xref: /dragonfly/sys/vfs/tmpfs/tmpfs_vnops.c (revision 2b3f93ea)
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/uio.h>
41 #include <sys/fcntl.h>
42 #include <sys/lockf.h>
43 #include <sys/caps.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/sysctl.h>
50 #include <sys/unistd.h>
51 #include <sys/vfsops.h>
52 #include <sys/vnode.h>
53 #include <sys/mountctl.h>
54 
55 #include <vm/vm.h>
56 #include <vm/vm_extern.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pageout.h>
60 #include <vm/vm_pager.h>
61 #include <vm/swap_pager.h>
62 
63 #include <sys/buf2.h>
64 #include <vm/vm_page2.h>
65 
66 #include <vfs/fifofs/fifo.h>
67 #include <vfs/tmpfs/tmpfs_vnops.h>
68 #include "tmpfs.h"
69 
70 static void tmpfs_strategy_done(struct bio *bio);
71 static void tmpfs_move_pages(vm_object_t src, vm_object_t dst, int movflags);
72 
73 /*
74  * bufcache_mode:
75  *	0	Normal page queue operation on flush.  Run through the buffer
76  *		cache if free memory is under the minimum.
77  *
78  *	1	Try to keep in memory, but run through the buffer cache if
79  *		the system is under memory pressure (though this might just
80  *		require inactive cleaning).
81  *
82  *	2	Be a bit more aggressive when running writes through the
83  *		buffer cache when the system is under memory pressure.
84  *
85  *	3	Always run tmpfs writes through the buffer cache, thus forcing
86  *		them out to swap.
87  */
88 __read_mostly static int tmpfs_cluster_rd_enable = 1;
89 __read_mostly static int tmpfs_cluster_wr_enable = 1;
90 __read_mostly int tmpfs_bufcache_mode = 0;
91 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "TMPFS filesystem");
92 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, cluster_rd_enable, CTLFLAG_RW,
93 		&tmpfs_cluster_rd_enable, 0, "");
94 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, cluster_wr_enable, CTLFLAG_RW,
95 		&tmpfs_cluster_wr_enable, 0, "");
96 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, bufcache_mode, CTLFLAG_RW,
97 		&tmpfs_bufcache_mode, 0, "");
98 
99 #define TMPFS_MOVF_FROMBACKING	0x0001
100 #define TMPFS_MOVF_DEACTIVATE	0x0002
101 
102 
103 static __inline
104 void
tmpfs_knote(struct vnode * vp,int flags)105 tmpfs_knote(struct vnode *vp, int flags)
106 {
107 	if (flags)
108 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
109 }
110 
111 
112 /* --------------------------------------------------------------------- */
113 
114 static int
tmpfs_nresolve(struct vop_nresolve_args * ap)115 tmpfs_nresolve(struct vop_nresolve_args *ap)
116 {
117 	struct vnode *dvp = ap->a_dvp;
118 	struct vnode *vp = NULL;
119 	struct namecache *ncp = ap->a_nch->ncp;
120 	struct tmpfs_node *tnode;
121 	struct tmpfs_dirent *de;
122 	struct tmpfs_node *dnode;
123 	int error;
124 
125 	dnode = VP_TO_TMPFS_DIR(dvp);
126 
127 	TMPFS_NODE_LOCK_SH(dnode);
128 loop:
129 	de = tmpfs_dir_lookup(dnode, NULL, ncp);
130 	if (de == NULL) {
131 		error = ENOENT;
132 	} else {
133 		/*
134 		 * Allocate a vnode for the node we found.  Use
135 		 * tmpfs_alloc_vp()'s deadlock handling mode.
136 		 */
137 		tnode = de->td_node;
138 		error = tmpfs_alloc_vp(dvp->v_mount, dnode, tnode,
139 				       LK_EXCLUSIVE | LK_RETRY, &vp);
140 		if (error == EAGAIN)
141 			goto loop;
142 		if (error)
143 			goto out;
144 		KKASSERT(vp);
145 	}
146 
147 out:
148 	TMPFS_NODE_UNLOCK(dnode);
149 
150 	if ((dnode->tn_status & TMPFS_NODE_ACCESSED) == 0) {
151 		TMPFS_NODE_LOCK(dnode);
152 		dnode->tn_status |= TMPFS_NODE_ACCESSED;
153 		TMPFS_NODE_UNLOCK(dnode);
154 	}
155 
156 	/*
157 	 * Store the result of this lookup in the cache.  Avoid this if the
158 	 * request was for creation, as it does not improve timings on
159 	 * emprical tests.
160 	 */
161 	if (vp) {
162 		vn_unlock(vp);
163 		cache_setvp(ap->a_nch, vp);
164 		vrele(vp);
165 	} else if (error == ENOENT) {
166 		cache_setvp(ap->a_nch, NULL);
167 	}
168 	return (error);
169 }
170 
171 static int
tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args * ap)172 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
173 {
174 	struct vnode *dvp = ap->a_dvp;
175 	struct vnode **vpp = ap->a_vpp;
176 	struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
177 	struct ucred *cred = ap->a_cred;
178 	int error;
179 
180 	*vpp = NULL;
181 
182 	/* Check accessibility of requested node as a first step. */
183 	error = VOP_ACCESS(dvp, VEXEC, cred);
184 	if (error != 0)
185 		return error;
186 
187 	if (dnode->tn_dir.tn_parent != NULL) {
188 		/* Allocate a new vnode on the matching entry. */
189 		error = tmpfs_alloc_vp(dvp->v_mount,
190 				       NULL, dnode->tn_dir.tn_parent,
191 				       LK_EXCLUSIVE | LK_RETRY, vpp);
192 
193 		if (*vpp)
194 			vn_unlock(*vpp);
195 	}
196 	return (*vpp == NULL) ? ENOENT : 0;
197 }
198 
199 /* --------------------------------------------------------------------- */
200 
201 static int
tmpfs_ncreate(struct vop_ncreate_args * ap)202 tmpfs_ncreate(struct vop_ncreate_args *ap)
203 {
204 	struct vnode *dvp = ap->a_dvp;
205 	struct vnode **vpp = ap->a_vpp;
206 	struct namecache *ncp = ap->a_nch->ncp;
207 	struct vattr *vap = ap->a_vap;
208 	struct ucred *cred = ap->a_cred;
209 	int error;
210 
211 	KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
212 
213 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
214 	if (error == 0) {
215 		cache_setunresolved(ap->a_nch);
216 		cache_setvp(ap->a_nch, *vpp);
217 		tmpfs_knote(dvp, NOTE_WRITE);
218 	}
219 	return (error);
220 }
221 /* --------------------------------------------------------------------- */
222 
223 static int
tmpfs_nmknod(struct vop_nmknod_args * ap)224 tmpfs_nmknod(struct vop_nmknod_args *ap)
225 {
226 	struct vnode *dvp = ap->a_dvp;
227 	struct vnode **vpp = ap->a_vpp;
228 	struct namecache *ncp = ap->a_nch->ncp;
229 	struct vattr *vap = ap->a_vap;
230 	struct ucred *cred = ap->a_cred;
231 	int error;
232 
233 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
234 	    vap->va_type != VFIFO) {
235 		return (EINVAL);
236 	}
237 
238 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
239 	if (error == 0) {
240 		cache_setunresolved(ap->a_nch);
241 		cache_setvp(ap->a_nch, *vpp);
242 		tmpfs_knote(dvp, NOTE_WRITE);
243 	}
244 	return error;
245 }
246 
247 /* --------------------------------------------------------------------- */
248 
249 static int
tmpfs_open(struct vop_open_args * ap)250 tmpfs_open(struct vop_open_args *ap)
251 {
252 	struct vnode *vp = ap->a_vp;
253 	int mode = ap->a_mode;
254 	struct tmpfs_node *node;
255 	int error;
256 
257 	node = VP_TO_TMPFS_NODE(vp);
258 
259 #if 0
260 	/* The file is still active but all its names have been removed
261 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
262 	 * it is about to die. */
263 	if (node->tn_links < 1)
264 		return (ENOENT);
265 #endif
266 
267 	/* If the file is marked append-only, deny write requests. */
268 	if ((node->tn_flags & APPEND) &&
269 	    (mode & (FWRITE | O_APPEND)) == FWRITE) {
270 		error = EPERM;
271 	} else {
272 		if (node->tn_reg.tn_pages_in_aobj) {
273 			TMPFS_NODE_LOCK(node);
274 			if (node->tn_reg.tn_pages_in_aobj) {
275 				tmpfs_move_pages(node->tn_reg.tn_aobj,
276 						 vp->v_object,
277 						 TMPFS_MOVF_FROMBACKING);
278 				node->tn_reg.tn_pages_in_aobj = 0;
279 			}
280 			TMPFS_NODE_UNLOCK(node);
281 		}
282 		error = vop_stdopen(ap);
283 	}
284 
285 	return (error);
286 }
287 
288 /* --------------------------------------------------------------------- */
289 
290 static int
tmpfs_close(struct vop_close_args * ap)291 tmpfs_close(struct vop_close_args *ap)
292 {
293 	struct vnode *vp = ap->a_vp;
294 	struct tmpfs_node *node;
295 	int error;
296 
297 	node = VP_TO_TMPFS_NODE(vp);
298 
299 	if (node->tn_links > 0) {
300 		/*
301 		 * Update node times.  No need to do it if the node has
302 		 * been deleted, because it will vanish after we return.
