xref: /netbsd-src/sys/ufs/lfs/lfs_vnops.c (revision 413d532bcc3f62d122e56d92e13ac64825a40baf)
1 /*	$NetBSD: lfs_vnops.c,v 1.262 2014/03/24 13:42:40 hannken 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 /*
32  * Copyright (c) 1986, 1989, 1991, 1993, 1995
33  *	The Regents of the University of California.  All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. Neither the name of the University nor the names of its contributors
44  *    may be used to endorse or promote products derived from this software
45  *    without specific prior written permission.
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
57  * SUCH DAMAGE.
58  *
59  *	@(#)lfs_vnops.c	8.13 (Berkeley) 6/10/95
60  */
61 
62 #include <sys/cdefs.h>
63 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.262 2014/03/24 13:42:40 hannken Exp $");
64 
65 #ifdef _KERNEL_OPT
66 #include "opt_compat_netbsd.h"
67 #include "opt_uvm_page_trkown.h"
68 #endif
69 
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/namei.h>
73 #include <sys/resourcevar.h>
74 #include <sys/kernel.h>
75 #include <sys/file.h>
76 #include <sys/stat.h>
77 #include <sys/buf.h>
78 #include <sys/proc.h>
79 #include <sys/mount.h>
80 #include <sys/vnode.h>
81 #include <sys/pool.h>
82 #include <sys/signalvar.h>
83 #include <sys/kauth.h>
84 #include <sys/syslog.h>
85 #include <sys/fstrans.h>
86 
87 #include <miscfs/fifofs/fifo.h>
88 #include <miscfs/genfs/genfs.h>
89 #include <miscfs/specfs/specdev.h>
90 
91 #include <ufs/lfs/ulfs_inode.h>
92 #include <ufs/lfs/ulfsmount.h>
93 #include <ufs/lfs/ulfs_bswap.h>
94 #include <ufs/lfs/ulfs_extern.h>
95 
96 #include <uvm/uvm.h>
97 #include <uvm/uvm_pmap.h>
98 #include <uvm/uvm_stat.h>
99 #include <uvm/uvm_pager.h>
100 
101 #include <ufs/lfs/lfs.h>
102 #include <ufs/lfs/lfs_kernel.h>
103 #include <ufs/lfs/lfs_extern.h>
104 
105 extern pid_t lfs_writer_daemon;
106 int lfs_ignore_lazy_sync = 1;
107 
108 static int lfs_openextattr(void *v);
109 static int lfs_closeextattr(void *v);
110 static int lfs_getextattr(void *v);
111 static int lfs_setextattr(void *v);
112 static int lfs_listextattr(void *v);
113 static int lfs_deleteextattr(void *v);
114 
115 /* Global vfs data structures for lfs. */
116 int (**lfs_vnodeop_p)(void *);
117 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = {
118 	{ &vop_default_desc, vn_default_error },
119 	{ &vop_lookup_desc, ulfs_lookup },		/* lookup */
120 	{ &vop_create_desc, lfs_create },		/* create */
121 	{ &vop_whiteout_desc, ulfs_whiteout },		/* whiteout */
122 	{ &vop_mknod_desc, lfs_mknod },			/* mknod */
123 	{ &vop_open_desc, ulfs_open },			/* open */
124 	{ &vop_close_desc, lfs_close },			/* close */
125 	{ &vop_access_desc, ulfs_access },		/* access */
126 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
127 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
128 	{ &vop_read_desc, lfs_read },			/* read */
129 	{ &vop_write_desc, lfs_write },			/* write */
130 	{ &vop_ioctl_desc, ulfs_ioctl },		/* ioctl */
131 	{ &vop_fcntl_desc, lfs_fcntl },			/* fcntl */
132 	{ &vop_poll_desc, ulfs_poll },			/* poll */
133 	{ &vop_kqfilter_desc, genfs_kqfilter },		/* kqfilter */
134 	{ &vop_revoke_desc, ulfs_revoke },		/* revoke */
135 	{ &vop_mmap_desc, lfs_mmap },			/* mmap */
136 	{ &vop_fsync_desc, lfs_fsync },			/* fsync */
137 	{ &vop_seek_desc, ulfs_seek },			/* seek */
138 	{ &vop_remove_desc, lfs_remove },		/* remove */
139 	{ &vop_link_desc, lfs_link },			/* link */
140 	{ &vop_rename_desc, lfs_rename },		/* rename */
141 	{ &vop_mkdir_desc, lfs_mkdir },			/* mkdir */
142 	{ &vop_rmdir_desc, lfs_rmdir },			/* rmdir */
143 	{ &vop_symlink_desc, lfs_symlink },		/* symlink */
144 	{ &vop_readdir_desc, ulfs_readdir },		/* readdir */
145 	{ &vop_readlink_desc, ulfs_readlink },		/* readlink */
146 	{ &vop_abortop_desc, ulfs_abortop },		/* abortop */
147 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
148 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
149 	{ &vop_lock_desc, ulfs_lock },			/* lock */
150 	{ &vop_unlock_desc, ulfs_unlock },		/* unlock */
151 	{ &vop_bmap_desc, ulfs_bmap },			/* bmap */
152 	{ &vop_strategy_desc, lfs_strategy },		/* strategy */
153 	{ &vop_print_desc, ulfs_print },		/* print */
154 	{ &vop_islocked_desc, ulfs_islocked },		/* islocked */
155 	{ &vop_pathconf_desc, ulfs_pathconf },		/* pathconf */
156 	{ &vop_advlock_desc, ulfs_advlock },		/* advlock */
157 	{ &vop_bwrite_desc, lfs_bwrite },		/* bwrite */
158 	{ &vop_getpages_desc, lfs_getpages },		/* getpages */
159 	{ &vop_putpages_desc, lfs_putpages },		/* putpages */
160 	{ &vop_openextattr_desc, lfs_openextattr },	/* openextattr */
161 	{ &vop_closeextattr_desc, lfs_closeextattr },	/* closeextattr */
162 	{ &vop_getextattr_desc, lfs_getextattr },	/* getextattr */
163 	{ &vop_setextattr_desc, lfs_setextattr },	/* setextattr */
164 	{ &vop_listextattr_desc, lfs_listextattr },	/* listextattr */
165 	{ &vop_deleteextattr_desc, lfs_deleteextattr },	/* deleteextattr */
166 	{ NULL, NULL }
167 };
168 const struct vnodeopv_desc lfs_vnodeop_opv_desc =
169 	{ &lfs_vnodeop_p, lfs_vnodeop_entries };
170 
171 int (**lfs_specop_p)(void *);
172 const struct vnodeopv_entry_desc lfs_specop_entries[] = {
173 	{ &vop_default_desc, vn_default_error },
174 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
175 	{ &vop_create_desc, spec_create },		/* create */
176 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
177 	{ &vop_open_desc, spec_open },			/* open */
178 	{ &vop_close_desc, lfsspec_close },		/* close */
179 	{ &vop_access_desc, ulfs_access },		/* access */
180 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
181 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
182 	{ &vop_read_desc, ulfsspec_read },		/* read */
183 	{ &vop_write_desc, ulfsspec_write },		/* write */
184 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
185 	{ &vop_fcntl_desc, ulfs_fcntl },		/* fcntl */
186 	{ &vop_poll_desc, spec_poll },			/* poll */
187 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
188 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
189 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
190 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
191 	{ &vop_seek_desc, spec_seek },			/* seek */
192 	{ &vop_remove_desc, spec_remove },		/* remove */
193 	{ &vop_link_desc, spec_link },			/* link */
194 	{ &vop_rename_desc, spec_rename },		/* rename */
195 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
196 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
197 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
198 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
199 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
200 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
201 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
202 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
203 	{ &vop_lock_desc, ulfs_lock },			/* lock */
204 	{ &vop_unlock_desc, ulfs_unlock },		/* unlock */
205 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
206 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
207 	{ &vop_print_desc, ulfs_print },		/* print */
208 	{ &vop_islocked_desc, ulfs_islocked },		/* islocked */
209 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
210 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
211 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
212 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
213 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
214 	{ &vop_openextattr_desc, lfs_openextattr },	/* openextattr */
215 	{ &vop_closeextattr_desc, lfs_closeextattr },	/* closeextattr */
216 	{ &vop_getextattr_desc, lfs_getextattr },	/* getextattr */
217 	{ &vop_setextattr_desc, lfs_setextattr },	/* setextattr */
218 	{ &vop_listextattr_desc, lfs_listextattr },	/* listextattr */
219 	{ &vop_deleteextattr_desc, lfs_deleteextattr },	/* deleteextattr */
220 	{ NULL, NULL }
221 };
222 const struct vnodeopv_desc lfs_specop_opv_desc =
223 	{ &lfs_specop_p, lfs_specop_entries };
224 
225 int (**lfs_fifoop_p)(void *);
226 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = {
227 	{ &vop_default_desc, vn_default_error },
228 	{ &vop_lookup_desc, vn_fifo_bypass },		/* lookup */
229 	{ &vop_create_desc, vn_fifo_bypass },		/* create */
230 	{ &vop_mknod_desc, vn_fifo_bypass },		/* mknod */
231 	{ &vop_open_desc, vn_fifo_bypass },		/* open */
232 	{ &vop_close_desc, lfsfifo_close },		/* close */
233 	{ &vop_access_desc, ulfs_access },		/* access */
234 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
235 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
236 	{ &vop_read_desc, ulfsfifo_read },		/* read */
237 	{ &vop_write_desc, ulfsfifo_write },		/* write */
238 	{ &vop_ioctl_desc, vn_fifo_bypass },		/* ioctl */
239 	{ &vop_fcntl_desc, ulfs_fcntl },		/* fcntl */
240 	{ &vop_poll_desc, vn_fifo_bypass },		/* poll */
241 	{ &vop_kqfilter_desc, vn_fifo_bypass },		/* kqfilter */
242 	{ &vop_revoke_desc, vn_fifo_bypass },		/* revoke */
243 	{ &vop_mmap_desc, vn_fifo_bypass },		/* mmap */
244 	{ &vop_fsync_desc, vn_fifo_bypass },		/* fsync */
245 	{ &vop_seek_desc, vn_fifo_bypass },		/* seek */
246 	{ &vop_remove_desc, vn_fifo_bypass },		/* remove */
247 	{ &vop_link_desc, vn_fifo_bypass },		/* link */
248 	{ &vop_rename_desc, vn_fifo_bypass },		/* rename */
249 	{ &vop_mkdir_desc, vn_fifo_bypass },		/* mkdir */
250 	{ &vop_rmdir_desc, vn_fifo_bypass },		/* rmdir */
251 	{ &vop_symlink_desc, vn_fifo_bypass },		/* symlink */
252 	{ &vop_readdir_desc, vn_fifo_bypass },		/* readdir */
253 	{ &vop_readlink_desc, vn_fifo_bypass },		/* readlink */
254 	{ &vop_abortop_desc, vn_fifo_bypass },		/* abortop */
255 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
256 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
257 	{ &vop_lock_desc, ulfs_lock },			/* lock */
258 	{ &vop_unlock_desc, ulfs_unlock },		/* unlock */
259 	{ &vop_bmap_desc, vn_fifo_bypass },		/* bmap */
260 	{ &vop_strategy_desc, vn_fifo_bypass },		/* strategy */
261 	{ &vop_print_desc, ulfs_print },		/* print */
262 	{ &vop_islocked_desc, ulfs_islocked },		/* islocked */
263 	{ &vop_pathconf_desc, vn_fifo_bypass },		/* pathconf */
264 	{ &vop_advlock_desc, vn_fifo_bypass },		/* advlock */
265 	{ &vop_bwrite_desc, lfs_bwrite },		/* bwrite */
266 	{ &vop_putpages_desc, vn_fifo_bypass },		/* putpages */
267 	{ &vop_openextattr_desc, lfs_openextattr },	/* openextattr */
268 	{ &vop_closeextattr_desc, lfs_closeextattr },	/* closeextattr */
269 	{ &vop_getextattr_desc, lfs_getextattr },	/* getextattr */
270 	{ &vop_setextattr_desc, lfs_setextattr },	/* setextattr */
271 	{ &vop_listextattr_desc, lfs_listextattr },	/* listextattr */
272 	{ &vop_deleteextattr_desc, lfs_deleteextattr },	/* deleteextattr */
273 	{ NULL, NULL }
274 };
275 const struct vnodeopv_desc lfs_fifoop_opv_desc =
276 	{ &lfs_fifoop_p, lfs_fifoop_entries };
277 
278 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int, struct vm_page **);
279 
280 #define	LFS_READWRITE
281 #include <ufs/lfs/ulfs_readwrite.c>
282 #undef	LFS_READWRITE
283 
284 /*
285  * Synch an open file.
