xref: /netbsd-src/sys/ufs/lfs/lfs_vnops.c (revision 7c3f385475147b6e1c4753f2bee961630e2dfc40)
1 /*	$NetBSD: lfs_vnops.c,v 1.215 2008/01/25 14:32:17 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Konrad E. Schroder <perseant@hhhh.org>.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 /*
39  * Copyright (c) 1986, 1989, 1991, 1993, 1995
40  *	The Regents of the University of California.  All rights reserved.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)lfs_vnops.c	8.13 (Berkeley) 6/10/95
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.215 2008/01/25 14:32:17 ad Exp $");
71 
72 #ifdef _KERNEL_OPT
73 #include "opt_compat_netbsd.h"
74 #endif
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/namei.h>
79 #include <sys/resourcevar.h>
80 #include <sys/kernel.h>
81 #include <sys/file.h>
82 #include <sys/stat.h>
83 #include <sys/buf.h>
84 #include <sys/proc.h>
85 #include <sys/mount.h>
86 #include <sys/vnode.h>
87 #include <sys/pool.h>
88 #include <sys/signalvar.h>
89 #include <sys/kauth.h>
90 #include <sys/syslog.h>
91 #include <sys/fstrans.h>
92 
93 #include <miscfs/fifofs/fifo.h>
94 #include <miscfs/genfs/genfs.h>
95 #include <miscfs/specfs/specdev.h>
96 
97 #include <ufs/ufs/inode.h>
98 #include <ufs/ufs/dir.h>
99 #include <ufs/ufs/ufsmount.h>
100 #include <ufs/ufs/ufs_extern.h>
101 
102 #include <uvm/uvm.h>
103 #include <uvm/uvm_pmap.h>
104 #include <uvm/uvm_stat.h>
105 #include <uvm/uvm_pager.h>
106 
107 #include <ufs/lfs/lfs.h>
108 #include <ufs/lfs/lfs_extern.h>
109 
110 extern pid_t lfs_writer_daemon;
111 int lfs_ignore_lazy_sync = 1;
112 
113 /* Global vfs data structures for lfs. */
114 int (**lfs_vnodeop_p)(void *);
115 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = {
116 	{ &vop_default_desc, vn_default_error },
117 	{ &vop_lookup_desc, ufs_lookup },		/* lookup */
118 	{ &vop_create_desc, lfs_create },		/* create */
119 	{ &vop_whiteout_desc, ufs_whiteout },		/* whiteout */
120 	{ &vop_mknod_desc, lfs_mknod },			/* mknod */
121 	{ &vop_open_desc, ufs_open },			/* open */
122 	{ &vop_close_desc, lfs_close },			/* close */
123 	{ &vop_access_desc, ufs_access },		/* access */
124 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
125 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
126 	{ &vop_read_desc, lfs_read },			/* read */
127 	{ &vop_write_desc, lfs_write },			/* write */
128 	{ &vop_ioctl_desc, ufs_ioctl },			/* ioctl */
129 	{ &vop_fcntl_desc, lfs_fcntl },			/* fcntl */
130 	{ &vop_poll_desc, ufs_poll },			/* poll */
131 	{ &vop_kqfilter_desc, genfs_kqfilter },		/* kqfilter */
132 	{ &vop_revoke_desc, ufs_revoke },		/* revoke */
133 	{ &vop_mmap_desc, lfs_mmap },			/* mmap */
134 	{ &vop_fsync_desc, lfs_fsync },			/* fsync */
135 	{ &vop_seek_desc, ufs_seek },			/* seek */
136 	{ &vop_remove_desc, lfs_remove },		/* remove */
137 	{ &vop_link_desc, lfs_link },			/* link */
138 	{ &vop_rename_desc, lfs_rename },		/* rename */
139 	{ &vop_mkdir_desc, lfs_mkdir },			/* mkdir */
140 	{ &vop_rmdir_desc, lfs_rmdir },			/* rmdir */
141 	{ &vop_symlink_desc, lfs_symlink },		/* symlink */
142 	{ &vop_readdir_desc, ufs_readdir },		/* readdir */
143 	{ &vop_readlink_desc, ufs_readlink },		/* readlink */
144 	{ &vop_abortop_desc, ufs_abortop },		/* abortop */
145 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
146 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
147 	{ &vop_lock_desc, ufs_lock },			/* lock */
148 	{ &vop_unlock_desc, ufs_unlock },		/* unlock */
149 	{ &vop_bmap_desc, ufs_bmap },			/* bmap */
150 	{ &vop_strategy_desc, lfs_strategy },		/* strategy */
151 	{ &vop_print_desc, ufs_print },			/* print */
152 	{ &vop_islocked_desc, ufs_islocked },		/* islocked */
153 	{ &vop_pathconf_desc, ufs_pathconf },		/* pathconf */
154 	{ &vop_advlock_desc, ufs_advlock },		/* advlock */
155 	{ &vop_bwrite_desc, lfs_bwrite },		/* bwrite */
156 	{ &vop_getpages_desc, lfs_getpages },		/* getpages */
157 	{ &vop_putpages_desc, lfs_putpages },		/* putpages */
158 	{ NULL, NULL }
159 };
160 const struct vnodeopv_desc lfs_vnodeop_opv_desc =
161 	{ &lfs_vnodeop_p, lfs_vnodeop_entries };
162 
163 int (**lfs_specop_p)(void *);
164 const struct vnodeopv_entry_desc lfs_specop_entries[] = {
165 	{ &vop_default_desc, vn_default_error },
166 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
167 	{ &vop_create_desc, spec_create },		/* create */
168 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
169 	{ &vop_open_desc, spec_open },			/* open */
170 	{ &vop_close_desc, lfsspec_close },		/* close */
171 	{ &vop_access_desc, ufs_access },		/* access */
172 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
173 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
174 	{ &vop_read_desc, ufsspec_read },		/* read */
175 	{ &vop_write_desc, ufsspec_write },		/* write */
176 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
177 	{ &vop_fcntl_desc, ufs_fcntl },			/* fcntl */
178 	{ &vop_poll_desc, spec_poll },			/* poll */
179 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
180 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
181 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
182 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
183 	{ &vop_seek_desc, spec_seek },			/* seek */
184 	{ &vop_remove_desc, spec_remove },		/* remove */
185 	{ &vop_link_desc, spec_link },			/* link */
186 	{ &vop_rename_desc, spec_rename },		/* rename */
187 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
188 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
189 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
190 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
191 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
192 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
193 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
194 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
195 	{ &vop_lock_desc, ufs_lock },			/* lock */
196 	{ &vop_unlock_desc, ufs_unlock },		/* unlock */
197 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
198 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
199 	{ &vop_print_desc, ufs_print },			/* print */
200 	{ &vop_islocked_desc, ufs_islocked },		/* islocked */
201 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
202 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
203 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
204 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
205 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
206 	{ NULL, NULL }
207 };
208 const struct vnodeopv_desc lfs_specop_opv_desc =
209 	{ &lfs_specop_p, lfs_specop_entries };
210 
211 int (**lfs_fifoop_p)(void *);
212 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = {
213 	{ &vop_default_desc, vn_default_error },
214 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
215 	{ &vop_create_desc, fifo_create },		/* create */
216 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
217 	{ &vop_open_desc, fifo_open },			/* open */
218 	{ &vop_close_desc, lfsfifo_close },		/* close */
219 	{ &vop_access_desc, ufs_access },		/* access */
220 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
221 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
222 	{ &vop_read_desc, ufsfifo_read },		/* read */
223 	{ &vop_write_desc, ufsfifo_write },		/* write */
224 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
225 	{ &vop_fcntl_desc, ufs_fcntl },			/* fcntl */
226 	{ &vop_poll_desc, fifo_poll },			/* poll */
227 	{ &vop_kqfilter_desc, fifo_kqfilter },		/* kqfilter */
228 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
229 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
230 	{ &vop_fsync_desc, fifo_fsync },		/* fsync */
231 	{ &vop_seek_desc, fifo_seek },			/* seek */
232 	{ &vop_remove_desc, fifo_remove },		/* remove */
233 	{ &vop_link_desc, fifo_link },			/* link */
234 	{ &vop_rename_desc, fifo_rename },		/* rename */
235 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
236 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
237 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
238 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
239 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
240 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
241 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
242 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
243 	{ &vop_lock_desc, ufs_lock },			/* lock */
244 	{ &vop_unlock_desc, ufs_unlock },		/* unlock */
245 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
246 	{ &vop_strategy_desc, fifo_strategy },		/* strategy */
247 	{ &vop_print_desc, ufs_print },			/* print */
248 	{ &vop_islocked_desc, ufs_islocked },		/* islocked */
249 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
250 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
251 	{ &vop_bwrite_desc, lfs_bwrite },		/* bwrite */
252 	{ &vop_putpages_desc, fifo_putpages },		/* putpages */
253 	{ NULL, NULL }
254 };
255 const struct vnodeopv_desc lfs_fifoop_opv_desc =
256 	{ &lfs_fifoop_p, lfs_fifoop_entries };
257 
258 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int, struct vm_page **);
259 
260 #define	LFS_READWRITE
261 #include <ufs/ufs/ufs_readwrite.c>
262 #undef	LFS_READWRITE
263 
264 /*
265  * Synch an open file.
266  */
267 /* ARGSUSED */
268 int
269 lfs_fsync(void *v)
270 {
271 	struct vop_fsync_args /* {
272 		struct vnode *a_vp;
273 		kauth_cred_t a_cred;
274 		int a_flags;
275 		off_t offlo;
276 		off_t offhi;
277 	} */ *ap = v;
278 	struct vnode *vp = ap->a_vp;
279 	int error, wait;
280 	struct inode *ip = VTOI(vp);
281 	struct lfs *fs = ip->i_lfs;
282 
283 	/* If we're mounted read-only, don't try to sync. */
284 	if (fs->lfs_ronly)
285 		return 0;
286 
287 	/*
288 	 * Trickle sync simply adds this vnode to the pager list, as if
289 	 * the pagedaemon had requested a pageout.
290 	 */
291 	if (ap->a_flags & FSYNC_LAZY) {
292 		if (lfs_ignore_lazy_sync == 0) {
293 			mutex_enter(&lfs_lock);
294 			if (!(ip->i_flags & IN_PAGING)) {
295 				ip->i_flags |= IN_PAGING;
296 				TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip,
297 						  i_lfs_pchain);
298 			}
299 			wakeup(&lfs_writer_daemon);
300 			mutex_exit(&lfs_lock);
301 		}
302 		return 0;
303 	}
304 
305 	/*
306 	 * If a vnode is bring cleaned, flush it out before we try to
307 	 * reuse it.  This prevents the cleaner from writing files twice
308 	 * in the same partial segment, causing an accounting underflow.
