xref: /netbsd-src/sys/miscfs/genfs/genfs_vnops.c (revision 27527e67bbdf8d9ec84fd58803048ed6d181ece2)
1 /*	$NetBSD: genfs_vnops.c,v 1.121 2006/01/16 19:45:00 reinoud Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  */
32 
33 #include <sys/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: genfs_vnops.c,v 1.121 2006/01/16 19:45:00 reinoud Exp $");
35 
36 #if defined(_KERNEL_OPT)
37 #include "opt_nfsserver.h"
38 #endif
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/proc.h>
43 #include <sys/kernel.h>
44 #include <sys/mount.h>
45 #include <sys/namei.h>
46 #include <sys/vnode.h>
47 #include <sys/fcntl.h>
48 #include <sys/malloc.h>
49 #include <sys/poll.h>
50 #include <sys/mman.h>
51 #include <sys/file.h>
52 
53 #include <miscfs/genfs/genfs.h>
54 #include <miscfs/genfs/genfs_node.h>
55 #include <miscfs/specfs/specdev.h>
56 
57 #include <uvm/uvm.h>
58 #include <uvm/uvm_pager.h>
59 
60 #ifdef NFSSERVER
61 #include <nfs/rpcv2.h>
62 #include <nfs/nfsproto.h>
63 #include <nfs/nfs.h>
64 #include <nfs/nqnfs.h>
65 #include <nfs/nfs_var.h>
66 #endif
67 
68 static inline void genfs_rel_pages(struct vm_page **, int);
69 static void filt_genfsdetach(struct knote *);
70 static int filt_genfsread(struct knote *, long);
71 static int filt_genfsvnode(struct knote *, long);
72 
73 #define MAX_READ_PAGES	16 	/* XXXUBC 16 */
74 
75 int
76 genfs_poll(void *v)
77 {
78 	struct vop_poll_args /* {
79 		struct vnode *a_vp;
80 		int a_events;
81 		struct lwp *a_l;
82 	} */ *ap = v;
83 
84 	return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
85 }
86 
87 int
88 genfs_seek(void *v)
89 {
90 	struct vop_seek_args /* {
91 		struct vnode *a_vp;
92 		off_t a_oldoff;
93 		off_t a_newoff;
94 		struct ucred *a_ucred;
95 	} */ *ap = v;
96 
97 	if (ap->a_newoff < 0)
98 		return (EINVAL);
99 
100 	return (0);
101 }
102 
103 int
104 genfs_abortop(void *v)
105 {
106 	struct vop_abortop_args /* {
107 		struct vnode *a_dvp;
108 		struct componentname *a_cnp;
109 	} */ *ap = v;
110 
111 	if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
112 		PNBUF_PUT(ap->a_cnp->cn_pnbuf);
113 	return (0);
114 }
115 
116 int
117 genfs_fcntl(void *v)
118 {
119 	struct vop_fcntl_args /* {
120 		struct vnode *a_vp;
121 		u_int a_command;
122 		caddr_t a_data;
123 		int a_fflag;
124 		struct ucred *a_cred;
125 		struct lwp *a_l;
126 	} */ *ap = v;
127 
128 	if (ap->a_command == F_SETFL)
129 		return (0);
130 	else
131 		return (EOPNOTSUPP);
132 }
133 
134 /*ARGSUSED*/
135 int
136 genfs_badop(void *v)
137 {
138 
139 	panic("genfs: bad op");
140 }
141 
142 /*ARGSUSED*/
143 int
144 genfs_nullop(void *v)
145 {
146 
147 	return (0);
148 }
149 
150 /*ARGSUSED*/
151 int
152 genfs_einval(void *v)
153 {
154 
155 	return (EINVAL);
156 }
157 
158 /*
159  * Called when an fs doesn't support a particular vop.
160  * This takes care to vrele, vput, or vunlock passed in vnodes.
161  */
162 int
163 genfs_eopnotsupp(void *v)
164 {
165 	struct vop_generic_args /*
166 		struct vnodeop_desc *a_desc;
167 		/ * other random data follows, presumably * /
168 	} */ *ap = v;
169 	struct vnodeop_desc *desc = ap->a_desc;
170 	struct vnode *vp, *vp_last = NULL;
171 	int flags, i, j, offset;
172 
173 	flags = desc->vdesc_flags;
174 	for (i = 0; i < VDESC_MAX_VPS; flags >>=1, i++) {
175 		if ((offset = desc->vdesc_vp_offsets[i]) == VDESC_NO_OFFSET)
176 			break;	/* stop at end of list */
177 		if ((j = flags & VDESC_VP0_WILLPUT)) {
178 			vp = *VOPARG_OFFSETTO(struct vnode **, offset, ap);
179 
180 			/* Skip if NULL */
181 			if (!vp)
182 				continue;
183 
184 			switch (j) {
185 			case VDESC_VP0_WILLPUT:
186 				/* Check for dvp == vp cases */
187 				if (vp == vp_last)
188 					vrele(vp);
189 				else {
190 					vput(vp);
191 					vp_last = vp;
192 				}
193 				break;
194 			case VDESC_VP0_WILLUNLOCK:
195 				VOP_UNLOCK(vp, 0);
196 				break;
197 			case VDESC_VP0_WILLRELE:
198 				vrele(vp);
199 				break;
200 			}
201 		}
202 	}
203 
204 	return (EOPNOTSUPP);
205 }
206 
207 /*ARGSUSED*/
208 int
209 genfs_ebadf(void *v)
210 {
211 
212 	return (EBADF);
213 }
214 
215 /* ARGSUSED */
216 int
217 genfs_enoioctl(void *v)
218 {
219 
220 	return (EPASSTHROUGH);
221 }
222 
223 
224 /*
225  * Eliminate all activity associated with the requested vnode
226  * and with all vnodes aliased to the requested vnode.
227  */
228 int
229 genfs_revoke(void *v)
230 {
231 	struct vop_revoke_args /* {
232 		struct vnode *a_vp;
233 		int a_flags;
234 	} */ *ap = v;
235 	struct vnode *vp, *vq;
236 	struct lwp *l = curlwp;		/* XXX */
237 
238 #ifdef DIAGNOSTIC
239 	if ((ap->a_flags & REVOKEALL) == 0)
240 		panic("genfs_revoke: not revokeall");
241 #endif
242 
243 	vp = ap->a_vp;
244 	simple_lock(&vp->v_interlock);
245 
246 	if (vp->v_flag & VALIASED) {
247 		/*
248 		 * If a vgone (or vclean) is already in progress,
249 		 * wait until it is done and return.
250 		 */
251 		if (vp->v_flag & VXLOCK) {
252 			vp->v_flag |= VXWANT;
253 			ltsleep(vp, PINOD|PNORELOCK, "vop_revokeall", 0,
254 				&vp->v_interlock);
255 			return (0);
256 		}
257 		/*
258 		 * Ensure that vp will not be vgone'd while we
259 		 * are eliminating its aliases.
260 		 */
261 		vp->v_flag |= VXLOCK;
262 		simple_unlock(&vp->v_interlock);
263 		while (vp->v_flag & VALIASED) {
264 			simple_lock(&spechash_slock);
265 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
266 				if (vq->v_rdev != vp->v_rdev ||
267 				    vq->v_type != vp->v_type || vp == vq)
268 					continue;
269 				simple_unlock(&spechash_slock);
270 				vgone(vq);
271 				break;
272 			}
273 			if (vq == NULLVP)
274 				simple_unlock(&spechash_slock);
275 		}
276 		/*
277 		 * Remove the lock so that vgone below will
278 		 * really eliminate the vnode after which time
279 		 * vgone will awaken any sleepers.
280 		 */
281 		simple_lock(&vp->v_interlock);
282 		vp->v_flag &= ~VXLOCK;
283 	}
284 	vgonel(vp, l);
285 	return (0);
286 }
287 
288 /*
289  * Lock the node.
290  */
291 int
292 genfs_lock(void *v)
293 {
294 	struct vop_lock_args /* {
295 		struct vnode *a_vp;
296 		int a_flags;
297 	} */ *ap = v;
298 	struct vnode *vp = ap->a_vp;
299 
300 	return (lockmgr(vp->v_vnlock, ap->a_flags, &vp->v_interlock));
301 }
302 
303 /*
304  * Unlock the node.