303 		 */
304 		tmpfs_update(vp);
305 	}
306 
307 	error = vop_stdclose(ap);
308 
309 	return (error);
310 }
311 
312 /* --------------------------------------------------------------------- */
313 
314 int
tmpfs_access(struct vop_access_args * ap)315 tmpfs_access(struct vop_access_args *ap)
316 {
317 	struct vnode *vp = ap->a_vp;
318 	int error;
319 	struct tmpfs_node *node;
320 
321 	node = VP_TO_TMPFS_NODE(vp);
322 
323 	switch (vp->v_type) {
324 	case VDIR:
325 		/* FALLTHROUGH */
326 	case VLNK:
327 		/* FALLTHROUGH */
328 	case VREG:
329 		if ((ap->a_mode & VWRITE) &&
330 	            (vp->v_mount->mnt_flag & MNT_RDONLY)) {
331 			error = EROFS;
332 			goto out;
333 		}
334 		break;
335 
336 	case VBLK:
337 		/* FALLTHROUGH */
338 	case VCHR:
339 		/* FALLTHROUGH */
340 	case VSOCK:
341 		/* FALLTHROUGH */
342 	case VFIFO:
343 		break;
344 
345 	default:
346 		error = EINVAL;
347 		goto out;
348 	}
349 
350 	if ((ap->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
351 		error = EPERM;
352 		goto out;
353 	}
354 
355 	error = vop_helper_access(ap, node->tn_uid, node->tn_gid,
356 			          node->tn_mode, 0);
357 out:
358 	return error;
359 }
360 
361 /* --------------------------------------------------------------------- */
362 
363 int
tmpfs_getattr(struct vop_getattr_args * ap)364 tmpfs_getattr(struct vop_getattr_args *ap)
365 {
366 	struct vnode *vp = ap->a_vp;
367 	struct vattr *vap = ap->a_vap;
368 	struct tmpfs_node *node;
369 
370 	node = VP_TO_TMPFS_NODE(vp);
371 
372 	tmpfs_update(vp);
373 
374 	vap->va_type = vp->v_type;
375 	vap->va_mode = node->tn_mode;
376 	vap->va_nlink = node->tn_links;
377 	vap->va_uid = node->tn_uid;
378 	vap->va_gid = node->tn_gid;
379 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
380 	vap->va_fileid = node->tn_id;
381 	vap->va_size = node->tn_size;
382 	vap->va_blocksize = PAGE_SIZE;
383 	vap->va_atime.tv_sec = node->tn_atime;
384 	vap->va_atime.tv_nsec = node->tn_atimensec;
385 	vap->va_mtime.tv_sec = node->tn_mtime;
386 	vap->va_mtime.tv_nsec = node->tn_mtimensec;
387 	vap->va_ctime.tv_sec = node->tn_ctime;
388 	vap->va_ctime.tv_nsec = node->tn_ctimensec;
389 	vap->va_gen = node->tn_gen;
390 	vap->va_flags = node->tn_flags;
391 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
392 		vap->va_rmajor = umajor(node->tn_rdev);
393 		vap->va_rminor = uminor(node->tn_rdev);
394 	}
395 	vap->va_bytes = round_page(node->tn_size);
396 	vap->va_filerev = 0;
397 
398 	return 0;
399 }
400 
401 /* --------------------------------------------------------------------- */
402 
403 int
tmpfs_getattr_lite(struct vop_getattr_lite_args * ap)404 tmpfs_getattr_lite(struct vop_getattr_lite_args *ap)
405 {
406 	struct vnode *vp = ap->a_vp;
407 	struct vattr_lite *lvap = ap->a_lvap;
408 	struct tmpfs_node *node;
409 
410 	node = VP_TO_TMPFS_NODE(vp);
411 
412 	tmpfs_update(vp);
413 
414 	lvap->va_type = vp->v_type;
415 	lvap->va_mode = node->tn_mode;
416 	lvap->va_nlink = node->tn_links;
417 	lvap->va_uid = node->tn_uid;
418 	lvap->va_gid = node->tn_gid;
419 #if 0
420 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
421 	vap->va_fileid = node->tn_id;
422 #endif
423 	lvap->va_size = node->tn_size;
424 #if 0
425 	vap->va_blocksize = PAGE_SIZE;
426 	vap->va_gen = node->tn_gen;
427 #endif
428 	lvap->va_flags = node->tn_flags;
429 #if 0
430 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
431 		vap->va_rmajor = umajor(node->tn_rdev);
432 		vap->va_rminor = uminor(node->tn_rdev);
433 	}
434 	vap->va_bytes = -1;
435 	vap->va_filerev = 0;
436 #endif
437 
438 	return 0;
439 }
440 
441 
442 /* --------------------------------------------------------------------- */
443 
444 int
tmpfs_setattr(struct vop_setattr_args * ap)445 tmpfs_setattr(struct vop_setattr_args *ap)
446 {
447 	struct vnode *vp = ap->a_vp;
448 	struct vattr *vap = ap->a_vap;
449 	struct ucred *cred = ap->a_cred;
450 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
451 	int error = 0;
452 	int kflags = 0;
453 
454 	TMPFS_NODE_LOCK(node);
455 	if (error == 0 && (vap->va_flags != VNOVAL)) {
456 		error = tmpfs_chflags(vp, vap->va_flags, cred);
457 		kflags |= NOTE_ATTRIB;
458 	}
459 
460 	if (error == 0 && (vap->va_size != VNOVAL)) {
461 		/* restore any saved pages before proceeding */
462 		if (node->tn_reg.tn_pages_in_aobj) {
463 			tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object,
464 					 TMPFS_MOVF_FROMBACKING |
465 					 TMPFS_MOVF_DEACTIVATE);
466 			node->tn_reg.tn_pages_in_aobj = 0;
467 		}
468 		if (vap->va_size > node->tn_size)
469 			kflags |= NOTE_WRITE | NOTE_EXTEND;
470 		else
471 			kflags |= NOTE_WRITE;
472 		error = tmpfs_chsize(vp, vap->va_size, cred);
473 	}
474 
475 	if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
476 			   vap->va_gid != (gid_t)VNOVAL)) {
477 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
478 		kflags |= NOTE_ATTRIB;
479 	}
480 
481 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
482 		error = tmpfs_chmod(vp, vap->va_mode, cred);
483 		kflags |= NOTE_ATTRIB;
484 	}
485 
486 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
487 	    vap->va_atime.tv_nsec != VNOVAL) ||
488 	    (vap->va_mtime.tv_sec != VNOVAL &&
489 	    vap->va_mtime.tv_nsec != VNOVAL) )) {
490 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
491 				      vap->va_vaflags, cred);
492 		kflags |= NOTE_ATTRIB;
493 	}
494 
495 	/*
496 	 * Update the node times.  We give preference to the error codes
497 	 * generated by this function rather than the ones that may arise
498 	 * from tmpfs_update.
499 	 */
500 	tmpfs_update(vp);
501 	TMPFS_NODE_UNLOCK(node);
502 	tmpfs_knote(vp, kflags);
503 
504 	return (error);
505 }
506 
507 /* --------------------------------------------------------------------- */
508 
509 /*
510  * fsync is usually a NOP, but we must take action when unmounting or
511  * when recycling.
512  */
513 static int
tmpfs_fsync(struct vop_fsync_args * ap)514 tmpfs_fsync(struct vop_fsync_args *ap)
515 {
516 	struct tmpfs_node *node;
517 	struct vnode *vp = ap->a_vp;
518 
519 	node = VP_TO_TMPFS_NODE(vp);
520 
521 	/*
522 	 * tmpfs vnodes typically remain dirty, avoid long syncer scans
523 	 * by forcing removal from the syncer list.
524 	 */
525 	vn_syncer_remove(vp, 1);
526 
527 	tmpfs_update(vp);
528 	if (vp->v_type == VREG) {
529 		if (vp->v_flag & VRECLAIMED) {
530 			if (node->tn_links == 0)
531 				tmpfs_truncate(vp, 0);
532 			else
533 				vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL);
534 		}
535 	}
536 
537 	return 0;
538 }
539 
540 /* --------------------------------------------------------------------- */
541 
542 static int
tmpfs_read(struct vop_read_args * ap)543 tmpfs_read(struct vop_read_args *ap)
544 {
545 	struct buf *bp;
546 	struct vnode *vp = ap->a_vp;
547 	struct uio *uio = ap->a_uio;
548 	struct tmpfs_node *node;
549 	off_t base_offset;
550 	size_t offset;
551 	size_t len;
552 	size_t resid;
553 	int error;
554 	int seqcount;
555 
556 	/*
557 	 * Check the basics
558 	 */
559 	if (uio->uio_offset < 0)
560 		return (EINVAL);
561 	if (vp->v_type != VREG)
562 		return (EINVAL);
563 
564 	/*
565 	 * Extract node, try to shortcut the operation through
566 	 * the VM page cache, allowing us to avoid buffer cache
567 	 * overheads.
568 	 */
569 	node = VP_TO_TMPFS_NODE(vp);
570         resid = uio->uio_resid;
571 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
572         error = vop_helper_read_shortcut(ap);
573         if (error)
574                 return error;
575         if (uio->uio_resid == 0) {
576 		if (resid)
577 			goto finished;
578 		return error;
579 	}
580 
581 	/*
582 	 * restore any saved pages before proceeding
583 	 */
584 	if (node->tn_reg.tn_pages_in_aobj) {
585 		TMPFS_NODE_LOCK(node);
586 		if (node->tn_reg.tn_pages_in_aobj) {
587 			tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object,
588 					 TMPFS_MOVF_FROMBACKING);
589 			node->tn_reg.tn_pages_in_aobj = 0;
590 		}
591 		TMPFS_NODE_UNLOCK(node);
592 	}
593 
594 	/*
595 	 * Fall-through to our normal read code.
596 	 */
597 	while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
598 		/*
599 		 * Use buffer cache I/O (via tmpfs_strategy)
600 		 */
601 		offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64;
602 		base_offset = (off_t)uio->uio_offset - offset;
603 		bp = getcacheblk(vp, base_offset,
604 				 node->tn_blksize, GETBLK_KVABIO);
605 		if (bp == NULL) {
606 			if (tmpfs_cluster_rd_enable) {
607 				error = cluster_readx(vp, node->tn_size,
608 						     base_offset,
609 						     node->tn_blksize,
610 						     B_NOTMETA | B_KVABIO,
611 						     uio->uio_resid,
612 						     seqcount * MAXBSIZE,
613 						     &bp);
614 			} else {
615 				error = bread_kvabio(vp, base_offset,
616 						     node->tn_blksize, &bp);
617 			}
618 			if (error) {
619 				brelse(bp);
620 				kprintf("tmpfs_read bread error %d\n", error);
621 				break;
622 			}
623 
624 			/*
625 			 * tmpfs pretty much fiddles directly with the VM
626 			 * system, don't let it exhaust it or we won't play
627 			 * nice with other processes.
