec = node->tn_atimensec;
323 	vap->va_mtime.tv_sec = node->tn_mtime;
324 	vap->va_mtime.tv_nsec = node->tn_mtimensec;
325 	vap->va_ctime.tv_sec = node->tn_ctime;
326 	vap->va_ctime.tv_nsec = node->tn_ctimensec;
327 	vap->va_gen = node->tn_gen;
328 	vap->va_flags = node->tn_flags;
329 	if (vp->v_type == VBLK || vp->v_type == VCHR)
330 	{
331 		vap->va_rmajor = umajor(node->tn_rdev);
332 		vap->va_rminor = uminor(node->tn_rdev);
333 	}
334 	vap->va_bytes = round_page(node->tn_size);
335 	vap->va_filerev = 0;
336 
337 	lwkt_reltoken(&vp->v_mount->mnt_token);
338 
339 	return 0;
340 }
341 
342 /* --------------------------------------------------------------------- */
343 
344 int
345 tmpfs_setattr(struct vop_setattr_args *v)
346 {
347 	struct vnode *vp = v->a_vp;
348 	struct vattr *vap = v->a_vap;
349 	struct ucred *cred = v->a_cred;
350 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
351 	int error = 0;
352 	int kflags = 0;
353 
354 	if (error == 0 && (vap->va_flags != VNOVAL)) {
355 		error = tmpfs_chflags(vp, vap->va_flags, cred);
356 		kflags |= NOTE_ATTRIB;
357 	}
358 
359 	if (error == 0 && (vap->va_size != VNOVAL)) {
360 		if (vap->va_size > node->tn_size)
361 			kflags |= NOTE_WRITE | NOTE_EXTEND;
362 		else
363 			kflags |= NOTE_WRITE;
364 		error = tmpfs_chsize(vp, vap->va_size, cred);
365 	}
366 
367 	if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
368 			   vap->va_gid != (gid_t)VNOVAL)) {
369 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
370 		kflags |= NOTE_ATTRIB;
371 	}
372 
373 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
374 		error = tmpfs_chmod(vp, vap->va_mode, cred);
375 		kflags |= NOTE_ATTRIB;
376 	}
377 
378 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
379 	    vap->va_atime.tv_nsec != VNOVAL) ||
380 	    (vap->va_mtime.tv_sec != VNOVAL &&
381 	    vap->va_mtime.tv_nsec != VNOVAL) )) {
382 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
383 				      vap->va_vaflags, cred);
384 		kflags |= NOTE_ATTRIB;
385 	}
386 
387 	/* Update the node times.  We give preference to the error codes
388 	 * generated by this function rather than the ones that may arise
389 	 * from tmpfs_update. */
390 	tmpfs_update(vp);
391 	tmpfs_knote(vp, kflags);
392 
393 	return error;
394 }
395 
396 /* --------------------------------------------------------------------- */
397 
398 /*
399  * fsync is usually a NOP, but we must take action when unmounting or
400  * when recycling.
401  */
402 static int
403 tmpfs_fsync(struct vop_fsync_args *v)
404 {
405 	struct tmpfs_node *node;
406 	struct vnode *vp = v->a_vp;
407 
408 	node = VP_TO_TMPFS_NODE(vp);
409 
410 	tmpfs_update(vp);
411 	if (vp->v_type == VREG) {
412 		if (vp->v_flag & VRECLAIMED) {
413 			if (node->tn_links == 0)
414 				tmpfs_truncate(vp, 0);
415 			else
416 				vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
417 		}
418 	}
419 	return 0;
420 }
421 
422 /* --------------------------------------------------------------------- */
423 
424 static int
425 tmpfs_read (struct vop_read_args *ap)
426 {
427 	struct buf *bp;
428 	struct vnode *vp = ap->a_vp;
429 	struct uio *uio = ap->a_uio;
430 	struct tmpfs_node *node;
431 	off_t base_offset;
432 	size_t offset;
433 	size_t len;
434 	int error;
435 
436 	error = 0;
437 	if (uio->uio_resid == 0) {
438 		return error;
439 	}
440 
441 	node = VP_TO_TMPFS_NODE(vp);
442 
443 	if (uio->uio_offset < 0)
444 		return (EINVAL);
445 	if (vp->v_type != VREG)
446 		return (EINVAL);
447 
448 	while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
449 		/*
450 		 * Use buffer cache I/O (via tmpfs_strategy)
451 		 */
452 		offset = (size_t)uio->uio_offset & BMASK;
453 		base_offset = (off_t)uio->uio_offset - offset;
454 		bp = getcacheblk(vp, base_offset, BSIZE, 0);
455 		if (bp == NULL) {
456 			lwkt_gettoken(&vp->v_mount->mnt_token);
457 			error = bread(vp, base_offset, BSIZE, &bp);
458 			if (error) {
459 				brelse(bp);
460 				lwkt_reltoken(&vp->v_mount->mnt_token);
461 				kprintf("tmpfs_read bread error %d\n", error);
462 				break;
463 			}
464 			lwkt_reltoken(&vp->v_mount->mnt_token);
465 		}
466 
467 		/*
468 		 * Figure out how many bytes we can actually copy this loop.