305  */
306 int
307 genfs_unlock(void *v)
308 {
309 	struct vop_unlock_args /* {
310 		struct vnode *a_vp;
311 		int a_flags;
312 	} */ *ap = v;
313 	struct vnode *vp = ap->a_vp;
314 
315 	return (lockmgr(vp->v_vnlock, ap->a_flags | LK_RELEASE,
316 	    &vp->v_interlock));
317 }
318 
319 /*
320  * Return whether or not the node is locked.
321  */
322 int
323 genfs_islocked(void *v)
324 {
325 	struct vop_islocked_args /* {
326 		struct vnode *a_vp;
327 	} */ *ap = v;
328 	struct vnode *vp = ap->a_vp;
329 
330 	return (lockstatus(vp->v_vnlock));
331 }
332 
333 /*
334  * Stubs to use when there is no locking to be done on the underlying object.
335  */
336 int
337 genfs_nolock(void *v)
338 {
339 	struct vop_lock_args /* {
340 		struct vnode *a_vp;
341 		int a_flags;
342 		struct lwp *a_l;
343 	} */ *ap = v;
344 
345 	/*
346 	 * Since we are not using the lock manager, we must clear
347 	 * the interlock here.
348 	 */
349 	if (ap->a_flags & LK_INTERLOCK)
350 		simple_unlock(&ap->a_vp->v_interlock);
351 	return (0);
352 }
353 
354 int
355 genfs_nounlock(void *v)
356 {
357 
358 	return (0);
359 }
360 
361 int
362 genfs_noislocked(void *v)
363 {
364 
365 	return (0);
366 }
367 
368 /*
369  * Local lease check for NFS servers.  Just set up args and let
370  * nqsrv_getlease() do the rest.  If NFSSERVER is not in the kernel,
371  * this is a null operation.
372  */
373 int
374 genfs_lease_check(void *v)
375 {
376 #ifdef NFSSERVER
377 	struct vop_lease_args /* {
378 		struct vnode *a_vp;
379 		struct lwp *a_l;
380 		struct ucred *a_cred;
381 		int a_flag;
382 	} */ *ap = v;
383 	u_int32_t duration = 0;
384 	int cache;
385 	u_quad_t frev;
386 
387 	(void) nqsrv_getlease(ap->a_vp, &duration, ND_CHECK | ap->a_flag,
388 	    NQLOCALSLP, ap->a_l, (struct mbuf *)0, &cache, &frev, ap->a_cred);
389 	return (0);
390 #else
391 	return (0);
392 #endif /* NFSSERVER */
393 }
394 
395 int
396 genfs_mmap(void *v)
397 {
398 
399 	return (0);
400 }
401 
402 static inline void
403 genfs_rel_pages(struct vm_page **pgs, int npages)
404 {
405 	int i;
406 
407 	for (i = 0; i < npages; i++) {
408 		struct vm_page *pg = pgs[i];
409 
410 		if (pg == NULL)
411 			continue;
412 		if (pg->flags & PG_FAKE) {
413 			pg->flags |= PG_RELEASED;
414 		}
415 	}
416 	uvm_lock_pageq();
417 	uvm_page_unbusy(pgs, npages);
418 	uvm_unlock_pageq();
419 }
420 
421 /*
422  * generic VM getpages routine.
423  * Return PG_BUSY pages for the given range,
424  * reading from backing store if necessary.
425  */
426 
427 int
428 genfs_getpages(void *v)
429 {
430 	struct vop_getpages_args /* {
431 		struct vnode *a_vp;
432 		voff_t a_offset;
433 		struct vm_page **a_m;
434 		int *a_count;
435 		int a_centeridx;
436 		vm_prot_t a_access_type;
437 		int a_advice;
438 		int a_flags;
439 	} */ *ap = v;
440 
441 	off_t newsize, diskeof, memeof;
442 	off_t offset, origoffset, startoffset, endoffset, raoffset;
443 	daddr_t lbn, blkno;
444 	int i, error, npages, orignpages, npgs, run, ridx, pidx, pcount;
445 	int fs_bshift, fs_bsize, dev_bshift;
446 	int flags = ap->a_flags;
447 	size_t bytes, iobytes, tailbytes, totalbytes, skipbytes;
448 	vaddr_t kva;
449 	struct buf *bp, *mbp;
450 	struct vnode *vp = ap->a_vp;
451 	struct vnode *devvp;
452 	struct genfs_node *gp = VTOG(vp);
453 	struct uvm_object *uobj = &vp->v_uobj;
454 	struct vm_page *pg, **pgs, *pgs_onstack[MAX_READ_PAGES];
455 	int pgs_size;
456 	struct ucred *cred = curproc->p_ucred;		/* XXXUBC curlwp */
457 	boolean_t async = (flags & PGO_SYNCIO) == 0;
458 	boolean_t write = (ap->a_access_type & VM_PROT_WRITE) != 0;
459 	boolean_t sawhole = FALSE;
460 	boolean_t overwrite = (flags & PGO_OVERWRITE) != 0;
461 	boolean_t blockalloc = write && (flags & PGO_NOBLOCKALLOC) == 0;
462 	UVMHIST_FUNC("genfs_getpages"); UVMHIST_CALLED(ubchist);
463 
464 	UVMHIST_LOG(ubchist, "vp %p off 0x%x/%x count %d",
465 	    vp, ap->a_offset >> 32, ap->a_offset, *ap->a_count);
466 
467 	KASSERT(vp->v_type == VREG || vp->v_type == VDIR ||
468 	    vp->v_type == VLNK || vp->v_type == VBLK);
469 
470 	/* XXXUBC temp limit */
471 	if (*ap->a_count > MAX_READ_PAGES) {
472 		panic("genfs_getpages: too many pages");
473 	}
474 
475 	error = 0;
476 	origoffset = ap->a_offset;
477 	orignpages = *ap->a_count;
478 	GOP_SIZE(vp, vp->v_size, &diskeof, GOP_SIZE_READ);
479 	if (flags & PGO_PASTEOF) {
480 		newsize = MAX(vp->v_size,
481 		    origoffset + (orignpages << PAGE_SHIFT));
482 		GOP_SIZE(vp, newsize, &memeof, GOP_SIZE_READ|GOP_SIZE_MEM);
483 	} else {
484 		GOP_SIZE(vp, vp->v_size, &memeof, GOP_SIZE_READ|GOP_SIZE_MEM);
485 	}
486 	KASSERT(ap->a_centeridx >= 0 || ap->a_centeridx <= orignpages);
487 	KASSERT((origoffset & (PAGE_SIZE - 1)) == 0 && origoffset >= 0);
488 	KASSERT(orignpages > 0);
489 
490 	/*
491 	 * Bounds-check the request.
492 	 */
493 
494 	if (origoffset + (ap->a_centeridx << PAGE_SHIFT) >= memeof) {
495 		if ((flags & PGO_LOCKED) == 0) {
496 			simple_unlock(&uobj->vmobjlock);
497 		}
498 		UVMHIST_LOG(ubchist, "off 0x%x count %d goes past EOF 0x%x",
499 		    origoffset, *ap->a_count, memeof,0);
500 		return (EINVAL);
501 	}
502 
503 	/* uobj is locked */
504 
505 	if ((flags & PGO_NOTIMESTAMP) == 0 &&
506 	    (vp->v_type != VBLK ||
507 	    (vp->v_mount->mnt_flag & MNT_NODEVMTIME) == 0)) {
508 		int updflags = 0;
509 
510 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0) {
511 			updflags = GOP_UPDATE_ACCESSED;
512 		}
513 		if (write) {
514 			updflags |= GOP_UPDATE_MODIFIED;
515 		}
516 		if (updflags != 0) {
517 			GOP_MARKUPDATE(vp, updflags);
518 		}
519 	}
520 
521 	if (write) {
522 		gp->g_dirtygen++;
523 		if ((vp->v_flag & VONWORKLST) == 0) {
524 			vn_syncer_add_to_worklist(vp, filedelay);
525 		}
526 		if ((vp->v_flag & (VWRITEMAP|VWRITEMAPDIRTY)) == VWRITEMAP) {
527 			vp->v_flag |= VWRITEMAPDIRTY;
528 		}
529 	}
530 
531 	/*
532 	 * For PGO_LOCKED requests, just return whatever's in memory.
533 	 */
534 
535 	if (flags & PGO_LOCKED) {
536 		uvn_findpages(uobj, origoffset, ap->a_count, ap->a_m,
537 		    UFP_NOWAIT|UFP_NOALLOC| (write ? UFP_NORDONLY : 0));
538 
539 		return (ap->a_m[ap->a_centeridx] == NULL ? EBUSY : 0);
540 	}
541 
542 	/*
543 	 * find the requested pages and make some simple checks.