286  */
287 /* ARGSUSED */
288 int
289 lfs_fsync(void *v)
290 {
291 	struct vop_fsync_args /* {
292 		struct vnode *a_vp;
293 		kauth_cred_t a_cred;
294 		int a_flags;
295 		off_t offlo;
296 		off_t offhi;
297 	} */ *ap = v;
298 	struct vnode *vp = ap->a_vp;
299 	int error, wait;
300 	struct inode *ip = VTOI(vp);
301 	struct lfs *fs = ip->i_lfs;
302 
303 	/* If we're mounted read-only, don't try to sync. */
304 	if (fs->lfs_ronly)
305 		return 0;
306 
307 	/* If a removed vnode is being cleaned, no need to sync here. */
308 	if ((ap->a_flags & FSYNC_RECLAIM) != 0 && ip->i_mode == 0)
309 		return 0;
310 
311 	/*
312 	 * Trickle sync simply adds this vnode to the pager list, as if
313 	 * the pagedaemon had requested a pageout.
314 	 */
315 	if (ap->a_flags & FSYNC_LAZY) {
316 		if (lfs_ignore_lazy_sync == 0) {
317 			mutex_enter(&lfs_lock);
318 			if (!(ip->i_flags & IN_PAGING)) {
319 				ip->i_flags |= IN_PAGING;
320 				TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip,
321 						  i_lfs_pchain);
322 			}
323 			wakeup(&lfs_writer_daemon);
324 			mutex_exit(&lfs_lock);
325 		}
326 		return 0;
327 	}
328 
329 	/*
330 	 * If a vnode is bring cleaned, flush it out before we try to
331 	 * reuse it.  This prevents the cleaner from writing files twice
332 	 * in the same partial segment, causing an accounting underflow.
333 	 */
334 	if (ap->a_flags & FSYNC_RECLAIM && ip->i_flags & IN_CLEANING) {
335 		lfs_vflush(vp);
336 	}
337 
338 	wait = (ap->a_flags & FSYNC_WAIT);
339 	do {
340 		mutex_enter(vp->v_interlock);
341 		error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo),
342 				     round_page(ap->a_offhi),
343 				     PGO_CLEANIT | (wait ? PGO_SYNCIO : 0));
344 		if (error == EAGAIN) {
345 			mutex_enter(&lfs_lock);
346 			mtsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_fsync",
347 				hz / 100 + 1, &lfs_lock);
348 			mutex_exit(&lfs_lock);
349 		}
350 	} while (error == EAGAIN);
351 	if (error)
352 		return error;
353 
354 	if ((ap->a_flags & FSYNC_DATAONLY) == 0)
355 		error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0);
356 
357 	if (error == 0 && ap->a_flags & FSYNC_CACHE) {
358 		int l = 0;
359 		error = VOP_IOCTL(ip->i_devvp, DIOCCACHESYNC, &l, FWRITE,
360 				  curlwp->l_cred);
361 	}
362 	if (wait && !VPISEMPTY(vp))
363 		LFS_SET_UINO(ip, IN_MODIFIED);
364 
365 	return error;
366 }
367 
368 /*
369  * Take IN_ADIROP off, then call ulfs_inactive.
370  */
371 int
372 lfs_inactive(void *v)
373 {
374 	struct vop_inactive_args /* {
375 		struct vnode *a_vp;
376 	} */ *ap = v;
377 
378 	lfs_unmark_vnode(ap->a_vp);
379 
380 	/*
381 	 * The Ifile is only ever inactivated on unmount.
382 	 * Streamline this process by not giving it more dirty blocks.
383 	 */
384 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) {
385 		mutex_enter(&lfs_lock);
386 		LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD);
387 		mutex_exit(&lfs_lock);
388 		VOP_UNLOCK(ap->a_vp);
389 		return 0;
390 	}
391 
392 #ifdef DEBUG
393 	/*
394 	 * This might happen on unmount.
395 	 * XXX If it happens at any other time, it should be a panic.
396 	 */
397 	if (ap->a_vp->v_uflag & VU_DIROP) {
398 		struct inode *ip = VTOI(ap->a_vp);
399 		printf("lfs_inactive: inactivating VU_DIROP? ino = %d\n", (int)ip->i_number);
400 	}
401 #endif /* DIAGNOSTIC */
402 
403 	return ulfs_inactive(v);
404 }
405 
406 int
407 lfs_set_dirop(struct vnode *dvp, struct vnode *vp)
408 {
409 	struct lfs *fs;
410 	int error;
411 
412 	KASSERT(VOP_ISLOCKED(dvp));
413 	KASSERT(vp == NULL || VOP_ISLOCKED(vp));
414 
415 	fs = VTOI(dvp)->i_lfs;
416 
417 	ASSERT_NO_SEGLOCK(fs);
418 	/*
419 	 * LFS_NRESERVE calculates direct and indirect blocks as well
420 	 * as an inode block; an overestimate in most cases.
421 	 */
422 	if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0)
423 		return (error);
424 
425     restart:
426 	mutex_enter(&lfs_lock);
427 	if (fs->lfs_dirops == 0) {
428 		mutex_exit(&lfs_lock);
429 		lfs_check(dvp, LFS_UNUSED_LBN, 0);
430 		mutex_enter(&lfs_lock);
431 	}
432 	while (fs->lfs_writer) {
433 		error = mtsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH,
434 		    "lfs_sdirop", 0, &lfs_lock);
435 		if (error == EINTR) {
436 			mutex_exit(&lfs_lock);
437 			goto unreserve;
438 		}
439 	}
440 	if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) {
441 		wakeup(&lfs_writer_daemon);
442 		mutex_exit(&lfs_lock);
443 		preempt();
444 		goto restart;
445 	}
446 
447 	if (lfs_dirvcount > LFS_MAX_DIROP) {
448 		DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, "
449 		      "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount));
450 		if ((error = mtsleep(&lfs_dirvcount,
451 		    PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0,
452 		    &lfs_lock)) != 0) {
453 			goto unreserve;
454 		}
455 		goto restart;
456 	}
457 
458 	++fs->lfs_dirops;
459 	/* fs->lfs_doifile = 1; */ /* XXX why? --ks */
460 	mutex_exit(&lfs_lock);
461 
462 	/* Hold a reference so SET_ENDOP will be happy */
463 	vref(dvp);
464 	if (vp) {
465 		vref(vp);
466 		MARK_VNODE(vp);
467 	}
468 
469 	MARK_VNODE(dvp);
470 	return 0;
471 
472   unreserve:
473 	lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs));
474 	return error;
475 }
476 
477 /*
478  * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock
479  * in getnewvnode(), if we have a stacked filesystem mounted on top
480  * of us.
481  *
482  * NB: this means we have to clear the new vnodes on error.  Fortunately
483  * SET_ENDOP is there to do that for us.
484  */
485 int
486 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp)
487 {
488 	int error;
489 	struct lfs *fs;
490 
491 	fs = VFSTOULFS(dvp->v_mount)->um_lfs;
492 	ASSERT_NO_SEGLOCK(fs);
493 	if (fs->lfs_ronly)
494 		return EROFS;
495 	if (vpp == NULL) {
496 		return lfs_set_dirop(dvp, NULL);
497 	}
498 	error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, NULL, vpp);
499 	if (error) {
500 		DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n",
501 		      dvp, error));
502 		return error;
503 	}
504 	if ((error = lfs_set_dirop(dvp, NULL)) != 0) {
505 		ungetnewvnode(*vpp);
506 		*vpp = NULL;
507 		return error;
508 	}
509 	return 0;
510 }
511 
512 void
513 lfs_mark_vnode(struct vnode *vp)
514 {
515 	struct inode *ip = VTOI(vp);
516 	struct lfs *fs = ip->i_lfs;
517 
518 	mutex_enter(&lfs_lock);
519 	if (!(ip->i_flag & IN_ADIROP)) {
520 		if (!(vp->v_uflag & VU_DIROP)) {
521 			mutex_exit(&lfs_lock);
522 			mutex_enter(vp->v_interlock);
523 			if (lfs_vref(vp) != 0)
524 				panic("lfs_mark_vnode: could not vref");
525 			mutex_enter(&lfs_lock);
526 			++lfs_dirvcount;
527 			++fs->lfs_dirvcount;
528 			TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain);
529 			vp->v_uflag |= VU_DIROP;
530 		}
531 		++fs->lfs_nadirop;
532 		ip->i_flag &= ~IN_CDIROP;
533 		ip->i_flag |= IN_ADIROP;
534 	} else
535 		KASSERT(vp->v_uflag & VU_DIROP);
536 	mutex_exit(&lfs_lock);
537 }
538 
539 void
540 lfs_unmark_vnode(struct vnode *vp)
541 {
542 	struct inode *ip = VTOI(vp);
543 
544 	mutex_enter(&lfs_lock);
545 	if (ip && (ip->i_flag & IN_ADIROP)) {
546 		KASSERT(vp->v_uflag & VU_DIROP);
547 		--ip->i_lfs->lfs_nadirop;
548 		ip->i_flag &= ~IN_ADIROP;
549 	}
550 	mutex_exit(&lfs_lock);
551 }
552 
553 int
554 lfs_symlink(void *v)
555 {
556 	struct vop_symlink_v3_args /* {
557 		struct vnode *a_dvp;
558 		struct vnode **a_vpp;
559 		struct componentname *a_cnp;
560 		struct vattr *a_vap;
561 		char *a_target;
562 	} */ *ap = v;
563 	int error;
564 
565 	if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) {
566 		return error;
567 	}
568 	error = ulfs_symlink(ap);
569 	SET_ENDOP_CREATE_AP(ap, "symlink");
570 	return (error);
571 }
572 
573 int
574 lfs_mknod(void *v)
575 {
576 	struct vop_mknod_v3_args	/* {
577 		struct vnode *a_dvp;
578 		struct vnode **a_vpp;
579 		struct componentname *a_cnp;
580 		struct vattr *a_vap;
581 	} */ *ap = v;
582 	struct vattr *vap;
583 	struct vnode **vpp;
584 	struct inode *ip;
585 	int error;
586 	struct mount	*mp;
587 	ino_t		ino;
588 	struct ulfs_lookup_results *ulr;
589 
590 	vap = ap->a_vap;
591 	vpp = ap->a_vpp;
592 
593 	/* XXX should handle this material another way */
594 	ulr = &VTOI(ap->a_dvp)->i_crap;
595 	ULFS_CHECK_CRAPCOUNTER(VTOI(ap->a_dvp));
596 
597 	if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) {
598 		return error;
599 	}
600 
601 	fstrans_start(ap->a_dvp->v_mount, FSTRANS_SHARED);
602 	error = ulfs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
603 			      ap->a_dvp, ulr, vpp, ap->a_cnp);
604 
605 	/* Either way we're done with the dirop at this point */
606 	SET_ENDOP_CREATE_AP(ap, "mknod");
607 
608 	if (error) {
609 		fstrans_done(ap->a_dvp->v_mount);
610 		*vpp = NULL;
611 		return (error);
612 	}
613 
614 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
615 	ip = VTOI(*vpp);
616 	mp  = (*vpp)->v_mount;
617 	ino = ip->i_number;
618 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
619 	if (vap->va_rdev != VNOVAL) {
620 		struct ulfsmount *ump = ip->i_ump;
621 		struct lfs *fs = ip->i_lfs;
622 		/*
623 		 * Want to be able to use this to make badblock
624 		 * inodes, so don't truncate the dev number.