628 			 *
629 			 * Only do this if the VOP is coming from a normal
630 			 * read/write.  The VM system handles the case for
631 			 * UIO_NOCOPY.
632 			 */
633 			if (uio->uio_segflg != UIO_NOCOPY)
634 				vm_wait_nominal();
635 		}
636 		bp->b_flags |= B_CLUSTEROK;
637 		bkvasync(bp);
638 
639 		/*
640 		 * Figure out how many bytes we can actually copy this loop.
641 		 */
642 		len = node->tn_blksize - offset;
643 		if (len > uio->uio_resid)
644 			len = uio->uio_resid;
645 		if (len > node->tn_size - uio->uio_offset)
646 			len = (size_t)(node->tn_size - uio->uio_offset);
647 
648 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
649 		bqrelse(bp);
650 		if (error) {
651 			kprintf("tmpfs_read uiomove error %d\n", error);
652 			break;
653 		}
654 	}
655 
656 finished:
657 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
658 		TMPFS_NODE_LOCK(node);
659 		node->tn_status |= TMPFS_NODE_ACCESSED;
660 		TMPFS_NODE_UNLOCK(node);
661 	}
662 	return (error);
663 }
664 
665 static int
tmpfs_write(struct vop_write_args * ap)666 tmpfs_write(struct vop_write_args *ap)
667 {
668 	struct buf *bp;
669 	struct vnode *vp = ap->a_vp;
670 	struct uio *uio = ap->a_uio;
671 	struct thread *td = uio->uio_td;
672 	struct tmpfs_node *node;
673 	boolean_t extended;
674 	off_t oldsize;
675 	int error;
676 	off_t base_offset;
677 	size_t offset;
678 	size_t len;
679 	struct rlimit limit;
680 	int trivial = 0;
681 	int kflags = 0;
682 	int seqcount;
683 
684 	error = 0;
685 	if (uio->uio_resid == 0) {
686 		return error;
687 	}
688 
689 	node = VP_TO_TMPFS_NODE(vp);
690 
691 	if (vp->v_type != VREG)
692 		return (EINVAL);
693 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
694 
695 	TMPFS_NODE_LOCK(node);
696 
697 	/*
698 	 * restore any saved pages before proceeding
699 	 */
700 	if (node->tn_reg.tn_pages_in_aobj) {
701 		tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object,
702 				 TMPFS_MOVF_FROMBACKING);
703 		node->tn_reg.tn_pages_in_aobj = 0;
704 	}
705 
706 	oldsize = node->tn_size;
707 	if (ap->a_ioflag & IO_APPEND)
708 		uio->uio_offset = node->tn_size;
709 
710 	/*
711 	 * Check for illegal write offsets.
712 	 */
713 	if (uio->uio_offset + uio->uio_resid >
714 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
715 		error = EFBIG;
716 		goto done;
717 	}
718 
719 	/*
720 	 * NOTE: Ignore if UIO does not come from a user thread (e.g. VN).
721 	 */
722 	if (vp->v_type == VREG && td != NULL && td->td_lwp != NULL) {
723 		error = kern_getrlimit(RLIMIT_FSIZE, &limit);
724 		if (error)
725 			goto done;
726 		if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
727 			ksignal(td->td_proc, SIGXFSZ);
728 			error = EFBIG;
729 			goto done;
730 		}
731 	}
732 
733 	/*
734 	 * Extend the file's size if necessary
735 	 */
736 	extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
737 
738 	while (uio->uio_resid > 0) {
739 		/*
740 		 * Don't completely blow out running buffer I/O
741 		 * when being hit from the pageout daemon.
742 		 */
743 		if (uio->uio_segflg == UIO_NOCOPY &&
744 		    (ap->a_ioflag & IO_RECURSE) == 0) {
745 			bwillwrite(node->tn_blksize);
746 		}
747 
748 		/*
749 		 * Use buffer cache I/O (via tmpfs_strategy)
750 		 *
751 		 * Calculate the maximum bytes we can write to the buffer at
752 		 * this offset (after resizing).
753 		 */
754 		offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64;
755 		base_offset = (off_t)uio->uio_offset - offset;
756 		len = uio->uio_resid;
757 		if (len > TMPFS_BLKSIZE - offset)
758 			len = TMPFS_BLKSIZE - offset;
759 
760 		if ((uio->uio_offset + len) > node->tn_size) {
761 			trivial = (uio->uio_offset <= node->tn_size);
762 			error = tmpfs_reg_resize(vp, uio->uio_offset + len,
763 						 trivial);
764 			if (error)
765 				break;
766 		}
767 
768 		/*
769 		 * Read to fill in any gaps.  Theoretically we could
770 		 * optimize this if the write covers the entire buffer
771 		 * and is not a UIO_NOCOPY write, however this can lead
772 		 * to a security violation exposing random kernel memory
773 		 * (whatever junk was in the backing VM pages before).
774 		 *
775 		 * So just use bread() to do the right thing.
776 		 */
777 		error = bread_kvabio(vp, base_offset, node->tn_blksize, &bp);
778 		bkvasync(bp);
779 		error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
780 		if (error) {
781 			kprintf("tmpfs_write uiomove error %d\n", error);
782 			brelse(bp);
783 			break;
784 		}
785 
786 		if (uio->uio_offset > node->tn_size) {
787 			node->tn_size = uio->uio_offset;
788 			kflags |= NOTE_EXTEND;
789 		}
790 		kflags |= NOTE_WRITE;
791 
792 		/*
793 		 * UIO_NOCOPY is a sensitive state due to potentially being
794 		 * issued from the pageout daemon while in a low-memory
795 		 * situation.  However, in order to cluster the I/O nicely
796 		 * (e.g. 64KB+ writes instead of 16KB writes), we still try
797 		 * to follow the same semantics that any other filesystem
798 		 * might use.
799 		 *
800 		 * For the normal case we buwrite(), dirtying the underlying
801 		 * VM pages instead of dirtying the buffer and releasing the
802 		 * buffer as a clean buffer.  This allows tmpfs to use
803 		 * essentially all available memory to cache file data.
804 		 * If we used bdwrite() the buffer cache would wind up
805 		 * flushing the data to swap too quickly.
806 		 *
807 		 * But because tmpfs can seriously load the VM system we
808 		 * fall-back to using bdwrite() when free memory starts
809 		 * to get low.  This shifts the load away from the VM system
810 		 * and makes tmpfs act more like a normal filesystem with
811 		 * regards to disk activity.
812 		 *
813 		 * tmpfs pretty much fiddles directly with the VM
814 		 * system, don't let it exhaust it or we won't play
815 		 * nice with other processes.  Only do this if the
816 		 * VOP is coming from a normal read/write.  The VM system
817 		 * handles the case for UIO_NOCOPY.
818 		 */
819 		bp->b_flags |= B_CLUSTEROK;
820 		if (uio->uio_segflg == UIO_NOCOPY) {
821 			/*
822 			 * Flush from the pageout daemon, deal with potentially
823 			 * very heavy tmpfs write activity causing long stalls
824 			 * in the pageout daemon before pages get to free/cache.
825 			 *
826 			 * We have to be careful not to bypass the page queues
827 			 * entirely or we can cause write-read thrashing and
828 			 * delay the paging of data that is more pageable then
829 			 * our current data.
830 			 *
831 			 * (a) Under severe pressure setting B_DIRECT will
832 			 *     cause a buffer release to try to free the
833 			 *     underlying pages.
834 			 *
835 			 * (b) Under modest memory pressure the B_AGE flag
836 			 *     we retire the buffer and its underlying pages
837 			 *     more quickly than normal.
838 			 *
839 			 *     We could also force this by setting B_NOTMETA
840 			 *     but that might have other unintended side-
841 			 *     effects (e.g. setting PG_NOTMETA on the VM page).
842 			 *
843 			 * (c) For the pageout->putpages->generic_putpages->
844 			 *     UIO_NOCOPY-write (here), issuing an immediate
845 			 *     write prevents any real clustering from
846 			 *     happening because the buffers probably aren't
847 			 *     (yet) marked dirty, or lost due to prior use
848 			 *     of buwrite().  Try to use the normal
849 			 *     cluster_write() mechanism for performance.
850 			 *
851 			 * Hopefully this will unblock the VM system more
852 			 * quickly under extreme tmpfs write load.
853 			 */
854 			if (tmpfs_bufcache_mode >= 2) {
855 				if (vm_paging_min_dnc(vm_page_free_hysteresis))
856 					bp->b_flags |= B_DIRECT | B_TTC;
857 				if (vm_pages_needed || vm_paging_start(0))
858 					bp->b_flags |= B_AGE;
859 			}
860 			bp->b_flags |= B_RELBUF;
861 			bp->b_act_count = 0;	/* buffer->deactivate pgs */
862 			if (tmpfs_cluster_wr_enable &&
863 			    (ap->a_ioflag & (IO_SYNC | IO_DIRECT)) == 0) {
864 				cluster_write(bp, node->tn_size,
865 					      node->tn_blksize, seqcount);
866 			} else {
867 				cluster_awrite(bp);
868 			}
869 		} else if (vm_paging_min() ||
870 			   ((vm_pages_needed || vm_paging_start(0)) &&
871 			    tmpfs_bufcache_mode >= 1)) {
872 			/*
873 			 * If the pageout daemon is running we cycle the
874 			 * write through the buffer cache normally to
875 			 * pipeline the flush, thus avoiding adding any
876 			 * more memory pressure to the pageout daemon.
877 			 */
878 			bp->b_act_count = 0;	/* buffer->deactivate pgs */
879 			if (tmpfs_cluster_wr_enable) {
880 				cluster_write(bp, node->tn_size,
881 					      node->tn_blksize, seqcount);
882 			} else {
883 				bdwrite(bp);
884 			}
885 		} else {
886 			/*
887 			 * Otherwise run the buffer directly through to the
888 			 * backing VM store, leaving the buffer clean so
889 			 * buffer limits do not force early flushes to swap.