544 	 * leave space in the page array for a whole block.
545 	 */
546 
547 	if (vp->v_type != VBLK) {
548 		fs_bshift = vp->v_mount->mnt_fs_bshift;
549 		dev_bshift = vp->v_mount->mnt_dev_bshift;
550 	} else {
551 		fs_bshift = DEV_BSHIFT;
552 		dev_bshift = DEV_BSHIFT;
553 	}
554 	fs_bsize = 1 << fs_bshift;
555 
556 	orignpages = MIN(orignpages,
557 	    round_page(memeof - origoffset) >> PAGE_SHIFT);
558 	npages = orignpages;
559 	startoffset = origoffset & ~(fs_bsize - 1);
560 	endoffset = round_page((origoffset + (npages << PAGE_SHIFT) +
561 	    fs_bsize - 1) & ~(fs_bsize - 1));
562 	endoffset = MIN(endoffset, round_page(memeof));
563 	ridx = (origoffset - startoffset) >> PAGE_SHIFT;
564 
565 	pgs_size = sizeof(struct vm_page *) *
566 	    ((endoffset - startoffset) >> PAGE_SHIFT);
567 	if (pgs_size > sizeof(pgs_onstack)) {
568 		pgs = malloc(pgs_size, M_DEVBUF, M_NOWAIT | M_ZERO);
569 		if (pgs == NULL) {
570 			simple_unlock(&uobj->vmobjlock);
571 			return (ENOMEM);
572 		}
573 	} else {
574 		pgs = pgs_onstack;
575 		memset(pgs, 0, pgs_size);
576 	}
577 	UVMHIST_LOG(ubchist, "ridx %d npages %d startoff %ld endoff %ld",
578 	    ridx, npages, startoffset, endoffset);
579 	if (uvn_findpages(uobj, origoffset, &npages, &pgs[ridx],
580 	    async ? UFP_NOWAIT : UFP_ALL) != orignpages) {
581 		KASSERT(async != 0);
582 		genfs_rel_pages(&pgs[ridx], orignpages);
583 		simple_unlock(&uobj->vmobjlock);
584 		if (pgs != pgs_onstack)
585 			free(pgs, M_DEVBUF);
586 		return (EBUSY);
587 	}
588 
589 	/*
590 	 * if the pages are already resident, just return them.
591 	 */
592 
593 	for (i = 0; i < npages; i++) {
594 		struct vm_page *pg1 = pgs[ridx + i];
595 
596 		if ((pg1->flags & PG_FAKE) ||
597 		    (blockalloc && (pg1->flags & PG_RDONLY))) {
598 			break;
599 		}
600 	}
601 	if (i == npages) {
602 		UVMHIST_LOG(ubchist, "returning cached pages", 0,0,0,0);
603 		raoffset = origoffset + (orignpages << PAGE_SHIFT);
604 		npages += ridx;
605 		goto out;
606 	}
607 
608 	/*
609 	 * if PGO_OVERWRITE is set, don't bother reading the pages.
610 	 */
611 
612 	if (flags & PGO_OVERWRITE) {
613 		UVMHIST_LOG(ubchist, "PGO_OVERWRITE",0,0,0,0);
614 
615 		for (i = 0; i < npages; i++) {
616 			struct vm_page *pg1 = pgs[ridx + i];
617 
618 			pg1->flags &= ~(PG_RDONLY|PG_CLEAN);
619 		}
620 		npages += ridx;
621 		goto out;
622 	}
623 
624 	/*
625 	 * the page wasn't resident and we're not overwriting,
626 	 * so we're going to have to do some i/o.
627 	 * find any additional pages needed to cover the expanded range.
628 	 */
629 
630 	npages = (endoffset - startoffset) >> PAGE_SHIFT;
631 	if (startoffset != origoffset || npages != orignpages) {
632 
633 		/*
634 		 * we need to avoid deadlocks caused by locking
635 		 * additional pages at lower offsets than pages we
636 		 * already have locked.  unlock them all and start over.
637 		 */
638 
639 		genfs_rel_pages(&pgs[ridx], orignpages);
640 		memset(pgs, 0, pgs_size);
641 
642 		UVMHIST_LOG(ubchist, "reset npages start 0x%x end 0x%x",
643 		    startoffset, endoffset, 0,0);
644 		npgs = npages;
645 		if (uvn_findpages(uobj, startoffset, &npgs, pgs,
646 		    async ? UFP_NOWAIT : UFP_ALL) != npages) {
647 			KASSERT(async != 0);
648 			genfs_rel_pages(pgs, npages);
649 			simple_unlock(&uobj->vmobjlock);
650 			if (pgs != pgs_onstack)
651 				free(pgs, M_DEVBUF);
652 			return (EBUSY);
653 		}
654 	}
655 	simple_unlock(&uobj->vmobjlock);
656 
657 	/*
658 	 * read the desired page(s).
659 	 */
660 
661 	totalbytes = npages << PAGE_SHIFT;
662 	bytes = MIN(totalbytes, MAX(diskeof - startoffset, 0));
663 	tailbytes = totalbytes - bytes;
664 	skipbytes = 0;
665 
666 	kva = uvm_pagermapin(pgs, npages,
667 	    UVMPAGER_MAPIN_READ | UVMPAGER_MAPIN_WAITOK);
668 
669 	mbp = getiobuf();
670 	mbp->b_bufsize = totalbytes;
671 	mbp->b_data = (void *)kva;
672 	mbp->b_resid = mbp->b_bcount = bytes;
673 	mbp->b_flags = B_BUSY|B_READ| (async ? B_CALL|B_ASYNC : 0);
674 	mbp->b_iodone = (async ? uvm_aio_biodone : 0);
675 	mbp->b_vp = vp;
676 	if (async)
677 		BIO_SETPRIO(mbp, BPRIO_TIMELIMITED);
678 	else
679 		BIO_SETPRIO(mbp, BPRIO_TIMECRITICAL);
680 
681 	/*
682 	 * if EOF is in the middle of the range, zero the part past EOF.
683 	 * if the page including EOF is not PG_FAKE, skip over it since
684 	 * in that case it has valid data that we need to preserve.
685 	 */
686 
687 	if (tailbytes > 0) {
688 		size_t tailstart = bytes;
689 
690 		if ((pgs[bytes >> PAGE_SHIFT]->flags & PG_FAKE) == 0) {
691 			tailstart = round_page(tailstart);
692 			tailbytes -= tailstart - bytes;
693 		}
694 		UVMHIST_LOG(ubchist, "tailbytes %p 0x%x 0x%x",
695 		    kva, tailstart, tailbytes,0);
696 		memset((void *)(kva + tailstart), 0, tailbytes);
697 	}
698 
699 	/*
700 	 * now loop over the pages, reading as needed.
701 	 */
702 
703 	if (blockalloc) {
704 		lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL);
705 	} else {
706 		lockmgr(&gp->g_glock, LK_SHARED, NULL);
707 	}
708 
709 	bp = NULL;
710 	for (offset = startoffset;
711 	    bytes > 0;
712 	    offset += iobytes, bytes -= iobytes) {
713 
714 		/*
715 		 * skip pages which don't need to be read.
716 		 */
717 
718 		pidx = (offset - startoffset) >> PAGE_SHIFT;
719 		while ((pgs[pidx]->flags & PG_FAKE) == 0) {
720 			size_t b;
721 
722 			KASSERT((offset & (PAGE_SIZE - 1)) == 0);
723 			if ((pgs[pidx]->flags & PG_RDONLY)) {
724 				sawhole = TRUE;
725 			}
726 			b = MIN(PAGE_SIZE, bytes);
727 			offset += b;
728 			bytes -= b;
729 			skipbytes += b;
730 			pidx++;
731 			UVMHIST_LOG(ubchist, "skipping, new offset 0x%x",
732 			    offset, 0,0,0);
733 			if (bytes == 0) {
734 				goto loopdone;
735 			}
736 		}
737 
738 		/*
739 		 * bmap the file to find out the blkno to read from and
740 		 * how much we can read in one i/o.  if bmap returns an error,
741 		 * skip the rest of the top-level i/o.
742 		 */
743 
744 		lbn = offset >> fs_bshift;
745 		error = VOP_BMAP(vp, lbn, &devvp, &blkno, &run);
746 		if (error) {
747 			UVMHIST_LOG(ubchist, "VOP_BMAP lbn 0x%x -> %d\n",
748 			    lbn, error,0,0);
749 			skipbytes += bytes;
750 			goto loopdone;
751 		}
752 
753 		/*
754 		 * see how many pages can be read with this i/o.