309 	 */
310 	if (ap->a_flags & FSYNC_RECLAIM && ip->i_flags & IN_CLEANING) {
311 		lfs_vflush(vp);
312 	}
313 
314 	wait = (ap->a_flags & FSYNC_WAIT);
315 	do {
316 		mutex_enter(&vp->v_interlock);
317 		error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo),
318 				     round_page(ap->a_offhi),
319 				     PGO_CLEANIT | (wait ? PGO_SYNCIO : 0));
320 		if (error == EAGAIN) {
321 			mutex_enter(&lfs_lock);
322 			mtsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_fsync",
323 				hz / 100 + 1, &lfs_lock);
324 			mutex_exit(&lfs_lock);
325 		}
326 	} while (error == EAGAIN);
327 	if (error)
328 		return error;
329 
330 	if ((ap->a_flags & FSYNC_DATAONLY) == 0)
331 		error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0);
332 
333 	if (error == 0 && ap->a_flags & FSYNC_CACHE) {
334 		int l = 0;
335 		error = VOP_IOCTL(ip->i_devvp, DIOCCACHESYNC, &l, FWRITE,
336 				  curlwp->l_cred);
337 	}
338 	if (wait && !VPISEMPTY(vp))
339 		LFS_SET_UINO(ip, IN_MODIFIED);
340 
341 	return error;
342 }
343 
344 /*
345  * Take IN_ADIROP off, then call ufs_inactive.
346  */
347 int
348 lfs_inactive(void *v)
349 {
350 	struct vop_inactive_args /* {
351 		struct vnode *a_vp;
352 	} */ *ap = v;
353 
354 	KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink);
355 
356 	lfs_unmark_vnode(ap->a_vp);
357 
358 	/*
359 	 * The Ifile is only ever inactivated on unmount.
360 	 * Streamline this process by not giving it more dirty blocks.
361 	 */
362 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) {
363 		mutex_enter(&lfs_lock);
364 		LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD);
365 		mutex_exit(&lfs_lock);
366 		VOP_UNLOCK(ap->a_vp, 0);
367 		return 0;
368 	}
369 
370 	return ufs_inactive(v);
371 }
372 
373 /*
374  * These macros are used to bracket UFS directory ops, so that we can
375  * identify all the pages touched during directory ops which need to
376  * be ordered and flushed atomically, so that they may be recovered.
377  *
378  * Because we have to mark nodes VU_DIROP in order to prevent
379  * the cache from reclaiming them while a dirop is in progress, we must
380  * also manage the number of nodes so marked (otherwise we can run out).
381  * We do this by setting lfs_dirvcount to the number of marked vnodes; it
382  * is decremented during segment write, when VU_DIROP is taken off.
383  */
384 #define	MARK_VNODE(vp)			lfs_mark_vnode(vp)
385 #define	UNMARK_VNODE(vp)		lfs_unmark_vnode(vp)
386 #define	SET_DIROP_CREATE(dvp, vpp)	lfs_set_dirop_create((dvp), (vpp))
387 #define	SET_DIROP_REMOVE(dvp, vp)	lfs_set_dirop((dvp), (vp))
388 static int lfs_set_dirop_create(struct vnode *, struct vnode **);
389 static int lfs_set_dirop(struct vnode *, struct vnode *);
390 
391 static int
392 lfs_set_dirop(struct vnode *dvp, struct vnode *vp)
393 {
394 	struct lfs *fs;
395 	int error;
396 
397 	KASSERT(VOP_ISLOCKED(dvp));
398 	KASSERT(vp == NULL || VOP_ISLOCKED(vp));
399 
400 	fs = VTOI(dvp)->i_lfs;
401 
402 	ASSERT_NO_SEGLOCK(fs);
403 	/*
404 	 * LFS_NRESERVE calculates direct and indirect blocks as well
405 	 * as an inode block; an overestimate in most cases.
406 	 */
407 	if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0)
408 		return (error);
409 
410     restart:
411 	mutex_enter(&lfs_lock);
412 	if (fs->lfs_dirops == 0) {
413 		mutex_exit(&lfs_lock);
414 		lfs_check(dvp, LFS_UNUSED_LBN, 0);
415 		mutex_enter(&lfs_lock);
416 	}
417 	while (fs->lfs_writer) {
418 		error = mtsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH,
419 		    "lfs_sdirop", 0, &lfs_lock);
420 		if (error == EINTR) {
421 			mutex_exit(&lfs_lock);
422 			goto unreserve;
423 		}
424 	}
425 	if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) {
426 		wakeup(&lfs_writer_daemon);
427 		mutex_exit(&lfs_lock);
428 		preempt();
429 		goto restart;
430 	}
431 
432 	if (lfs_dirvcount > LFS_MAX_DIROP) {
433 		mutex_exit(&lfs_lock);
434 		DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, "
435 		      "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount));
436 		if ((error = mtsleep(&lfs_dirvcount,
437 		    PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0,
438 		    &lfs_lock)) != 0) {
439 			goto unreserve;
440 		}
441 		goto restart;
442 	}
443 
444 	++fs->lfs_dirops;
445 	fs->lfs_doifile = 1;
446 	mutex_exit(&lfs_lock);
447 
448 	/* Hold a reference so SET_ENDOP will be happy */
449 	vref(dvp);
450 	if (vp) {
451 		vref(vp);
452 		MARK_VNODE(vp);
453 	}
454 
455 	MARK_VNODE(dvp);
456 	return 0;
457 
458   unreserve:
459 	lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs));
460 	return error;
461 }
462 
463 /*
464  * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock
465  * in getnewvnode(), if we have a stacked filesystem mounted on top
466  * of us.
467  *
468  * NB: this means we have to clear the new vnodes on error.  Fortunately
469  * SET_ENDOP is there to do that for us.
470  */
471 static int
472 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp)
473 {
474 	int error;
475 	struct lfs *fs;
476 
477 	fs = VFSTOUFS(dvp->v_mount)->um_lfs;
478 	ASSERT_NO_SEGLOCK(fs);
479 	if (fs->lfs_ronly)
480 		return EROFS;
481 	if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) {
482 		DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n",
483 		      dvp, error));
484 		return error;
485 	}
486 	if ((error = lfs_set_dirop(dvp, NULL)) != 0) {
487 		if (vpp) {
488 			ungetnewvnode(*vpp);
489 			*vpp = NULL;
490 		}
491 		return error;
492 	}
493 	return 0;
494 }
495 
496 #define	SET_ENDOP_BASE(fs, dvp, str)					\
497 	do {								\
498 		mutex_enter(&lfs_lock);				\
499 		--(fs)->lfs_dirops;					\
500 		if (!(fs)->lfs_dirops) {				\
501 			if ((fs)->lfs_nadirop) {			\
502 				panic("SET_ENDOP: %s: no dirops but "	\
503 					" nadirop=%d", (str),		\
504 					(fs)->lfs_nadirop);		\
505 			}						\
506 			wakeup(&(fs)->lfs_writer);			\
507 			mutex_exit(&lfs_lock);				\
508 			lfs_check((dvp), LFS_UNUSED_LBN, 0);		\
509 		} else							\
510 			mutex_exit(&lfs_lock);				\
511 	} while(0)
512 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str)				\
513 	do {								\
514 		UNMARK_VNODE(dvp);					\
515 		if (nvpp && *nvpp)					\
516 			UNMARK_VNODE(*nvpp);				\
517 		/* Check for error return to stem vnode leakage */	\
518 		if (nvpp && *nvpp && !((*nvpp)->v_uflag & VU_DIROP))	\
519 			ungetnewvnode(*(nvpp));				\
520 		SET_ENDOP_BASE((fs), (dvp), (str));			\
521 		lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs));	\
522 		vrele(dvp);						\
523 	} while(0)
524 #define SET_ENDOP_CREATE_AP(ap, str)					\
525 	SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp,		\
526 			 (ap)->a_vpp, (str))
527 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str)				\
528 	do {								\
529 		UNMARK_VNODE(dvp);					\
530 		if (ovp)						\
531 			UNMARK_VNODE(ovp);				\
532 		SET_ENDOP_BASE((fs), (dvp), (str));			\
533 		lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs));	\
534 		vrele(dvp);						\
535 		if (ovp)						\
536 			vrele(ovp);					\
537 	} while(0)
538 
539 void
540 lfs_mark_vnode(struct vnode *vp)
541 {
542 	struct inode *ip = VTOI(vp);
543 	struct lfs *fs = ip->i_lfs;
544 
545 	mutex_enter(&lfs_lock);
546 	if (!(ip->i_flag & IN_ADIROP)) {
547 		if (!(vp->v_uflag & VU_DIROP)) {
548 			mutex_enter(&vp->v_interlock);
549 			(void)lfs_vref(vp);
550 			++lfs_dirvcount;
551 			++fs->lfs_dirvcount;
552 			TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain);
553 			vp->v_uflag |= VU_DIROP;
554 		}
555 		++fs->lfs_nadirop;
556 		ip->i_flag |= IN_ADIROP;
557 	} else
558 		KASSERT(vp->v_uflag & VU_DIROP);
559 	mutex_exit(&lfs_lock);
560 }
561 
562 void
563 lfs_unmark_vnode(struct vnode *vp)
564 {
565 	struct inode *ip = VTOI(vp);
566 
567 	if (ip && (ip->i_flag & IN_ADIROP)) {
568 		KASSERT(vp->v_uflag & VU_DIROP);
569 		mutex_enter(&lfs_lock);
570 		--ip->i_lfs->lfs_nadirop;
571 		mutex_exit(&lfs_lock);
572 		ip->i_flag &= ~IN_ADIROP;
573 	}
574 }
575 
576 int
577 lfs_symlink(void *v)
578 {
579 	struct vop_symlink_args /* {
580 		struct vnode *a_dvp;
581 		struct vnode **a_vpp;
582 		struct componentname *a_cnp;
583 		struct vattr *a_vap;
584 		char *a_target;
585 	} */ *ap = v;
586 	int error;
587 
588 	if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) {
589 		vput(ap->a_dvp);
590 		return error;
591 	}
592 	error = ufs_symlink(ap);
593 	SET_ENDOP_CREATE_AP(ap, "symlink");
594 	return (error);
595 }
596 
597 int
598 lfs_mknod(void *v)
599 {
600 	struct vop_mknod_args	/* {
601 		struct vnode *a_dvp;
602 		struct vnode **a_vpp;
603 		struct componentname *a_cnp;
604 		struct vattr *a_vap;
605 	} */ *ap = v;
606 	struct vattr *vap = ap->a_vap;
607 	struct vnode **vpp = ap->a_vpp;
608 	struct inode *ip;
609 	int error;
610 	struct mount	*mp;
611 	ino_t		ino;
612 
613 	if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) {
614 		vput(ap->a_dvp);
615 		return error;
616 	}
617 	error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
618 			      ap->a_dvp, vpp, ap->a_cnp);
619 
620 	/* Either way we're done with the dirop at this point */
621 	SET_ENDOP_CREATE_AP(ap, "mknod");
622 
623 	if (error)
624 		return (error);
625 
626 	ip = VTOI(*vpp);
627 	mp  = (*vpp)->v_mount;
628 	ino = ip->i_number;
629 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
630 	if (vap->va_rdev != VNOVAL) {
631 		/*
632 		 * Want to be able to use this to make badblock
633 		 * inodes, so don't truncate the dev number.