625 		 */
626 		if (ump->um_fstype == ULFS1)
627 			ip->i_ffs1_rdev = ulfs_rw32(vap->va_rdev,
628 			    ULFS_MPNEEDSWAP(fs));
629 		else
630 			ip->i_ffs2_rdev = ulfs_rw64(vap->va_rdev,
631 			    ULFS_MPNEEDSWAP(fs));
632 	}
633 
634 	/*
635 	 * Call fsync to write the vnode so that we don't have to deal with
636 	 * flushing it when it's marked VU_DIROP or reclaiming.
637 	 *
638 	 * XXX KS - If we can't flush we also can't call vgone(), so must
639 	 * return.  But, that leaves this vnode in limbo, also not good.
640 	 * Can this ever happen (barring hardware failure)?
641 	 */
642 	if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0)) != 0) {
643 		panic("lfs_mknod: couldn't fsync (ino %llu)",
644 		      (unsigned long long)ino);
645 		/* return (error); */
646 	}
647 	/*
648 	 * Remove vnode so that it will be reloaded by VFS_VGET and
649 	 * checked to see if it is an alias of an existing entry in
650 	 * the inode cache.
651 	 */
652 	/* Used to be vput, but that causes us to call VOP_INACTIVE twice. */
653 
654 	(*vpp)->v_type = VNON;
655 	VOP_UNLOCK(*vpp);
656 	vgone(*vpp);
657 	error = VFS_VGET(mp, ino, vpp);
658 
659 	fstrans_done(ap->a_dvp->v_mount);
660 	if (error != 0) {
661 		*vpp = NULL;
662 		return (error);
663 	}
664 	VOP_UNLOCK(*vpp);
665 	return (0);
666 }
667 
668 int
669 lfs_create(void *v)
670 {
671 	struct vop_create_v3_args	/* {
672 		struct vnode *a_dvp;
673 		struct vnode **a_vpp;
674 		struct componentname *a_cnp;
675 		struct vattr *a_vap;
676 	} */ *ap = v;
677 	int error;
678 
679 	if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) {
680 		return error;
681 	}
682 	error = ulfs_create(ap);
683 	SET_ENDOP_CREATE_AP(ap, "create");
684 	return (error);
685 }
686 
687 int
688 lfs_mkdir(void *v)
689 {
690 	struct vop_mkdir_v3_args	/* {
691 		struct vnode *a_dvp;
692 		struct vnode **a_vpp;
693 		struct componentname *a_cnp;
694 		struct vattr *a_vap;
695 	} */ *ap = v;
696 	int error;
697 
698 	if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) {
699 		return error;
700 	}
701 	error = ulfs_mkdir(ap);
702 	SET_ENDOP_CREATE_AP(ap, "mkdir");
703 	return (error);
704 }
705 
706 int
707 lfs_remove(void *v)
708 {
709 	struct vop_remove_args	/* {
710 		struct vnode *a_dvp;
711 		struct vnode *a_vp;
712 		struct componentname *a_cnp;
713 	} */ *ap = v;
714 	struct vnode *dvp, *vp;
715 	struct inode *ip;
716 	int error;
717 
718 	dvp = ap->a_dvp;
719 	vp = ap->a_vp;
720 	ip = VTOI(vp);
721 	if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) {
722 		if (dvp == vp)
723 			vrele(vp);
724 		else
725 			vput(vp);
726 		vput(dvp);
727 		return error;
728 	}
729 	error = ulfs_remove(ap);
730 	if (ip->i_nlink == 0)
731 		lfs_orphan(ip->i_lfs, ip->i_number);
732 	SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove");
733 	return (error);
734 }
735 
736 int
737 lfs_rmdir(void *v)
738 {
739 	struct vop_rmdir_args	/* {
740 		struct vnodeop_desc *a_desc;
741 		struct vnode *a_dvp;
742 		struct vnode *a_vp;
743 		struct componentname *a_cnp;
744 	} */ *ap = v;
745 	struct vnode *vp;
746 	struct inode *ip;
747 	int error;
748 
749 	vp = ap->a_vp;
750 	ip = VTOI(vp);
751 	if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) {
752 		if (ap->a_dvp == vp)
753 			vrele(ap->a_dvp);
754 		else
755 			vput(ap->a_dvp);
756 		vput(vp);
757 		return error;
758 	}
759 	error = ulfs_rmdir(ap);
760 	if (ip->i_nlink == 0)
761 		lfs_orphan(ip->i_lfs, ip->i_number);
762 	SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir");
763 	return (error);
764 }
765 
766 int
767 lfs_link(void *v)
768 {
769 	struct vop_link_args	/* {
770 		struct vnode *a_dvp;
771 		struct vnode *a_vp;
772 		struct componentname *a_cnp;
773 	} */ *ap = v;
774 	int error;
775 	struct vnode **vpp = NULL;
776 
777 	if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) {
778 		vput(ap->a_dvp);
779 		return error;
780 	}
781 	error = ulfs_link(ap);
782 	SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link");
783 	return (error);
784 }
785 
786 /* XXX hack to avoid calling ITIMES in getattr */
787 int
788 lfs_getattr(void *v)
789 {
790 	struct vop_getattr_args /* {
791 		struct vnode *a_vp;
792 		struct vattr *a_vap;
793 		kauth_cred_t a_cred;
794 	} */ *ap = v;
795 	struct vnode *vp = ap->a_vp;
796 	struct inode *ip = VTOI(vp);
797 	struct vattr *vap = ap->a_vap;
798 	struct lfs *fs = ip->i_lfs;
799 
800 	fstrans_start(vp->v_mount, FSTRANS_SHARED);
801 	/*
802 	 * Copy from inode table
803 	 */
804 	vap->va_fsid = ip->i_dev;
805 	vap->va_fileid = ip->i_number;
806 	vap->va_mode = ip->i_mode & ~LFS_IFMT;
807 	vap->va_nlink = ip->i_nlink;
808 	vap->va_uid = ip->i_uid;
809 	vap->va_gid = ip->i_gid;
810 	vap->va_rdev = (dev_t)ip->i_ffs1_rdev;
811 	vap->va_size = vp->v_size;
812 	vap->va_atime.tv_sec = ip->i_ffs1_atime;
813 	vap->va_atime.tv_nsec = ip->i_ffs1_atimensec;
814 	vap->va_mtime.tv_sec = ip->i_ffs1_mtime;
815 	vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec;
816 	vap->va_ctime.tv_sec = ip->i_ffs1_ctime;
817 	vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec;
818 	vap->va_flags = ip->i_flags;
819 	vap->va_gen = ip->i_gen;
820 	/* this doesn't belong here */
821 	if (vp->v_type == VBLK)
822 		vap->va_blocksize = BLKDEV_IOSIZE;
823 	else if (vp->v_type == VCHR)
824 		vap->va_blocksize = MAXBSIZE;
825 	else
826 		vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
827 	vap->va_bytes = lfs_fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks);
828 	vap->va_type = vp->v_type;
829 	vap->va_filerev = ip->i_modrev;
830 	fstrans_done(vp->v_mount);
831 	return (0);
832 }
833 
834 /*
835  * Check to make sure the inode blocks won't choke the buffer
836  * cache, then call ulfs_setattr as usual.
837  */
838 int
839 lfs_setattr(void *v)
840 {
841 	struct vop_setattr_args /* {
842 		struct vnode *a_vp;
843 		struct vattr *a_vap;
844 		kauth_cred_t a_cred;
845 	} */ *ap = v;
846 	struct vnode *vp = ap->a_vp;
847 
848 	lfs_check(vp, LFS_UNUSED_LBN, 0);
849 	return ulfs_setattr(v);
850 }
851 
852 /*
853  * Release the block we hold on lfs_newseg wrapping.  Called on file close,
854  * or explicitly from LFCNWRAPGO.  Called with the interlock held.
855  */
856 static int
857 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor)
858 {
859 	if (fs->lfs_stoplwp != curlwp)
860 		return EBUSY;
861 
862 	fs->lfs_stoplwp = NULL;
863 	cv_signal(&fs->lfs_stopcv);
864 
865 	KASSERT(fs->lfs_nowrap > 0);
866 	if (fs->lfs_nowrap <= 0) {
867 		return 0;
868 	}
869 
870 	if (--fs->lfs_nowrap == 0) {
871 		log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt);
872 		wakeup(&fs->lfs_wrappass);
873 		lfs_wakeup_cleaner(fs);
874 	}
875 	if (waitfor) {
876 		mtsleep(&fs->lfs_nextseg, PCATCH | PUSER, "segment",
877 		    0, &lfs_lock);
878 	}
879 
880 	return 0;
881 }
882 
883 /*
884  * Close called.
885  *
886  * Update the times on the inode.