755 		 * reduce the i/o size if necessary to avoid
756 		 * overwriting pages with valid data.
757 		 */
758 
759 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
760 		    bytes);
761 		if (offset + iobytes > round_page(offset)) {
762 			pcount = 1;
763 			while (pidx + pcount < npages &&
764 			    pgs[pidx + pcount]->flags & PG_FAKE) {
765 				pcount++;
766 			}
767 			iobytes = MIN(iobytes, (pcount << PAGE_SHIFT) -
768 			    (offset - trunc_page(offset)));
769 		}
770 
771 		/*
772 		 * if this block isn't allocated, zero it instead of
773 		 * reading it.  unless we are going to allocate blocks,
774 		 * mark the pages we zeroed PG_RDONLY.
775 		 */
776 
777 		if (blkno < 0) {
778 			int holepages = (round_page(offset + iobytes) -
779 			    trunc_page(offset)) >> PAGE_SHIFT;
780 			UVMHIST_LOG(ubchist, "lbn 0x%x -> HOLE", lbn,0,0,0);
781 
782 			sawhole = TRUE;
783 			memset((char *)kva + (offset - startoffset), 0,
784 			    iobytes);
785 			skipbytes += iobytes;
786 
787 			for (i = 0; i < holepages; i++) {
788 				if (write) {
789 					pgs[pidx + i]->flags &= ~PG_CLEAN;
790 				}
791 				if (!blockalloc) {
792 					pgs[pidx + i]->flags |= PG_RDONLY;
793 				}
794 			}
795 			continue;
796 		}
797 
798 		/*
799 		 * allocate a sub-buf for this piece of the i/o
800 		 * (or just use mbp if there's only 1 piece),
801 		 * and start it going.
802 		 */
803 
804 		if (offset == startoffset && iobytes == bytes) {
805 			bp = mbp;
806 		} else {
807 			bp = getiobuf();
808 			nestiobuf_setup(mbp, bp, offset - startoffset, iobytes);
809 		}
810 		bp->b_lblkno = 0;
811 
812 		/* adjust physical blkno for partial blocks */
813 		bp->b_blkno = blkno + ((offset - ((off_t)lbn << fs_bshift)) >>
814 		    dev_bshift);
815 
816 		UVMHIST_LOG(ubchist,
817 		    "bp %p offset 0x%x bcount 0x%x blkno 0x%x",
818 		    bp, offset, iobytes, bp->b_blkno);
819 
820 		VOP_STRATEGY(devvp, bp);
821 	}
822 
823 loopdone:
824 	nestiobuf_done(mbp, skipbytes, error);
825 	if (async) {
826 		UVMHIST_LOG(ubchist, "returning 0 (async)",0,0,0,0);
827 		lockmgr(&gp->g_glock, LK_RELEASE, NULL);
828 		if (pgs != pgs_onstack)
829 			free(pgs, M_DEVBUF);
830 		return (0);
831 	}
832 	if (bp != NULL) {
833 		error = biowait(mbp);
834 	}
835 	putiobuf(mbp);
836 	uvm_pagermapout(kva, npages);
837 	raoffset = startoffset + totalbytes;
838 
839 	/*
840 	 * if this we encountered a hole then we have to do a little more work.
841 	 * for read faults, we marked the page PG_RDONLY so that future
842 	 * write accesses to the page will fault again.
843 	 * for write faults, we must make sure that the backing store for
844 	 * the page is completely allocated while the pages are locked.
845 	 */
846 
847 	if (!error && sawhole && blockalloc) {
848 		error = GOP_ALLOC(vp, startoffset, npages << PAGE_SHIFT, 0,
849 		    cred);
850 		UVMHIST_LOG(ubchist, "gop_alloc off 0x%x/0x%x -> %d",
851 		    startoffset, npages << PAGE_SHIFT, error,0);
852 		if (!error) {
853 			for (i = 0; i < npages; i++) {
854 				if (pgs[i] == NULL) {
855 					continue;
856 				}
857 				pgs[i]->flags &= ~(PG_CLEAN|PG_RDONLY);
858 				UVMHIST_LOG(ubchist, "mark dirty pg %p",
859 				    pgs[i],0,0,0);
860 			}
861 		}
862 	}
863 	lockmgr(&gp->g_glock, LK_RELEASE, NULL);
864 	simple_lock(&uobj->vmobjlock);
865 
866 	/*
867 	 * we're almost done!  release the pages...
868 	 * for errors, we free the pages.
869 	 * otherwise we activate them and mark them as valid and clean.
870 	 * also, unbusy pages that were not actually requested.
871 	 */
872 
873 	if (error) {
874 		for (i = 0; i < npages; i++) {
875 			if (pgs[i] == NULL) {
876 				continue;
877 			}
878 			UVMHIST_LOG(ubchist, "examining pg %p flags 0x%x",
879 			    pgs[i], pgs[i]->flags, 0,0);
880 			if (pgs[i]->flags & PG_FAKE) {
881 				pgs[i]->flags |= PG_RELEASED;
882 			}
883 		}
884 		uvm_lock_pageq();
885 		uvm_page_unbusy(pgs, npages);
886 		uvm_unlock_pageq();
887 		simple_unlock(&uobj->vmobjlock);
888 		UVMHIST_LOG(ubchist, "returning error %d", error,0,0,0);
889 		if (pgs != pgs_onstack)
890 			free(pgs, M_DEVBUF);
891 		return (error);
892 	}
893 
894 out:
895 	UVMHIST_LOG(ubchist, "succeeding, npages %d", npages,0,0,0);
896 	uvm_lock_pageq();
897 	for (i = 0; i < npages; i++) {
898 		pg = pgs[i];
899 		if (pg == NULL) {
900 			continue;
901 		}
902 		UVMHIST_LOG(ubchist, "examining pg %p flags 0x%x",
903 		    pg, pg->flags, 0,0);
904 		if (pg->flags & PG_FAKE && !overwrite) {
905 			pg->flags &= ~(PG_FAKE);
906 			pmap_clear_modify(pgs[i]);
907 		}
908 		KASSERT(!write || !blockalloc || (pg->flags & PG_RDONLY) == 0);
909 		if (i < ridx || i >= ridx + orignpages || async) {
910 			UVMHIST_LOG(ubchist, "unbusy pg %p offset 0x%x",
911 			    pg, pg->offset,0,0);
912 			if (pg->flags & PG_WANTED) {
913 				wakeup(pg);
914 			}
915 			if (pg->flags & PG_FAKE) {
916 				KASSERT(overwrite);
917 				uvm_pagezero(pg);
918 			}
919 			if (pg->flags & PG_RELEASED) {
920 				uvm_pagefree(pg);
921 				continue;
922 			}
923 			uvm_pageactivate(pg);
924 			pg->flags &= ~(PG_WANTED|PG_BUSY|PG_FAKE);
925 			UVM_PAGE_OWN(pg, NULL);
926 		}
927 	}
928 	uvm_unlock_pageq();
929 	simple_unlock(&uobj->vmobjlock);
930 	if (ap->a_m != NULL) {
931 		memcpy(ap->a_m, &pgs[ridx],
932 		    orignpages * sizeof(struct vm_page *));
933 	}
934 	if (pgs != pgs_onstack)
935 		free(pgs, M_DEVBUF);
936 	return (0);
937 }
938 
939 /*
940  * generic VM putpages routine.
941  * Write the given range of pages to backing store.
942  *
943  * => "offhi == 0" means flush all pages at or after "offlo".
944  * => object should be locked by caller.   we may _unlock_ the object
945  *	if (and only if) we need to clean a page (PGO_CLEANIT), or
946  *	if PGO_SYNCIO is set and there are pages busy.
947  *	we return with the object locked.
948  * => if PGO_CLEANIT or PGO_SYNCIO is set, we may block (due to I/O).
949  *	thus, a caller might want to unlock higher level resources
950  *	(e.g. vm_map) before calling flush.
951  * => if neither PGO_CLEANIT nor PGO_SYNCIO is set, then we will neither
952  *	unlock the object nor block.
953  * => if PGO_ALLPAGES is set, then all pages in the object will be processed.
954  * => NOTE: we rely on the fact that the object's memq is a TAILQ and
955  *	that new pages are inserted on the tail end of the list.   thus,
956  *	we can make a complete pass through the object in one go by starting
957  *	at the head and working towards the tail (new pages are put in
958  *	front of us).