469 		 */
470 		len = BSIZE - offset;
471 		if (len > uio->uio_resid)
472 			len = uio->uio_resid;
473 		if (len > node->tn_size - uio->uio_offset)
474 			len = (size_t)(node->tn_size - uio->uio_offset);
475 
476 		error = uiomove((char *)bp->b_data + offset, len, uio);
477 		bqrelse(bp);
478 		if (error) {
479 			kprintf("tmpfs_read uiomove error %d\n", error);
480 			break;
481 		}
482 	}
483 
484 	TMPFS_NODE_LOCK(node);
485 	node->tn_status |= TMPFS_NODE_ACCESSED;
486 	TMPFS_NODE_UNLOCK(node);
487 
488 	return(error);
489 }
490 
491 static int
492 tmpfs_write (struct vop_write_args *ap)
493 {
494 	struct buf *bp;
495 	struct vnode *vp = ap->a_vp;
496 	struct uio *uio = ap->a_uio;
497 	struct thread *td = uio->uio_td;
498 	struct tmpfs_node *node;
499 	boolean_t extended;
500 	off_t oldsize;
501 	int error;
502 	off_t base_offset;
503 	size_t offset;
504 	size_t len;
505 	struct rlimit limit;
506 	int trivial = 0;
507 	int kflags = 0;
508 
509 	error = 0;
510 	if (uio->uio_resid == 0) {
511 		return error;
512 	}
513 
514 	node = VP_TO_TMPFS_NODE(vp);
515 
516 	if (vp->v_type != VREG)
517 		return (EINVAL);
518 
519 	lwkt_gettoken(&vp->v_mount->mnt_token);
520 
521 	oldsize = node->tn_size;
522 	if (ap->a_ioflag & IO_APPEND)
523 		uio->uio_offset = node->tn_size;
524 
525 	/*
526 	 * Check for illegal write offsets.
527 	 */
528 	if (uio->uio_offset + uio->uio_resid >
529 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
530 		lwkt_reltoken(&vp->v_mount->mnt_token);
531 		return (EFBIG);
532 	}
533 
534 	if (vp->v_type == VREG && td != NULL) {
535 		error = kern_getrlimit(RLIMIT_FSIZE, &limit);
536 		if (error != 0) {
537 			lwkt_reltoken(&vp->v_mount->mnt_token);
538 			return error;
539 		}
540 		if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
541 			ksignal(td->td_proc, SIGXFSZ);
542 			lwkt_reltoken(&vp->v_mount->mnt_token);
543 			return (EFBIG);
544 		}
545 	}
546 
547 
548 	/*
549 	 * Extend the file's size if necessary
550 	 */
551 	extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
552 
553 	while (uio->uio_resid > 0) {
554 		/*
555 		 * Use buffer cache I/O (via tmpfs_strategy)
556 		 */
557 		offset = (size_t)uio->uio_offset & BMASK;
558 		base_offset = (off_t)uio->uio_offset - offset;
559 		len = BSIZE - offset;
560 		if (len > uio->uio_resid)
561 			len = uio->uio_resid;
562 
563 		if ((uio->uio_offset + len) > node->tn_size) {
564 			trivial = (uio->uio_offset <= node->tn_size);
565 			error = tmpfs_reg_resize(vp, uio->uio_offset + len,  trivial);
566 			if (error)
567 				break;
568 		}
569 
570 		/*
571 		 * Read to fill in any gaps.  Theoretically we could
572 		 * optimize this if the write covers the entire buffer
573 		 * and is not a UIO_NOCOPY write, however this can lead
574 		 * to a security violation exposing random kernel memory
575 		 * (whatever junk was in the backing VM pages before).
576 		 *
577 		 * So just use bread() to do the right thing.
578 		 */
579 		error = bread(vp, base_offset, BSIZE, &bp);
580 		error = uiomove((char *)bp->b_data + offset, len, uio);
581 		if (error) {
582 			kprintf("tmpfs_write uiomove error %d\n", error);
583 			brelse(bp);
584 			break;
585 		}
586 
587 		if (uio->uio_offset > node->tn_size) {
588 			node->tn_size = uio->uio_offset;
589 			kflags |= NOTE_EXTEND;
590 		}
591 		kflags |= NOTE_WRITE;
592 
593 		/*
594 		 * Always try to flush the page if the request is coming
595 		 * from the pageout daemon (IO_ASYNC), else buwrite() the
596 		 * buffer.
597 		 *
598 		 * buwrite() dirties the underlying VM pages instead of
599 		 * dirtying the buffer, releasing the buffer as a clean
600 		 * buffer.  This allows tmpfs to use essentially all
601 		 * available memory to cache file data.  If we used bdwrite()
602 		 * the buffer cache would wind up flushing the data to
603 		 * swap too quickly.
604 		 */
605 		bp->b_flags |= B_AGE;
606 		if (ap->a_ioflag & IO_ASYNC) {
607 			bawrite(bp);
608 		} else {
609 			buwrite(bp);
610 		}
611 
612 		if (bp->b_error) {
613 			kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
614 			break;
615 		}
616 	}
617 
618 	if (error) {
619 		if (extended) {
620 			(void)tmpfs_reg_resize(vp, oldsize, trivial);
621 			kflags &= ~NOTE_EXTEND;
622 		}
623 		goto done;
624 	}
625 
626 	/*
627 	 * Currently we don't set the mtime on files modified via mmap()
628 	 * because we can't tell the difference between those modifications
629 	 * and an attempt by the pageout daemon to flush tmpfs pages to
630 	 * swap.
631 	 *
632 	 * This is because in order to defer flushes as long as possible
633 	 * buwrite() works by marking the underlying VM pages dirty in
634 	 * order to be able to dispose of the buffer cache buffer without
635 	 * flushing it.
636 	 */
637 	TMPFS_NODE_LOCK(node);
638 	if (uio->uio_segflg != UIO_NOCOPY)
639 		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED;
640 	if (extended)
641 		node->tn_status |= TMPFS_NODE_CHANGED;
642 
643 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
644 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
645 			node->tn_mode &= ~(S_ISUID | S_ISGID);
646 	}
647 	TMPFS_NODE_UNLOCK(node);
648 done:
649 
650 	tmpfs_knote(vp, kflags);
651 
652 
653 	lwkt_reltoken(&vp->v_mount->mnt_token);
654 	return(error);
655 }
656 
657 static int
658 tmpfs_advlock (struct vop_advlock_args *ap)
659 {
660 	struct tmpfs_node *node;
661 	struct vnode *vp = ap->a_vp;
662 
663 	node = VP_TO_TMPFS_NODE(vp);
664 
665 	return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
666 }
667 
668 /*
669  * The strategy function is typically only called when memory pressure
670  * forces the system to attempt to pageout pages.  It can also be called
671  * by [n]vtruncbuf() when a truncation cuts a page in half.  Normal write
672  * operations
673  */
674 static int
675 tmpfs_strategy(struct vop_strategy_args *ap)
676 {
677 	struct bio *bio = ap->a_bio;
678 	struct bio *nbio;
679 	struct buf *bp = bio->bio_buf;
680 	struct vnode *vp = ap->a_vp;
681 	struct tmpfs_node *node;
682 	vm_object_t uobj;
683 	vm_page_t m;
684 	int i;
685 
686 	if (vp->v_type != VREG) {
687 		bp->b_resid = bp->b_bcount;
688 		bp->b_flags |= B_ERROR | B_INVAL;
689 		bp->b_error = EINVAL;
690 		biodone(bio);
691 		return(0);
692 	}
693 
694 	lwkt_gettoken(&vp->v_mount->mnt_token);
695 	node = VP_TO_TMPFS_NODE(vp);
696 
697 	uobj = node->tn_reg.tn_aobj;
698 
699 	/*
700 	 * Don't bother flushing to swap if there is no swap, just
701 	 * ensure that the pages are marked as needing a commit (still).