890 			 */
891 			buwrite(bp);
892 			/*vm_wait_nominal();*/
893 		}
894 
895 		if (bp->b_error) {
896 			kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
897 			break;
898 		}
899 	}
900 
901 	if (error) {
902 		if (extended) {
903 			(void)tmpfs_reg_resize(vp, oldsize, trivial);
904 			kflags &= ~NOTE_EXTEND;
905 		}
906 		goto done;
907 	}
908 
909 	/*
910 	 * Currently we don't set the mtime on files modified via mmap()
911 	 * because we can't tell the difference between those modifications
912 	 * and an attempt by the pageout daemon to flush tmpfs pages to
913 	 * swap.
914 	 *
915 	 * This is because in order to defer flushes as long as possible
916 	 * buwrite() works by marking the underlying VM pages dirty in
917 	 * order to be able to dispose of the buffer cache buffer without
918 	 * flushing it.
919 	 */
920 	if (uio->uio_segflg == UIO_NOCOPY) {
921 		if (vp->v_flag & VLASTWRITETS) {
922 			node->tn_mtime = vp->v_lastwrite_ts.tv_sec;
923 			node->tn_mtimensec = vp->v_lastwrite_ts.tv_nsec;
924 		}
925 	} else {
926 		node->tn_status |= TMPFS_NODE_MODIFIED;
927 		vclrflags(vp, VLASTWRITETS);
928 	}
929 
930 	if (extended)
931 		node->tn_status |= TMPFS_NODE_CHANGED;
932 
933 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
934 		if (caps_priv_check(ap->a_cred, SYSCAP_NOVFS_RETAINSUGID))
935 			node->tn_mode &= ~(S_ISUID | S_ISGID);
936 	}
937 done:
938 	TMPFS_NODE_UNLOCK(node);
939 	if (kflags)
940 		tmpfs_knote(vp, kflags);
941 
942 	return(error);
943 }
944 
945 static int
tmpfs_advlock(struct vop_advlock_args * ap)946 tmpfs_advlock(struct vop_advlock_args *ap)
947 {
948 	struct tmpfs_node *node;
949 	struct vnode *vp = ap->a_vp;
950 	int error;
951 
952 	node = VP_TO_TMPFS_NODE(vp);
953 	error = (lf_advlock(ap, &node->tn_advlock, node->tn_size));
954 
955 	return (error);
956 }
957 
958 /*
959  * The strategy function is typically only called when memory pressure
960  * forces the system to attempt to pageout pages.  It can also be called
961  * by [n]vtruncbuf() when a truncation cuts a page in half.  Normal write
962  * operations
963  *
964  * We set VKVABIO for VREG files so bp->b_data may not be synchronized to
965  * our cpu.  swap_pager_strategy() is all we really use, and it directly
966  * supports this.
967  */
968 static int
tmpfs_strategy(struct vop_strategy_args * ap)969 tmpfs_strategy(struct vop_strategy_args *ap)
970 {
971 	struct bio *bio = ap->a_bio;
972 	struct bio *nbio;
973 	struct buf *bp = bio->bio_buf;
974 	struct vnode *vp = ap->a_vp;
975 	struct tmpfs_node *node;
976 	vm_object_t uobj;
977 	vm_page_t m;
978 	int i;
979 
980 	if (vp->v_type != VREG) {
981 		bp->b_resid = bp->b_bcount;
982 		bp->b_flags |= B_ERROR | B_INVAL;
983 		bp->b_error = EINVAL;
984 		biodone(bio);
985 		return(0);
986 	}
987 
988 	node = VP_TO_TMPFS_NODE(vp);
989 
990 	uobj = node->tn_reg.tn_aobj;
991 
992 	/*
993 	 * Don't bother flushing to swap if there is no swap, just
994 	 * ensure that the pages are marked as needing a commit (still).
995 	 */
996 	if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
997 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
998 			m = bp->b_xio.xio_pages[i];
999 			vm_page_need_commit(m);
1000 		}
1001 		bp->b_resid = 0;
1002 		bp->b_error = 0;
1003 		biodone(bio);
1004 	} else {
1005 #if 0
1006 		/*
1007 		 * XXX removed, this does not work well because under heavy
1008 		 * filesystem loads it often
1009 		 * forces the data to be read right back in again after
1010 		 * being written due to bypassing normal LRU operation.
1011 		 *
1012 		 * Tell the buffer cache to try to recycle the pages
1013 		 * to PQ_CACHE on release.
1014 		 */
1015 		if (tmpfs_bufcache_mode >= 2 ||
1016 		    (tmpfs_bufcache_mode == 1 && vm_paging_needed(0))) {
1017 			bp->b_flags |= B_TTC;
1018 		}
1019 #endif
1020 		nbio = push_bio(bio);
1021 		nbio->bio_done = tmpfs_strategy_done;
1022 		nbio->bio_offset = bio->bio_offset;
1023 		swap_pager_strategy(uobj, nbio);
1024 	}
1025 	return 0;
1026 }
1027 
1028 /*
1029  * If we were unable to commit the pages to swap make sure they are marked
1030  * as needing a commit (again).  If we were, clear the flag to allow the
1031  * pages to be freed.
1032  *
1033  * Do not error-out the buffer.  In particular, vinvalbuf() needs to
1034  * always work.
1035  */
1036 static void
tmpfs_strategy_done(struct bio * bio)1037 tmpfs_strategy_done(struct bio *bio)
1038 {
1039 	struct buf *bp;
1040 	vm_page_t m;
1041 	int i;
1042 
1043 	bp = bio->bio_buf;
1044 
1045 	if (bp->b_flags & B_ERROR) {
1046 		bp->b_flags &= ~B_ERROR;
1047 		bp->b_error = 0;
1048 		bp->b_resid = 0;
1049 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
1050 			m = bp->b_xio.xio_pages[i];
1051 			vm_page_need_commit(m);
1052 		}
1053 	} else {
1054 		for (i = 0; i < bp->b_xio.xio_npages; ++i) {
1055 			m = bp->b_xio.xio_pages[i];
1056 			vm_page_clear_commit(m);
1057 		}
1058 	}
1059 	bio = pop_bio(bio);
1060 	biodone(bio);
1061 }
1062 
1063 /*
1064  * To make write clustering work well make the backing store look
1065  * contiguous to the cluster_*() code.  The swap_strategy() function
1066  * will take it from there.
1067  *
1068  * Use MAXBSIZE-sized chunks as a micro-optimization to make random
1069  * flushes leave full-sized gaps.
1070  */
1071 static int
tmpfs_bmap(struct vop_bmap_args * ap)1072 tmpfs_bmap(struct vop_bmap_args *ap)
1073 {
1074 	if (ap->a_doffsetp != NULL)
1075 		*ap->a_doffsetp = ap->a_loffset;
1076 	if (ap->a_runp != NULL)
1077 		*ap->a_runp = MAXBSIZE - (ap->a_loffset & (MAXBSIZE - 1));
1078 	if (ap->a_runb != NULL)
1079 		*ap->a_runb = ap->a_loffset & (MAXBSIZE - 1);
1080 
1081 	return 0;
1082 }
1083 
1084 /* --------------------------------------------------------------------- */
1085 
1086 static int
tmpfs_nremove(struct vop_nremove_args * ap)1087 tmpfs_nremove(struct vop_nremove_args *ap)
1088 {
1089 	struct vnode *dvp = ap->a_dvp;
1090 	struct namecache *ncp = ap->a_nch->ncp;
1091 	struct vnode *vp;
1092 	int error;
1093 	struct tmpfs_dirent *de;
1094 	struct tmpfs_mount *tmp;
1095 	struct tmpfs_node *dnode;
1096 	struct tmpfs_node *node;
1097 
1098 	/*
1099 	 * We have to acquire the vp from ap->a_nch because we will likely
1100 	 * unresolve the namecache entry, and a vrele/vput is needed to
1101 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1102 	 *
1103 	 * We have to use vget to clear any inactive state on the vnode,
1104 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1105 	 * will not get called when we release it.