959  * => NOTE: we are allowed to lock the page queues, so the caller
960  *	must not be holding the page queue lock.
961  *
962  * note on "cleaning" object and PG_BUSY pages:
963  *	this routine is holding the lock on the object.   the only time
964  *	that it can run into a PG_BUSY page that it does not own is if
965  *	some other process has started I/O on the page (e.g. either
966  *	a pagein, or a pageout).    if the PG_BUSY page is being paged
967  *	in, then it can not be dirty (!PG_CLEAN) because no one has
968  *	had a chance to modify it yet.    if the PG_BUSY page is being
969  *	paged out then it means that someone else has already started
970  *	cleaning the page for us (how nice!).    in this case, if we
971  *	have syncio specified, then after we make our pass through the
972  *	object we need to wait for the other PG_BUSY pages to clear
973  *	off (i.e. we need to do an iosync).   also note that once a
974  *	page is PG_BUSY it must stay in its object until it is un-busyed.
975  *
976  * note on page traversal:
977  *	we can traverse the pages in an object either by going down the
978  *	linked list in "uobj->memq", or we can go over the address range
979  *	by page doing hash table lookups for each address.    depending
980  *	on how many pages are in the object it may be cheaper to do one
981  *	or the other.   we set "by_list" to true if we are using memq.
982  *	if the cost of a hash lookup was equal to the cost of the list
983  *	traversal we could compare the number of pages in the start->stop
984  *	range to the total number of pages in the object.   however, it
985  *	seems that a hash table lookup is more expensive than the linked
986  *	list traversal, so we multiply the number of pages in the
987  *	range by an estimate of the relatively higher cost of the hash lookup.
988  */
989 
990 int
991 genfs_putpages(void *v)
992 {
993 	struct vop_putpages_args /* {
994 		struct vnode *a_vp;
995 		voff_t a_offlo;
996 		voff_t a_offhi;
997 		int a_flags;
998 	} */ *ap = v;
999 	struct vnode *vp = ap->a_vp;
1000 	struct uvm_object *uobj = &vp->v_uobj;
1001 	struct simplelock *slock = &uobj->vmobjlock;
1002 	off_t startoff = ap->a_offlo;
1003 	off_t endoff = ap->a_offhi;
1004 	off_t off;
1005 	int flags = ap->a_flags;
1006 	/* Even for strange MAXPHYS, the shift rounds down to a page */
1007 	const int maxpages = MAXPHYS >> PAGE_SHIFT;
1008 	int i, s, error, npages, nback;
1009 	int freeflag;
1010 	struct vm_page *pgs[maxpages], *pg, *nextpg, *tpg, curmp, endmp;
1011 	boolean_t wasclean, by_list, needs_clean, yld;
1012 	boolean_t async = (flags & PGO_SYNCIO) == 0;
1013 	boolean_t pagedaemon = curproc == uvm.pagedaemon_proc;
1014 	struct lwp *l = curlwp ? curlwp : &lwp0;
1015 	struct genfs_node *gp = VTOG(vp);
1016 	int dirtygen;
1017 	boolean_t modified = FALSE;
1018 	boolean_t cleanall;
1019 
1020 	UVMHIST_FUNC("genfs_putpages"); UVMHIST_CALLED(ubchist);
1021 
1022 	KASSERT(flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE));
1023 	KASSERT((startoff & PAGE_MASK) == 0 && (endoff & PAGE_MASK) == 0);
1024 	KASSERT(startoff < endoff || endoff == 0);
1025 
1026 	UVMHIST_LOG(ubchist, "vp %p pages %d off 0x%x len 0x%x",
1027 	    vp, uobj->uo_npages, startoff, endoff - startoff);
1028 
1029 	KASSERT((vp->v_flag & VONWORKLST) != 0 ||
1030 	    (vp->v_flag & VWRITEMAPDIRTY) == 0);
1031 	if (uobj->uo_npages == 0) {
1032 		s = splbio();
1033 		if (vp->v_flag & VONWORKLST) {
1034 			vp->v_flag &= ~VWRITEMAPDIRTY;
1035 			if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
1036 				vp->v_flag &= ~VONWORKLST;
1037 				LIST_REMOVE(vp, v_synclist);
1038 			}
1039 		}
1040 		splx(s);
1041 		simple_unlock(slock);
1042 		return (0);
1043 	}
1044 
1045 	/*
1046 	 * the vnode has pages, set up to process the request.
1047 	 */
1048 
1049 	error = 0;
1050 	s = splbio();
1051 	simple_lock(&global_v_numoutput_slock);
1052 	wasclean = (vp->v_numoutput == 0);
1053 	simple_unlock(&global_v_numoutput_slock);
1054 	splx(s);
1055 	off = startoff;
1056 	if (endoff == 0 || flags & PGO_ALLPAGES) {
1057 		endoff = trunc_page(LLONG_MAX);
1058 	}
1059 	by_list = (uobj->uo_npages <=
1060 	    ((endoff - startoff) >> PAGE_SHIFT) * UVM_PAGE_HASH_PENALTY);
1061 
1062 #if !defined(DEBUG)
1063 	/*
1064 	 * if this vnode is known not to have dirty pages,
1065 	 * don't bother to clean it out.
1066 	 */
1067 
1068 	if ((vp->v_flag & VONWORKLST) == 0) {
1069 		if ((flags & (PGO_FREE|PGO_DEACTIVATE)) == 0) {
1070 			goto skip_scan;
1071 		}
1072 		flags &= ~PGO_CLEANIT;
1073 	}
1074 #endif /* !defined(DEBUG) */
1075 
1076 	/*
1077 	 * start the loop.  when scanning by list, hold the last page
1078 	 * in the list before we start.  pages allocated after we start
1079 	 * will be added to the end of the list, so we can stop at the
1080 	 * current last page.
1081 	 */
1082 
1083 	cleanall = (flags & PGO_CLEANIT) != 0 && wasclean &&
1084 	    startoff == 0 && endoff == trunc_page(LLONG_MAX) &&
1085 	    (vp->v_flag & VONWORKLST) != 0;
1086 	dirtygen = gp->g_dirtygen;
1087 	freeflag = pagedaemon ? PG_PAGEOUT : PG_RELEASED;
1088 	if (by_list) {
1089 		curmp.uobject = uobj;
1090 		curmp.offset = (voff_t)-1;
1091 		curmp.flags = PG_BUSY;
1092 		endmp.uobject = uobj;
1093 		endmp.offset = (voff_t)-1;
1094 		endmp.flags = PG_BUSY;
1095 		pg = TAILQ_FIRST(&uobj->memq);
1096 		TAILQ_INSERT_TAIL(&uobj->memq, &endmp, listq);
1097 		PHOLD(l);
1098 	} else {
1099 		pg = uvm_pagelookup(uobj, off);
1100 	}
1101 	nextpg = NULL;
1102 	while (by_list || off < endoff) {
1103 
1104 		/*
1105 		 * if the current page is not interesting, move on to the next.
1106 		 */
1107 
1108 		KASSERT(pg == NULL || pg->uobject == uobj);
1109 		KASSERT(pg == NULL ||
1110 		    (pg->flags & (PG_RELEASED|PG_PAGEOUT)) == 0 ||
1111 		    (pg->flags & PG_BUSY) != 0);
1112 		if (by_list) {
1113 			if (pg == &endmp) {
1114 				break;
1115 			}
1116 			if (pg->offset < startoff || pg->offset >= endoff ||
1117 			    pg->flags & (PG_RELEASED|PG_PAGEOUT)) {
1118 				if (pg->flags & (PG_RELEASED|PG_PAGEOUT)) {
1119 					wasclean = FALSE;
1120 				}
1121 				pg = TAILQ_NEXT(pg, listq);
1122 				continue;
1123 			}
1124 			off = pg->offset;
1125 		} else if (pg == NULL || pg->flags & (PG_RELEASED|PG_PAGEOUT)) {
1126 			if (pg != NULL) {
1127 				wasclean = FALSE;
1128 			}
1129 			off += PAGE_SIZE;
1130 			if (off < endoff) {
1131 				pg = uvm_pagelookup(uobj, off);
1132 			}
1133 			continue;
1134 		}
1135 
1136 		/*
1137 		 * if the current page needs to be cleaned and it's busy,
1138 		 * wait for it to become unbusy.