702 	 */
703 	if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
704 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
705 			m = bp->b_xio.xio_pages[i];
706 			vm_page_need_commit(m);
707 		}
708 		bp->b_resid = 0;
709 		bp->b_error = 0;
710 		biodone(bio);
711 	} else {
712 		nbio = push_bio(bio);
713 		nbio->bio_done = tmpfs_strategy_done;
714 		nbio->bio_offset = bio->bio_offset;
715 		swap_pager_strategy(uobj, nbio);
716 	}
717 
718 	lwkt_reltoken(&vp->v_mount->mnt_token);
719 	return 0;
720 }
721 
722 /*
723  * If we were unable to commit the pages to swap make sure they are marked
724  * as needing a commit (again).  If we were, clear the flag to allow the
725  * pages to be freed.
726  */
727 static void
728 tmpfs_strategy_done(struct bio *bio)
729 {
730 	struct buf *bp;
731 	vm_page_t m;
732 	int i;
733 
734 	bp = bio->bio_buf;
735 
736 	if (bp->b_flags & B_ERROR) {
737 		bp->b_flags &= ~B_ERROR;
738 		bp->b_error = 0;
739 		bp->b_resid = 0;
740 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
741 			m = bp->b_xio.xio_pages[i];
742 			vm_page_need_commit(m);
743 		}
744 	} else {
745 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
746 			m = bp->b_xio.xio_pages[i];
747 			vm_page_clear_commit(m);
748 		}
749 	}
750 	bio = pop_bio(bio);
751 	biodone(bio);
752 }
753 
754 static int
755 tmpfs_bmap(struct vop_bmap_args *ap)
756 {
757 	if (ap->a_doffsetp != NULL)
758 		*ap->a_doffsetp = ap->a_loffset;
759 	if (ap->a_runp != NULL)
760 		*ap->a_runp = 0;
761 	if (ap->a_runb != NULL)
762 		*ap->a_runb = 0;
763 
764 	return 0;
765 }
766 
767 /* --------------------------------------------------------------------- */
768 
769 static int
770 tmpfs_nremove(struct vop_nremove_args *v)
771 {
772 	struct vnode *dvp = v->a_dvp;
773 	struct namecache *ncp = v->a_nch->ncp;
774 	struct vnode *vp;
775 	int error;
776 	struct tmpfs_dirent *de;
777 	struct tmpfs_mount *tmp;
778 	struct tmpfs_node *dnode;
779 	struct tmpfs_node *node;
780 
781 	/*
782 	 * We have to acquire the vp from v->a_nch because we will likely
783 	 * unresolve the namecache entry, and a vrele/vput is needed to
784 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
785 	 *
786 	 * We have to use vget to clear any inactive state on the vnode,
787 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
788 	 * will not get called when we release it.
789 	 */
790 	error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
791 	KKASSERT(error == 0);
792 	vn_unlock(vp);
793 
794 	if (vp->v_type == VDIR) {
795 		error = EISDIR;
796 		goto out;
797 	}
798 
799 	dnode = VP_TO_TMPFS_DIR(dvp);
800 	node = VP_TO_TMPFS_NODE(vp);
801 	tmp = VFS_TO_TMPFS(vp->v_mount);
802 	de = tmpfs_dir_lookup(dnode, node, ncp);
803 	if (de == NULL) {
804 		error = ENOENT;
805 		goto out;
806 	}
807 
808 	/* Files marked as immutable or append-only cannot be deleted. */
809 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
810 	    (dnode->tn_flags & APPEND)) {
811 		error = EPERM;
812 		goto out;
813 	}
814 
815 	/* Remove the entry from the directory; as it is a file, we do not
816 	 * have to change the number of hard links of the directory. */
817 	tmpfs_dir_detach(dnode, de);
818 
819 	/* Free the directory entry we just deleted.  Note that the node
820 	 * referred by it will not be removed until the vnode is really
821 	 * reclaimed. */
822 	tmpfs_free_dirent(tmp, de);
823 
824 	if (node->tn_links > 0) {
825 	        TMPFS_NODE_LOCK(node);
826 		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
827 	                TMPFS_NODE_MODIFIED;
828 	        TMPFS_NODE_UNLOCK(node);
829 	}
830 
831 	cache_setunresolved(v->a_nch);
832 	cache_setvp(v->a_nch, NULL);
833 	tmpfs_knote(vp, NOTE_DELETE);
834 	/*cache_inval_vp(vp, CINV_DESTROY);*/
835 	tmpfs_knote(dvp, NOTE_WRITE);
836 	error = 0;
837 
838 out:
839 	vrele(vp);
840 
841 	return error;
842 }
843 
844 /* --------------------------------------------------------------------- */
845 
846 static int
847 tmpfs_nlink(struct vop_nlink_args *v)
848 {
849 	struct vnode *dvp = v->a_dvp;
850 	struct vnode *vp = v->a_vp;
851 	struct namecache *ncp = v->a_nch->ncp;
852 	struct tmpfs_dirent *de;
853 	struct tmpfs_node *node;
854 	struct tmpfs_node *dnode;
855 	int error;
856 
857 	KKASSERT(dvp != vp); /* XXX When can this be false? */
858 
859 	node = VP_TO_TMPFS_NODE(vp);
860 	dnode = VP_TO_TMPFS_NODE(dvp);
861 
862 	/* XXX: Why aren't the following two tests done by the caller? */
863 
864 	/* Hard links of directories are forbidden. */
865 	if (vp->v_type == VDIR) {
866 		error = EPERM;
867 		goto out;
868 	}
869 
870 	/* Cannot create cross-device links. */
871 	if (dvp->v_mount != vp->v_mount) {
872 		error = EXDEV;
873 		goto out;
874 	}
875 
876 	/* Ensure that we do not overflow the maximum number of links imposed