887  */
888 /* ARGSUSED */
889 int
890 lfs_close(void *v)
891 {
892 	struct vop_close_args /* {
893 		struct vnode *a_vp;
894 		int  a_fflag;
895 		kauth_cred_t a_cred;
896 	} */ *ap = v;
897 	struct vnode *vp = ap->a_vp;
898 	struct inode *ip = VTOI(vp);
899 	struct lfs *fs = ip->i_lfs;
900 
901 	if ((ip->i_number == ULFS_ROOTINO || ip->i_number == LFS_IFILE_INUM) &&
902 	    fs->lfs_stoplwp == curlwp) {
903 		mutex_enter(&lfs_lock);
904 		log(LOG_NOTICE, "lfs_close: releasing log wrap control\n");
905 		lfs_wrapgo(fs, ip, 0);
906 		mutex_exit(&lfs_lock);
907 	}
908 
909 	if (vp == ip->i_lfs->lfs_ivnode &&
910 	    vp->v_mount->mnt_iflag & IMNT_UNMOUNT)
911 		return 0;
912 
913 	fstrans_start(vp->v_mount, FSTRANS_SHARED);
914 	if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) {
915 		LFS_ITIMES(ip, NULL, NULL, NULL);
916 	}
917 	fstrans_done(vp->v_mount);
918 	return (0);
919 }
920 
921 /*
922  * Close wrapper for special devices.
923  *
924  * Update the times on the inode then do device close.
925  */
926 int
927 lfsspec_close(void *v)
928 {
929 	struct vop_close_args /* {
930 		struct vnode	*a_vp;
931 		int		a_fflag;
932 		kauth_cred_t	a_cred;
933 	} */ *ap = v;
934 	struct vnode	*vp;
935 	struct inode	*ip;
936 
937 	vp = ap->a_vp;
938 	ip = VTOI(vp);
939 	if (vp->v_usecount > 1) {
940 		LFS_ITIMES(ip, NULL, NULL, NULL);
941 	}
942 	return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap));
943 }
944 
945 /*
946  * Close wrapper for fifo's.
947  *
948  * Update the times on the inode then do device close.
949  */
950 int
951 lfsfifo_close(void *v)
952 {
953 	struct vop_close_args /* {
954 		struct vnode	*a_vp;
955 		int		a_fflag;
956 		kauth_cred_	a_cred;
957 	} */ *ap = v;
958 	struct vnode	*vp;
959 	struct inode	*ip;
960 
961 	vp = ap->a_vp;
962 	ip = VTOI(vp);
963 	if (ap->a_vp->v_usecount > 1) {
964 		LFS_ITIMES(ip, NULL, NULL, NULL);
965 	}
966 	return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap));
967 }
968 
969 /*
970  * Reclaim an inode so that it can be used for other purposes.
971  */
972 
973 int
974 lfs_reclaim(void *v)
975 {
976 	struct vop_reclaim_args /* {
977 		struct vnode *a_vp;
978 	} */ *ap = v;
979 	struct vnode *vp = ap->a_vp;
980 	struct inode *ip = VTOI(vp);
981 	struct lfs *fs = ip->i_lfs;
982 	int error;
983 
984 	/*
985 	 * The inode must be freed and updated before being removed
986 	 * from its hash chain.  Other threads trying to gain a hold
987 	 * or lock on the inode will be stalled.
988 	 */
989 	if (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
990 		lfs_vfree(vp, ip->i_number, ip->i_omode);
991 
992 	mutex_enter(&lfs_lock);
993 	LFS_CLR_UINO(ip, IN_ALLMOD);
994 	mutex_exit(&lfs_lock);
995 	if ((error = ulfs_reclaim(vp)))
996 		return (error);
997 
998 	/*
999 	 * Take us off the paging and/or dirop queues if we were on them.
1000 	 * We shouldn't be on them.
1001 	 */
1002 	mutex_enter(&lfs_lock);
1003 	if (ip->i_flags & IN_PAGING) {
1004 		log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n",
1005 		    fs->lfs_fsmnt);
1006 		ip->i_flags &= ~IN_PAGING;
1007 		TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain);
1008 	}
1009 	if (vp->v_uflag & VU_DIROP) {
1010 		panic("reclaimed vnode is VU_DIROP");
1011 		vp->v_uflag &= ~VU_DIROP;
1012 		TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
1013 	}
1014 	mutex_exit(&lfs_lock);
1015 
1016 	pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din);
1017 	lfs_deregister_all(vp);
1018 	pool_put(&lfs_inoext_pool, ip->inode_ext.lfs);
1019 	ip->inode_ext.lfs = NULL;
1020 	genfs_node_destroy(vp);
1021 	pool_put(&lfs_inode_pool, vp->v_data);
1022 	vp->v_data = NULL;
1023 	return (0);
1024 }
1025 
1026 /*
1027  * Read a block from a storage device.
1028  *
1029  * Calculate the logical to physical mapping if not done already,
1030  * then call the device strategy routine.
1031  *
1032  * In order to avoid reading blocks that are in the process of being
1033  * written by the cleaner---and hence are not mutexed by the normal
1034  * buffer cache / page cache mechanisms---check for collisions before
1035  * reading.
1036  *
1037  * We inline ulfs_strategy to make sure that the VOP_BMAP occurs *before*
1038  * the active cleaner test.
1039  *
1040  * XXX This code assumes that lfs_markv makes synchronous checkpoints.
1041  */
1042 int
1043 lfs_strategy(void *v)
1044 {
1045 	struct vop_strategy_args /* {
1046 		struct vnode *a_vp;
1047 		struct buf *a_bp;
1048 	} */ *ap = v;
1049 	struct buf	*bp;
1050 	struct lfs	*fs;
1051 	struct vnode	*vp;
1052 	struct inode	*ip;
1053 	daddr_t		tbn;
1054 #define MAXLOOP 25
1055 	int		i, sn, error, slept, loopcount;
1056 
1057 	bp = ap->a_bp;
1058 	vp = ap->a_vp;
1059 	ip = VTOI(vp);
1060 	fs = ip->i_lfs;
1061 
1062 	/* lfs uses its strategy routine only for read */
1063 	KASSERT(bp->b_flags & B_READ);
1064 
1065 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1066 		panic("lfs_strategy: spec");
1067 	KASSERT(bp->b_bcount != 0);
1068 	if (bp->b_blkno == bp->b_lblkno) {
1069 		error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno,
1070 				 NULL);
1071 		if (error) {
1072 			bp->b_error = error;
1073 			bp->b_resid = bp->b_bcount;
1074 			biodone(bp);
1075 			return (error);
1076 		}
1077 		if ((long)bp->b_blkno == -1) /* no valid data */
1078 			clrbuf(bp);
1079 	}
1080 	if ((long)bp->b_blkno < 0) { /* block is not on disk */
1081 		bp->b_resid = bp->b_bcount;
1082 		biodone(bp);
1083 		return (0);
1084 	}
1085 
1086 	slept = 1;
1087 	loopcount = 0;
1088 	mutex_enter(&lfs_lock);
1089 	while (slept && fs->lfs_seglock) {
1090 		mutex_exit(&lfs_lock);
1091 		/*
1092 		 * Look through list of intervals.
1093 		 * There will only be intervals to look through
1094 		 * if the cleaner holds the seglock.
1095 		 * Since the cleaner is synchronous, we can trust
1096 		 * the list of intervals to be current.
1097 		 */
1098 		tbn = LFS_DBTOFSB(fs, bp->b_blkno);
1099 		sn = lfs_dtosn(fs, tbn);
1100 		slept = 0;
1101 		for (i = 0; i < fs->lfs_cleanind; i++) {
1102 			if (sn == lfs_dtosn(fs, fs->lfs_cleanint[i]) &&
1103 			    tbn >= fs->lfs_cleanint[i]) {
1104 				DLOG((DLOG_CLEAN,
1105 				      "lfs_strategy: ino %d lbn %" PRId64
1106 				      " ind %d sn %d fsb %" PRIx32
1107 				      " given sn %d fsb %" PRIx64 "\n",
1108 				      ip->i_number, bp->b_lblkno, i,
1109 				      lfs_dtosn(fs, fs->lfs_cleanint[i]),
1110 				      fs->lfs_cleanint[i], sn, tbn));
1111 				DLOG((DLOG_CLEAN,
1112 				      "lfs_strategy: sleeping on ino %d lbn %"
1113 				      PRId64 "\n", ip->i_number, bp->b_lblkno));
1114 				mutex_enter(&lfs_lock);
1115 				if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) {
1116 					/*
1117 					 * Cleaner can't wait for itself.
1118 					 * Instead, wait for the blocks
1119 					 * to be written to disk.
1120 					 * XXX we need pribio in the test
1121 					 * XXX here.
1122 					 */
1123  					mtsleep(&fs->lfs_iocount,
1124  						(PRIBIO + 1) | PNORELOCK,
1125 						"clean2", hz/10 + 1,
1126  						&lfs_lock);
1127 					slept = 1;
1128 					++loopcount;
1129 					break;
1130 				} else if (fs->lfs_seglock) {
1131 					mtsleep(&fs->lfs_seglock,
1132 						(PRIBIO + 1) | PNORELOCK,
1133 						"clean1", 0,
1134 						&lfs_lock);
1135 					slept = 1;
1136 					break;
1137 				}
1138 				mutex_exit(&lfs_lock);
1139 			}
1140 		}
1141 		mutex_enter(&lfs_lock);
1142 		if (loopcount > MAXLOOP) {
1143 			printf("lfs_strategy: breaking out of clean2 loop\n");
1144 			break;
1145 		}
1146 	}
1147 	mutex_exit(&lfs_lock);
1148 
1149 	vp = ip->i_devvp;
1150 	return VOP_STRATEGY(vp, bp);
1151 }
1152 
1153 /*
1154  * Inline lfs_segwrite/lfs_writevnodes, but just for dirops.
1155  * Technically this is a checkpoint (the on-disk state is valid)
1156  * even though we are leaving out all the file data.
1157  */
1158 int
1159 lfs_flush_dirops(struct lfs *fs)
1160 {
1161 	struct inode *ip, *nip;
1162 	struct vnode *vp;
1163 	extern int lfs_dostats;
1164 	struct segment *sp;
1165 	int flags = 0;
1166 	int error = 0;
1167 
1168 	ASSERT_MAYBE_SEGLOCK(fs);
1169 	KASSERT(fs->lfs_nadirop == 0);
1170 
1171 	if (fs->lfs_ronly)
1172 		return EROFS;
1173 
1174 	mutex_enter(&lfs_lock);
1175 	if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) {
1176 		mutex_exit(&lfs_lock);
1177 		return 0;
1178 	} else
1179 		mutex_exit(&lfs_lock);
1180 
1181 	if (lfs_dostats)
1182 		++lfs_stats.flush_invoked;
1183 
1184 	lfs_imtime(fs);
1185 	lfs_seglock(fs, flags);
1186 	sp = fs->lfs_sp;
1187 
1188 	/*
1189 	 * lfs_writevnodes, optimized to get dirops out of the way.