1139 		 */
1140 
1141 		yld = (l->l_cpu->ci_schedstate.spc_flags &
1142 		    SPCF_SHOULDYIELD) && !pagedaemon;
1143 		if (pg->flags & PG_BUSY || yld) {
1144 			UVMHIST_LOG(ubchist, "busy %p", pg,0,0,0);
1145 			if (flags & PGO_BUSYFAIL && pg->flags & PG_BUSY) {
1146 				UVMHIST_LOG(ubchist, "busyfail %p", pg, 0,0,0);
1147 				error = EDEADLK;
1148 				break;
1149 			}
1150 			KASSERT(!pagedaemon);
1151 			if (by_list) {
1152 				TAILQ_INSERT_BEFORE(pg, &curmp, listq);
1153 				UVMHIST_LOG(ubchist, "curmp next %p",
1154 				    TAILQ_NEXT(&curmp, listq), 0,0,0);
1155 			}
1156 			if (yld) {
1157 				simple_unlock(slock);
1158 				preempt(1);
1159 				simple_lock(slock);
1160 			} else {
1161 				pg->flags |= PG_WANTED;
1162 				UVM_UNLOCK_AND_WAIT(pg, slock, 0, "genput", 0);
1163 				simple_lock(slock);
1164 			}
1165 			if (by_list) {
1166 				UVMHIST_LOG(ubchist, "after next %p",
1167 				    TAILQ_NEXT(&curmp, listq), 0,0,0);
1168 				pg = TAILQ_NEXT(&curmp, listq);
1169 				TAILQ_REMOVE(&uobj->memq, &curmp, listq);
1170 			} else {
1171 				pg = uvm_pagelookup(uobj, off);
1172 			}
1173 			continue;
1174 		}
1175 
1176 		/*
1177 		 * if we're freeing, remove all mappings of the page now.
1178 		 * if we're cleaning, check if the page is needs to be cleaned.
1179 		 */
1180 
1181 		if (flags & PGO_FREE) {
1182 			pmap_page_protect(pg, VM_PROT_NONE);
1183 		} else if (flags & PGO_CLEANIT) {
1184 
1185 			/*
1186 			 * if we still have some hope to pull this vnode off
1187 			 * from the syncer queue, write-protect the page.
1188 			 */
1189 
1190 			if (cleanall && wasclean &&
1191 			    gp->g_dirtygen == dirtygen) {
1192 
1193 				/*
1194 				 * uobj pages get wired only by uvm_fault
1195 				 * where uobj is locked.
1196 				 */
1197 
1198 				if (pg->wire_count == 0) {
1199 					pmap_page_protect(pg,
1200 					    VM_PROT_READ|VM_PROT_EXECUTE);
1201 				} else {
1202 					cleanall = FALSE;
1203 				}
1204 			}
1205 		}
1206 
1207 		if (flags & PGO_CLEANIT) {
1208 			needs_clean = pmap_clear_modify(pg) ||
1209 			    (pg->flags & PG_CLEAN) == 0;
1210 			pg->flags |= PG_CLEAN;
1211 		} else {
1212 			needs_clean = FALSE;
1213 		}
1214 
1215 		/*
1216 		 * if we're cleaning, build a cluster.
1217 		 * the cluster will consist of pages which are currently dirty,
1218 		 * but they will be returned to us marked clean.
1219 		 * if not cleaning, just operate on the one page.
1220 		 */
1221 
1222 		if (needs_clean) {
1223 			KDASSERT((vp->v_flag & VONWORKLST));
1224 			wasclean = FALSE;
1225 			memset(pgs, 0, sizeof(pgs));
1226 			pg->flags |= PG_BUSY;
1227 			UVM_PAGE_OWN(pg, "genfs_putpages");
1228 
1229 			/*
1230 			 * first look backward.
1231 			 */
1232 
1233 			npages = MIN(maxpages >> 1, off >> PAGE_SHIFT);
1234 			nback = npages;
1235 			uvn_findpages(uobj, off - PAGE_SIZE, &nback, &pgs[0],
1236 			    UFP_NOWAIT|UFP_NOALLOC|UFP_DIRTYONLY|UFP_BACKWARD);
1237 			if (nback) {
1238 				memmove(&pgs[0], &pgs[npages - nback],
1239 				    nback * sizeof(pgs[0]));
1240 				if (npages - nback < nback)
1241 					memset(&pgs[nback], 0,
1242 					    (npages - nback) * sizeof(pgs[0]));
1243 				else
1244 					memset(&pgs[npages - nback], 0,
1245 					    nback * sizeof(pgs[0]));
1246 			}
1247 
1248 			/*
1249 			 * then plug in our page of interest.
1250 			 */
1251 
1252 			pgs[nback] = pg;
1253 
1254 			/*
1255 			 * then look forward to fill in the remaining space in
1256 			 * the array of pages.
1257 			 */
1258 
1259 			npages = maxpages - nback - 1;
1260 			uvn_findpages(uobj, off + PAGE_SIZE, &npages,
1261 			    &pgs[nback + 1],
1262 			    UFP_NOWAIT|UFP_NOALLOC|UFP_DIRTYONLY);
1263 			npages += nback + 1;
1264 		} else {
1265 			pgs[0] = pg;
1266 			npages = 1;
1267 			nback = 0;
1268 		}
1269 
1270 		/*
1271 		 * apply FREE or DEACTIVATE options if requested.
1272 		 */
1273 
1274 		if (flags & (PGO_DEACTIVATE|PGO_FREE)) {
1275 			uvm_lock_pageq();
1276 		}
1277 		for (i = 0; i < npages; i++) {
1278 			tpg = pgs[i];
1279 			KASSERT(tpg->uobject == uobj);
1280 			if (by_list && tpg == TAILQ_NEXT(pg, listq))
1281 				pg = tpg;
1282 			if (tpg->offset < startoff || tpg->offset >= endoff)
1283 				continue;
1284 			if (flags & PGO_DEACTIVATE &&
1285 			    (tpg->pqflags & PQ_INACTIVE) == 0 &&
1286 			    tpg->wire_count == 0) {
1287 				(void) pmap_clear_reference(tpg);
1288 				uvm_pagedeactivate(tpg);
1289 			} else if (flags & PGO_FREE) {
1290 				pmap_page_protect(tpg, VM_PROT_NONE);
1291 				if (tpg->flags & PG_BUSY) {
1292 					tpg->flags |= freeflag;
1293 					if (pagedaemon) {
1294 						uvmexp.paging++;
1295 						uvm_pagedequeue(tpg);
1296 					}
1297 				} else {
1298 
1299 					/*
1300 					 * ``page is not busy''
1301 					 * implies that npages is 1
1302 					 * and needs_clean is false.
1303 					 */
1304 
1305 					nextpg = TAILQ_NEXT(tpg, listq);
1306 					uvm_pagefree(tpg);
1307 					if (pagedaemon)
1308 						uvmexp.pdfreed++;
1309 				}
1310 			}
1311 		}
1312 		if (flags & (PGO_DEACTIVATE|PGO_FREE)) {
1313 			uvm_unlock_pageq();
1314 		}
1315 		if (needs_clean) {
1316 			modified = TRUE;
1317 
1318 			/*
1319 			 * start the i/o.  if we're traversing by list,
1320 			 * keep our place in the list with a marker page.
1321 			 */
1322 
1323 			if (by_list) {
1324 				TAILQ_INSERT_AFTER(&uobj->memq, pg, &curmp,
1325 				    listq);
1326 			}
1327 			simple_unlock(slock);
1328 			error = GOP_WRITE(vp, pgs, npages, flags);
1329 			simple_lock(slock);
1330 			if (by_list) {
1331 				pg = TAILQ_NEXT(&curmp, listq);
1332 				TAILQ_REMOVE(&uobj->memq, &curmp, listq);
1333 			}
1334 			if (error) {
1335 				break;
1336 			}
1337 			if (by_list) {
1338 				continue;
1339 			}
1340 		}
1341 
1342 		/*
1343 		 * find the next page and continue if there was no error.
1344 		 */
1345 
1346 		if (by_list) {
1347 			if (nextpg) {
1348 				pg = nextpg;
1349 				nextpg = NULL;
1350 			} else {
1351 				pg = TAILQ_NEXT(pg, listq);
1352 			}
1353 		} else {
1354 			off += (npages - nback) << PAGE_SHIFT;
1355 			if (off < endoff) {
1356 				pg = uvm_pagelookup(uobj, off);
1357 			}
1358 		}
1359 	}
1360 	if (by_list) {
1361 		TAILQ_REMOVE(&uobj->memq, &endmp, listq);
1362 		PRELE(l);
1363 	}
1364 
1365 	if (modified && (vp->v_flag & VWRITEMAPDIRTY) != 0 &&
1366 	    (vp->v_type != VBLK ||
1367 	    (vp->v_mount->mnt_flag & MNT_NODEVMTIME) == 0)) {
1368 		GOP_MARKUPDATE(vp, GOP_UPDATE_MODIFIED);
1369 	}
1370 
1371 	/*
1372 	 * if we're cleaning and there was nothing to clean,
1373 	 * take us off the syncer list.  if we started any i/o
1374 	 * and we're doing sync i/o, wait for all writes to finish.