877 	 * by the system. */
878 	KKASSERT(node->tn_links <= LINK_MAX);
879 	if (node->tn_links == LINK_MAX) {
880 		error = EMLINK;
881 		goto out;
882 	}
883 
884 	/* We cannot create links of files marked immutable or append-only. */
885 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
886 		error = EPERM;
887 		goto out;
888 	}
889 
890 	/* Allocate a new directory entry to represent the node. */
891 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
892 	    ncp->nc_name, ncp->nc_nlen, &de);
893 	if (error != 0)
894 		goto out;
895 
896 	/* Insert the new directory entry into the appropriate directory. */
897 	tmpfs_dir_attach(dnode, de);
898 
899 	/* vp link count has changed, so update node times. */
900 
901 	TMPFS_NODE_LOCK(node);
902 	node->tn_status |= TMPFS_NODE_CHANGED;
903 	TMPFS_NODE_UNLOCK(node);
904 	tmpfs_update(vp);
905 
906 	tmpfs_knote(vp, NOTE_LINK);
907 	cache_setunresolved(v->a_nch);
908 	cache_setvp(v->a_nch, vp);
909 	tmpfs_knote(dvp, NOTE_WRITE);
910 	error = 0;
911 
912 out:
913 	return error;
914 }
915 
916 /* --------------------------------------------------------------------- */
917 
918 static int
919 tmpfs_nrename(struct vop_nrename_args *v)
920 {
921 	struct vnode *fdvp = v->a_fdvp;
922 	struct namecache *fncp = v->a_fnch->ncp;
923 	struct vnode *fvp = fncp->nc_vp;
924 	struct vnode *tdvp = v->a_tdvp;
925 	struct namecache *tncp = v->a_tnch->ncp;
926 	struct vnode *tvp;
927 	struct tmpfs_dirent *de, *tde;
928 	struct tmpfs_mount *tmp;
929 	struct tmpfs_node *fdnode;
930 	struct tmpfs_node *fnode;
931 	struct tmpfs_node *tnode;
932 	struct tmpfs_node *tdnode;
933 	char *newname;
934 	char *oldname;
935 	int error;
936 
937 	/*
938 	 * Because tvp can get overwritten we have to vget it instead of
939 	 * just vref or use it, otherwise it's VINACTIVE flag may not get
940 	 * cleared and the node won't get destroyed.
941 	 */
942 	error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp);
943 	if (error == 0) {
944 		tnode = VP_TO_TMPFS_NODE(tvp);
945 		vn_unlock(tvp);
946 	} else {
947 		tnode = NULL;
948 	}
949 
950 	/* Disallow cross-device renames.
951 	 * XXX Why isn't this done by the caller? */
952 	if (fvp->v_mount != tdvp->v_mount ||
953 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
954 		error = EXDEV;
955 		goto out;
956 	}
957 
958 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
959 	tdnode = VP_TO_TMPFS_DIR(tdvp);
960 
961 	/* If source and target are the same file, there is nothing to do. */
962 	if (fvp == tvp) {
963 		error = 0;
964 		goto out;
965 	}
966 
967 	fdnode = VP_TO_TMPFS_DIR(fdvp);
968 	fnode = VP_TO_TMPFS_NODE(fvp);
969 	de = tmpfs_dir_lookup(fdnode, fnode, fncp);
970 
971 	/* Avoid manipulating '.' and '..' entries. */
972 	if (de == NULL) {
973 		error = ENOENT;
974 		goto out_locked;
975 	}
976 	KKASSERT(de->td_node == fnode);
977 
978 	/*
979 	 * If replacing an entry in the target directory and that entry
980 	 * is a directory, it must be empty.
981 	 *
982 	 * Kern_rename gurantees the destination to be a directory
983 	 * if the source is one (it does?).
984 	 */
985 	if (tvp != NULL) {
986 		KKASSERT(tnode != NULL);
987 
988 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
989 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
990 			error = EPERM;
991 			goto out_locked;
992 		}
993 
994 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
995 			if (tnode->tn_size > 0) {
996 				error = ENOTEMPTY;
997 				goto out_locked;
998 			}
999 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1000 			error = ENOTDIR;
1001 			goto out_locked;
1002 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1003 			error = EISDIR;
1004 			goto out_locked;
1005 		} else {
1006 			KKASSERT(fnode->tn_type != VDIR &&
1007 				tnode->tn_type != VDIR);
1008 		}
1009 	}
1010 
1011 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1012 	    (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1013 		error = EPERM;
1014 		goto out_locked;
1015 	}
1016 
1017 	/*
1018 	 * Ensure that we have enough memory to hold the new name, if it
1019 	 * has to be changed.
1020 	 */
1021 	if (fncp->nc_nlen != tncp->nc_nlen ||
1022 	    bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1023 		newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1024 				  M_WAITOK | M_NULLOK);
1025 		if (newname == NULL) {
1026 			error = ENOSPC;
1027 			goto out_locked;
1028 		}
1029 		bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1030 		newname[tncp->nc_nlen] = '\0';
1031 	} else {
1032 		newname = NULL;
1033 	}
1034 
1035 	/*
1036 	 * Unlink entry from source directory.  Note that the kernel has
1037 	 * already checked for illegal recursion cases (renaming a directory
1038 	 * into a subdirectory of itself).