1190 	 * Only write dirops, and don't flush files' pages, only
1191 	 * blocks from the directories.
1192 	 *
1193 	 * We don't need to vref these files because they are
1194 	 * dirops and so hold an extra reference until the
1195 	 * segunlock clears them of that status.
1196 	 *
1197 	 * We don't need to check for IN_ADIROP because we know that
1198 	 * no dirops are active.
1199 	 *
1200 	 */
1201 	mutex_enter(&lfs_lock);
1202 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
1203 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
1204 		mutex_exit(&lfs_lock);
1205 		vp = ITOV(ip);
1206 		mutex_enter(vp->v_interlock);
1207 
1208 		KASSERT((ip->i_flag & IN_ADIROP) == 0);
1209 		KASSERT(vp->v_uflag & VU_DIROP);
1210 		KASSERT(vdead_check(vp, VDEAD_NOWAIT) == 0);
1211 
1212 		/*
1213 		 * All writes to directories come from dirops; all
1214 		 * writes to files' direct blocks go through the page
1215 		 * cache, which we're not touching.  Reads to files
1216 		 * and/or directories will not be affected by writing
1217 		 * directory blocks inodes and file inodes.  So we don't
1218 		 * really need to lock.
1219 		 */
1220 		if (vdead_check(vp, VDEAD_NOWAIT) != 0) {
1221 			mutex_exit(vp->v_interlock);
1222 			mutex_enter(&lfs_lock);
1223 			continue;
1224 		}
1225 		mutex_exit(vp->v_interlock);
1226 		/* XXX see below
1227 		 * waslocked = VOP_ISLOCKED(vp);
1228 		 */
1229 		if (vp->v_type != VREG &&
1230 		    ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) {
1231 			error = lfs_writefile(fs, sp, vp);
1232 			if (!VPISEMPTY(vp) && !WRITEINPROG(vp) &&
1233 			    !(ip->i_flag & IN_ALLMOD)) {
1234 			    	mutex_enter(&lfs_lock);
1235 				LFS_SET_UINO(ip, IN_MODIFIED);
1236 			    	mutex_exit(&lfs_lock);
1237 			}
1238 			if (error && (sp->seg_flags & SEGM_SINGLE)) {
1239 				mutex_enter(&lfs_lock);
1240 				error = EAGAIN;
1241 				break;
1242 			}
1243 		}
1244 		KDASSERT(ip->i_number != LFS_IFILE_INUM);
1245 		error = lfs_writeinode(fs, sp, ip);
1246 		mutex_enter(&lfs_lock);
1247 		if (error && (sp->seg_flags & SEGM_SINGLE)) {
1248 			error = EAGAIN;
1249 			break;
1250 		}
1251 
1252 		/*
1253 		 * We might need to update these inodes again,
1254 		 * for example, if they have data blocks to write.
1255 		 * Make sure that after this flush, they are still
1256 		 * marked IN_MODIFIED so that we don't forget to
1257 		 * write them.
1258 		 */
1259 		/* XXX only for non-directories? --KS */
1260 		LFS_SET_UINO(ip, IN_MODIFIED);
1261 	}
1262 	mutex_exit(&lfs_lock);
1263 	/* We've written all the dirops there are */
1264 	((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT);
1265 	lfs_finalize_fs_seguse(fs);
1266 	(void) lfs_writeseg(fs, sp);
1267 	lfs_segunlock(fs);
1268 
1269 	return error;
1270 }
1271 
1272 /*
1273  * Flush all vnodes for which the pagedaemon has requested pageouts.
1274  * Skip over any files that are marked VU_DIROP (since lfs_flush_dirop()
1275  * has just run, this would be an error).  If we have to skip a vnode
1276  * for any reason, just skip it; if we have to wait for the cleaner,
1277  * abort.  The writer daemon will call us again later.
1278  */
1279 int
1280 lfs_flush_pchain(struct lfs *fs)
1281 {
1282 	struct inode *ip, *nip;
1283 	struct vnode *vp;
1284 	extern int lfs_dostats;
1285 	struct segment *sp;
1286 	int error, error2;
1287 
1288 	ASSERT_NO_SEGLOCK(fs);
1289 
1290 	if (fs->lfs_ronly)
1291 		return EROFS;
1292 
1293 	mutex_enter(&lfs_lock);
1294 	if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) {
1295 		mutex_exit(&lfs_lock);
1296 		return 0;
1297 	} else
1298 		mutex_exit(&lfs_lock);
1299 
1300 	/* Get dirops out of the way */
1301 	if ((error = lfs_flush_dirops(fs)) != 0)
1302 		return error;
1303 
1304 	if (lfs_dostats)
1305 		++lfs_stats.flush_invoked;
1306 
1307 	/*
1308 	 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts.
1309 	 */
1310 	lfs_imtime(fs);
1311 	lfs_seglock(fs, 0);
1312 	sp = fs->lfs_sp;
1313 
1314 	/*
1315 	 * lfs_writevnodes, optimized to clear pageout requests.
1316 	 * Only write non-dirop files that are in the pageout queue.
1317 	 * We're very conservative about what we write; we want to be
1318 	 * fast and async.
1319 	 */
1320 	mutex_enter(&lfs_lock);
1321     top:
1322 	for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) {
1323 		nip = TAILQ_NEXT(ip, i_lfs_pchain);
1324 		vp = ITOV(ip);
1325 
1326 		if (!(ip->i_flags & IN_PAGING))
1327 			goto top;
1328 
1329 		mutex_enter(vp->v_interlock);
1330 		if (vdead_check(vp, VDEAD_NOWAIT) != 0 ||
1331 		    (vp->v_uflag & VU_DIROP) != 0) {
1332 			mutex_exit(vp->v_interlock);
1333 			continue;
1334 		}
1335 		if (vp->v_type != VREG) {
1336 			mutex_exit(vp->v_interlock);
1337 			continue;
1338 		}
1339 		if (lfs_vref(vp))
1340 			continue;
1341 		mutex_exit(&lfs_lock);
1342 
1343 		if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_RETRY) != 0) {
1344 			lfs_vunref(vp);
1345 			mutex_enter(&lfs_lock);
1346 			continue;
1347 		}
1348 
1349 		error = lfs_writefile(fs, sp, vp);
1350 		if (!VPISEMPTY(vp) && !WRITEINPROG(vp) &&
1351 		    !(ip->i_flag & IN_ALLMOD)) {
1352 		    	mutex_enter(&lfs_lock);
1353 			LFS_SET_UINO(ip, IN_MODIFIED);
1354 		    	mutex_exit(&lfs_lock);
1355 		}
1356 		KDASSERT(ip->i_number != LFS_IFILE_INUM);
1357 		error2 = lfs_writeinode(fs, sp, ip);
1358 
1359 		VOP_UNLOCK(vp);
1360 		lfs_vunref(vp);
1361 
1362 		if (error == EAGAIN || error2 == EAGAIN) {
1363 			lfs_writeseg(fs, sp);
1364 			mutex_enter(&lfs_lock);
1365 			break;
1366 		}
1367 		mutex_enter(&lfs_lock);
1368 	}
1369 	mutex_exit(&lfs_lock);
1370 	(void) lfs_writeseg(fs, sp);
1371 	lfs_segunlock(fs);
1372 
1373 	return 0;
1374 }
1375 
1376 /*
1377  * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}.
1378  */
1379 int
1380 lfs_fcntl(void *v)
1381 {
1382 	struct vop_fcntl_args /* {
1383 		struct vnode *a_vp;
1384 		u_int a_command;
1385 		void * a_data;
1386 		int  a_fflag;
1387 		kauth_cred_t a_cred;
1388 	} */ *ap = v;
1389 	struct timeval tv;
1390 	struct timeval *tvp;
1391 	BLOCK_INFO *blkiov;
1392 	CLEANERINFO *cip;
1393 	SEGUSE *sup;
1394 	int blkcnt, error;
1395 	size_t fh_size;
1396 	struct lfs_fcntl_markv blkvp;
1397 	struct lwp *l;
1398 	fsid_t *fsidp;
1399 	struct lfs *fs;
1400 	struct buf *bp;
1401 	fhandle_t *fhp;
1402 	daddr_t off;
1403 	int oclean;
1404 
1405 	/* Only respect LFS fcntls on fs root or Ifile */
1406 	if (VTOI(ap->a_vp)->i_number != ULFS_ROOTINO &&
1407 	    VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) {
1408 		return ulfs_fcntl(v);
1409 	}
1410 
1411 	/* Avoid locking a draining lock */
1412 	if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) {
1413 		return ESHUTDOWN;
1414 	}
1415 
1416 	/* LFS control and monitoring fcntls are available only to root */
1417 	l = curlwp;
1418 	if (((ap->a_command & 0xff00) >> 8) == 'L' &&
1419 	    (error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
1420 	     KAUTH_REQ_SYSTEM_LFS_FCNTL, NULL, NULL, NULL)) != 0)
1421 		return (error);
1422 
1423 	fs = VTOI(ap->a_vp)->i_lfs;
1424 	fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx;
1425 
1426 	error = 0;
1427 	switch ((int)ap->a_command) {
1428 	    case LFCNSEGWAITALL_COMPAT_50:
1429 	    case LFCNSEGWAITALL_COMPAT:
1430 		fsidp = NULL;
1431 		/* FALLSTHROUGH */
1432 	    case LFCNSEGWAIT_COMPAT_50:
1433 	    case LFCNSEGWAIT_COMPAT:
1434 		{
1435 			struct timeval50 *tvp50
1436 				= (struct timeval50 *)ap->a_data;
1437 			timeval50_to_timeval(tvp50, &tv);
1438 			tvp = &tv;
1439 		}
1440 		goto segwait_common;
1441 	    case LFCNSEGWAITALL:
1442 		fsidp = NULL;
1443 		/* FALLSTHROUGH */
1444 	    case LFCNSEGWAIT:
1445 		tvp = (struct timeval *)ap->a_data;
1446 segwait_common:
1447 		mutex_enter(&lfs_lock);
1448 		++fs->lfs_sleepers;
1449 		mutex_exit(&lfs_lock);
1450 
1451 		error = lfs_segwait(fsidp, tvp);
1452 
1453 		mutex_enter(&lfs_lock);
1454 		if (--fs->lfs_sleepers == 0)
1455 			wakeup(&fs->lfs_sleepers);
1456 		mutex_exit(&lfs_lock);
1457 		return error;
1458 
1459 	    case LFCNBMAPV:
1460 	    case LFCNMARKV:
1461 		blkvp = *(struct lfs_fcntl_markv *)ap->a_data;
1462 
1463 		blkcnt = blkvp.blkcnt;
1464 		if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT)
1465 			return (EINVAL);
1466 		blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV);
1467 		if ((error = copyin(blkvp.blkiov, blkiov,
1468 		     blkcnt * sizeof(BLOCK_INFO))) != 0) {
1469 			lfs_free(fs, blkiov, LFS_NB_BLKIOV);
1470 			return error;
1471 		}
1472 
1473 		mutex_enter(&lfs_lock);
1474 		++fs->lfs_sleepers;
1475 		mutex_exit(&lfs_lock);
1476 		if (ap->a_command == LFCNBMAPV)
1477 			error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt);
1478 		else /* LFCNMARKV */
1479 			error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt);
1480 		if (error == 0)
1481 			error = copyout(blkiov, blkvp.blkiov,
1482 					blkcnt * sizeof(BLOCK_INFO));
1483 		mutex_enter(&lfs_lock);
1484 		if (--fs->lfs_sleepers == 0)
1485 			wakeup(&fs->lfs_sleepers);
1486 		mutex_exit(&lfs_lock);
1487 		lfs_free(fs, blkiov, LFS_NB_BLKIOV);
1488 		return error;
1489 
1490 	    case LFCNRECLAIM:
1491 		/*
1492 		 * Flush dirops and write Ifile, allowing empty segments
1493 		 * to be immediately reclaimed.