1106 	 */
1107 	error = cache_vget(ap->a_nch, ap->a_cred, LK_SHARED, &vp);
1108 	KKASSERT(vp->v_mount == dvp->v_mount);
1109 	KKASSERT(error == 0);
1110 	vn_unlock(vp);
1111 
1112 	if (vp->v_type == VDIR) {
1113 		error = EISDIR;
1114 		goto out2;
1115 	}
1116 
1117 	dnode = VP_TO_TMPFS_DIR(dvp);
1118 	node = VP_TO_TMPFS_NODE(vp);
1119 	tmp = VFS_TO_TMPFS(vp->v_mount);
1120 
1121 	TMPFS_NODE_LOCK(dnode);
1122 	TMPFS_NODE_LOCK(node);
1123 	de = tmpfs_dir_lookup(dnode, node, ncp);
1124 	if (de == NULL) {
1125 		error = ENOENT;
1126 		TMPFS_NODE_UNLOCK(node);
1127 		TMPFS_NODE_UNLOCK(dnode);
1128 		goto out;
1129 	}
1130 
1131 	/* Files marked as immutable or append-only cannot be deleted. */
1132 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
1133 	    (dnode->tn_flags & APPEND)) {
1134 		error = EPERM;
1135 		TMPFS_NODE_UNLOCK(node);
1136 		TMPFS_NODE_UNLOCK(dnode);
1137 		goto out;
1138 	}
1139 
1140 	/* Remove the entry from the directory; as it is a file, we do not
1141 	 * have to change the number of hard links of the directory. */
1142 	tmpfs_dir_detach_locked(dnode, de);
1143 	TMPFS_NODE_UNLOCK(dnode);
1144 
1145 	/* Free the directory entry we just deleted.  Note that the node
1146 	 * referred by it will not be removed until the vnode is really
1147 	 * reclaimed. */
1148 	tmpfs_free_dirent(tmp, de);
1149 
1150 	if (node->tn_links > 0)
1151 		node->tn_status |= TMPFS_NODE_CHANGED;
1152 	TMPFS_NODE_UNLOCK(node);
1153 
1154 	cache_unlink(ap->a_nch);
1155 	tmpfs_knote(vp, NOTE_DELETE);
1156 	error = 0;
1157 
1158 out:
1159 	if (error == 0)
1160 		tmpfs_knote(dvp, NOTE_WRITE);
1161 out2:
1162 	vrele(vp);
1163 
1164 	return error;
1165 }
1166 
1167 /* --------------------------------------------------------------------- */
1168 
1169 static int
tmpfs_nlink(struct vop_nlink_args * ap)1170 tmpfs_nlink(struct vop_nlink_args *ap)
1171 {
1172 	struct vnode *dvp = ap->a_dvp;
1173 	struct vnode *vp = ap->a_vp;
1174 	struct tmpfs_mount *tmp = VFS_TO_TMPFS(vp->v_mount);
1175 	struct namecache *ncp = ap->a_nch->ncp;
1176 	struct tmpfs_dirent *de;
1177 	struct tmpfs_node *node;
1178 	struct tmpfs_node *dnode;
1179 	int error;
1180 
1181 	KKASSERT(dvp != vp); /* XXX When can this be false? */
1182 
1183 	node = VP_TO_TMPFS_NODE(vp);
1184 	dnode = VP_TO_TMPFS_NODE(dvp);
1185 	TMPFS_NODE_LOCK(dnode);
1186 
1187 	/* XXX: Why aren't the following two tests done by the caller? */
1188 
1189 	/* Hard links of directories are forbidden. */
1190 	if (vp->v_type == VDIR) {
1191 		error = EPERM;
1192 		goto out;
1193 	}
1194 
1195 	/* Cannot create cross-device links. */
1196 	if (dvp->v_mount != vp->v_mount) {
1197 		error = EXDEV;
1198 		goto out;
1199 	}
1200 
1201 	/* Cannot hard-link into a deleted directory */
1202 	if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL) {
1203 		error = ENOENT;
1204 		goto out;
1205 	}
1206 
1207 	/* Ensure that we do not overflow the maximum number of links imposed
1208 	 * by the system. */
1209 	KKASSERT(node->tn_links <= LINK_MAX);
1210 	if (node->tn_links >= LINK_MAX) {
1211 		error = EMLINK;
1212 		goto out;
1213 	}
1214 
1215 	/* We cannot create links of files marked immutable or append-only. */
1216 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
1217 		error = EPERM;
1218 		goto out;
1219 	}
1220 
1221 	/* Allocate a new directory entry to represent the node. */
1222 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
1223 				   ncp->nc_name, ncp->nc_nlen, &de);
1224 	if (error != 0)
1225 		goto out;
1226 
1227 	/* Insert the new directory entry into the appropriate directory. */
1228 	tmpfs_dir_attach_locked(dnode, de);
1229 
1230 	/* vp link count has changed, so update node times. */
1231 
1232 	TMPFS_NODE_LOCK(node);
1233 	node->tn_status |= TMPFS_NODE_CHANGED;
1234 	TMPFS_NODE_UNLOCK(node);
1235 	tmpfs_update(vp);
1236 
1237 	tmpfs_knote(vp, NOTE_LINK);
1238 	cache_setunresolved(ap->a_nch);
1239 	cache_setvp(ap->a_nch, vp);
1240 	error = 0;
1241 
1242 out:
1243 	TMPFS_NODE_UNLOCK(dnode);
1244 	if (error == 0)
1245 		tmpfs_knote(dvp, NOTE_WRITE);
1246 	return error;
1247 }
1248 
1249 /* --------------------------------------------------------------------- */
1250 
1251 static int
tmpfs_nrename(struct vop_nrename_args * ap)1252 tmpfs_nrename(struct vop_nrename_args *ap)
1253 {
1254 	struct vnode *fdvp = ap->a_fdvp;
1255 	struct namecache *fncp = ap->a_fnch->ncp;
1256 	struct vnode *fvp = fncp->nc_vp;
1257 	struct vnode *tdvp = ap->a_tdvp;
1258 	struct namecache *tncp = ap->a_tnch->ncp;
1259 	struct vnode *tvp;
1260 	struct tmpfs_dirent *de, *tde, *de2;
1261 	struct tmpfs_mount *tmp;
1262 	struct tmpfs_node *fdnode;
1263 	struct tmpfs_node *tdnode;
1264 	struct tmpfs_node *fnode;
1265 	struct tmpfs_node *tnode;
1266 	char *newname;
1267 	char *oldname;
1268 	int error;
1269 
1270 	KKASSERT(fdvp->v_mount == fvp->v_mount);
1271 
1272 	/*
1273 	 * Because tvp can get overwritten we have to vget it instead of
1274 	 * just vref or use it, otherwise it's VINACTIVE flag may not get
1275 	 * cleared and the node won't get destroyed.
1276 	 */
1277 	error = cache_vget(ap->a_tnch, ap->a_cred, LK_SHARED, &tvp);
1278 	if (error == 0) {
1279 		tnode = VP_TO_TMPFS_NODE(tvp);
1280 		vn_unlock(tvp);
1281 	} else {
1282 		tnode = NULL;
1283 	}
1284 
1285 	/* Disallow cross-device renames.
1286 	 * XXX Why isn't this done by the caller? */
1287 	if (fvp->v_mount != tdvp->v_mount ||
1288 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
1289 		error = EXDEV;
1290 		goto out;
1291 	}
1292 
1293 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
1294 	tdnode = VP_TO_TMPFS_DIR(tdvp);
1295 
1296 	/* If source and target are the same file, there is nothing to do. */
1297 	if (fvp == tvp) {
1298 		error = 0;
1299 		goto out;
1300 	}
1301 
1302 	fdnode = VP_TO_TMPFS_DIR(fdvp);
1303 	fnode = VP_TO_TMPFS_NODE(fvp);
1304 
1305 	tmpfs_lock4(fdnode, tdnode, fnode, tnode);
1306 
1307 	/*
1308 	 * Cannot rename into a deleted directory
1309 	 */
1310 	if (tdnode != tmp->tm_root && tdnode->tn_dir.tn_parent == NULL) {
1311 		error = ENOENT;
1312 		goto out_locked;
1313 	}
1314 
1315 	/* Avoid manipulating '.' and '..' entries. */
1316 	de = tmpfs_dir_lookup(fdnode, fnode, fncp);
1317 	if (de == NULL) {
1318 		error = ENOENT;
1319 		goto out_locked;
1320 	}
1321 	KKASSERT(de->td_node == fnode);
1322 
1323 	/*
1324 	 * If replacing an entry in the target directory and that entry
1325 	 * is a directory, it must be empty.
1326 	 *
1327 	 * Kern_rename gurantees the destination to be a directory
1328 	 * if the source is one (it does?).
1329 	 */
1330 	if (tvp != NULL) {
1331 		KKASSERT(tnode != NULL);
1332 
1333 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1334 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1335 			error = EPERM;
1336 			goto out_locked;
1337 		}
1338 
1339 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1340 			if (tnode->tn_size > 0) {
1341 				error = ENOTEMPTY;
1342 				goto out_locked;
1343 			}
1344 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1345 			error = ENOTDIR;
1346 			goto out_locked;
1347 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1348 			error = EISDIR;
1349 			goto out_locked;
1350 		} else {
1351 			KKASSERT(fnode->tn_type != VDIR &&
1352 				tnode->tn_type != VDIR);
1353 		}
1354 	}
1355 
1356 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1357 	    (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1358 		error = EPERM;
1359 		goto out_locked;
1360 	}
1361 
1362 	/*
1363 	 * Ensure that we have enough memory to hold the new name, if it
1364 	 * has to be changed.
1365 	 */
1366 	if (fncp->nc_nlen != tncp->nc_nlen ||
1367 	    bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1368 		newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1369 				  M_WAITOK | M_NULLOK);
1370 		if (newname == NULL) {
1371 			error = ENOSPC;
1372 			goto out_locked;
1373 		}
1374 		bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1375 		newname[tncp->nc_nlen] = '\0';
1376 	} else {
1377 		newname = NULL;
1378 	}
1379 
1380 	/*
1381 	 * Unlink entry from source directory.  Note that the kernel has
1382 	 * already checked for illegal recursion cases (renaming a directory
1383 	 * into a subdirectory of itself).
1384 	 */
1385 	if (fdnode != tdnode) {
1386 		tmpfs_dir_detach_locked(fdnode, de);
1387 	} else {
1388 		/* XXX depend on namecache lock */
1389 		KKASSERT(de == tmpfs_dir_lookup(fdnode, fnode, fncp));
1390 		RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de);
1391 		RB_REMOVE(tmpfs_dirtree_cookie,
1392 			  &fdnode->tn_dir.tn_cookietree, de);
1393 	}
1394 
1395 	/*
1396 	 * Handle any name change.  Swap with newname, we will
1397 	 * deallocate it at the end.
1398 	 */
1399 	if (newname != NULL) {
1400 		oldname = de->td_name;
1401 		de->td_name = newname;
1402 		de->td_namelen = (uint16_t)tncp->nc_nlen;
1403 		newname = oldname;
1404 	}
1405 
1406 	/*
1407 	 * If we are overwriting an entry, we have to remove the old one
1408 	 * from the target directory.
1409 	 */
1410 	if (tvp != NULL) {
1411 		/* Remove the old entry from the target directory. */
1412 		tde = tmpfs_dir_lookup(tdnode, tnode, tncp);
1413 		tmpfs_dir_detach_locked(tdnode, tde);
1414 		tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1415 
1416 		/*
1417 		 * Free the directory entry we just deleted.  Note that the
1418 		 * node referred by it will not be removed until the vnode is
1419 		 * really reclaimed.
1420 		 */
1421 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1422 		/*cache_inval_vp(tvp, CINV_DESTROY);*/
1423 	}
1424 
1425 	/*
1426 	 * Link entry to target directory.  If the entry
1427 	 * represents a directory move the parent linkage
1428 	 * as well.