634 		 */
635 #if 0
636 		ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev,
637 					   UFS_MPNEEDSWAP((*vpp)->v_mount));
638 #else
639 		ip->i_ffs1_rdev = vap->va_rdev;
640 #endif
641 	}
642 
643 	/*
644 	 * Call fsync to write the vnode so that we don't have to deal with
645 	 * flushing it when it's marked VU_DIROP|VI_XLOCK.
646 	 *
647 	 * XXX KS - If we can't flush we also can't call vgone(), so must
648 	 * return.  But, that leaves this vnode in limbo, also not good.
649 	 * Can this ever happen (barring hardware failure)?
650 	 */
651 	if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0)) != 0) {
652 		panic("lfs_mknod: couldn't fsync (ino %llu)",
653 		      (unsigned long long)ino);
654 		/* return (error); */
655 	}
656 	/*
657 	 * Remove vnode so that it will be reloaded by VFS_VGET and
658 	 * checked to see if it is an alias of an existing entry in
659 	 * the inode cache.
660 	 */
661 	/* Used to be vput, but that causes us to call VOP_INACTIVE twice. */
662 
663 	VOP_UNLOCK(*vpp, 0);
664 	(*vpp)->v_type = VNON;
665 	vgone(*vpp);
666 	error = VFS_VGET(mp, ino, vpp);
667 
668 	if (error != 0) {
669 		*vpp = NULL;
670 		return (error);
671 	}
672 	return (0);
673 }
674 
675 int
676 lfs_create(void *v)
677 {
678 	struct vop_create_args	/* {
679 		struct vnode *a_dvp;
680 		struct vnode **a_vpp;
681 		struct componentname *a_cnp;
682 		struct vattr *a_vap;
683 	} */ *ap = v;
684 	int error;
685 
686 	if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) {
687 		vput(ap->a_dvp);
688 		return error;
689 	}
690 	error = ufs_create(ap);
691 	SET_ENDOP_CREATE_AP(ap, "create");
692 	return (error);
693 }
694 
695 int
696 lfs_mkdir(void *v)
697 {
698 	struct vop_mkdir_args	/* {
699 		struct vnode *a_dvp;
700 		struct vnode **a_vpp;
701 		struct componentname *a_cnp;
702 		struct vattr *a_vap;
703 	} */ *ap = v;
704 	int error;
705 
706 	if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) {
707 		vput(ap->a_dvp);
708 		return error;
709 	}
710 	error = ufs_mkdir(ap);
711 	SET_ENDOP_CREATE_AP(ap, "mkdir");
712 	return (error);
713 }
714 
715 int
716 lfs_remove(void *v)
717 {
718 	struct vop_remove_args	/* {
719 		struct vnode *a_dvp;
720 		struct vnode *a_vp;
721 		struct componentname *a_cnp;
722 	} */ *ap = v;
723 	struct vnode *dvp, *vp;
724 	struct inode *ip;
725 	int error;
726 
727 	dvp = ap->a_dvp;
728 	vp = ap->a_vp;
729 	ip = VTOI(vp);
730 	if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) {
731 		if (dvp == vp)
732 			vrele(vp);
733 		else
734 			vput(vp);
735 		vput(dvp);
736 		return error;
737 	}
738 	error = ufs_remove(ap);
739 	if (ip->i_nlink == 0)
740 		lfs_orphan(ip->i_lfs, ip->i_number);
741 	SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove");
742 	return (error);
743 }
744 
745 int
746 lfs_rmdir(void *v)
747 {
748 	struct vop_rmdir_args	/* {
749 		struct vnodeop_desc *a_desc;
750 		struct vnode *a_dvp;
751 		struct vnode *a_vp;
752 		struct componentname *a_cnp;
753 	} */ *ap = v;
754 	struct vnode *vp;
755 	struct inode *ip;
756 	int error;
757 
758 	vp = ap->a_vp;
759 	ip = VTOI(vp);
760 	if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) {
761 		if (ap->a_dvp == vp)
762 			vrele(ap->a_dvp);
763 		else
764 			vput(ap->a_dvp);
765 		vput(vp);
766 		return error;
767 	}
768 	error = ufs_rmdir(ap);
769 	if (ip->i_nlink == 0)
770 		lfs_orphan(ip->i_lfs, ip->i_number);
771 	SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir");
772 	return (error);
773 }
774 
775 int
776 lfs_link(void *v)
777 {
778 	struct vop_link_args	/* {
779 		struct vnode *a_dvp;
780 		struct vnode *a_vp;
781 		struct componentname *a_cnp;
782 	} */ *ap = v;
783 	int error;
784 	struct vnode **vpp = NULL;
785 
786 	if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) {
787 		vput(ap->a_dvp);
788 		return error;
789 	}
790 	error = ufs_link(ap);
791 	SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link");
792 	return (error);
793 }
794 
795 int
796 lfs_rename(void *v)
797 {
798 	struct vop_rename_args	/* {
799 		struct vnode *a_fdvp;
800 		struct vnode *a_fvp;
801 		struct componentname *a_fcnp;
802 		struct vnode *a_tdvp;
803 		struct vnode *a_tvp;
804 		struct componentname *a_tcnp;
805 	} */ *ap = v;
806 	struct vnode *tvp, *fvp, *tdvp, *fdvp;
807 	struct componentname *tcnp, *fcnp;
808 	int error;
809 	struct lfs *fs;
810 
811 	fs = VTOI(ap->a_fdvp)->i_lfs;
812 	tvp = ap->a_tvp;
813 	tdvp = ap->a_tdvp;
814 	tcnp = ap->a_tcnp;
815 	fvp = ap->a_fvp;
816 	fdvp = ap->a_fdvp;
817 	fcnp = ap->a_fcnp;
818 
819 	/*
820 	 * Check for cross-device rename.
821 	 * If it is, we don't want to set dirops, just error out.
822 	 * (In particular note that MARK_VNODE(tdvp) will DTWT on
823 	 * a cross-device rename.)
824 	 *
825 	 * Copied from ufs_rename.
826 	 */
827 	if ((fvp->v_mount != tdvp->v_mount) ||
828 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
829 		error = EXDEV;
830 		goto errout;
831 	}
832 
833 	/*
834 	 * Check to make sure we're not renaming a vnode onto itself
835 	 * (deleting a hard link by renaming one name onto another);
836 	 * if we are we can't recursively call VOP_REMOVE since that
837 	 * would leave us with an unaccounted-for number of live dirops.
838 	 *
839 	 * Inline the relevant section of ufs_rename here, *before*
840 	 * calling SET_DIROP_REMOVE.
841 	 */
842 	if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) ||
843 		    (VTOI(tdvp)->i_flags & APPEND))) {
844 		error = EPERM;
845 		goto errout;
846 	}
847 	if (fvp == tvp) {
848 		if (fvp->v_type == VDIR) {
849 			error = EINVAL;
850 			goto errout;
851 		}
852 
853 		/* Release destination completely. */
854 		VOP_ABORTOP(tdvp, tcnp);
855 		vput(tdvp);
856 		vput(tvp);
857 
858 		/* Delete source. */
859 		vrele(fvp);
860 		fcnp->cn_flags &= ~(MODMASK | SAVESTART);
861 		fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
862 		fcnp->cn_nameiop = DELETE;
863 		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
864 		if ((error = relookup(fdvp, &fvp, fcnp))) {
865 			vput(fdvp);
866 			return (error);
867 		}
868 		return (VOP_REMOVE(fdvp, fvp, fcnp));
869 	}
870 
871 	if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0)
872 		goto errout;
873 	MARK_VNODE(fdvp);
874 	MARK_VNODE(fvp);
875 
876 	error = ufs_rename(ap);
877 	UNMARK_VNODE(fdvp);
878 	UNMARK_VNODE(fvp);
879 	SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename");
880 	return (error);
881 
882   errout:
883 	VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */
884 	if (tdvp == tvp)
885 		vrele(tdvp);
886 	else
887 		vput(tdvp);
888 	if (tvp)
889 		vput(tvp);
890 	VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */
891 	vrele(fdvp);
892 	vrele(fvp);
893 	return (error);
894 }
895 
896 /* XXX hack to avoid calling ITIMES in getattr */
897 int
898 lfs_getattr(void *v)
899 {
900 	struct vop_getattr_args /* {
901 		struct vnode *a_vp;
902 		struct vattr *a_vap;
903 		kauth_cred_t a_cred;
904 	} */ *ap = v;
905 	struct vnode *vp = ap->a_vp;
906 	struct inode *ip = VTOI(vp);
907 	struct vattr *vap = ap->a_vap;
908 	struct lfs *fs = ip->i_lfs;
909 	/*
910 	 * Copy from inode table
911 	 */
912 	vap->va_fsid = ip->i_dev;
913 	vap->va_fileid = ip->i_number;
914 	vap->va_mode = ip->i_mode & ~IFMT;
915 	vap->va_nlink = ip->i_nlink;
916 	vap->va_uid = ip->i_uid;
917 	vap->va_gid = ip->i_gid;
918 	vap->va_rdev = (dev_t)ip->i_ffs1_rdev;
919 	vap->va_size = vp->v_size;
920 	vap->va_atime.tv_sec = ip->i_ffs1_atime;
921 	vap->va_atime.tv_nsec = ip->i_ffs1_atimensec;
922 	vap->va_mtime.tv_sec = ip->i_ffs1_mtime;
923 	vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec;
924 	vap->va_ctime.tv_sec = ip->i_ffs1_ctime;
925 	vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec;
926 	vap->va_flags = ip->i_flags;
927 	vap->va_gen = ip->i_gen;
928 	/* this doesn't belong here */
929 	if (vp->v_type == VBLK)
930 		vap->va_blocksize = BLKDEV_IOSIZE;
931 	else if (vp->v_type == VCHR)
932 		vap->va_blocksize = MAXBSIZE;
933 	else
934 		vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
935 	vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks);
936 	vap->va_type = vp->v_type;
937 	vap->va_filerev = ip->i_modrev;
938 	return (0);
939 }
940 
941 /*
942  * Check to make sure the inode blocks won't choke the buffer
943  * cache, then call ufs_setattr as usual.
944  */
945 int
946 lfs_setattr(void *v)
947 {
948 	struct vop_setattr_args /* {
949 		struct vnode *a_vp;
950 		struct vattr *a_vap;
951 		kauth_cred_t a_cred;
952 	} */ *ap = v;
953 	struct vnode *vp = ap->a_vp;
954 
955 	lfs_check(vp, LFS_UNUSED_LBN, 0);
956 	return ufs_setattr(v);
957 }
958 
959 /*
960  * Release the block we hold on lfs_newseg wrapping.  Called on file close,
961  * or explicitly from LFCNWRAPGO.  Called with the interlock held.