1375 	 */
1376 
1377 	s = splbio();
1378 	if (cleanall && wasclean && gp->g_dirtygen == dirtygen &&
1379 	    (vp->v_flag & VONWORKLST) != 0) {
1380 		vp->v_flag &= ~VWRITEMAPDIRTY;
1381 		if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
1382 			vp->v_flag &= ~VONWORKLST;
1383 			LIST_REMOVE(vp, v_synclist);
1384 		}
1385 	}
1386 	splx(s);
1387 
1388 #if !defined(DEBUG)
1389 skip_scan:
1390 #endif /* !defined(DEBUG) */
1391 	if (!wasclean && !async) {
1392 		s = splbio();
1393 		/*
1394 		 * XXX - we want simple_unlock(&global_v_numoutput_slock);
1395 		 *	 but the slot in ltsleep() is taken!
1396 		 * XXX - try to recover from missed wakeups with a timeout..
1397 		 *	 must think of something better.
1398 		 */
1399 		while (vp->v_numoutput != 0) {
1400 			vp->v_flag |= VBWAIT;
1401 			UVM_UNLOCK_AND_WAIT(&vp->v_numoutput, slock, FALSE,
1402 			    "genput2", hz);
1403 			simple_lock(slock);
1404 		}
1405 		splx(s);
1406 	}
1407 	simple_unlock(&uobj->vmobjlock);
1408 	return (error);
1409 }
1410 
1411 int
1412 genfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags)
1413 {
1414 	int s, error, run;
1415 	int fs_bshift, dev_bshift;
1416 	vaddr_t kva;
1417 	off_t eof, offset, startoffset;
1418 	size_t bytes, iobytes, skipbytes;
1419 	daddr_t lbn, blkno;
1420 	struct vm_page *pg;
1421 	struct buf *mbp, *bp;
1422 	struct vnode *devvp;
1423 	boolean_t async = (flags & PGO_SYNCIO) == 0;
1424 	UVMHIST_FUNC("genfs_gop_write"); UVMHIST_CALLED(ubchist);
1425 
1426 	UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1427 	    vp, pgs, npages, flags);
1428 
1429 	GOP_SIZE(vp, vp->v_size, &eof, GOP_SIZE_WRITE);
1430 	if (vp->v_type != VBLK) {
1431 		fs_bshift = vp->v_mount->mnt_fs_bshift;
1432 		dev_bshift = vp->v_mount->mnt_dev_bshift;
1433 	} else {
1434 		fs_bshift = DEV_BSHIFT;
1435 		dev_bshift = DEV_BSHIFT;
1436 	}
1437 	error = 0;
1438 	pg = pgs[0];
1439 	startoffset = pg->offset;
1440 	bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1441 	skipbytes = 0;
1442 	KASSERT(bytes != 0);
1443 
1444 	kva = uvm_pagermapin(pgs, npages,
1445 	    UVMPAGER_MAPIN_WRITE | UVMPAGER_MAPIN_WAITOK);
1446 
1447 	s = splbio();
1448 	simple_lock(&global_v_numoutput_slock);
1449 	vp->v_numoutput += 2;
1450 	simple_unlock(&global_v_numoutput_slock);
1451 	splx(s);
1452 	mbp = getiobuf();
1453 	UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
1454 	    vp, mbp, vp->v_numoutput, bytes);
1455 	mbp->b_bufsize = npages << PAGE_SHIFT;
1456 	mbp->b_data = (void *)kva;
1457 	mbp->b_resid = mbp->b_bcount = bytes;
1458 	mbp->b_flags = B_BUSY|B_WRITE|B_AGE| (async ? (B_CALL|B_ASYNC) : 0);
1459 	mbp->b_iodone = uvm_aio_biodone;
1460 	mbp->b_vp = vp;
1461 	if (curproc == uvm.pagedaemon_proc)
1462 		BIO_SETPRIO(mbp, BPRIO_TIMELIMITED);
1463 	else if (async)
1464 		BIO_SETPRIO(mbp, BPRIO_TIMENONCRITICAL);
1465 	else
1466 		BIO_SETPRIO(mbp, BPRIO_TIMECRITICAL);
1467 
1468 	bp = NULL;
1469 	for (offset = startoffset;
1470 	    bytes > 0;
1471 	    offset += iobytes, bytes -= iobytes) {
1472 		lbn = offset >> fs_bshift;
1473 		error = VOP_BMAP(vp, lbn, &devvp, &blkno, &run);
1474 		if (error) {
1475 			UVMHIST_LOG(ubchist, "VOP_BMAP() -> %d", error,0,0,0);
1476 			skipbytes += bytes;
1477 			bytes = 0;
1478 			break;
1479 		}
1480 
1481 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
1482 		    bytes);
1483 		if (blkno == (daddr_t)-1) {
1484 			skipbytes += iobytes;
1485 			continue;
1486 		}
1487 
1488 		/* if it's really one i/o, don't make a second buf */
1489 		if (offset == startoffset && iobytes == bytes) {
1490 			bp = mbp;
1491 		} else {
1492 			UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
1493 			    vp, bp, vp->v_numoutput, 0);
1494 			bp = getiobuf();
1495 			nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes);
1496 		}
1497 		bp->b_lblkno = 0;
1498 
1499 		/* adjust physical blkno for partial blocks */
1500 		bp->b_blkno = blkno + ((offset - ((off_t)lbn << fs_bshift)) >>
1501 		    dev_bshift);
1502 		UVMHIST_LOG(ubchist,
1503 		    "vp %p offset 0x%x bcount 0x%x blkno 0x%x",
1504 		    vp, offset, bp->b_bcount, bp->b_blkno);
1505 
1506 		VOP_STRATEGY(devvp, bp);
1507 	}
1508 	if (skipbytes) {
1509 		UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
1510 	}
1511 	nestiobuf_done(mbp, skipbytes, error);
1512 	if (async) {
1513 		UVMHIST_LOG(ubchist, "returning 0 (async)", 0,0,0,0);
1514 		return (0);
1515 	}
1516 	UVMHIST_LOG(ubchist, "waiting for mbp %p", mbp,0,0,0);
1517 	error = biowait(mbp);
1518 	uvm_aio_aiodone(mbp);
1519 	UVMHIST_LOG(ubchist, "returning, error %d", error,0,0,0);
1520 	return (error);
1521 }
1522 
1523 /*
1524  * VOP_PUTPAGES() for vnodes which never have pages.