1429 	 */
1430 	if (fdnode != tdnode) {
1431 		if (de->td_node->tn_type == VDIR) {
1432 			TMPFS_VALIDATE_DIR(fnode);
1433 		}
1434 		tmpfs_dir_attach_locked(tdnode, de);
1435 	} else {
1436 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1437 		de2 = RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de);
1438 		KASSERT(de2 == NULL,
1439 			("tmpfs_nrenameA: duplicate insertion of %p, has %p\n",
1440 			de, de2));
1441 		de2 = RB_INSERT(tmpfs_dirtree_cookie,
1442 				&tdnode->tn_dir.tn_cookietree, de);
1443 		KASSERT(de2 == NULL,
1444 			("tmpfs_nrenameB: duplicate insertion of %p, has %p\n",
1445 			de, de2));
1446 	}
1447 	tmpfs_unlock4(fdnode, tdnode, fnode, tnode);
1448 
1449 	/*
1450 	 * Finish up
1451 	 */
1452 	if (newname) {
1453 		kfree(newname, tmp->tm_name_zone);
1454 		newname = NULL;
1455 	}
1456 	cache_rename(ap->a_fnch, ap->a_tnch);
1457 	tmpfs_knote(ap->a_fdvp, NOTE_WRITE);
1458 	tmpfs_knote(ap->a_tdvp, NOTE_WRITE);
1459 	if (fnode->tn_vnode)
1460 		tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1461 	if (tvp)
1462 		vrele(tvp);
1463 	return 0;
1464 
1465 out_locked:
1466 	tmpfs_unlock4(fdnode, tdnode, fnode, tnode);
1467 out:
1468 	if (tvp)
1469 		vrele(tvp);
1470 	return error;
1471 }
1472 
1473 /* --------------------------------------------------------------------- */
1474 
1475 static int
tmpfs_nmkdir(struct vop_nmkdir_args * ap)1476 tmpfs_nmkdir(struct vop_nmkdir_args *ap)
1477 {
1478 	struct vnode *dvp = ap->a_dvp;
1479 	struct vnode **vpp = ap->a_vpp;
1480 	struct namecache *ncp = ap->a_nch->ncp;
1481 	struct vattr *vap = ap->a_vap;
1482 	struct ucred *cred = ap->a_cred;
1483 	int error;
1484 
1485 	KKASSERT(vap->va_type == VDIR);
1486 
1487 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1488 	if (error == 0) {
1489 		cache_setunresolved(ap->a_nch);
1490 		cache_setvp(ap->a_nch, *vpp);
1491 		tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1492 	}
1493 	return error;
1494 }
1495 
1496 /* --------------------------------------------------------------------- */
1497 
1498 static int
tmpfs_nrmdir(struct vop_nrmdir_args * ap)1499 tmpfs_nrmdir(struct vop_nrmdir_args *ap)
1500 {
1501 	struct vnode *dvp = ap->a_dvp;
1502 	struct namecache *ncp = ap->a_nch->ncp;
1503 	struct vnode *vp;
1504 	struct tmpfs_dirent *de;
1505 	struct tmpfs_mount *tmp;
1506 	struct tmpfs_node *dnode;
1507 	struct tmpfs_node *node;
1508 	int error;
1509 
1510 	/*
1511 	 * We have to acquire the vp from ap->a_nch because we will likely
1512 	 * unresolve the namecache entry, and a vrele/vput is needed to
1513 	 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1514 	 *
1515 	 * We have to use vget to clear any inactive state on the vnode,
1516 	 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1517 	 * will not get called when we release it.
1518 	 */
1519 	error = cache_vget(ap->a_nch, ap->a_cred, LK_SHARED, &vp);
1520 	KKASSERT(error == 0);
1521 	vn_unlock(vp);
1522 
1523 	/*
1524 	 * Prevalidate so we don't hit an assertion later
1525 	 */
1526 	if (vp->v_type != VDIR) {
1527 		error = ENOTDIR;
1528 		goto out;
1529 	}
1530 
1531 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1532 	dnode = VP_TO_TMPFS_DIR(dvp);
1533 	node = VP_TO_TMPFS_DIR(vp);
1534 
1535 	/*
1536 	 *
1537 	 */
1538 	TMPFS_NODE_LOCK(dnode);
1539 	TMPFS_NODE_LOCK(node);
1540 
1541 	/*
1542 	 * Only empty directories can be removed.
1543 	 */
1544 	if (node->tn_size > 0) {
1545 		error = ENOTEMPTY;
1546 		goto out_locked;
1547 	}
1548 
1549 	if ((dnode->tn_flags & APPEND)
1550 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1551 		error = EPERM;
1552 		goto out_locked;
1553 	}
1554 
1555 	/*
1556 	 * This invariant holds only if we are not trying to
1557 	 * remove "..".  We checked for that above so this is safe now.
1558 	 */
1559 	KKASSERT(node->tn_dir.tn_parent == dnode);
1560 
1561 	/*
1562 	 * Get the directory entry associated with node (vp)
1563 	 */
1564 	de = tmpfs_dir_lookup(dnode, node, ncp);
1565 	KKASSERT(TMPFS_DIRENT_MATCHES(de, ncp->nc_name, ncp->nc_nlen));
1566 
1567 	/* Check flags to see if we are allowed to remove the directory. */
1568 	if ((dnode->tn_flags & APPEND) ||
1569 	    node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1570 		error = EPERM;
1571 		goto out_locked;
1572 	}
1573 
1574 	/* Detach the directory entry from the directory (dnode). */
1575 	tmpfs_dir_detach_locked(dnode, de);
1576 
1577 	/*
1578 	 * Must set parent linkage to NULL (tested by ncreate to disallow
1579 	 * the creation of new files/dirs in a deleted directory)
1580 	 */
1581 	node->tn_status |= TMPFS_NODE_CHANGED;
1582 
1583 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1584 			    TMPFS_NODE_MODIFIED;
1585 
1586 	/* Free the directory entry we just deleted.  Note that the node
1587 	 * referred by it will not be removed until the vnode is really
1588 	 * reclaimed. */
1589 	tmpfs_free_dirent(tmp, de);
1590 
1591 	/* Release the deleted vnode (will destroy the node, notify
1592 	 * interested parties and clean it from the cache). */
1593 
1594 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1595 
1596 	TMPFS_NODE_UNLOCK(node);
1597 	TMPFS_NODE_UNLOCK(dnode);
1598 
1599 	tmpfs_update(dvp);
1600 	cache_unlink(ap->a_nch);
1601 	tmpfs_knote(vp, NOTE_DELETE);
1602 	tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1603 	vrele(vp);
1604 	return 0;
1605 
1606 out_locked:
1607 	TMPFS_NODE_UNLOCK(node);
1608 	TMPFS_NODE_UNLOCK(dnode);
1609 
1610 out:
1611 	vrele(vp);
1612 
1613 	return error;
1614 }
1615 
1616 /* --------------------------------------------------------------------- */
1617 
1618 static int
tmpfs_nsymlink(struct vop_nsymlink_args * ap)1619 tmpfs_nsymlink(struct vop_nsymlink_args *ap)
1620 {
1621 	struct vnode *dvp = ap->a_dvp;
1622 	struct vnode **vpp = ap->a_vpp;
1623 	struct namecache *ncp = ap->a_nch->ncp;
1624 	struct vattr *vap = ap->a_vap;
1625 	struct ucred *cred = ap->a_cred;
1626 	char *target = ap->a_target;
1627 	int error;
1628 
1629 	vap->va_type = VLNK;
1630 	error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1631 	if (error == 0) {
1632 		tmpfs_knote(*vpp, NOTE_WRITE);
1633 		cache_setunresolved(ap->a_nch);
1634 		cache_setvp(ap->a_nch, *vpp);
1635 	}
1636 	return error;
1637 }
1638 
1639 /* --------------------------------------------------------------------- */
1640 
1641 static int
tmpfs_readdir(struct vop_readdir_args * ap)1642 tmpfs_readdir(struct vop_readdir_args *ap)
1643 {
1644 	struct vnode *vp = ap->a_vp;
1645 	struct uio *uio = ap->a_uio;
1646 	int *eofflag = ap->a_eofflag;
1647 	off_t **cookies = ap->a_cookies;
1648 	int *ncookies = ap->a_ncookies;
1649 	struct tmpfs_mount *tmp;
1650 	int error;
1651 	off_t startoff;
1652 	off_t cnt = 0;
1653 	struct tmpfs_node *node;
1654 
1655 	/* This operation only makes sense on directory nodes. */
1656 	if (vp->v_type != VDIR) {
1657 		return ENOTDIR;
1658 	}
1659 
1660 	tmp = VFS_TO_TMPFS(vp->v_mount);
1661 	node = VP_TO_TMPFS_DIR(vp);
1662 	startoff = uio->uio_offset;
1663 
1664 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1665 		error = tmpfs_dir_getdotdent(node, uio);
1666 		if (error && error != EINVAL) {
1667 			TMPFS_NODE_LOCK_SH(node);
1668 			goto outok;
1669 		}
1670 		cnt++;
1671 	}
1672 
1673 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1674 		/* may lock parent, cannot hold node lock */
1675 		error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1676 		if (error && error != EINVAL) {
1677 			TMPFS_NODE_LOCK_SH(node);
1678 			goto outok;
1679 		}
1680 		cnt++;
1681 	}
1682 
1683 	TMPFS_NODE_LOCK_SH(node);
1684 	error = tmpfs_dir_getdents(node, uio, &cnt);
1685 
1686 outok:
1687 	KKASSERT(error >= -1);
1688 
1689 	if (error == -1)
1690 		error = 0;
1691 
1692 	if (eofflag != NULL)
1693 		*eofflag =
1694 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1695 
1696 	/* Update NFS-related variables. */
1697 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1698 		off_t i;
1699 		off_t off = startoff;
1700 		struct tmpfs_dirent *de = NULL;
1701 
1702 		*ncookies = cnt;
1703 		*cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1704 
1705 		for (i = 0; i < cnt; i++) {
1706 			KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1707 			if (off == TMPFS_DIRCOOKIE_DOT) {
1708 				off = TMPFS_DIRCOOKIE_DOTDOT;
1709 			} else {
1710 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1711 					de = RB_MIN(tmpfs_dirtree_cookie,
1712 						&node->tn_dir.tn_cookietree);
1713 				} else if (de != NULL) {
1714 					de = RB_NEXT(tmpfs_dirtree_cookie,
1715 					       &node->tn_dir.tn_cookietree, de);
1716 				} else {
1717 					de = tmpfs_dir_lookupbycookie(node, off,
1718 								      1);
1719 					KKASSERT(de != NULL);
1720 					de = RB_NEXT(tmpfs_dirtree_cookie,
1721 					       &node->tn_dir.tn_cookietree, de);
1722 				}
1723 				if (de == NULL)
1724 					off = TMPFS_DIRCOOKIE_EOF;
1725 				else
1726 					off = tmpfs_dircookie(de);
1727 			}
1728 			(*cookies)[i] = off;
1729 		}
1730 		KKASSERT(uio->uio_offset == off);
1731 	}
1732 	TMPFS_NODE_UNLOCK(node);
1733 
1734 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
1735 		TMPFS_NODE_LOCK(node);
1736 		node->tn_status |= TMPFS_NODE_ACCESSED;
1737 		TMPFS_NODE_UNLOCK(node);
1738 	}
1739 	return error;
1740 }
1741 
1742 /* --------------------------------------------------------------------- */
1743 
1744 static int
tmpfs_readlink(struct vop_readlink_args * ap)1745 tmpfs_readlink(struct vop_readlink_args *ap)
1746 {
1747 	struct vnode *vp = ap->a_vp;
1748 	struct uio *uio = ap->a_uio;
1749 	int error;
1750 	struct tmpfs_node *node;
1751 
1752 	KKASSERT(uio->uio_offset == 0);
1753 	KKASSERT(vp->v_type == VLNK);
1754 
1755 	node = VP_TO_TMPFS_NODE(vp);
1756 	TMPFS_NODE_LOCK_SH(node);
1757 	error = uiomove(node->tn_link,
1758 			MIN(node->tn_size, uio->uio_resid), uio);
1759 	TMPFS_NODE_UNLOCK(node);
1760 	if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) {
1761 		TMPFS_NODE_LOCK(node);
1762 		node->tn_status |= TMPFS_NODE_ACCESSED;
1763 		TMPFS_NODE_UNLOCK(node);
1764 	}
1765 	return error;
1766 }
1767 
1768 /* --------------------------------------------------------------------- */
1769 
1770 static int
tmpfs_inactive(struct vop_inactive_args * ap)1771 tmpfs_inactive(struct vop_inactive_args *ap)
1772 {
1773 	struct vnode *vp = ap->a_vp;
1774 	struct tmpfs_node *node;
1775 	struct mount *mp;
1776 
1777 	mp = vp->v_mount;
1778 	lwkt_gettoken(&mp->mnt_token);
1779 	node = VP_TO_TMPFS_NODE(vp);
1780 
1781 	/*
1782 	 * Degenerate case
1783 	 */
1784 	if (node == NULL) {
1785 		vrecycle(vp);
1786 		lwkt_reltoken(&mp->mnt_token);
1787 		return(0);
1788 	}
1789 
1790 	/*
1791 	 * Get rid of unreferenced deleted vnodes sooner rather than
1792 	 * later so the data memory can be recovered immediately.