1494 		 */
1495 		lfs_writer_enter(fs, "pndirop");
1496 		off = fs->lfs_offset;
1497 		lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP);
1498 		lfs_flush_dirops(fs);
1499 		LFS_CLEANERINFO(cip, fs, bp);
1500 		oclean = cip->clean;
1501 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
1502 		lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP);
1503 		fs->lfs_sp->seg_flags |= SEGM_PROT;
1504 		lfs_segunlock(fs);
1505 		lfs_writer_leave(fs);
1506 
1507 #ifdef DEBUG
1508 		LFS_CLEANERINFO(cip, fs, bp);
1509 		DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64
1510 		      " blocks, cleaned %" PRId32 " segments (activesb %d)\n",
1511 		      fs->lfs_offset - off, cip->clean - oclean,
1512 		      fs->lfs_activesb));
1513 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 0);
1514 #else
1515 		__USE(oclean);
1516 		__USE(off);
1517 #endif
1518 
1519 		return 0;
1520 
1521 	    case LFCNIFILEFH_COMPAT:
1522 		/* Return the filehandle of the Ifile */
1523 		if ((error = kauth_authorize_system(l->l_cred,
1524 		    KAUTH_SYSTEM_FILEHANDLE, 0, NULL, NULL, NULL)) != 0)
1525 			return (error);
1526 		fhp = (struct fhandle *)ap->a_data;
1527 		fhp->fh_fsid = *fsidp;
1528 		fh_size = 16;	/* former VFS_MAXFIDSIZ */
1529 		return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size);
1530 
1531 	    case LFCNIFILEFH_COMPAT2:
1532 	    case LFCNIFILEFH:
1533 		/* Return the filehandle of the Ifile */
1534 		fhp = (struct fhandle *)ap->a_data;
1535 		fhp->fh_fsid = *fsidp;
1536 		fh_size = sizeof(struct lfs_fhandle) -
1537 		    offsetof(fhandle_t, fh_fid);
1538 		return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size);
1539 
1540 	    case LFCNREWIND:
1541 		/* Move lfs_offset to the lowest-numbered segment */
1542 		return lfs_rewind(fs, *(int *)ap->a_data);
1543 
1544 	    case LFCNINVAL:
1545 		/* Mark a segment SEGUSE_INVAL */
1546 		LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp);
1547 		if (sup->su_nbytes > 0) {
1548 			brelse(bp, 0);
1549 			lfs_unset_inval_all(fs);
1550 			return EBUSY;
1551 		}
1552 		sup->su_flags |= SEGUSE_INVAL;
1553 		VOP_BWRITE(bp->b_vp, bp);
1554 		return 0;
1555 
1556 	    case LFCNRESIZE:
1557 		/* Resize the filesystem */
1558 		return lfs_resize_fs(fs, *(int *)ap->a_data);
1559 
1560 	    case LFCNWRAPSTOP:
1561 	    case LFCNWRAPSTOP_COMPAT:
1562 		/*
1563 		 * Hold lfs_newseg at segment 0; if requested, sleep until
1564 		 * the filesystem wraps around.  To support external agents
1565 		 * (dump, fsck-based regression test) that need to look at
1566 		 * a snapshot of the filesystem, without necessarily
1567 		 * requiring that all fs activity stops.
1568 		 */
1569 		if (fs->lfs_stoplwp == curlwp)
1570 			return EALREADY;
1571 
1572 		mutex_enter(&lfs_lock);
1573 		while (fs->lfs_stoplwp != NULL)
1574 			cv_wait(&fs->lfs_stopcv, &lfs_lock);
1575 		fs->lfs_stoplwp = curlwp;
1576 		if (fs->lfs_nowrap == 0)
1577 			log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt);
1578 		++fs->lfs_nowrap;
1579 		if (*(int *)ap->a_data == 1
1580 		    || ap->a_command == LFCNWRAPSTOP_COMPAT) {
1581 			log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n");
1582 			error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER,
1583 				"segwrap", 0, &lfs_lock);
1584 			log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n");
1585 			if (error) {
1586 				lfs_wrapgo(fs, VTOI(ap->a_vp), 0);
1587 			}
1588 		}
1589 		mutex_exit(&lfs_lock);
1590 		return 0;
1591 
1592 	    case LFCNWRAPGO:
1593 	    case LFCNWRAPGO_COMPAT:
1594 		/*
1595 		 * Having done its work, the agent wakes up the writer.
1596 		 * If the argument is 1, it sleeps until a new segment
1597 		 * is selected.
1598 		 */
1599 		mutex_enter(&lfs_lock);
1600 		error = lfs_wrapgo(fs, VTOI(ap->a_vp),
1601 				   ap->a_command == LFCNWRAPGO_COMPAT ? 1 :
1602 				    *((int *)ap->a_data));
1603 		mutex_exit(&lfs_lock);
1604 		return error;
1605 
1606 	    case LFCNWRAPPASS:
1607 		if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT))
1608 			return EALREADY;
1609 		mutex_enter(&lfs_lock);
1610 		if (fs->lfs_stoplwp != curlwp) {
1611 			mutex_exit(&lfs_lock);
1612 			return EALREADY;
1613 		}
1614 		if (fs->lfs_nowrap == 0) {
1615 			mutex_exit(&lfs_lock);
1616 			return EBUSY;
1617 		}
1618 		fs->lfs_wrappass = 1;
1619 		wakeup(&fs->lfs_wrappass);
1620 		/* Wait for the log to wrap, if asked */
1621 		if (*(int *)ap->a_data) {
1622 			mutex_enter(ap->a_vp->v_interlock);
1623 			if (lfs_vref(ap->a_vp) != 0)
1624 				panic("LFCNWRAPPASS: lfs_vref failed");
1625 			VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT;
1626 			log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n");
1627 			error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER,
1628 				"segwrap", 0, &lfs_lock);
1629 			log(LOG_NOTICE, "LFCNPASS done waiting\n");
1630 			VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT;
1631 			lfs_vunref(ap->a_vp);
1632 		}
1633 		mutex_exit(&lfs_lock);
1634 		return error;
1635 
1636 	    case LFCNWRAPSTATUS:
1637 		mutex_enter(&lfs_lock);
1638 		*(int *)ap->a_data = fs->lfs_wrapstatus;
1639 		mutex_exit(&lfs_lock);
1640 		return 0;
1641 
1642 	    default:
1643 		return ulfs_fcntl(v);
1644 	}
1645 	return 0;
1646 }
1647 
1648 int
1649 lfs_getpages(void *v)
1650 {
1651 	struct vop_getpages_args /* {
1652 		struct vnode *a_vp;
1653 		voff_t a_offset;
1654 		struct vm_page **a_m;
1655 		int *a_count;
1656 		int a_centeridx;
1657 		vm_prot_t a_access_type;
1658 		int a_advice;
1659 		int a_flags;
1660 	} */ *ap = v;
1661 
1662 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM &&
1663 	    (ap->a_access_type & VM_PROT_WRITE) != 0) {
1664 		return EPERM;
1665 	}
1666 	if ((ap->a_access_type & VM_PROT_WRITE) != 0) {
1667 		mutex_enter(&lfs_lock);
1668 		LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED);
1669 		mutex_exit(&lfs_lock);
1670 	}
1671 
1672 	/*
1673 	 * we're relying on the fact that genfs_getpages() always read in
1674 	 * entire filesystem blocks.
1675 	 */
1676 	return genfs_getpages(v);
1677 }
1678 
1679 /*
1680  * Wait for a page to become unbusy, possibly printing diagnostic messages
1681  * as well.
1682  *
1683  * Called with vp->v_interlock held; return with it held.
1684  */
1685 static void
1686 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label)
1687 {
1688 	KASSERT(mutex_owned(vp->v_interlock));
1689 	if ((pg->flags & PG_BUSY) == 0)
1690 		return;		/* Nothing to wait for! */
1691 
1692 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN)
1693 	static struct vm_page *lastpg;
1694 
1695 	if (label != NULL && pg != lastpg) {
1696 		if (pg->owner_tag) {
1697 			printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n",
1698 			       curproc->p_pid, curlwp->l_lid, label,
1699 			       pg, pg->owner, pg->lowner, pg->owner_tag);
1700 		} else {
1701 			printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n",
1702 			       curproc->p_pid, curlwp->l_lid, label, pg);
1703 		}
1704 	}
1705 	lastpg = pg;
1706 #endif
1707 
1708 	pg->flags |= PG_WANTED;
1709 	UVM_UNLOCK_AND_WAIT(pg, vp->v_interlock, 0, "lfsput", 0);
1710 	mutex_enter(vp->v_interlock);
1711 }
1712 
1713 /*
1714  * This routine is called by lfs_putpages() when it can't complete the
1715  * write because a page is busy.  This means that either (1) someone,
1716  * possibly the pagedaemon, is looking at this page, and will give it up
1717  * presently; or (2) we ourselves are holding the page busy in the
1718  * process of being written (either gathered or actually on its way to
1719  * disk).  We don't need to give up the segment lock, but we might need
1720  * to call lfs_writeseg() to expedite the page's journey to disk.
1721  *
1722  * Called with vp->v_interlock held; return with it held.