1525  */
1526 
1527 int
1528 genfs_null_putpages(void *v)
1529 {
1530 	struct vop_putpages_args /* {
1531 		struct vnode *a_vp;
1532 		voff_t a_offlo;
1533 		voff_t a_offhi;
1534 		int a_flags;
1535 	} */ *ap = v;
1536 	struct vnode *vp = ap->a_vp;
1537 
1538 	KASSERT(vp->v_uobj.uo_npages == 0);
1539 	simple_unlock(&vp->v_interlock);
1540 	return (0);
1541 }
1542 
1543 void
1544 genfs_node_init(struct vnode *vp, const struct genfs_ops *ops)
1545 {
1546 	struct genfs_node *gp = VTOG(vp);
1547 
1548 	lockinit(&gp->g_glock, PINOD, "glock", 0, 0);
1549 	gp->g_op = ops;
1550 }
1551 
1552 void
1553 genfs_size(struct vnode *vp, off_t size, off_t *eobp, int flags)
1554 {
1555 	int bsize;
1556 
1557 	bsize = 1 << vp->v_mount->mnt_fs_bshift;
1558 	*eobp = (size + bsize - 1) & ~(bsize - 1);
1559 }
1560 
1561 int
1562 genfs_compat_getpages(void *v)
1563 {
1564 	struct vop_getpages_args /* {
1565 		struct vnode *a_vp;
1566 		voff_t a_offset;
1567 		struct vm_page **a_m;
1568 		int *a_count;
1569 		int a_centeridx;
1570 		vm_prot_t a_access_type;
1571 		int a_advice;
1572 		int a_flags;
1573 	} */ *ap = v;
1574 
1575 	off_t origoffset;
1576 	struct vnode *vp = ap->a_vp;
1577 	struct uvm_object *uobj = &vp->v_uobj;
1578 	struct vm_page *pg, **pgs;
1579 	vaddr_t kva;
1580 	int i, error, orignpages, npages;
1581 	struct iovec iov;
1582 	struct uio uio;
1583 	struct ucred *cred = curproc->p_ucred;
1584 	boolean_t write = (ap->a_access_type & VM_PROT_WRITE) != 0;
1585 
1586 	error = 0;
1587 	origoffset = ap->a_offset;
1588 	orignpages = *ap->a_count;
1589 	pgs = ap->a_m;
1590 
1591 	if (write && (vp->v_flag & VONWORKLST) == 0) {
1592 		vn_syncer_add_to_worklist(vp, filedelay);
1593 	}
1594 	if (ap->a_flags & PGO_LOCKED) {
1595 		uvn_findpages(uobj, origoffset, ap->a_count, ap->a_m,
1596 		    UFP_NOWAIT|UFP_NOALLOC| (write ? UFP_NORDONLY : 0));
1597 
1598 		return (ap->a_m[ap->a_centeridx] == NULL ? EBUSY : 0);
1599 	}
1600 	if (origoffset + (ap->a_centeridx << PAGE_SHIFT) >= vp->v_size) {
1601 		simple_unlock(&uobj->vmobjlock);
1602 		return (EINVAL);
1603 	}
1604 	if ((ap->a_flags & PGO_SYNCIO) == 0) {
1605 		simple_unlock(&uobj->vmobjlock);
1606 		return 0;
1607 	}
1608 	npages = orignpages;
1609 	uvn_findpages(uobj, origoffset, &npages, pgs, UFP_ALL);
1610 	simple_unlock(&uobj->vmobjlock);
1611 	kva = uvm_pagermapin(pgs, npages,
1612 	    UVMPAGER_MAPIN_READ | UVMPAGER_MAPIN_WAITOK);
1613 	for (i = 0; i < npages; i++) {
1614 		pg = pgs[i];
1615 		if ((pg->flags & PG_FAKE) == 0) {
1616 			continue;
1617 		}
1618 		iov.iov_base = (char *)kva + (i << PAGE_SHIFT);
1619 		iov.iov_len = PAGE_SIZE;
1620 		uio.uio_iov = &iov;
1621 		uio.uio_iovcnt = 1;
1622 		uio.uio_offset = origoffset + (i << PAGE_SHIFT);
1623 		uio.uio_segflg = UIO_SYSSPACE;
1624 		uio.uio_rw = UIO_READ;
1625 		uio.uio_resid = PAGE_SIZE;
1626 		uio.uio_lwp = NULL;
1627 		/* XXX vn_lock */
1628 		error = VOP_READ(vp, &uio, 0, cred);
1629 		if (error) {
1630 			break;
1631 		}
1632 		if (uio.uio_resid) {
1633 			memset(iov.iov_base, 0, uio.uio_resid);
1634 		}
1635 	}
1636 	uvm_pagermapout(kva, npages);
1637 	simple_lock(&uobj->vmobjlock);
1638 	uvm_lock_pageq();
1639 	for (i = 0; i < npages; i++) {
1640 		pg = pgs[i];
1641 		if (error && (pg->flags & PG_FAKE) != 0) {
1642 			pg->flags |= PG_RELEASED;
1643 		} else {
1644 			pmap_clear_modify(pg);
1645 			uvm_pageactivate(pg);
1646 		}
1647 	}
1648 	if (error) {
1649 		uvm_page_unbusy(pgs, npages);
1650 	}
1651 	uvm_unlock_pageq();
1652 	simple_unlock(&uobj->vmobjlock);
1653 	return (error);
1654 }
1655 
1656 int
1657 genfs_compat_gop_write(struct vnode *vp, struct vm_page **pgs, int npages,
1658     int flags)
1659 {
1660 	off_t offset;
1661 	struct iovec iov;
1662 	struct uio uio;
1663 	struct ucred *cred = curproc->p_ucred;
1664 	struct buf *bp;
1665 	vaddr_t kva;
1666 	int s, error;
1667 
1668 	offset = pgs[0]->offset;
1669 	kva = uvm_pagermapin(pgs, npages,
1670 	    UVMPAGER_MAPIN_WRITE | UVMPAGER_MAPIN_WAITOK);
1671 
1672 	iov.iov_base = (void *)kva;
1673 	iov.iov_len = npages << PAGE_SHIFT;
1674 	uio.uio_iov = &iov;
1675 	uio.uio_iovcnt = 1;
1676 	uio.uio_offset = offset;
1677 	uio.uio_segflg = UIO_SYSSPACE;
1678 	uio.uio_rw = UIO_WRITE;
1679 	uio.uio_resid = npages << PAGE_SHIFT;
1680 	uio.uio_lwp = NULL;
1681 	/* XXX vn_lock */
1682 	error = VOP_WRITE(vp, &uio, 0, cred);
1683 
1684 	s = splbio();
1685 	V_INCR_NUMOUTPUT(vp);
1686 	splx(s);
1687 
1688 	bp = getiobuf();
1689 	bp->b_flags = B_BUSY | B_WRITE | B_AGE;
1690 	bp->b_vp = vp;
1691 	bp->b_lblkno = offset >> vp->v_mount->mnt_fs_bshift;
1692 	bp->b_data = (char *)kva;
1693 	bp->b_bcount = npages << PAGE_SHIFT;
1694 	bp->b_bufsize = npages << PAGE_SHIFT;
1695 	bp->b_resid = 0;
1696 	if (error) {
1697 		bp->b_flags |= B_ERROR;
1698 		bp->b_error = error;
1699 	}
1700 	uvm_aio_aiodone(bp);
1701 	return (error);
1702 }
1703 
1704 static void
1705 filt_genfsdetach(struct knote *kn)
1706 {
1707 	struct vnode *vp = (struct vnode *)kn->kn_hook;
1708 
1709 	/* XXXLUKEM lock the struct? */
1710 	SLIST_REMOVE(&vp->v_klist, kn, knote, kn_selnext);
1711 }
1712 
1713 static int
1714 filt_genfsread(struct knote *kn, long hint)
1715 {
1716 	struct vnode *vp = (struct vnode *)kn->kn_hook;
1717 
1718 	/*
1719 	 * filesystem is gone, so set the EOF flag and schedule
1720 	 * the knote for deletion.
1721 	 */
1722 	if (hint == NOTE_REVOKE) {
1723 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1724 		return (1);
1725 	}
1726 
1727 	/* XXXLUKEM lock the struct? */
1728 	kn->kn_data = vp->v_size - kn->kn_fp->f_offset;
1729         return (kn->kn_data != 0);
1730 }
1731 
1732 static int
1733 filt_genfsvnode(struct knote *kn, long hint)
1734 {
1735 
1736 	if (kn->kn_sfflags & hint)
1737 		kn->kn_fflags |= hint;
1738 	if (hint == NOTE_REVOKE) {
1739 		kn->kn_flags |= EV_EOF;
1740 		return (1);
1741 	}
1742 	return (kn->kn_fflags != 0);
1743 }
1744 
1745 static const struct filterops genfsread_filtops =
1746 	{ 1, NULL, filt_genfsdetach, filt_genfsread };
1747 static const struct filterops genfsvnode_filtops =
1748 	{ 1, NULL, filt_genfsdetach, filt_genfsvnode };
1749 
1750 int
1751 genfs_kqfilter(void *v)
1752 {
1753 	struct vop_kqfilter_args /* {
1754 		struct vnode	*a_vp;
1755 		struct knote	*a_kn;
1756 	} */ *ap = v;
1757 	struct vnode *vp;
1758 	struct knote *kn;
1759 
1760 	vp = ap->a_vp;
1761 	kn = ap->a_kn;
1762 	switch (kn->kn_filter) {
1763 	case EVFILT_READ:
1764 		kn->kn_fop = &genfsread_filtops;
1765 		break;
1766 	case EVFILT_VNODE:
1767 		kn->kn_fop = &genfsvnode_filtops;
1768 		break;
1769 	default:
1770 		return (1);
1771 	}
1772 
1773 	kn->kn_hook = vp;
1774 
1775 	/* XXXLUKEM lock the struct? */
1776 	SLIST_INSERT_HEAD(&vp->v_klist, kn, kn_selnext);
1777 
1778 	return (0);
1779 }
1780