962  */
963 static int
964 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor)
965 {
966 	if (fs->lfs_stoplwp != curlwp)
967 		return EBUSY;
968 
969 	fs->lfs_stoplwp = NULL;
970 	cv_signal(&fs->lfs_stopcv);
971 
972 	KASSERT(fs->lfs_nowrap > 0);
973 	if (fs->lfs_nowrap <= 0) {
974 		return 0;
975 	}
976 
977 	if (--fs->lfs_nowrap == 0) {
978 		log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt);
979 		wakeup(&fs->lfs_wrappass);
980 		lfs_wakeup_cleaner(fs);
981 	}
982 	if (waitfor) {
983 		mtsleep(&fs->lfs_nextseg, PCATCH | PUSER, "segment",
984 		    0, &lfs_lock);
985 	}
986 
987 	return 0;
988 }
989 
990 /*
991  * Close called
992  */
993 /* ARGSUSED */
994 int
995 lfs_close(void *v)
996 {
997 	struct vop_close_args /* {
998 		struct vnode *a_vp;
999 		int  a_fflag;
1000 		kauth_cred_t a_cred;
1001 	} */ *ap = v;
1002 	struct vnode *vp = ap->a_vp;
1003 	struct inode *ip = VTOI(vp);
1004 	struct lfs *fs = ip->i_lfs;
1005 
1006 	if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) &&
1007 	    fs->lfs_stoplwp == curlwp) {
1008 		mutex_enter(&lfs_lock);
1009 		log(LOG_NOTICE, "lfs_close: releasing log wrap control\n");
1010 		lfs_wrapgo(fs, ip, 0);
1011 		mutex_exit(&lfs_lock);
1012 	}
1013 
1014 	if (vp == ip->i_lfs->lfs_ivnode &&
1015 	    vp->v_mount->mnt_iflag & IMNT_UNMOUNT)
1016 		return 0;
1017 
1018 	if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) {
1019 		LFS_ITIMES(ip, NULL, NULL, NULL);
1020 	}
1021 	return (0);
1022 }
1023 
1024 /*
1025  * Close wrapper for special devices.
1026  *
1027  * Update the times on the inode then do device close.
1028  */
1029 int
1030 lfsspec_close(void *v)
1031 {
1032 	struct vop_close_args /* {
1033 		struct vnode	*a_vp;
1034 		int		a_fflag;
1035 		kauth_cred_t	a_cred;
1036 	} */ *ap = v;
1037 	struct vnode	*vp;
1038 	struct inode	*ip;
1039 
1040 	vp = ap->a_vp;
1041 	ip = VTOI(vp);
1042 	if (vp->v_usecount > 1) {
1043 		LFS_ITIMES(ip, NULL, NULL, NULL);
1044 	}
1045 	return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap));
1046 }
1047 
1048 /*
1049  * Close wrapper for fifo's.
1050  *
1051  * Update the times on the inode then do device close.
1052  */
1053 int
1054 lfsfifo_close(void *v)
1055 {
1056 	struct vop_close_args /* {
1057 		struct vnode	*a_vp;
1058 		int		a_fflag;
1059 		kauth_cred_	a_cred;
1060 	} */ *ap = v;
1061 	struct vnode	*vp;
1062 	struct inode	*ip;
1063 
1064 	vp = ap->a_vp;
1065 	ip = VTOI(vp);
1066 	if (ap->a_vp->v_usecount > 1) {
1067 		LFS_ITIMES(ip, NULL, NULL, NULL);
1068 	}
1069 	return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap));
1070 }
1071 
1072 /*
1073  * Reclaim an inode so that it can be used for other purposes.
1074  */
1075 
1076 int
1077 lfs_reclaim(void *v)
1078 {
1079 	struct vop_reclaim_args /* {
1080 		struct vnode *a_vp;
1081 	} */ *ap = v;
1082 	struct vnode *vp = ap->a_vp;
1083 	struct inode *ip = VTOI(vp);
1084 	struct lfs *fs = ip->i_lfs;
1085 	int error;
1086 
1087 	KASSERT(ip->i_nlink == ip->i_ffs_effnlink);
1088 
1089 	mutex_enter(&lfs_lock);
1090 	LFS_CLR_UINO(ip, IN_ALLMOD);
1091 	mutex_exit(&lfs_lock);
1092 	if ((error = ufs_reclaim(vp)))
1093 		return (error);
1094 
1095 	/*
1096 	 * Take us off the paging and/or dirop queues if we were on them.
1097 	 * We shouldn't be on them.
1098 	 */
1099 	mutex_enter(&lfs_lock);
1100 	if (ip->i_flags & IN_PAGING) {
1101 		log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n",
1102 		    fs->lfs_fsmnt);
1103 		ip->i_flags &= ~IN_PAGING;
1104 		TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain);
1105 	}
1106 	if (vp->v_uflag & VU_DIROP) {
1107 		panic("reclaimed vnode is VU_DIROP");
1108 		vp->v_uflag &= ~VU_DIROP;
1109 		TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
1110 	}
1111 	mutex_exit(&lfs_lock);
1112 
1113 	pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din);
1114 	lfs_deregister_all(vp);
1115 	pool_put(&lfs_inoext_pool, ip->inode_ext.lfs);
1116 	ip->inode_ext.lfs = NULL;
1117 	genfs_node_destroy(vp);
1118 	pool_put(&lfs_inode_pool, vp->v_data);
1119 	vp->v_data = NULL;
1120 	return (0);
1121 }
1122 
1123 /*
1124  * Read a block from a storage device.
1125  * In order to avoid reading blocks that are in the process of being
1126  * written by the cleaner---and hence are not mutexed by the normal
1127  * buffer cache / page cache mechanisms---check for collisions before
1128  * reading.
1129  *
1130  * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before*
1131  * the active cleaner test.
1132  *
1133  * XXX This code assumes that lfs_markv makes synchronous checkpoints.
1134  */
1135 int
1136 lfs_strategy(void *v)
1137 {
1138 	struct vop_strategy_args /* {
1139 		struct vnode *a_vp;
1140 		struct buf *a_bp;
1141 	} */ *ap = v;
1142 	struct buf	*bp;
1143 	struct lfs	*fs;
1144 	struct vnode	*vp;
1145 	struct inode	*ip;
1146 	daddr_t		tbn;
1147 	int		i, sn, error, slept;
1148 
1149 	bp = ap->a_bp;
1150 	vp = ap->a_vp;
1151 	ip = VTOI(vp);
1152 	fs = ip->i_lfs;
1153 
1154 	/* lfs uses its strategy routine only for read */
1155 	KASSERT(bp->b_flags & B_READ);
1156 
1157 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1158 		panic("lfs_strategy: spec");
1159 	KASSERT(bp->b_bcount != 0);
1160 	if (bp->b_blkno == bp->b_lblkno) {
1161 		error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno,
1162 				 NULL);
1163 		if (error) {
1164 			bp->b_error = error;
1165 			bp->b_resid = bp->b_bcount;
1166 			biodone(bp);
1167 			return (error);
1168 		}
1169 		if ((long)bp->b_blkno == -1) /* no valid data */
1170 			clrbuf(bp);
1171 	}
1172 	if ((long)bp->b_blkno < 0) { /* block is not on disk */
1173 		bp->b_resid = bp->b_bcount;
1174 		biodone(bp);
1175 		return (0);
1176 	}
1177 
1178 	slept = 1;
1179 	mutex_enter(&lfs_lock);
1180 	while (slept && fs->lfs_seglock) {
1181 		mutex_exit(&lfs_lock);
1182 		/*
1183 		 * Look through list of intervals.
1184 		 * There will only be intervals to look through
1185 		 * if the cleaner holds the seglock.
1186 		 * Since the cleaner is synchronous, we can trust
1187 		 * the list of intervals to be current.
1188 		 */
1189 		tbn = dbtofsb(fs, bp->b_blkno);
1190 		sn = dtosn(fs, tbn);
1191 		slept = 0;
1192 		for (i = 0; i < fs->lfs_cleanind; i++) {
1193 			if (sn == dtosn(fs, fs->lfs_cleanint[i]) &&
1194 			    tbn >= fs->lfs_cleanint[i]) {
1195 				DLOG((DLOG_CLEAN,
1196 				      "lfs_strategy: ino %d lbn %" PRId64
1197 				      " ind %d sn %d fsb %" PRIx32
1198 				      " given sn %d fsb %" PRIx64 "\n",
1199 				      ip->i_number, bp->b_lblkno, i,
1200 				      dtosn(fs, fs->lfs_cleanint[i]),
1201 				      fs->lfs_cleanint[i], sn, tbn));
1202 				DLOG((DLOG_CLEAN,
1203 				      "lfs_strategy: sleeping on ino %d lbn %"
1204 				      PRId64 "\n", ip->i_number, bp->b_lblkno));
1205 				mutex_enter(&lfs_lock);
1206 				if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) {
1207 					/* Cleaner can't wait for itself */
1208 					mtsleep(&fs->lfs_iocount,
1209 						(PRIBIO + 1) | PNORELOCK,
1210 						"clean2", 0,
1211 						&lfs_lock);
1212 					slept = 1;
1213 					break;
1214 				} else if (fs->lfs_seglock) {
1215 					mtsleep(&fs->lfs_seglock,
1216 						(PRIBIO + 1) | PNORELOCK,
1217 						"clean1", 0,
1218 						&lfs_lock);
1219 					slept = 1;
1220 					break;
1221 				}
1222 				mutex_exit(&lfs_lock);
1223 			}
1224 		}
1225 		mutex_enter(&lfs_lock);
1226 	}
1227 	mutex_exit(&lfs_lock);
1228 
1229 	vp = ip->i_devvp;
1230 	VOP_STRATEGY(vp, bp);
1231 	return (0);
1232 }
1233 
1234 void
1235 lfs_flush_dirops(struct lfs *fs)
1236 {
1237 	struct inode *ip, *nip;
1238 	struct vnode *vp;
1239 	extern int lfs_dostats;
1240 	struct segment *sp;
1241 	int waslocked;
1242 
1243 	ASSERT_MAYBE_SEGLOCK(fs);
1244 	KASSERT(fs->lfs_nadirop == 0);
1245 
1246 	if (fs->lfs_ronly)
1247 		return;
1248 
1249 	mutex_enter(&lfs_lock);
1250 	if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) {
1251 		mutex_exit(&lfs_lock);
1252 		return;
1253 	} else
1254 		mutex_exit(&lfs_lock);
1255 
1256 	if (lfs_dostats)
1257 		++lfs_stats.flush_invoked;
1258 
1259 	/*
1260 	 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops.
1261 	 * Technically this is a checkpoint (the on-disk state is valid)
1262 	 * even though we are leaving out all the file data.
1263 	 */
1264 	lfs_imtime(fs);
1265 	lfs_seglock(fs, SEGM_CKP);
1266 	sp = fs->lfs_sp;
1267 
1268 	/*
1269 	 * lfs_writevnodes, optimized to get dirops out of the way.
1270 	 * Only write dirops, and don't flush files' pages, only
1271 	 * blocks from the directories.
1272 	 *
1273 	 * We don't need to vref these files because they are
1274 	 * dirops and so hold an extra reference until the
1275 	 * segunlock clears them of that status.
1276 	 *
1277 	 * We don't need to check for IN_ADIROP because we know that
1278 	 * no dirops are active.