1039 	 */
1040 	if (fdnode != tdnode)
1041 		tmpfs_dir_detach(fdnode, de);
1042 	else {
1043 		RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de);
1044 	}
1045 
1046 	/*
1047 	 * Handle any name change.  Swap with newname, we will
1048 	 * deallocate it at the end.
1049 	 */
1050 	if (newname != NULL) {
1051 #if 0
1052 		TMPFS_NODE_LOCK(fnode);
1053 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1054 		TMPFS_NODE_UNLOCK(fnode);
1055 #endif
1056 		oldname = de->td_name;
1057 		de->td_name = newname;
1058 		de->td_namelen = (uint16_t)tncp->nc_nlen;
1059 		newname = oldname;
1060 	}
1061 
1062 	/*
1063 	 * If we are overwriting an entry, we have to remove the old one
1064 	 * from the target directory.
1065 	 */
1066 	if (tvp != NULL) {
1067 		/* Remove the old entry from the target directory. */
1068 		tde = tmpfs_dir_lookup(tdnode, tnode, tncp);
1069 		tmpfs_dir_detach(tdnode, tde);
1070 		tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1071 
1072 		/*
1073 		 * Free the directory entry we just deleted.  Note that the
1074 		 * node referred by it will not be removed until the vnode is
1075 		 * really reclaimed.
1076 		 */
1077 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1078 		/*cache_inval_vp(tvp, CINV_DESTROY);*/
1079 	}
1080 
1081 	/*
1082 	 * Link entry to target directory.  If the entry
1083 	 * represents a directory move the parent linkage
1084 	 * as well.
1085 	 */
1086 	if (fdnode != tdnode) {
1087 		if (de->td_node->tn_type == VDIR) {
1088 			TMPFS_VALIDATE_DIR(fnode);
1089 
1090 			TMPFS_NODE_LOCK(tdnode);
1091 			tdnode->tn_links++;
1092 			tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1093 			TMPFS_NODE_UNLOCK(tdnode);
1094 
1095 			TMPFS_NODE_LOCK(fnode);
1096 			fnode->tn_dir.tn_parent = tdnode;
1097 			fnode->tn_status |= TMPFS_NODE_CHANGED;
1098 			TMPFS_NODE_UNLOCK(fnode);
1099 
1100 			TMPFS_NODE_LOCK(fdnode);
1101 			fdnode->tn_links--;
1102 			fdnode->tn_status |= TMPFS_NODE_MODIFIED;
1103 			TMPFS_NODE_UNLOCK(fdnode);
1104 		}
1105 		tmpfs_dir_attach(tdnode, de);
1106 	} else {
1107 		TMPFS_NODE_LOCK(tdnode);
1108 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1109 		RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de);
1110 		TMPFS_NODE_UNLOCK(tdnode);
1111 	}
1112 
1113 	/*
1114 	 * Finish up
1115 	 */
1116 	if (newname) {
1117 		kfree(newname, tmp->tm_name_zone);
1118 		newname = NULL;
1119 	}
1120 	cache_rename(v->a_fnch, v->a_tnch);
1121 	tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1122 	tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1123 	if (fnode->tn_vnode)
1124 		tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1125 	error = 0;
1126 
1127 out_locked:
1128 	;
1129 
1130 out:
1131 	if (tvp)
1132 		vrele(tvp);
1133 
1134 	return error;
1135 }
1136 
1137 /* --------------------------------------------------------------------- */
1138 
1139 static int
1140 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1141 {
1142 	struct vnode *dvp = v->a_dvp;
1143 	struct vnode **vpp = v->a_vpp;
1144 	struct namecache *ncp = v->a_nch->ncp;
1145 	struct vattr *vap = v->a_vap;
1146 	struct ucred *cred = v->a_cred;
1147 	int error;
1148 
1149 	KKASSERT(vap->va_type == VDIR);
1150 
1151 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1152 	if (error == 0) {
1153 		cache_setunresolved(v->a_nch);
1154 		cache_setvp(v->a_nch, *vpp);
1155 		tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1156 	}
1157 
1158 	return error;
1159 }
1160 
1161 /* --------------------------------------------------------------------- */
1162 
1163 static int
1164 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1165 {
1166 	struct vnode *dvp = v->a_dvp;
1167 	struct namecache *ncp = v->a_nch->ncp;
1168 	struct vnode *vp;
1169 	struct tmpfs_dirent *de;
1170 	struct tmpfs_mount *tmp;
1171 	struct tmpfs_node *dnode;
1172 	struct tmpfs_node *node;
1173 	int error;
1174 
1175 	/*
1176 	 * We have to acquire the vp from v->a_nch because we will likely
1177 	 * unresolve the namecache entry, and a vrele/vput is needed to
1178 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1179 	 *
1180 	 * We have to use vget to clear any inactive state on the vnode,
1181 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1182 	 * will not get called when we release it.
1183 	 */
1184 	error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
1185 	KKASSERT(error == 0);
1186 	vn_unlock(vp);
1187 
1188 	/*
1189 	 * Prevalidate so we don't hit an assertion later
1190 	 */
1191 	if (vp->v_type != VDIR) {
1192 		error = ENOTDIR;
1193 		goto out;
1194 	}
1195 
1196 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1197 	dnode = VP_TO_TMPFS_DIR(dvp);
1198 	node = VP_TO_TMPFS_DIR(vp);
1199 
1200 	/* Directories with more than two entries ('.' and '..') cannot be
1201 	 * removed. */
1202 	 if (node->tn_size > 0) {
1203 		 error = ENOTEMPTY;
1204 		 goto out;
1205 	 }
1206 
1207 	if ((dnode->tn_flags & APPEND)
1208 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1209 		error = EPERM;
1210 		goto out;
1211 	}
1212 
1213 	/* This invariant holds only if we are not trying to remove "..".