1723  */
1724 /* #define BUSYWAIT */
1725 static void
1726 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg,
1727 	       int seglocked, const char *label)
1728 {
1729 	KASSERT(mutex_owned(vp->v_interlock));
1730 #ifndef BUSYWAIT
1731 	struct inode *ip = VTOI(vp);
1732 	struct segment *sp = fs->lfs_sp;
1733 	int count = 0;
1734 
1735 	if (pg == NULL)
1736 		return;
1737 
1738 	while (pg->flags & PG_BUSY &&
1739 	    pg->uobject == &vp->v_uobj) {
1740 		mutex_exit(vp->v_interlock);
1741 		if (sp->cbpp - sp->bpp > 1) {
1742 			/* Write gathered pages */
1743 			lfs_updatemeta(sp);
1744 			lfs_release_finfo(fs);
1745 			(void) lfs_writeseg(fs, sp);
1746 
1747 			/*
1748 			 * Reinitialize FIP
1749 			 */
1750 			KASSERT(sp->vp == vp);
1751 			lfs_acquire_finfo(fs, ip->i_number,
1752 					  ip->i_gen);
1753 		}
1754 		++count;
1755 		mutex_enter(vp->v_interlock);
1756 		wait_for_page(vp, pg, label);
1757 	}
1758 	if (label != NULL && count > 1) {
1759 		DLOG((DLOG_PAGE, "lfs_putpages[%d]: %s: %sn = %d\n",
1760 		      curproc->p_pid, label, (count > 0 ? "looping, " : ""),
1761 		      count));
1762 	}
1763 #else
1764 	preempt(1);
1765 #endif
1766 	KASSERT(mutex_owned(vp->v_interlock));
1767 }
1768 
1769 /*
1770  * Make sure that for all pages in every block in the given range,
1771  * either all are dirty or all are clean.  If any of the pages
1772  * we've seen so far are dirty, put the vnode on the paging chain,
1773  * and mark it IN_PAGING.
1774  *
1775  * If checkfirst != 0, don't check all the pages but return at the
1776  * first dirty page.
1777  */
1778 static int
1779 check_dirty(struct lfs *fs, struct vnode *vp,
1780 	    off_t startoffset, off_t endoffset, off_t blkeof,
1781 	    int flags, int checkfirst, struct vm_page **pgp)
1782 {
1783 	int by_list;
1784 	struct vm_page *curpg = NULL; /* XXX: gcc */
1785 	struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg;
1786 	off_t soff = 0; /* XXX: gcc */
1787 	voff_t off;
1788 	int i;
1789 	int nonexistent;
1790 	int any_dirty;	/* number of dirty pages */
1791 	int dirty;	/* number of dirty pages in a block */
1792 	int tdirty;
1793 	int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT;
1794 	int pagedaemon = (curlwp == uvm.pagedaemon_lwp);
1795 
1796 	KASSERT(mutex_owned(vp->v_interlock));
1797 	ASSERT_MAYBE_SEGLOCK(fs);
1798   top:
1799 	by_list = (vp->v_uobj.uo_npages <=
1800 		   ((endoffset - startoffset) >> PAGE_SHIFT) *
1801 		   UVM_PAGE_TREE_PENALTY);
1802 	any_dirty = 0;
1803 
1804 	if (by_list) {
1805 		curpg = TAILQ_FIRST(&vp->v_uobj.memq);
1806 	} else {
1807 		soff = startoffset;
1808 	}
1809 	while (by_list || soff < MIN(blkeof, endoffset)) {
1810 		if (by_list) {
1811 			/*
1812 			 * Find the first page in a block.  Skip
1813 			 * blocks outside our area of interest or beyond
1814 			 * the end of file.
1815 			 */
1816 			KASSERT(curpg == NULL
1817 			    || (curpg->flags & PG_MARKER) == 0);
1818 			if (pages_per_block > 1) {
1819 				while (curpg &&
1820 				    ((curpg->offset & fs->lfs_bmask) ||
1821 				    curpg->offset >= vp->v_size ||
1822 				    curpg->offset >= endoffset)) {
1823 					curpg = TAILQ_NEXT(curpg, listq.queue);
1824 					KASSERT(curpg == NULL ||
1825 					    (curpg->flags & PG_MARKER) == 0);
1826 				}
1827 			}
1828 			if (curpg == NULL)
1829 				break;
1830 			soff = curpg->offset;
1831 		}
1832 
1833 		/*
1834 		 * Mark all pages in extended range busy; find out if any
1835 		 * of them are dirty.
1836 		 */
1837 		nonexistent = dirty = 0;
1838 		for (i = 0; i == 0 || i < pages_per_block; i++) {
1839 			KASSERT(mutex_owned(vp->v_interlock));
1840 			if (by_list && pages_per_block <= 1) {
1841 				pgs[i] = pg = curpg;
1842 			} else {
1843 				off = soff + (i << PAGE_SHIFT);
1844 				pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off);
1845 				if (pg == NULL) {
1846 					++nonexistent;
1847 					continue;
1848 				}
1849 			}
1850 			KASSERT(pg != NULL);
1851 
1852 			/*
1853 			 * If we're holding the segment lock, we can deadlock
1854 			 * against a process that has our page and is waiting
1855 			 * for the cleaner, while the cleaner waits for the
1856 			 * segment lock.  Just bail in that case.
1857 			 */
1858 			if ((pg->flags & PG_BUSY) &&
1859 			    (pagedaemon || LFS_SEGLOCK_HELD(fs))) {
1860 				if (i > 0)
1861 					uvm_page_unbusy(pgs, i);
1862 				DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n"));
1863 				if (pgp)
1864 					*pgp = pg;
1865 				KASSERT(mutex_owned(vp->v_interlock));
1866 				return -1;
1867 			}
1868 
1869 			while (pg->flags & PG_BUSY) {
1870 				wait_for_page(vp, pg, NULL);
1871 				KASSERT(mutex_owned(vp->v_interlock));
1872 				if (i > 0)
1873 					uvm_page_unbusy(pgs, i);
1874 				KASSERT(mutex_owned(vp->v_interlock));
1875 				goto top;
1876 			}
1877 			pg->flags |= PG_BUSY;
1878 			UVM_PAGE_OWN(pg, "lfs_putpages");
1879 
1880 			pmap_page_protect(pg, VM_PROT_NONE);
1881 			tdirty = (pmap_clear_modify(pg) ||
1882 				  (pg->flags & PG_CLEAN) == 0);
1883 			dirty += tdirty;
1884 		}
1885 		if (pages_per_block > 0 && nonexistent >= pages_per_block) {
1886 			if (by_list) {
1887 				curpg = TAILQ_NEXT(curpg, listq.queue);
1888 			} else {
1889 				soff += fs->lfs_bsize;
1890 			}
1891 			continue;
1892 		}
1893 
1894 		any_dirty += dirty;
1895 		KASSERT(nonexistent == 0);
1896 		KASSERT(mutex_owned(vp->v_interlock));
1897 
1898 		/*
1899 		 * If any are dirty make all dirty; unbusy them,
1900 		 * but if we were asked to clean, wire them so that
1901 		 * the pagedaemon doesn't bother us about them while
1902 		 * they're on their way to disk.
1903 		 */
1904 		for (i = 0; i == 0 || i < pages_per_block; i++) {
1905 			KASSERT(mutex_owned(vp->v_interlock));
1906 			pg = pgs[i];
1907 			KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI)));
1908 			KASSERT(pg->flags & PG_BUSY);
1909 			if (dirty) {
1910 				pg->flags &= ~PG_CLEAN;
1911 				if (flags & PGO_FREE) {
1912 					/*
1913 					 * Wire the page so that
1914 					 * pdaemon doesn't see it again.
1915 					 */
1916 					mutex_enter(&uvm_pageqlock);
1917 					uvm_pagewire(pg);
1918 					mutex_exit(&uvm_pageqlock);
1919 
1920 					/* Suspended write flag */
1921 					pg->flags |= PG_DELWRI;
1922 				}
1923 			}
1924 			if (pg->flags & PG_WANTED)
1925 				wakeup(pg);
1926 			pg->flags &= ~(PG_WANTED|PG_BUSY);
1927 			UVM_PAGE_OWN(pg, NULL);
1928 		}
1929 
1930 		if (checkfirst && any_dirty)
1931 			break;
1932 
1933 		if (by_list) {
1934 			curpg = TAILQ_NEXT(curpg, listq.queue);
1935 		} else {
1936 			soff += MAX(PAGE_SIZE, fs->lfs_bsize);
1937 		}
1938 	}
1939 
1940 	KASSERT(mutex_owned(vp->v_interlock));
1941 	return any_dirty;
1942 }
1943 
1944 /*
1945  * lfs_putpages functions like genfs_putpages except that
1946  *
1947  * (1) It needs to bounds-check the incoming requests to ensure that
1948  *     they are block-aligned; if they are not, expand the range and
1949  *     do the right thing in case, e.g., the requested range is clean
1950  *     but the expanded range is dirty.
1951  *
1952  * (2) It needs to explicitly send blocks to be written when it is done.
1953  *     If VOP_PUTPAGES is called without the seglock held, we simply take
1954  *     the seglock and let lfs_segunlock wait for us.
1955  *     XXX There might be a bad situation if we have to flush a vnode while
1956  *     XXX lfs_markv is in operation.  As of this writing we panic in this
1957  *     XXX case.
1958  *
1959  * Assumptions:
1960  *
1961  * (1) The caller does not hold any pages in this vnode busy.  If it does,
1962  *     there is a danger that when we expand the page range and busy the
1963  *     pages we will deadlock.
1964  *
1965  * (2) We are called with vp->v_interlock held; we must return with it
1966  *     released.
1967  *
1968  * (3) We don't absolutely have to free pages right away, provided that
1969  *     the request does not have PGO_SYNCIO.  When the pagedaemon gives
1970  *     us a request with PGO_FREE, we take the pages out of the paging
1971  *     queue and wake up the writer, which will handle freeing them for us.
1972  *
1973  *     We ensure that for any filesystem block, all pages for that
1974  *     block are either resident or not, even if those pages are higher
1975  *     than EOF; that means that we will be getting requests to free
1976  *     "unused" pages above EOF all the time, and should ignore them.
1977  *
1978  * (4) If we are called with PGO_LOCKED, the finfo array we are to write
1979  *     into has been set up for us by lfs_writefile.  If not, we will
1980  *     have to handle allocating and/or freeing an finfo entry.
1981  *
1982  * XXX note that we're (ab)using PGO_LOCKED as "seglock held".
1983  */
1984 
1985 /* How many times to loop before we should start to worry */
1986 #define TOOMANY 4
1987 
1988 int
1989 lfs_putpages(void *v)
1990 {
1991 	int error;
1992 	struct vop_putpages_args /* {
1993 		struct vnode *a_vp;
1994 		voff_t a_offlo;
1995 		voff_t a_offhi;
1996 		int a_flags;
1997 	} */ *ap = v;
1998 	struct vnode *vp;
1999 	struct inode *ip;
2000 	struct lfs *fs;
2001 	struct segment *sp;
2002 	off_t origoffset, startoffset, endoffset, origendoffset, blkeof;
2003 	off_t off, max_endoffset;
2004 	bool seglocked, sync, pagedaemon, reclaim;
2005 	struct vm_page *pg, *busypg;
2006 	UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist);
2007 	int oreclaim = 0;
2008 	int donewriting = 0;
2009 #ifdef DEBUG
2010 	int debug_n_again, debug_n_dirtyclean;
2011 #endif
2012 
2013 	vp = ap->a_vp;
2014 	ip = VTOI(vp);
2015 	fs = ip->i_lfs;
2016 	sync = (ap->a_flags & PGO_SYNCIO) != 0;
2017 	reclaim = (ap->a_flags & PGO_RECLAIM) != 0;
2018 	pagedaemon = (curlwp == uvm.pagedaemon_lwp);
2019 
2020 	KASSERT(mutex_owned(vp->v_interlock));
2021 
2022 	/* Putpages does nothing for metadata. */
2023 	if (vp == fs->lfs_ivnode || vp->v_type != VREG) {
2024 		mutex_exit(vp->v_interlock);
2025 		return 0;
2026 	}
2027 
2028 	/*
2029 	 * If there are no pages, don't do anything.