1793 	 *
1794 	 * We must truncate the vnode to prevent the normal reclamation
1795 	 * path from flushing the data for the removed file to disk.
1796 	 */
1797 	TMPFS_NODE_LOCK(node);
1798 	if (node->tn_links == 0) {
1799 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1800 		TMPFS_NODE_UNLOCK(node);
1801 		if (node->tn_type == VREG)
1802 			tmpfs_truncate(vp, 0);
1803 		vrecycle(vp);
1804 	} else {
1805 		/*
1806 		 * We must retain any VM pages belonging to the vnode's
1807 		 * object as the vnode will destroy the object during a
1808 		 * later reclaim.  We call vinvalbuf(V_SAVE) to clean
1809 		 * out the buffer cache.
1810 		 *
1811 		 * On DragonFlyBSD, vnodes are not immediately deactivated
1812 		 * on the 1->0 refs, so this is a relatively optimal
1813 		 * operation.  We have to do this in tmpfs_inactive()
1814 		 * because the pages will have already been thrown away
1815 		 * at the time tmpfs_reclaim() is called.
1816 		 */
1817 		if (node->tn_type == VREG &&
1818 		    node->tn_reg.tn_pages_in_aobj == 0) {
1819 			vinvalbuf(vp, V_SAVE, 0, 0);
1820 			KKASSERT(RB_EMPTY(&vp->v_rbdirty_tree));
1821 			KKASSERT(RB_EMPTY(&vp->v_rbclean_tree));
1822 			tmpfs_move_pages(vp->v_object, node->tn_reg.tn_aobj,
1823 					 TMPFS_MOVF_DEACTIVATE);
1824 			node->tn_reg.tn_pages_in_aobj = 1;
1825 		}
1826 
1827 		TMPFS_NODE_UNLOCK(node);
1828 	}
1829 	lwkt_reltoken(&mp->mnt_token);
1830 
1831 	return 0;
1832 }
1833 
1834 /* --------------------------------------------------------------------- */
1835 
1836 int
tmpfs_reclaim(struct vop_reclaim_args * ap)1837 tmpfs_reclaim(struct vop_reclaim_args *ap)
1838 {
1839 	struct vnode *vp = ap->a_vp;
1840 	struct tmpfs_mount *tmp;
1841 	struct tmpfs_node *node;
1842 	struct mount *mp;
1843 
1844 	mp = vp->v_mount;
1845 	lwkt_gettoken(&mp->mnt_token);
1846 
1847 	node = VP_TO_TMPFS_NODE(vp);
1848 	tmp = VFS_TO_TMPFS(vp->v_mount);
1849 	KKASSERT(mp == tmp->tm_mount);
1850 
1851         TMPFS_NODE_LOCK(node);
1852 	KKASSERT(node->tn_vnode == vp);
1853         node->tn_vnode = NULL;
1854         vp->v_data = NULL;
1855 
1856 	/*
1857 	 * If the node referenced by this vnode was deleted by the
1858 	 * user, we must free its associated data structures now that
1859 	 * the vnode is being reclaimed.
1860 	 *
1861 	 * Directories have an extra link ref.
1862 	 */
1863 	if (node->tn_links == 0) {
1864 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1865 		tmpfs_free_node(tmp, node);
1866 		/* eats the lock */
1867 	} else {
1868 		TMPFS_NODE_UNLOCK(node);
1869 	}
1870 	lwkt_reltoken(&mp->mnt_token);
1871 
1872 	KKASSERT(vp->v_data == NULL);
1873 	return 0;
1874 }
1875 
1876 /* --------------------------------------------------------------------- */
1877 
1878 static int
tmpfs_mountctl(struct vop_mountctl_args * ap)1879 tmpfs_mountctl(struct vop_mountctl_args *ap)
1880 {
1881 	struct tmpfs_mount *tmp;
1882 	struct mount *mp;
1883 	int rc;
1884 
1885 	mp = ap->a_head.a_ops->head.vv_mount;
1886 	lwkt_gettoken(&mp->mnt_token);
1887 
1888 	switch (ap->a_op) {
1889 	case (MOUNTCTL_SET_EXPORT):
1890 		tmp = (struct tmpfs_mount *) mp->mnt_data;
1891 
1892 		if (ap->a_ctllen != sizeof(struct export_args))
1893 			rc = (EINVAL);
1894 		else
1895 			rc = vfs_export(mp, &tmp->tm_export,
1896 					(const struct export_args *) ap->a_ctl);
1897 		break;
1898 	default:
1899 		rc = vop_stdmountctl(ap);
1900 		break;
1901 	}
1902 
1903 	lwkt_reltoken(&mp->mnt_token);
1904 	return (rc);
1905 }
1906 
1907 /* --------------------------------------------------------------------- */
1908 
1909 static int
tmpfs_print(struct vop_print_args * ap)1910 tmpfs_print(struct vop_print_args *ap)
1911 {
1912 	struct vnode *vp = ap->a_vp;
1913 
1914 	struct tmpfs_node *node;
1915 
1916 	node = VP_TO_TMPFS_NODE(vp);
1917 
1918 	kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1919 	    node, node->tn_flags, node->tn_links);
1920 	kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1921 	    node->tn_mode, node->tn_uid, node->tn_gid,
1922 	    (uintmax_t)node->tn_size, node->tn_status);
1923 
1924 	if (vp->v_type == VFIFO)
1925 		fifo_printinfo(vp);
1926 
1927 	kprintf("\n");
1928 
1929 	return 0;
1930 }
1931 
1932 /* --------------------------------------------------------------------- */
1933 
1934 static int
tmpfs_pathconf(struct vop_pathconf_args * ap)1935 tmpfs_pathconf(struct vop_pathconf_args *ap)
1936 {
1937 	struct vnode *vp = ap->a_vp;
1938 	int name = ap->a_name;
1939 	register_t *retval = ap->a_retval;
1940 	struct tmpfs_mount *tmp;
1941 	int error;
1942 
1943 	error = 0;
1944 
1945 	switch (name) {
1946 	case _PC_CHOWN_RESTRICTED:
1947 		*retval = 1;
1948 		break;
1949 
1950 	case _PC_FILESIZEBITS:
1951 		tmp = VFS_TO_TMPFS(vp->v_mount);
1952 		*retval = max(32, flsll(tmp->tm_pages_max * PAGE_SIZE) + 1);
1953 		break;
1954 
1955 	case _PC_LINK_MAX:
1956 		*retval = LINK_MAX;
1957 		break;
1958 
1959 	case _PC_NAME_MAX:
1960 		*retval = NAME_MAX;
1961 		break;
1962 
1963 	case _PC_NO_TRUNC:
1964 		*retval = 1;
1965 		break;
1966 
1967 	case _PC_PATH_MAX:
1968 		*retval = PATH_MAX;
1969 		break;
1970 
1971 	case _PC_PIPE_BUF:
1972 		*retval = PIPE_BUF;
1973 		break;
1974 
1975 	case _PC_SYNC_IO:
1976 		*retval = 1;
1977 		break;
1978 
1979 	case _PC_2_SYMLINKS:
1980 		*retval = 1;
1981 		break;
1982 
1983 	default:
1984 		error = EINVAL;
1985 	}
1986 
1987 	return error;
1988 }
1989 
1990 /************************************************************************
1991  *                          KQFILTER OPS                                *
1992  ************************************************************************/
1993 
1994 static void filt_tmpfsdetach(struct knote *kn);
1995 static int filt_tmpfsread(struct knote *kn, long hint);
1996 static int filt_tmpfswrite(struct knote *kn, long hint);
1997 static int filt_tmpfsvnode(struct knote *kn, long hint);
1998 
1999 static struct filterops tmpfsread_filtops =
2000 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
2001 	  NULL, filt_tmpfsdetach, filt_tmpfsread };
2002 static struct filterops tmpfswrite_filtops =
2003 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
2004 	  NULL, filt_tmpfsdetach, filt_tmpfswrite };
2005 static struct filterops tmpfsvnode_filtops =
2006 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
2007 	  NULL, filt_tmpfsdetach, filt_tmpfsvnode };
2008 
2009 static int
tmpfs_kqfilter(struct vop_kqfilter_args * ap)2010 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
2011 {
2012 	struct vnode *vp = ap->a_vp;
2013 	struct knote *kn = ap->a_kn;
2014 
2015 	switch (kn->kn_filter) {
2016 	case EVFILT_READ:
2017 		kn->kn_fop = &tmpfsread_filtops;
2018 		break;
2019 	case EVFILT_WRITE:
2020 		kn->kn_fop = &tmpfswrite_filtops;
2021 		break;
2022 	case EVFILT_VNODE:
2023 		kn->kn_fop = &tmpfsvnode_filtops;
2024 		break;
2025 	default:
2026 		return (EOPNOTSUPP);
2027 	}
2028 
2029 	kn->kn_hook = (caddr_t)vp;
2030 
2031 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2032 
2033 	return(0);
2034 }
2035 
2036 static void
filt_tmpfsdetach(struct knote * kn)2037 filt_tmpfsdetach(struct knote *kn)
2038 {
2039 	struct vnode *vp = (void *)kn->kn_hook;
2040 
2041 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2042 }
2043 
2044 static int
filt_tmpfsread(struct knote * kn,long hint)2045 filt_tmpfsread(struct knote *kn, long hint)
2046 {
2047 	struct vnode *vp = (void *)kn->kn_hook;
2048 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
2049 	off_t off;
2050 
2051 	if (hint == NOTE_REVOKE) {
2052 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2053 		return(1);
2054 	}
2055 
2056 	/*
2057 	 * Interlock against MP races when performing this function.