1279 	 *
1280 	 */
1281 	mutex_enter(&lfs_lock);
1282 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
1283 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
1284 		mutex_exit(&lfs_lock);
1285 		vp = ITOV(ip);
1286 
1287 		KASSERT((ip->i_flag & IN_ADIROP) == 0);
1288 
1289 		/*
1290 		 * All writes to directories come from dirops; all
1291 		 * writes to files' direct blocks go through the page
1292 		 * cache, which we're not touching.  Reads to files
1293 		 * and/or directories will not be affected by writing
1294 		 * directory blocks inodes and file inodes.  So we don't
1295 		 * really need to lock.	 If we don't lock, though,
1296 		 * make sure that we don't clear IN_MODIFIED
1297 		 * unnecessarily.
1298 		 */
1299 		if (vp->v_iflag & VI_XLOCK) {
1300 			mutex_enter(&lfs_lock);
1301 			continue;
1302 		}
1303 		waslocked = VOP_ISLOCKED(vp);
1304 		if (vp->v_type != VREG &&
1305 		    ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) {
1306 			lfs_writefile(fs, sp, vp);
1307 			if (!VPISEMPTY(vp) && !WRITEINPROG(vp) &&
1308 			    !(ip->i_flag & IN_ALLMOD)) {
1309 			    	mutex_enter(&lfs_lock);
1310 				LFS_SET_UINO(ip, IN_MODIFIED);
1311 			    	mutex_exit(&lfs_lock);
1312 			}
1313 		}
1314 		KDASSERT(ip->i_number != LFS_IFILE_INUM);
1315 		(void) lfs_writeinode(fs, sp, ip);
1316 		mutex_enter(&lfs_lock);
1317 		if (waslocked == LK_EXCLOTHER)
1318 			LFS_SET_UINO(ip, IN_MODIFIED);
1319 	}
1320 	mutex_exit(&lfs_lock);
1321 	/* We've written all the dirops there are */
1322 	((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT);
1323 	lfs_finalize_fs_seguse(fs);
1324 	(void) lfs_writeseg(fs, sp);
1325 	lfs_segunlock(fs);
1326 }
1327 
1328 /*
1329  * Flush all vnodes for which the pagedaemon has requested pageouts.
1330  * Skip over any files that are marked VU_DIROP (since lfs_flush_dirop()
1331  * has just run, this would be an error).  If we have to skip a vnode
1332  * for any reason, just skip it; if we have to wait for the cleaner,
1333  * abort.  The writer daemon will call us again later.
1334  */
1335 void
1336 lfs_flush_pchain(struct lfs *fs)
1337 {
1338 	struct inode *ip, *nip;
1339 	struct vnode *vp;
1340 	extern int lfs_dostats;
1341 	struct segment *sp;
1342 	int error;
1343 
1344 	ASSERT_NO_SEGLOCK(fs);
1345 
1346 	if (fs->lfs_ronly)
1347 		return;
1348 
1349 	mutex_enter(&lfs_lock);
1350 	if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) {
1351 		mutex_exit(&lfs_lock);
1352 		return;
1353 	} else
1354 		mutex_exit(&lfs_lock);
1355 
1356 	/* Get dirops out of the way */
1357 	lfs_flush_dirops(fs);
1358 
1359 	if (lfs_dostats)
1360 		++lfs_stats.flush_invoked;
1361 
1362 	/*
1363 	 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts.
1364 	 */
1365 	lfs_imtime(fs);
1366 	lfs_seglock(fs, 0);
1367 	sp = fs->lfs_sp;
1368 
1369 	/*
1370 	 * lfs_writevnodes, optimized to clear pageout requests.
1371 	 * Only write non-dirop files that are in the pageout queue.
1372 	 * We're very conservative about what we write; we want to be
1373 	 * fast and async.
1374 	 */
1375 	mutex_enter(&lfs_lock);
1376     top:
1377 	for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) {
1378 		nip = TAILQ_NEXT(ip, i_lfs_pchain);
1379 		vp = ITOV(ip);
1380 
1381 		if (!(ip->i_flags & IN_PAGING))
1382 			goto top;
1383 
1384 		mutex_enter(&vp->v_interlock);
1385 		if ((vp->v_iflag & VI_XLOCK) || (vp->v_uflag & VU_DIROP) != 0) {
1386 			mutex_exit(&vp->v_interlock);
1387 			continue;
1388 		}
1389 		if (vp->v_type != VREG) {
1390 			mutex_exit(&vp->v_interlock);
1391 			continue;
1392 		}
1393 		if (lfs_vref(vp))
1394 			continue;
1395 		mutex_exit(&lfs_lock);
1396 
1397 		if (VOP_ISLOCKED(vp)) {
1398 			lfs_vunref(vp);
1399 			mutex_enter(&lfs_lock);
1400 			continue;
1401 		}
1402 
1403 		error = lfs_writefile(fs, sp, vp);
1404 		if (!VPISEMPTY(vp) && !WRITEINPROG(vp) &&
1405 		    !(ip->i_flag & IN_ALLMOD)) {
1406 		    	mutex_enter(&lfs_lock);
1407 			LFS_SET_UINO(ip, IN_MODIFIED);
1408 		    	mutex_exit(&lfs_lock);
1409 		}
1410 		KDASSERT(ip->i_number != LFS_IFILE_INUM);
1411 		(void) lfs_writeinode(fs, sp, ip);
1412 
1413 		lfs_vunref(vp);
1414 
1415 		if (error == EAGAIN) {
1416 			lfs_writeseg(fs, sp);
1417 			mutex_enter(&lfs_lock);
1418 			break;
1419 		}
1420 		mutex_enter(&lfs_lock);
1421 	}
1422 	mutex_exit(&lfs_lock);
1423 	(void) lfs_writeseg(fs, sp);
1424 	lfs_segunlock(fs);
1425 }
1426 
1427 /*
1428  * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}.
1429  */
1430 int
1431 lfs_fcntl(void *v)
1432 {
1433 	struct vop_fcntl_args /* {
1434 		struct vnode *a_vp;
1435 		u_long a_command;
1436 		void * a_data;
1437 		int  a_fflag;
1438 		kauth_cred_t a_cred;
1439 	} */ *ap = v;
1440 	struct timeval *tvp;
1441 	BLOCK_INFO *blkiov;
1442 	CLEANERINFO *cip;
1443 	SEGUSE *sup;
1444 	int blkcnt, error, oclean;
1445 	size_t fh_size;
1446 	struct lfs_fcntl_markv blkvp;
1447 	struct lwp *l;
1448 	fsid_t *fsidp;
1449 	struct lfs *fs;
1450 	struct buf *bp;
1451 	fhandle_t *fhp;
1452 	daddr_t off;
1453 
1454 	/* Only respect LFS fcntls on fs root or Ifile */
1455 	if (VTOI(ap->a_vp)->i_number != ROOTINO &&
1456 	    VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) {
1457 		return ufs_fcntl(v);
1458 	}
1459 
1460 	/* Avoid locking a draining lock */
1461 	if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) {
1462 		return ESHUTDOWN;
1463 	}
1464 
1465 	/* LFS control and monitoring fcntls are available only to root */
1466 	l = curlwp;
1467 	if (((ap->a_command & 0xff00) >> 8) == 'L' &&
1468 	    (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER,
1469 					     NULL)) != 0)
1470 		return (error);
1471 
1472 	fs = VTOI(ap->a_vp)->i_lfs;
1473 	fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx;
1474 
1475 	error = 0;
1476 	switch (ap->a_command) {
1477 	    case LFCNSEGWAITALL:
1478 	    case LFCNSEGWAITALL_COMPAT:
1479 		fsidp = NULL;
1480 		/* FALLSTHROUGH */
1481 	    case LFCNSEGWAIT:
1482 	    case LFCNSEGWAIT_COMPAT:
1483 		tvp = (struct timeval *)ap->a_data;
1484 		mutex_enter(&lfs_lock);
1485 		++fs->lfs_sleepers;
1486 		mutex_exit(&lfs_lock);
1487 
1488 		error = lfs_segwait(fsidp, tvp);
1489 
1490 		mutex_enter(&lfs_lock);
1491 		if (--fs->lfs_sleepers == 0)
1492 			wakeup(&fs->lfs_sleepers);
1493 		mutex_exit(&lfs_lock);
1494 		return error;
1495 
1496 	    case LFCNBMAPV:
1497 	    case LFCNMARKV:
1498 		blkvp = *(struct lfs_fcntl_markv *)ap->a_data;
1499 
1500 		blkcnt = blkvp.blkcnt;
1501 		if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT)
1502 			return (EINVAL);
1503 		blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV);
1504 		if ((error = copyin(blkvp.blkiov, blkiov,
1505 		     blkcnt * sizeof(BLOCK_INFO))) != 0) {
1506 			lfs_free(fs, blkiov, LFS_NB_BLKIOV);
1507 			return error;
1508 		}
1509 
1510 		mutex_enter(&lfs_lock);
1511 		++fs->lfs_sleepers;
1512 		mutex_exit(&lfs_lock);
1513 		if (ap->a_command == LFCNBMAPV)
1514 			error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt);
1515 		else /* LFCNMARKV */
1516 			error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt);
1517 		if (error == 0)
1518 			error = copyout(blkiov, blkvp.blkiov,
1519 					blkcnt * sizeof(BLOCK_INFO));
1520 		mutex_enter(&lfs_lock);
1521 		if (--fs->lfs_sleepers == 0)
1522 			wakeup(&fs->lfs_sleepers);
1523 		mutex_exit(&lfs_lock);
1524 		lfs_free(fs, blkiov, LFS_NB_BLKIOV);
1525 		return error;
1526 
1527 	    case LFCNRECLAIM:
1528 		/*
1529 		 * Flush dirops and write Ifile, allowing empty segments
1530 		 * to be immediately reclaimed.
1531 		 */
1532 		lfs_writer_enter(fs, "pndirop");
1533 		off = fs->lfs_offset;
1534 		lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP);
1535 		lfs_flush_dirops(fs);
1536 		LFS_CLEANERINFO(cip, fs, bp);
1537 		oclean = cip->clean;
1538 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
1539 		lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP);
1540 		fs->lfs_sp->seg_flags |= SEGM_PROT;
1541 		lfs_segunlock(fs);
1542 		lfs_writer_leave(fs);
1543 
1544 #ifdef DEBUG
1545 		LFS_CLEANERINFO(cip, fs, bp);
1546 		DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64
1547 		      " blocks, cleaned %" PRId32 " segments (activesb %d)\n",
1548 		      fs->lfs_offset - off, cip->clean - oclean,
1549 		      fs->lfs_activesb));
1550 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 0);
1551 #endif
1552 
1553 		return 0;
1554 
1555 #ifdef COMPAT_30
1556 	    case LFCNIFILEFH_COMPAT:
1557 		/* Return the filehandle of the Ifile */
1558 		if ((error = kauth_authorize_generic(l->l_cred,
1559 		    KAUTH_GENERIC_ISSUSER, NULL)) != 0)
1560 			return (error);
1561 		fhp = (struct fhandle *)ap->a_data;
1562 		fhp->fh_fsid = *fsidp;
1563 		fh_size = 16;	/* former VFS_MAXFIDSIZ */
1564 		return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size);
1565 #endif
1566 
1567 	    case LFCNIFILEFH_COMPAT2:
1568 	    case LFCNIFILEFH:
1569 		/* Return the filehandle of the Ifile */
1570 		fhp = (struct fhandle *)ap->a_data;
1571 		fhp->fh_fsid = *fsidp;
1572 		fh_size = sizeof(struct lfs_fhandle) -
1573 		    offsetof(fhandle_t, fh_fid);
1574 		return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size);
1575 
1576 	    case LFCNREWIND:
1577 		/* Move lfs_offset to the lowest-numbered segment */
1578 		return lfs_rewind(fs, *(int *)ap->a_data);
1579 
1580 	    case LFCNINVAL:
1581 		/* Mark a segment SEGUSE_INVAL */
1582 		LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp);
1583 		if (sup->su_nbytes > 0) {
1584 			brelse(bp, 0);
1585 			lfs_unset_inval_all(fs);
1586 			return EBUSY;
1587 		}
1588 		sup->su_flags |= SEGUSE_INVAL;
1589 		VOP_BWRITE(bp);
1590 		return 0;
1591 
1592 	    case LFCNRESIZE:
1593 		/* Resize the filesystem */
1594 		return lfs_resize_fs(fs, *(int *)ap->a_data);
1595 
1596 	    case LFCNWRAPSTOP:
1597 	    case LFCNWRAPSTOP_COMPAT:
1598 		/*
1599 		 * Hold lfs_newseg at segment 0; if requested, sleep until
1600 		 * the filesystem wraps around.  To support external agents
1601 		 * (dump, fsck-based regression test) that need to look at
1602 		 * a snapshot of the filesystem, without necessarily
1603 		 * requiring that all fs activity stops.