1214 	  * We checked for that above so this is safe now. */
1215 	KKASSERT(node->tn_dir.tn_parent == dnode);
1216 
1217 	/* Get the directory entry associated with node (vp).  This was
1218 	 * filled by tmpfs_lookup while looking up the entry. */
1219 	de = tmpfs_dir_lookup(dnode, node, ncp);
1220 	KKASSERT(TMPFS_DIRENT_MATCHES(de,
1221 	    ncp->nc_name,
1222 	    ncp->nc_nlen));
1223 
1224 	/* Check flags to see if we are allowed to remove the directory. */
1225 	if ((dnode->tn_flags & APPEND) ||
1226 	    node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1227 		error = EPERM;
1228 		goto out;
1229 	}
1230 
1231 
1232 	/* Detach the directory entry from the directory (dnode). */
1233 	tmpfs_dir_detach(dnode, de);
1234 
1235 	/* No vnode should be allocated for this entry from this point */
1236 	TMPFS_NODE_LOCK(node);
1237 	TMPFS_ASSERT_ELOCKED(node);
1238 	TMPFS_NODE_LOCK(dnode);
1239 	TMPFS_ASSERT_ELOCKED(dnode);
1240 
1241 #if 0
1242 	/* handled by tmpfs_free_node */
1243 	KKASSERT(node->tn_links > 0);
1244 	node->tn_links--;
1245 	node->tn_dir.tn_parent = NULL;
1246 #endif
1247 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1248 	    TMPFS_NODE_MODIFIED;
1249 
1250 #if 0
1251 	/* handled by tmpfs_free_node */
1252 	KKASSERT(dnode->tn_links > 0);
1253 	dnode->tn_links--;
1254 #endif
1255 	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1256 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1257 
1258 	TMPFS_NODE_UNLOCK(dnode);
1259 	TMPFS_NODE_UNLOCK(node);
1260 
1261 	/* Free the directory entry we just deleted.  Note that the node
1262 	 * referred by it will not be removed until the vnode is really
1263 	 * reclaimed. */
1264 	tmpfs_free_dirent(tmp, de);
1265 
1266 	/* Release the deleted vnode (will destroy the node, notify
1267 	 * interested parties and clean it from the cache). */
1268 
1269 	TMPFS_NODE_LOCK(dnode);
1270 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1271 	TMPFS_NODE_UNLOCK(dnode);
1272 	tmpfs_update(dvp);
1273 
1274 	cache_setunresolved(v->a_nch);
1275 	cache_setvp(v->a_nch, NULL);
1276 	/*cache_inval_vp(vp, CINV_DESTROY);*/
1277 	tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1278 	error = 0;
1279 
1280 out:
1281 	vrele(vp);
1282 
1283 	return error;
1284 }
1285 
1286 /* --------------------------------------------------------------------- */
1287 
1288 static int
1289 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1290 {
1291 	struct vnode *dvp = v->a_dvp;
1292 	struct vnode **vpp = v->a_vpp;
1293 	struct namecache *ncp = v->a_nch->ncp;
1294 	struct vattr *vap = v->a_vap;
1295 	struct ucred *cred = v->a_cred;
1296 	char *target = v->a_target;
1297 	int error;
1298 
1299 	vap->va_type = VLNK;
1300 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1301 	if (error == 0) {
1302 		tmpfs_knote(*vpp, NOTE_WRITE);
1303 		cache_setunresolved(v->a_nch);
1304 		cache_setvp(v->a_nch, *vpp);
1305 	}
1306 
1307 	return error;
1308 }
1309 
1310 /* --------------------------------------------------------------------- */
1311 
1312 static int
1313 tmpfs_readdir(struct vop_readdir_args *v)
1314 {
1315 	struct vnode *vp = v->a_vp;
1316 	struct uio *uio = v->a_uio;
1317 	int *eofflag = v->a_eofflag;
1318 	off_t **cookies = v->a_cookies;
1319 	int *ncookies = v->a_ncookies;
1320 	struct tmpfs_mount *tmp;
1321 	int error;
1322 	off_t startoff;
1323 	off_t cnt = 0;
1324 	struct tmpfs_node *node;
1325 
1326 	/* This operation only makes sense on directory nodes. */
1327 	if (vp->v_type != VDIR)
1328 		return ENOTDIR;
1329 
1330 	tmp = VFS_TO_TMPFS(vp->v_mount);
1331 	node = VP_TO_TMPFS_DIR(vp);
1332 	startoff = uio->uio_offset;
1333 
1334 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1335 		error = tmpfs_dir_getdotdent(node, uio);
1336 		if (error != 0)
1337 			goto outok;
1338 		cnt++;
1339 	}
1340 
1341 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1342 		error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1343 		if (error != 0)
1344 			goto outok;
1345 		cnt++;
1346 	}
1347 
1348 	error = tmpfs_dir_getdents(node, uio, &cnt);
1349 
1350 outok:
1351 	KKASSERT(error >= -1);
1352 
1353 	if (error == -1)
1354 		error = 0;
1355 
1356 	if (eofflag != NULL)
1357 		*eofflag =
1358 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1359 
1360 	/* Update NFS-related variables. */
1361 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1362 		off_t i;
1363 		off_t off = startoff;
1364 		struct tmpfs_dirent *de = NULL;
1365 
1366 		*ncookies = cnt;
1367 		*cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1368 
1369 		for (i = 0; i < cnt; i++) {
1370 			KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1371 			if (off == TMPFS_DIRCOOKIE_DOT) {
1372 				off = TMPFS_DIRCOOKIE_DOTDOT;
1373 			} else {
1374 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1375 					de = RB_MIN(tmpfs_dirtree, &node->tn_dir.tn_dirtree);
1376 				} else if (de != NULL) {
1377 					de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
1378 				} else {
1379 					de = tmpfs_dir_lookupbycookie(node,
1380 					    off);
1381 					KKASSERT(de != NULL);
1382 					de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
1383 				}
1384 				if (de == NULL)
1385 					off = TMPFS_DIRCOOKIE_EOF;
1386 				else
1387 					off = tmpfs_dircookie(de);
1388 			}
1389 
1390 			(*cookies)[i] = off;
1391 		}
1392 		KKASSERT(uio->uio_offset == off);
1393 	}
1394 
1395 	return error;
1396 }
1397 
1398 /* --------------------------------------------------------------------- */
1399 
1400 static int
1401 tmpfs_readlink(struct vop_readlink_args *v)
1402 {
1403 	struct vnode *vp = v->a_vp;
1404 	struct uio *uio = v->a_uio;
1405 
1406 	int error;
1407 	struct tmpfs_node *node;
1408 
1409 	KKASSERT(uio->uio_offset == 0);
1410 	KKASSERT(vp->v_type == VLNK);
1411 
1412 	node = VP_TO_TMPFS_NODE(vp);
1413 
1414 	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1415 	    uio);
1416 	TMPFS_NODE_LOCK(node);
1417 	node->tn_status |= TMPFS_NODE_ACCESSED;
1418 	TMPFS_NODE_UNLOCK(node);
1419 
1420 	return error;
1421 }
1422 
1423 /* --------------------------------------------------------------------- */
1424 
1425 static int
1426 tmpfs_inactive(struct vop_inactive_args *v)
1427 {
1428 	struct vnode *vp = v->a_vp;
1429 	struct tmpfs_node *node;
1430 
1431 	node = VP_TO_TMPFS_NODE(vp);
1432 
1433 	/*
1434 	 * Degenerate case
1435 	 */
1436 	if (node == NULL) {
1437 		vrecycle(vp);
1438 		return(0);
1439 	}
1440 
1441 	/*
1442 	 * Get rid of unreferenced deleted vnodes sooner rather than
1443 	 * later so the data memory can be recovered immediately.