2058 	 */
2059 	TMPFS_NODE_LOCK_SH(node);
2060 	off = node->tn_size - kn->kn_fp->f_offset;
2061 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2062 	if (kn->kn_sfflags & NOTE_OLDAPI) {
2063 		TMPFS_NODE_UNLOCK(node);
2064 		return(1);
2065 	}
2066 	if (kn->kn_data == 0) {
2067 		kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2068 	}
2069 	TMPFS_NODE_UNLOCK(node);
2070 	return (kn->kn_data != 0);
2071 }
2072 
2073 static int
filt_tmpfswrite(struct knote * kn,long hint)2074 filt_tmpfswrite(struct knote *kn, long hint)
2075 {
2076 	if (hint == NOTE_REVOKE)
2077 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2078 	kn->kn_data = 0;
2079 	return (1);
2080 }
2081 
2082 static int
filt_tmpfsvnode(struct knote * kn,long hint)2083 filt_tmpfsvnode(struct knote *kn, long hint)
2084 {
2085 	if (kn->kn_sfflags & hint)
2086 		kn->kn_fflags |= hint;
2087 	if (hint == NOTE_REVOKE) {
2088 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2089 		return (1);
2090 	}
2091 	return (kn->kn_fflags != 0);
2092 }
2093 
2094 /*
2095  * Helper to move VM pages between objects
2096  *
2097  * NOTE: The vm_page_rename() dirties the page, so we can clear the
2098  *	 PG_NEED_COMMIT flag.  If the pages are being moved into tn_aobj,
2099  *	 the pageout daemon will be able to page them out.
2100  */
2101 static int
tmpfs_move_pages_callback(vm_page_t p,void * data)2102 tmpfs_move_pages_callback(vm_page_t p, void *data)
2103 {
2104 	struct rb_vm_page_scan_info *info = data;
2105 	vm_pindex_t pindex;
2106 
2107 	/*
2108 	 * Take control of the page
2109 	 */
2110 	pindex = p->pindex;
2111 	if (vm_page_busy_try(p, TRUE)) {
2112 		vm_page_sleep_busy(p, TRUE, "tpgmov");
2113 		info->error = -1;
2114 		return -1;
2115 	}
2116 	if (p->object != info->object || p->pindex != pindex) {
2117 		vm_page_wakeup(p);
2118 		info->error = -1;
2119 		return -1;
2120 	}
2121 
2122 	/*
2123 	 * Make sure the page is not mapped.  These flags might also still be
2124 	 * set heuristically even if we know the page is not mapped and must
2125 	 * be properly cleaned up.
2126 	 */
2127 	if (__predict_false((p->flags & (PG_MAPPED|PG_WRITEABLE)) != 0))
2128 		vm_page_protect(p, VM_PROT_NONE);
2129 
2130 	/*
2131 	 * Free or rename the page as appropriate
2132 	 */
2133 	if ((info->pagerflags & TMPFS_MOVF_FROMBACKING) &&
2134 	    (p->flags & PG_SWAPPED) &&
2135 	    (p->flags & PG_NEED_COMMIT) == 0 &&
2136 	    p->dirty == 0) {
2137 		/*
2138 		 * If the page in the backing aobj was paged out to swap
2139 		 * it will be clean and it is better to free it rather
2140 		 * than re-dirty it.  We will assume that the page was
2141 		 * paged out to swap for a reason!
2142 		 *
2143 		 * This helps avoid unnecessary swap thrashing on the page.
2144 		 */
2145 		vm_page_free(p);
2146 	} else if ((info->pagerflags & TMPFS_MOVF_FROMBACKING) == 0 &&
2147 		   (p->flags & PG_NEED_COMMIT) == 0 &&
2148 		   p->dirty == 0) {
2149 		/*
2150 		 * If the page associated with the vnode was cleaned via
2151 		 * a tmpfs_strategy() call, it exists as a swap block in
2152 		 * aobj and it is again better to free it rather than
2153 		 * re-dirty it.  We will assume that the page was
2154 		 * paged out to swap for a reason!
2155 		 *
2156 		 * This helps avoid unnecessary swap thrashing on the page.
2157 		 */
2158 		vm_page_free(p);
2159 	} else {
2160 		/*
2161 		 * Rename the page, which will also ensure that it is flagged
2162 		 * as dirty and check whether a swap block association exists
2163 		 * in the target object or not, setting appropriate flags if
2164 		 * it does.
2165 		 */
2166 		vm_page_rename(p, info->dest_object, pindex);
2167 		vm_page_clear_commit(p);
2168 		if (info->pagerflags & TMPFS_MOVF_DEACTIVATE)
2169 			vm_page_deactivate(p);
2170 		vm_page_wakeup(p);
2171 		/* page automaticaly made dirty */
2172 	}
2173 
2174 	return 0;
2175 }
2176 
2177 static
2178 void
tmpfs_move_pages(vm_object_t src,vm_object_t dst,int movflags)2179 tmpfs_move_pages(vm_object_t src, vm_object_t dst, int movflags)
2180 {
2181 	struct rb_vm_page_scan_info info;
2182 
2183 	vm_object_hold(src);
2184 	vm_object_hold(dst);
2185 	info.object = src;
2186 	info.dest_object = dst;
2187 	info.pagerflags = movflags;
2188 	do {
2189 		if (src->paging_in_progress)
2190 			vm_object_pip_wait(src, "objtfs");
2191 		info.error = 1;
2192 		vm_page_rb_tree_RB_SCAN(&src->rb_memq, NULL,
2193 					tmpfs_move_pages_callback, &info);
2194 	} while (info.error < 0 || !RB_EMPTY(&src->rb_memq) ||
2195 		 src->paging_in_progress);
2196 	vm_object_drop(dst);
2197 	vm_object_drop(src);
2198 }
2199 
2200 /* --------------------------------------------------------------------- */
2201 
2202 /*
2203  * vnode operations vector used for files stored in a tmpfs file system.
2204  */
2205 struct vop_ops tmpfs_vnode_vops = {
2206 	.vop_default =			vop_defaultop,
2207 	.vop_getpages = 		vop_stdgetpages,
2208 	.vop_putpages = 		vop_stdputpages,
2209 	.vop_ncreate =			tmpfs_ncreate,
2210 	.vop_nresolve =			tmpfs_nresolve,
2211 	.vop_nlookupdotdot =		tmpfs_nlookupdotdot,
2212 	.vop_nmknod =			tmpfs_nmknod,
2213 	.vop_open =			tmpfs_open,
2214 	.vop_close =			tmpfs_close,
2215 	.vop_access =			tmpfs_access,
2216 	.vop_getattr =			tmpfs_getattr,
2217 	.vop_getattr_lite =		tmpfs_getattr_lite,
2218 	.vop_setattr =			tmpfs_setattr,
2219 	.vop_read =			tmpfs_read,
2220 	.vop_write =			tmpfs_write,
2221 	.vop_fsync =			tmpfs_fsync,
2222 	.vop_mountctl =			tmpfs_mountctl,
2223 	.vop_nremove =			tmpfs_nremove,
2224 	.vop_nlink =			tmpfs_nlink,
2225 	.vop_nrename =			tmpfs_nrename,
2226 	.vop_nmkdir =			tmpfs_nmkdir,
2227 	.vop_nrmdir =			tmpfs_nrmdir,
2228 	.vop_nsymlink =			tmpfs_nsymlink,
2229 	.vop_readdir =			tmpfs_readdir,
2230 	.vop_readlink =			tmpfs_readlink,
2231 	.vop_inactive =			tmpfs_inactive,
2232 	.vop_reclaim =			tmpfs_reclaim,
2233 	.vop_print =			tmpfs_print,
2234 	.vop_pathconf =			tmpfs_pathconf,
2235 	.vop_bmap =			tmpfs_bmap,
2236 	.vop_strategy =			tmpfs_strategy,
2237 	.vop_advlock =			tmpfs_advlock,
2238 	.vop_kqfilter =			tmpfs_kqfilter
2239 };
2240