1604 		 */
1605 		if (fs->lfs_stoplwp == curlwp)
1606 			return EALREADY;
1607 
1608 		mutex_enter(&lfs_lock);
1609 		while (fs->lfs_stoplwp != NULL)
1610 			cv_wait(&fs->lfs_stopcv, &lfs_lock);
1611 		fs->lfs_stoplwp = curlwp;
1612 		if (fs->lfs_nowrap == 0)
1613 			log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt);
1614 		++fs->lfs_nowrap;
1615 		if (*(int *)ap->a_data == 1 ||
1616 		    ap->a_command == LFCNWRAPSTOP_COMPAT) {
1617 			log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n");
1618 			error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER,
1619 				"segwrap", 0, &lfs_lock);
1620 			log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n");
1621 			if (error) {
1622 				lfs_wrapgo(fs, VTOI(ap->a_vp), 0);
1623 			}
1624 		}
1625 		mutex_exit(&lfs_lock);
1626 		return 0;
1627 
1628 	    case LFCNWRAPGO:
1629 	    case LFCNWRAPGO_COMPAT:
1630 		/*
1631 		 * Having done its work, the agent wakes up the writer.
1632 		 * If the argument is 1, it sleeps until a new segment
1633 		 * is selected.
1634 		 */
1635 		mutex_enter(&lfs_lock);
1636 		error = lfs_wrapgo(fs, VTOI(ap->a_vp),
1637 				   (ap->a_command == LFCNWRAPGO_COMPAT ? 1 :
1638 				    *((int *)ap->a_data)));
1639 		mutex_exit(&lfs_lock);
1640 		return error;
1641 
1642 	    case LFCNWRAPPASS:
1643 		if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT))
1644 			return EALREADY;
1645 		mutex_enter(&lfs_lock);
1646 		if (fs->lfs_stoplwp != curlwp) {
1647 			mutex_exit(&lfs_lock);
1648 			return EALREADY;
1649 		}
1650 		if (fs->lfs_nowrap == 0) {
1651 			mutex_exit(&lfs_lock);
1652 			return EBUSY;
1653 		}
1654 		fs->lfs_wrappass = 1;
1655 		wakeup(&fs->lfs_wrappass);
1656 		/* Wait for the log to wrap, if asked */
1657 		if (*(int *)ap->a_data) {
1658 			mutex_enter(&ap->a_vp->v_interlock);
1659 			lfs_vref(ap->a_vp);
1660 			VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT;
1661 			log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n");
1662 			error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER,
1663 				"segwrap", 0, &lfs_lock);
1664 			log(LOG_NOTICE, "LFCNPASS done waiting\n");
1665 			VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT;
1666 			lfs_vunref(ap->a_vp);
1667 		}
1668 		mutex_exit(&lfs_lock);
1669 		return error;
1670 
1671 	    case LFCNWRAPSTATUS:
1672 		mutex_enter(&lfs_lock);
1673 		*(int *)ap->a_data = fs->lfs_wrapstatus;
1674 		mutex_exit(&lfs_lock);
1675 		return 0;
1676 
1677 	    default:
1678 		return ufs_fcntl(v);
1679 	}
1680 	return 0;
1681 }
1682 
1683 int
1684 lfs_getpages(void *v)
1685 {
1686 	struct vop_getpages_args /* {
1687 		struct vnode *a_vp;
1688 		voff_t a_offset;
1689 		struct vm_page **a_m;
1690 		int *a_count;
1691 		int a_centeridx;
1692 		vm_prot_t a_access_type;
1693 		int a_advice;
1694 		int a_flags;
1695 	} */ *ap = v;
1696 
1697 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM &&
1698 	    (ap->a_access_type & VM_PROT_WRITE) != 0) {
1699 		return EPERM;
1700 	}
1701 	if ((ap->a_access_type & VM_PROT_WRITE) != 0) {
1702 		mutex_enter(&lfs_lock);
1703 		LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED);
1704 		mutex_exit(&lfs_lock);
1705 	}
1706 
1707 	/*
1708 	 * we're relying on the fact that genfs_getpages() always read in
1709 	 * entire filesystem blocks.
1710 	 */
1711 	return genfs_getpages(v);
1712 }
1713 
1714 /*
1715  * Wait for a page to become unbusy, possibly printing diagnostic messages
1716  * as well.
1717  *
1718  * Called with vp->v_interlock held; return with it held.
1719  */
1720 static void
1721 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label)
1722 {
1723 	if ((pg->flags & PG_BUSY) == 0)
1724 		return;		/* Nothing to wait for! */
1725 
1726 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN)
1727 	static struct vm_page *lastpg;
1728 
1729 	if (label != NULL && pg != lastpg) {
1730 		if (pg->owner_tag) {
1731 			printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n",
1732 			       curproc->p_pid, curlwp->l_lid, label,
1733 			       pg, pg->owner, pg->lowner, pg->owner_tag);
1734 		} else {
1735 			printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n",
1736 			       curproc->p_pid, curlwp->l_lid, label, pg);
1737 		}
1738 	}
1739 	lastpg = pg;
1740 #endif
1741 
1742 	pg->flags |= PG_WANTED;
1743 	UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, "lfsput", 0);
1744 	mutex_enter(&vp->v_interlock);
1745 }
1746 
1747 /*
1748  * This routine is called by lfs_putpages() when it can't complete the
1749  * write because a page is busy.  This means that either (1) someone,
1750  * possibly the pagedaemon, is looking at this page, and will give it up
1751  * presently; or (2) we ourselves are holding the page busy in the
1752  * process of being written (either gathered or actually on its way to
1753  * disk).  We don't need to give up the segment lock, but we might need
1754  * to call lfs_writeseg() to expedite the page's journey to disk.
1755  *
1756  * Called with vp->v_interlock held; return with it held.
1757  */
1758 /* #define BUSYWAIT */
1759 static void
1760 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg,
1761 	       int seglocked, const char *label)
1762 {
1763 #ifndef BUSYWAIT
1764 	struct inode *ip = VTOI(vp);
1765 	struct segment *sp = fs->lfs_sp;
1766 	int count = 0;
1767 
1768 	if (pg == NULL)
1769 		return;
1770 
1771 	while (pg->flags & PG_BUSY) {
1772 		mutex_exit(&vp->v_interlock);
1773 		if (sp->cbpp - sp->bpp > 1) {
1774 			/* Write gathered pages */
1775 			lfs_updatemeta(sp);
1776 			lfs_release_finfo(fs);
1777 			(void) lfs_writeseg(fs, sp);
1778 
1779 			/*
1780 			 * Reinitialize FIP
1781 			 */
1782 			KASSERT(sp->vp == vp);
1783 			lfs_acquire_finfo(fs, ip->i_number,
1784 					  ip->i_gen);
1785 		}
1786 		++count;
1787 		mutex_enter(&vp->v_interlock);
1788 		wait_for_page(vp, pg, label);
1789 	}
1790 	if (label != NULL && count > 1)
1791 		printf("lfs_putpages[%d]: %s: %sn = %d\n", curproc->p_pid,
1792 		       label, (count > 0 ? "looping, " : ""), count);
1793 #else
1794 	preempt(1);
1795 #endif
1796 }
1797 
1798 /*
1799  * Make sure that for all pages in every block in the given range,
1800  * either all are dirty or all are clean.  If any of the pages
1801  * we've seen so far are dirty, put the vnode on the paging chain,
1802  * and mark it IN_PAGING.
1803  *
1804  * If checkfirst != 0, don't check all the pages but return at the
1805  * first dirty page.
1806  */
1807 static int
1808 check_dirty(struct lfs *fs, struct vnode *vp,
1809 	    off_t startoffset, off_t endoffset, off_t blkeof,
1810 	    int flags, int checkfirst, struct vm_page **pgp)
1811 {
1812 	int by_list;
1813 	struct vm_page *curpg = NULL; /* XXX: gcc */
1814 	struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg;
1815 	off_t soff = 0; /* XXX: gcc */
1816 	voff_t off;
1817 	int i;
1818 	int nonexistent;
1819 	int any_dirty;	/* number of dirty pages */
1820 	int dirty;	/* number of dirty pages in a block */
1821 	int tdirty;
1822 	int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT;
1823 	int pagedaemon = (curlwp == uvm.pagedaemon_lwp);
1824 
1825 	ASSERT_MAYBE_SEGLOCK(fs);
1826   top:
1827 	by_list = (vp->v_uobj.uo_npages <=
1828 		   ((endoffset - startoffset) >> PAGE_SHIFT) *
1829 		   UVM_PAGE_HASH_PENALTY);
1830 	any_dirty = 0;
1831 
1832 	if (by_list) {
1833 		curpg = TAILQ_FIRST(&vp->v_uobj.memq);
1834 	} else {
1835 		soff = startoffset;
1836 	}
1837 	while (by_list || soff < MIN(blkeof, endoffset)) {
1838 		if (by_list) {
1839 			/*
1840 			 * Find the first page in a block.  Skip
1841 			 * blocks outside our area of interest or beyond
1842 			 * the end of file.
1843 			 */
1844 			if (pages_per_block > 1) {
1845 				while (curpg &&
1846 				       ((curpg->offset & fs->lfs_bmask) ||
1847 					curpg->offset >= vp->v_size ||
1848 					curpg->offset >= endoffset))
1849 					curpg = TAILQ_NEXT(curpg, listq);
1850 			}
1851 			if (curpg == NULL)
1852 				break;
1853 			soff = curpg->offset;
1854 		}
1855 
1856 		/*
1857 		 * Mark all pages in extended range busy; find out if any
1858 		 * of them are dirty.