1444 	 *
1445 	 * We must truncate the vnode to prevent the normal reclamation
1446 	 * path from flushing the data for the removed file to disk.
1447 	 */
1448 	TMPFS_NODE_LOCK(node);
1449 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1450 	    (node->tn_links == 0 ||
1451 	     (node->tn_links == 1 && node->tn_type == VDIR &&
1452 	      node->tn_dir.tn_parent)))
1453 	{
1454 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1455 		TMPFS_NODE_UNLOCK(node);
1456 		if (node->tn_type == VREG)
1457 			tmpfs_truncate(vp, 0);
1458 		vrecycle(vp);
1459 	} else {
1460 		TMPFS_NODE_UNLOCK(node);
1461 	}
1462 
1463 	return 0;
1464 }
1465 
1466 /* --------------------------------------------------------------------- */
1467 
1468 int
1469 tmpfs_reclaim(struct vop_reclaim_args *v)
1470 {
1471 	struct vnode *vp = v->a_vp;
1472 	struct tmpfs_mount *tmp;
1473 	struct tmpfs_node *node;
1474 
1475 	node = VP_TO_TMPFS_NODE(vp);
1476 	tmp = VFS_TO_TMPFS(vp->v_mount);
1477 
1478 	tmpfs_free_vp(vp);
1479 
1480 	/*
1481 	 * If the node referenced by this vnode was deleted by the
1482 	 * user, we must free its associated data structures now that
1483 	 * the vnode is being reclaimed.
1484 	 *
1485 	 * Directories have an extra link ref.
1486 	 */
1487 	TMPFS_NODE_LOCK(node);
1488 	if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1489 	    (node->tn_links == 0 ||
1490 	     (node->tn_links == 1 && node->tn_type == VDIR &&
1491 	      node->tn_dir.tn_parent)))
1492 	{
1493 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1494 		tmpfs_free_node(tmp, node);
1495 		/* eats the lock */
1496 	} else {
1497 		TMPFS_NODE_UNLOCK(node);
1498 	}
1499 
1500 	KKASSERT(vp->v_data == NULL);
1501 	return 0;
1502 }
1503 
1504 /* --------------------------------------------------------------------- */
1505 
1506 static int
1507 tmpfs_mountctl(struct vop_mountctl_args *ap)
1508 {
1509 	struct tmpfs_mount *tmp;
1510 	struct mount *mp;
1511 	int rc;
1512 
1513 	switch (ap->a_op) {
1514 	case (MOUNTCTL_SET_EXPORT):
1515 		mp = ap->a_head.a_ops->head.vv_mount;
1516 		tmp = (struct tmpfs_mount *) mp->mnt_data;
1517 
1518 		if (ap->a_ctllen != sizeof(struct export_args))
1519 			rc = (EINVAL);
1520 		else
1521 			rc = vfs_export(mp, &tmp->tm_export,
1522 					(const struct export_args *) ap->a_ctl);
1523 		break;
1524 	default:
1525 		rc = vop_stdmountctl(ap);
1526 		break;
1527 	}
1528 	return (rc);
1529 }
1530 
1531 /* --------------------------------------------------------------------- */
1532 
1533 static int
1534 tmpfs_print(struct vop_print_args *v)
1535 {
1536 	struct vnode *vp = v->a_vp;
1537 
1538 	struct tmpfs_node *node;
1539 
1540 	node = VP_TO_TMPFS_NODE(vp);
1541 
1542 	kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1543 	    node, node->tn_flags, node->tn_links);
1544 	kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1545 	    node->tn_mode, node->tn_uid, node->tn_gid,
1546 	    (uintmax_t)node->tn_size, node->tn_status);
1547 
1548 	if (vp->v_type == VFIFO)
1549 		fifo_printinfo(vp);
1550 
1551 	kprintf("\n");
1552 
1553 	return 0;
1554 }
1555 
1556 /* --------------------------------------------------------------------- */
1557 
1558 static int
1559 tmpfs_pathconf(struct vop_pathconf_args *v)
1560 {
1561 	int name = v->a_name;
1562 	register_t *retval = v->a_retval;
1563 
1564 	int error;
1565 
1566 	error = 0;
1567 
1568 	switch (name) {
1569 	case _PC_LINK_MAX:
1570 		*retval = LINK_MAX;
1571 		break;
1572 
1573 	case _PC_NAME_MAX:
1574 		*retval = NAME_MAX;
1575 		break;
1576 
1577 	case _PC_PATH_MAX:
1578 		*retval = PATH_MAX;
1579 		break;
1580 
1581 	case _PC_PIPE_BUF:
1582 		*retval = PIPE_BUF;
1583 		break;
1584 
1585 	case _PC_CHOWN_RESTRICTED:
1586 		*retval = 1;
1587 		break;
1588 
1589 	case _PC_NO_TRUNC:
1590 		*retval = 1;
1591 		break;
1592 
1593 	case _PC_SYNC_IO:
1594 		*retval = 1;
1595 		break;
1596 
1597 	case _PC_FILESIZEBITS:
1598 		*retval = 0; /* XXX Don't know which value should I return. */
1599 		break;
1600 
1601 	default:
1602 		error = EINVAL;
1603 	}
1604 
1605 	return error;
1606 }
1607 
1608 /************************************************************************
1609  *                          KQFILTER OPS                                *
1610  ************************************************************************/
1611 
1612 static void filt_tmpfsdetach(struct knote *kn);
1613 static int filt_tmpfsread(struct knote *kn, long hint);
1614 static int filt_tmpfswrite(struct knote *kn, long hint);
1615 static int filt_tmpfsvnode(struct knote *kn, long hint);
1616 
1617 static struct filterops tmpfsread_filtops =
1618 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1619 static struct filterops tmpfswrite_filtops =
1620 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1621 static struct filterops tmpfsvnode_filtops =
1622 	{ FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1623 
1624 static int
1625 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1626 {
1627 	struct vnode *vp = ap->a_vp;
1628 	struct knote *kn = ap->a_kn;
1629 
1630 	switch (kn->kn_filter) {
1631 	case EVFILT_READ:
1632 		kn->kn_fop = &tmpfsread_filtops;
1633 		break;
1634 	case EVFILT_WRITE:
1635 		kn->kn_fop = &tmpfswrite_filtops;
1636 		break;
1637 	case EVFILT_VNODE:
1638 		kn->kn_fop = &tmpfsvnode_filtops;
1639 		break;
1640 	default:
1641 		return (EOPNOTSUPP);
1642 	}
1643 
1644 	kn->kn_hook = (caddr_t)vp;
1645 
1646 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1647 
1648 	return(0);
1649 }
1650 
1651 static void
1652 filt_tmpfsdetach(struct knote *kn)
1653 {
1654 	struct vnode *vp = (void *)kn->kn_hook;
1655 
1656 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1657 }
1658 
1659 static int
1660 filt_tmpfsread(struct knote *kn, long hint)
1661 {
1662 	struct vnode *vp = (void *)kn->kn_hook;
1663 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1664 	off_t off;
1665 
1666 	if (hint == NOTE_REVOKE) {
1667 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1668 		return(1);
1669 	}
1670 
1671 	/*
1672 	 * Interlock against MP races when performing this function.
1673 	 */
1674 	lwkt_gettoken(&vp->v_mount->mnt_token);
1675 	off = node->tn_size - kn->kn_fp->f_offset;
1676 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1677 	if (kn->kn_sfflags & NOTE_OLDAPI) {
1678 		lwkt_reltoken(&vp->v_mount->mnt_token);
1679 		return(1);
1680 	}
1681 
1682 	if (kn->kn_data == 0) {
1683 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1684 	}
1685 	lwkt_reltoken(&vp->v_mount->mnt_token);
1686 	return (kn->kn_data != 0);
1687 }
1688 
1689 static int
1690 filt_tmpfswrite(struct knote *kn, long hint)
1691 {
1692 	if (hint == NOTE_REVOKE)
1693 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1694 	kn->kn_data = 0;
1695 	return (1);
1696 }
1697 
1698 static int
1699 filt_tmpfsvnode(struct knote *kn, long hint)
1700 {
1701 	if (kn->kn_sfflags & hint)
1702 		kn->kn_fflags |= hint;
1703 	if (hint == NOTE_REVOKE) {
1704 		kn->kn_flags |= (EV_EOF | EV_NODATA);
1705 		return (1);
1706 	}
1707 	return (kn->kn_fflags != 0);
1708 }
1709 
1710 
1711 /* --------------------------------------------------------------------- */
1712 
1713 /*
1714  * vnode operations vector used for files stored in a tmpfs file system.
1715  */
1716 struct vop_ops tmpfs_vnode_vops = {
1717 	.vop_default =			vop_defaultop,
1718 	.vop_getpages = 		vop_stdgetpages,
1719 	.vop_putpages = 		vop_stdputpages,
1720 	.vop_ncreate =			tmpfs_ncreate,
1721 	.vop_nresolve =			tmpfs_nresolve,
1722 	.vop_nlookupdotdot =		tmpfs_nlookupdotdot,
1723 	.vop_nmknod =			tmpfs_nmknod,
1724 	.vop_open =			tmpfs_open,
1725 	.vop_close =			tmpfs_close,
1726 	.vop_access =			tmpfs_access,
1727 	.vop_getattr =			tmpfs_getattr,
1728 	.vop_setattr =			tmpfs_setattr,
1729 	.vop_read =			tmpfs_read,
1730 	.vop_write =			tmpfs_write,
1731 	.vop_fsync =			tmpfs_fsync,
1732 	.vop_mountctl =			tmpfs_mountctl,
1733 	.vop_nremove =			tmpfs_nremove,
1734 	.vop_nlink =			tmpfs_nlink,
1735 	.vop_nrename =			tmpfs_nrename,
1736 	.vop_nmkdir =			tmpfs_nmkdir,
1737 	.vop_nrmdir =			tmpfs_nrmdir,
1738 	.vop_nsymlink =			tmpfs_nsymlink,
1739 	.vop_readdir =			tmpfs_readdir,
1740 	.vop_readlink =			tmpfs_readlink,
1741 	.vop_inactive =			tmpfs_inactive,
1742 	.vop_reclaim =			tmpfs_reclaim,
1743 	.vop_print =			tmpfs_print,
1744 	.vop_pathconf =			tmpfs_pathconf,
1745 	.vop_bmap =			tmpfs_bmap,
1746 	.vop_strategy =			tmpfs_strategy,
1747 	.vop_advlock =			tmpfs_advlock,
1748 	.vop_kqfilter =			tmpfs_kqfilter
1749 };
1750