1859 		 */
1860 		nonexistent = dirty = 0;
1861 		for (i = 0; i == 0 || i < pages_per_block; i++) {
1862 			if (by_list && pages_per_block <= 1) {
1863 				pgs[i] = pg = curpg;
1864 			} else {
1865 				off = soff + (i << PAGE_SHIFT);
1866 				pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off);
1867 				if (pg == NULL) {
1868 					++nonexistent;
1869 					continue;
1870 				}
1871 			}
1872 			KASSERT(pg != NULL);
1873 
1874 			/*
1875 			 * If we're holding the segment lock, we can deadlock
1876 			 * against a process that has our page and is waiting
1877 			 * for the cleaner, while the cleaner waits for the
1878 			 * segment lock.  Just bail in that case.
1879 			 */
1880 			if ((pg->flags & PG_BUSY) &&
1881 			    (pagedaemon || LFS_SEGLOCK_HELD(fs))) {
1882 				if (i > 0)
1883 					uvm_page_unbusy(pgs, i);
1884 				DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n"));
1885 				if (pgp)
1886 					*pgp = pg;
1887 				return -1;
1888 			}
1889 
1890 			while (pg->flags & PG_BUSY) {
1891 				wait_for_page(vp, pg, NULL);
1892 				if (i > 0)
1893 					uvm_page_unbusy(pgs, i);
1894 				goto top;
1895 			}
1896 			pg->flags |= PG_BUSY;
1897 			UVM_PAGE_OWN(pg, "lfs_putpages");
1898 
1899 			pmap_page_protect(pg, VM_PROT_NONE);
1900 			tdirty = (pmap_clear_modify(pg) ||
1901 				  (pg->flags & PG_CLEAN) == 0);
1902 			dirty += tdirty;
1903 		}
1904 		if (pages_per_block > 0 && nonexistent >= pages_per_block) {
1905 			if (by_list) {
1906 				curpg = TAILQ_NEXT(curpg, listq);
1907 			} else {
1908 				soff += fs->lfs_bsize;
1909 			}
1910 			continue;
1911 		}
1912 
1913 		any_dirty += dirty;
1914 		KASSERT(nonexistent == 0);
1915 
1916 		/*
1917 		 * If any are dirty make all dirty; unbusy them,
1918 		 * but if we were asked to clean, wire them so that
1919 		 * the pagedaemon doesn't bother us about them while
1920 		 * they're on their way to disk.
1921 		 */
1922 		for (i = 0; i == 0 || i < pages_per_block; i++) {
1923 			pg = pgs[i];
1924 			KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI)));
1925 			if (dirty) {
1926 				pg->flags &= ~PG_CLEAN;
1927 				if (flags & PGO_FREE) {
1928 					/*
1929 					 * Wire the page so that
1930 					 * pdaemon doesn't see it again.
1931 					 */
1932 					mutex_enter(&uvm_pageqlock);
1933 					uvm_pagewire(pg);
1934 					mutex_exit(&uvm_pageqlock);
1935 
1936 					/* Suspended write flag */
1937 					pg->flags |= PG_DELWRI;
1938 				}
1939 			}
1940 			if (pg->flags & PG_WANTED)
1941 				wakeup(pg);
1942 			pg->flags &= ~(PG_WANTED|PG_BUSY);
1943 			UVM_PAGE_OWN(pg, NULL);
1944 		}
1945 
1946 		if (checkfirst && any_dirty)
1947 			break;
1948 
1949 		if (by_list) {
1950 			curpg = TAILQ_NEXT(curpg, listq);
1951 		} else {
1952 			soff += MAX(PAGE_SIZE, fs->lfs_bsize);
1953 		}
1954 	}
1955 
1956 	return any_dirty;
1957 }
1958 
1959 /*
1960  * lfs_putpages functions like genfs_putpages except that
1961  *
1962  * (1) It needs to bounds-check the incoming requests to ensure that
1963  *     they are block-aligned; if they are not, expand the range and
1964  *     do the right thing in case, e.g., the requested range is clean
1965  *     but the expanded range is dirty.
1966  *
1967  * (2) It needs to explicitly send blocks to be written when it is done.
1968  *     If VOP_PUTPAGES is called without the seglock held, we simply take
1969  *     the seglock and let lfs_segunlock wait for us.
1970  *     XXX There might be a bad situation if we have to flush a vnode while
1971  *     XXX lfs_markv is in operation.  As of this writing we panic in this
1972  *     XXX case.
1973  *
1974  * Assumptions:
1975  *
1976  * (1) The caller does not hold any pages in this vnode busy.  If it does,
1977  *     there is a danger that when we expand the page range and busy the
1978  *     pages we will deadlock.
1979  *
1980  * (2) We are called with vp->v_interlock held; we must return with it
1981  *     released.
1982  *
1983  * (3) We don't absolutely have to free pages right away, provided that
1984  *     the request does not have PGO_SYNCIO.  When the pagedaemon gives
1985  *     us a request with PGO_FREE, we take the pages out of the paging
1986  *     queue and wake up the writer, which will handle freeing them for us.
1987  *
1988  *     We ensure that for any filesystem block, all pages for that
1989  *     block are either resident or not, even if those pages are higher
1990  *     than EOF; that means that we will be getting requests to free
1991  *     "unused" pages above EOF all the time, and should ignore them.
1992  *
1993  * (4) If we are called with PGO_LOCKED, the finfo array we are to write
1994  *     into has been set up for us by lfs_writefile.  If not, we will
1995  *     have to handle allocating and/or freeing an finfo entry.
1996  *
1997  * XXX note that we're (ab)using PGO_LOCKED as "seglock held".
1998  */
1999 
2000 /* How many times to loop before we should start to worry */
2001 #define TOOMANY 4
2002 
2003 int
2004 lfs_putpages(void *v)
2005 {
2006 	int error;
2007 	struct vop_putpages_args /* {
2008 		struct vnode *a_vp;
2009 		voff_t a_offlo;
2010 		voff_t a_offhi;
2011 		int a_flags;
2012 	} */ *ap = v;
2013 	struct vnode *vp;
2014 	struct inode *ip;
2015 	struct lfs *fs;
2016 	struct segment *sp;
2017 	off_t origoffset, startoffset, endoffset, origendoffset, blkeof;
2018 	off_t off, max_endoffset;
2019 	bool seglocked, sync, pagedaemon;
2020 	struct vm_page *pg, *busypg;
2021 	UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist);
2022 #ifdef DEBUG
2023 	int debug_n_again, debug_n_dirtyclean;
2024 #endif
2025 
2026 	vp = ap->a_vp;
2027 	ip = VTOI(vp);
2028 	fs = ip->i_lfs;
2029 	sync = (ap->a_flags & PGO_SYNCIO) != 0;
2030 	pagedaemon = (curlwp == uvm.pagedaemon_lwp);
2031 
2032 	/* Putpages does nothing for metadata. */
2033 	if (vp == fs->lfs_ivnode || vp->v_type != VREG) {
2034 		mutex_exit(&vp->v_interlock);
2035 		return 0;
2036 	}
2037 
2038 	/*
2039 	 * If there are no pages, don't do anything.
2040 	 */
2041 	if (vp->v_uobj.uo_npages == 0) {
2042 		if (TAILQ_EMPTY(&vp->v_uobj.memq) &&
2043 		    (vp->v_iflag & VI_ONWORKLST) &&
2044 		    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
2045 			vp->v_iflag &= ~VI_WRMAPDIRTY;
2046 			vn_syncer_remove_from_worklist(vp);
2047 		}
2048 		mutex_exit(&vp->v_interlock);
2049 
2050 		/* Remove us from paging queue, if we were on it */
2051 		mutex_enter(&lfs_lock);
2052 		if (ip->i_flags & IN_PAGING) {
2053 			ip->i_flags &= ~IN_PAGING;
2054 			TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain);
2055 		}
2056 		mutex_exit(&lfs_lock);
2057 		return 0;
2058 	}
2059 
2060 	blkeof = blkroundup(fs, ip->i_size);
2061 
2062 	/*
2063 	 * Ignore requests to free pages past EOF but in the same block
2064 	 * as EOF, unless the request is synchronous.  (If the request is
2065 	 * sync, it comes from lfs_truncate.)
2066 	 * XXXUBC Make these pages look "active" so the pagedaemon won't
2067 	 * XXXUBC bother us with them again.
2068 	 */
2069 	if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) {
2070 		origoffset = ap->a_offlo;
2071 		for (off = origoffset; off < blkeof; off += fs->lfs_bsize) {
2072 			pg = uvm_pagelookup(&vp->v_uobj, off);
2073 			KASSERT(pg != NULL);
2074 			while (pg->flags & PG_BUSY) {
2075 				pg->flags |= PG_WANTED;
2076 				UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0,
2077 						    "lfsput2", 0);
2078 				mutex_enter(&vp->v_interlock);
2079 			}
2080 			mutex_enter(&uvm_pageqlock);
2081 			uvm_pageactivate(pg);
2082 			mutex_exit(&uvm_pageqlock);
2083 		}
2084 		ap->a_offlo = blkeof;
2085 		if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) {
2086 			mutex_exit(&vp->v_interlock);
2087 			return 0;
2088 		}
2089 	}
2090 
2091 	/*
2092 	 * Extend page range to start and end at block boundaries.
2093 	 * (For the purposes of VOP_PUTPAGES, fragments don't exist.)
2094 	 */
2095 	origoffset = ap->a_offlo;
2096 	origendoffset = ap->a_offhi;
2097 	startoffset = origoffset & ~(fs->lfs_bmask);
2098 	max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift)
2099 					       << fs->lfs_bshift;
2100 
2101 	if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) {
2102 		endoffset = max_endoffset;
2103 		origendoffset = endoffset;
2104 	} else {
2105 		origendoffset = round_page(ap->a_offhi);
2106 		endoffset = round_page(blkroundup(fs, origendoffset));
2107 	}
2108 
2109 	KASSERT(startoffset > 0 || endoffset >= startoffset);
2110 	if (startoffset == endoffset) {
2111 		/* Nothing to do, why were we called? */
2112 		mutex_exit(&vp->v_interlock);
2113 		DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %"
2114 		      PRId64 "\n", startoffset));
2115 		return 0;
2116 	}
2117 
2118 	ap->a_offlo = startoffset;
2119 	ap->a_offhi = endoffset;
2120 
2121 	/*
2122 	 * If not cleaning, just send the pages through genfs_putpages
2123 	 * to be returned to the pool.
2124 	 */
2125 	if (!(ap->a_flags & PGO_CLEANIT))
2126 		return genfs_putpages(v);
2127 
2128 	/* Set PGO_BUSYFAIL to avoid deadlocks */
2129 	ap->a_flags |= PGO_BUSYFAIL;
2130 
2131 	/*
2132 	 * Likewise, if we are asked to clean but the pages are not
2133 	 * dirty, we can just free them using genfs_putpages.
2134 	 */
2135 #ifdef DEBUG
2136 	debug_n_dirtyclean = 0;
2137 #endif
2138 	do {
2139 		int r;
2140 
2141 		/* Count the number of dirty pages */
2142 		r = check_dirty(fs, vp, startoffset, endoffset, blkeof,
2143 				ap->a_flags, 1, NULL);
2144 		if (r < 0) {
2145 			/* Pages are busy with another process */
2146 			mutex_exit(&vp->v_interlock);
2147 			return EDEADLK;
2148 		}
2149 		if (r > 0) /* Some pages are dirty */
2150 			break;
2151 
2152 		/*
2153 		 * Sometimes pages are dirtied between the time that
2154 		 * we check and the time we try to clean them.
2155 		 * Instruct lfs_gop_write to return EDEADLK in this case
2156 		 * so we can write them properly.
2157 		 */
2158 		ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE;
2159 		r = genfs_do_putpages(vp, startoffset, endoffset,
2160 				       ap->a_flags, &busypg);
2161 		ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE;
2162 		if (r != EDEADLK)
2163 			return r;
2164 
2165 		/* One of the pages was busy.  Start over. */
2166 		mutex_enter(&vp->v_interlock);
2167 		wait_for_page(vp, busypg, "dirtyclean");
2168 #ifdef DEBUG
2169 		++debug_n_dirtyclean;
2170 #endif
2171 	} while(1);
2172 
2173 #ifdef DEBUG
2174 	if (debug_n_dirtyclean > TOOMANY)
2175 		printf("lfs_putpages: dirtyclean: looping, n = %d\n",
2176 		       debug_n_dirtyclean);
2177 #endif
2178 
2179 	/*
2180 	 * Dirty and asked to clean.
2181 	 *
2182 	 * Pagedaemon can't actually write LFS pages; wake up
2183 	 * the writer to take care of that.  The writer will
2184 	 * notice the pager inode queue and act on that.
2185 	 */
2186 	if (pagedaemon) {
2187 		mutex_enter(&lfs_lock);
2188 		if (!(ip->i_flags & IN_PAGING)) {
2189 			ip->i_flags |= IN_PAGING;
2190 			TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain);
2191 		}
2192 		wakeup(&lfs_writer_daemon);
2193 		mutex_exit(&lfs_lock);
2194 		mutex_exit(&vp->v_interlock);
2195 		preempt();
2196 		return EWOULDBLOCK;
2197 	}
2198 
2199 	/*
2200 	 * If this is a file created in a recent dirop, we can't flush its
2201 	 * inode until the dirop is complete.  Drain dirops, then flush the
2202 	 * filesystem (taking care of any other pending dirops while we're
2203 	 * at it).
2204 	 */
2205 	if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT &&
2206 	    (vp->v_uflag & VU_DIROP)) {
2207 		int locked;
2208 
2209 		DLOG((DLOG_PAGE, "lfs_putpages: flushing VU_DIROP\n"));
2210 		locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
2211 		mutex_exit(&vp->v_interlock);
2212 		lfs_writer_enter(fs, "ppdirop");
2213 		if (locked)
2214 			VOP_UNLOCK(vp, 0); /* XXX why? */
2215 
2216 		mutex_enter(&lfs_lock);
2217 		lfs_flush_fs(fs, sync ? SEGM_SYNC : 0);
2218 		mutex_exit(&lfs_lock);
2219 
2220 		mutex_enter(&vp->v_interlock);
2221 		if (locked) {
2222 			VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2223 			mutex_enter(&vp->v_interlock);
2224 		}
2225 		lfs_writer_leave(fs);
2226 
2227 		/* XXX the flush should have taken care of this one too! */
2228 	}
2229 
2230 	/*
2231 	 * This is it.	We are going to write some pages.  From here on
2232 	 * down it's all just mechanics.
2233 	 *
2234 	 * Don't let genfs_putpages wait; lfs_segunlock will wait for us.
2235 	 */
2236 	ap->a_flags &= ~PGO_SYNCIO;
2237 
2238 	/*
2239 	 * If we've already got the seglock, flush the node and return.
2240 	 * The FIP has already been set up for us by lfs_writefile,
2241 	 * and FIP cleanup and lfs_updatemeta will also be done there,
2242 	 * unless genfs_putpages returns EDEADLK; then we must flush
2243 	 * what we have, and correct FIP and segment header accounting.
2244 	 */
2245   get_seglock:
2246 	/*
2247 	 * If we are not called with the segment locked, lock it.
2248 	 * Account for a new FIP in the segment header, and set sp->vp.
2249 	 * (This should duplicate the setup at the top of lfs_writefile().)
2250 	 */
2251 	seglocked = (ap->a_flags & PGO_LOCKED) != 0;
2252 	if (!seglocked) {
2253 		mutex_exit(&vp->v_interlock);
2254 		error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0));
2255 		if (error != 0)
2256 			return error;
2257 		mutex_enter(&vp->v_interlock);
2258 		lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
2259 	}
2260 	sp = fs->lfs_sp;
2261 	KASSERT(sp->vp == NULL);
2262 	sp->vp = vp;
2263 
2264 	/*
2265 	 * Ensure that the partial segment is marked SS_DIROP if this
2266 	 * vnode is a DIROP.
2267 	 */
2268 	if (!seglocked && vp->v_uflag & VU_DIROP)
2269 		((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT);
2270 
2271 	/*
2272 	 * Loop over genfs_putpages until all pages are gathered.
2273 	 * genfs_putpages() drops the interlock, so reacquire it if necessary.
2274 	 * Whenever we lose the interlock we have to rerun check_dirty, as
2275 	 * well, since more pages might have been dirtied in our absence.
2276 	 */
2277 #ifdef DEBUG
2278 	debug_n_again = 0;
2279 #endif
2280 	do {
2281 		busypg = NULL;
2282 		if (check_dirty(fs, vp, startoffset, endoffset, blkeof,
2283 				ap->a_flags, 0, &busypg) < 0) {
2284 			mutex_exit(&vp->v_interlock);
2285 
2286 			mutex_enter(&vp->v_interlock);
2287 			write_and_wait(fs, vp, busypg, seglocked, NULL);
2288 			if (!seglocked) {
2289 				lfs_release_finfo(fs);
2290 				lfs_segunlock(fs);
2291 			}
2292 			sp->vp = NULL;
2293 			goto get_seglock;
2294 		}
2295 
2296 		busypg = NULL;
2297 		error = genfs_do_putpages(vp, startoffset, endoffset,
2298 					   ap->a_flags, &busypg);
2299 
2300 		if (error == EDEADLK || error == EAGAIN) {
2301 			DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned"
2302 			      " %d ino %d off %x (seg %d)\n", error,
2303 			      ip->i_number, fs->lfs_offset,
2304 			      dtosn(fs, fs->lfs_offset)));
2305 
2306 			mutex_enter(&vp->v_interlock);
2307 			write_and_wait(fs, vp, busypg, seglocked, "again");
2308 		}
2309 #ifdef DEBUG
2310 		++debug_n_again;
2311 #endif
2312 	} while (error == EDEADLK);
2313 #ifdef DEBUG
2314 	if (debug_n_again > TOOMANY)
2315 		printf("lfs_putpages: again: looping, n = %d\n", debug_n_again);
2316 #endif
2317 
2318 	KASSERT(sp != NULL && sp->vp == vp);
2319 	if (!seglocked) {
2320 		sp->vp = NULL;
2321 
2322 		/* Write indirect blocks as well */
2323 		lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir);
2324 		lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir);
2325 		lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir);
2326 
2327 		KASSERT(sp->vp == NULL);
2328 		sp->vp = vp;
2329 	}
2330 
2331 	/*
2332 	 * Blocks are now gathered into a segment waiting to be written.
2333 	 * All that's left to do is update metadata, and write them.
2334 	 */
2335 	lfs_updatemeta(sp);
2336 	KASSERT(sp->vp == vp);
2337 	sp->vp = NULL;
2338 
2339 	/*
2340 	 * If we were called from lfs_writefile, we don't need to clean up
2341 	 * the FIP or unlock the segment lock.	We're done.
2342 	 */
2343 	if (seglocked)
2344 		return error;
2345 
2346 	/* Clean up FIP and send it to disk. */
2347 	lfs_release_finfo(fs);
2348 	lfs_writeseg(fs, fs->lfs_sp);
2349 
2350 	/*
2351 	 * Remove us from paging queue if we wrote all our pages.
2352 	 */
2353 	if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) {
2354 		mutex_enter(&lfs_lock);
2355 		if (ip->i_flags & IN_PAGING) {
2356 			ip->i_flags &= ~IN_PAGING;
2357 			TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain);
2358 		}
2359 		mutex_exit(&lfs_lock);
2360 	}
2361 
2362 	/*
2363 	 * XXX - with the malloc/copy writeseg, the pages are freed by now
2364 	 * even if we don't wait (e.g. if we hold a nested lock).  This
2365 	 * will not be true if we stop using malloc/copy.
2366 	 */
2367 	KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT);
2368 	lfs_segunlock(fs);
2369 
2370 	/*
2371 	 * Wait for v_numoutput to drop to zero.  The seglock should
2372 	 * take care of this, but there is a slight possibility that
2373 	 * aiodoned might not have got around to our buffers yet.
2374 	 */
2375 	if (sync) {
2376 		mutex_enter(&vp->v_interlock);
2377 		while (vp->v_numoutput > 0) {
2378 			DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on"
2379 			      " num %d\n", ip->i_number, vp->v_numoutput));
2380 			cv_wait(&vp->v_cv, &vp->v_interlock);
2381 		}
2382 		mutex_exit(&vp->v_interlock);
2383 	}
2384 	return error;
2385 }
2386 
2387 /*
2388  * Return the last logical file offset that should be written for this file
2389  * if we're doing a write that ends at "size".	If writing, we need to know
2390  * about sizes on disk, i.e. fragments if there are any; if reading, we need
2391  * to know about entire blocks.
2392  */
2393 void
2394 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags)
2395 {
2396 	struct inode *ip = VTOI(vp);
2397 	struct lfs *fs = ip->i_lfs;
2398 	daddr_t olbn, nlbn;
2399 
2400 	olbn = lblkno(fs, ip->i_size);
2401 	nlbn = lblkno(fs, size);
2402 	if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) {
2403 		*eobp = fragroundup(fs, size);
2404 	} else {
2405 		*eobp = blkroundup(fs, size);
2406 	}
2407 }
2408 
2409 #ifdef DEBUG
2410 void lfs_dump_vop(void *);
2411 
2412 void
2413 lfs_dump_vop(void *v)
2414 {
2415 	struct vop_putpages_args /* {
2416 		struct vnode *a_vp;
2417 		voff_t a_offlo;
2418 		voff_t a_offhi;
2419 		int a_flags;
2420 	} */ *ap = v;
2421 
2422 #ifdef DDB
2423 	vfs_vnode_print(ap->a_vp, 0, printf);
2424 #endif
2425 	lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din);
2426 }
2427 #endif
2428 
2429 int
2430 lfs_mmap(void *v)
2431 {
2432 	struct vop_mmap_args /* {
2433 		const struct vnodeop_desc *a_desc;
2434 		struct vnode *a_vp;
2435 		vm_prot_t a_prot;
2436 		kauth_cred_t a_cred;
2437 	} */ *ap = v;
2438 
2439 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM)
2440 		return EOPNOTSUPP;
2441 	return ufs_mmap(v);
2442 }
2443