2030 	 */
2031 	if (vp->v_uobj.uo_npages == 0) {
2032 		if (TAILQ_EMPTY(&vp->v_uobj.memq) &&
2033 		    (vp->v_iflag & VI_ONWORKLST) &&
2034 		    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
2035 			vp->v_iflag &= ~VI_WRMAPDIRTY;
2036 			vn_syncer_remove_from_worklist(vp);
2037 		}
2038 		mutex_exit(vp->v_interlock);
2039 
2040 		/* Remove us from paging queue, if we were on it */
2041 		mutex_enter(&lfs_lock);
2042 		if (ip->i_flags & IN_PAGING) {
2043 			ip->i_flags &= ~IN_PAGING;
2044 			TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain);
2045 		}
2046 		mutex_exit(&lfs_lock);
2047 
2048 		KASSERT(!mutex_owned(vp->v_interlock));
2049 		return 0;
2050 	}
2051 
2052 	blkeof = lfs_blkroundup(fs, ip->i_size);
2053 
2054 	/*
2055 	 * Ignore requests to free pages past EOF but in the same block
2056 	 * as EOF, unless the vnode is being reclaimed or the request
2057 	 * is synchronous.  (If the request is sync, it comes from
2058 	 * lfs_truncate.)
2059 	 *
2060 	 * To avoid being flooded with this request, make these pages
2061 	 * look "active".
2062 	 */
2063 	if (!sync && !reclaim &&
2064 	    ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) {
2065 		origoffset = ap->a_offlo;
2066 		for (off = origoffset; off < blkeof; off += fs->lfs_bsize) {
2067 			pg = uvm_pagelookup(&vp->v_uobj, off);
2068 			KASSERT(pg != NULL);
2069 			while (pg->flags & PG_BUSY) {
2070 				pg->flags |= PG_WANTED;
2071 				UVM_UNLOCK_AND_WAIT(pg, vp->v_interlock, 0,
2072 						    "lfsput2", 0);
2073 				mutex_enter(vp->v_interlock);
2074 			}
2075 			mutex_enter(&uvm_pageqlock);
2076 			uvm_pageactivate(pg);
2077 			mutex_exit(&uvm_pageqlock);
2078 		}
2079 		ap->a_offlo = blkeof;
2080 		if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) {
2081 			mutex_exit(vp->v_interlock);
2082 			return 0;
2083 		}
2084 	}
2085 
2086 	/*
2087 	 * Extend page range to start and end at block boundaries.
2088 	 * (For the purposes of VOP_PUTPAGES, fragments don't exist.)
2089 	 */
2090 	origoffset = ap->a_offlo;
2091 	origendoffset = ap->a_offhi;
2092 	startoffset = origoffset & ~(fs->lfs_bmask);
2093 	max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift)
2094 					       << fs->lfs_bshift;
2095 
2096 	if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) {
2097 		endoffset = max_endoffset;
2098 		origendoffset = endoffset;
2099 	} else {
2100 		origendoffset = round_page(ap->a_offhi);
2101 		endoffset = round_page(lfs_blkroundup(fs, origendoffset));
2102 	}
2103 
2104 	KASSERT(startoffset > 0 || endoffset >= startoffset);
2105 	if (startoffset == endoffset) {
2106 		/* Nothing to do, why were we called? */
2107 		mutex_exit(vp->v_interlock);
2108 		DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %"
2109 		      PRId64 "\n", startoffset));
2110 		return 0;
2111 	}
2112 
2113 	ap->a_offlo = startoffset;
2114 	ap->a_offhi = endoffset;
2115 
2116 	/*
2117 	 * If not cleaning, just send the pages through genfs_putpages
2118 	 * to be returned to the pool.
2119 	 */
2120 	if (!(ap->a_flags & PGO_CLEANIT)) {
2121 		DLOG((DLOG_PAGE, "lfs_putpages: no cleanit vn %p ino %d (flags %x)\n",
2122 		      vp, (int)ip->i_number, ap->a_flags));
2123 		int r = genfs_putpages(v);
2124 		KASSERT(!mutex_owned(vp->v_interlock));
2125 		return r;
2126 	}
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 		KASSERT(mutex_owned(vp->v_interlock));
2141 
2142 		/* Count the number of dirty pages */
2143 		r = check_dirty(fs, vp, startoffset, endoffset, blkeof,
2144 				ap->a_flags, 1, NULL);
2145 		if (r < 0) {
2146 			/* Pages are busy with another process */
2147 			mutex_exit(vp->v_interlock);
2148 			return EDEADLK;
2149 		}
2150 		if (r > 0) /* Some pages are dirty */
2151 			break;
2152 
2153 		/*
2154 		 * Sometimes pages are dirtied between the time that
2155 		 * we check and the time we try to clean them.
2156 		 * Instruct lfs_gop_write to return EDEADLK in this case
2157 		 * so we can write them properly.
2158 		 */
2159 		ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE;
2160 		r = genfs_do_putpages(vp, startoffset, endoffset,
2161 				       ap->a_flags & ~PGO_SYNCIO, &busypg);
2162 		ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE;
2163 		if (r != EDEADLK) {
2164 			KASSERT(!mutex_owned(vp->v_interlock));
2165  			return r;
2166 		}
2167 
2168 		/* One of the pages was busy.  Start over. */
2169 		mutex_enter(vp->v_interlock);
2170 		wait_for_page(vp, busypg, "dirtyclean");
2171 #ifdef DEBUG
2172 		++debug_n_dirtyclean;
2173 #endif
2174 	} while(1);
2175 
2176 #ifdef DEBUG
2177 	if (debug_n_dirtyclean > TOOMANY)
2178 		DLOG((DLOG_PAGE, "lfs_putpages: dirtyclean: looping, n = %d\n",
2179 		      debug_n_dirtyclean));
2180 #endif
2181 
2182 	/*
2183 	 * Dirty and asked to clean.
2184 	 *
2185 	 * Pagedaemon can't actually write LFS pages; wake up
2186 	 * the writer to take care of that.  The writer will
2187 	 * notice the pager inode queue and act on that.
2188 	 *
2189 	 * XXX We must drop the vp->interlock before taking the lfs_lock or we
2190 	 * get a nasty deadlock with lfs_flush_pchain().
2191 	 */
2192 	if (pagedaemon) {
2193 		mutex_exit(vp->v_interlock);
2194 		mutex_enter(&lfs_lock);
2195 		if (!(ip->i_flags & IN_PAGING)) {
2196 			ip->i_flags |= IN_PAGING;
2197 			TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain);
2198 		}
2199 		wakeup(&lfs_writer_daemon);
2200 		mutex_exit(&lfs_lock);
2201 		preempt();
2202 		KASSERT(!mutex_owned(vp->v_interlock));
2203 		return EWOULDBLOCK;
2204 	}
2205 
2206 	/*
2207 	 * If this is a file created in a recent dirop, we can't flush its
2208 	 * inode until the dirop is complete.  Drain dirops, then flush the
2209 	 * filesystem (taking care of any other pending dirops while we're
2210 	 * at it).
2211 	 */
2212 	if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT &&
2213 	    (vp->v_uflag & VU_DIROP)) {
2214 		DLOG((DLOG_PAGE, "lfs_putpages: flushing VU_DIROP\n"));
2215 
2216  		lfs_writer_enter(fs, "ppdirop");
2217 
2218 		/* Note if we hold the vnode locked */
2219 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
2220 		{
2221 		    DLOG((DLOG_PAGE, "lfs_putpages: dirop inode already locked\n"));
2222 		} else {
2223 		    DLOG((DLOG_PAGE, "lfs_putpages: dirop inode not locked\n"));
2224 		}
2225 		mutex_exit(vp->v_interlock);
2226 
2227 		mutex_enter(&lfs_lock);
2228 		lfs_flush_fs(fs, sync ? SEGM_SYNC : 0);
2229 		mutex_exit(&lfs_lock);
2230 
2231 		mutex_enter(vp->v_interlock);
2232 		lfs_writer_leave(fs);
2233 
2234 		/* The flush will have cleaned out this vnode as well,
2235 		   no need to do more to it. */
2236 	}
2237 
2238 	/*
2239 	 * This is it.	We are going to write some pages.  From here on
2240 	 * down it's all just mechanics.
2241 	 *
2242 	 * Don't let genfs_putpages wait; lfs_segunlock will wait for us.
2243 	 */
2244 	ap->a_flags &= ~PGO_SYNCIO;
2245 
2246 	/*
2247 	 * If we've already got the seglock, flush the node and return.
2248 	 * The FIP has already been set up for us by lfs_writefile,
2249 	 * and FIP cleanup and lfs_updatemeta will also be done there,
2250 	 * unless genfs_putpages returns EDEADLK; then we must flush
2251 	 * what we have, and correct FIP and segment header accounting.
2252 	 */
2253   get_seglock:
2254 	/*
2255 	 * If we are not called with the segment locked, lock it.
2256 	 * Account for a new FIP in the segment header, and set sp->vp.
2257 	 * (This should duplicate the setup at the top of lfs_writefile().)
2258 	 */
2259 	seglocked = (ap->a_flags & PGO_LOCKED) != 0;
2260 	if (!seglocked) {
2261 		mutex_exit(vp->v_interlock);
2262 		error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0));
2263 		if (error != 0) {
2264 			KASSERT(!mutex_owned(vp->v_interlock));
2265  			return error;
2266 		}
2267 		mutex_enter(vp->v_interlock);
2268 		lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
2269 	}
2270 	sp = fs->lfs_sp;
2271 	KASSERT(sp->vp == NULL);
2272 	sp->vp = vp;
2273 
2274 	/* Note segments written by reclaim; only for debugging */
2275 	if (vdead_check(vp, VDEAD_NOWAIT) != 0) {
2276 		sp->seg_flags |= SEGM_RECLAIM;
2277 		fs->lfs_reclino = ip->i_number;
2278 	}
2279 
2280 	/*
2281 	 * Ensure that the partial segment is marked SS_DIROP if this
2282 	 * vnode is a DIROP.
2283 	 */
2284 	if (!seglocked && vp->v_uflag & VU_DIROP)
2285 		((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT);
2286 
2287 	/*
2288 	 * Loop over genfs_putpages until all pages are gathered.
2289 	 * genfs_putpages() drops the interlock, so reacquire it if necessary.
2290 	 * Whenever we lose the interlock we have to rerun check_dirty, as
2291 	 * well, since more pages might have been dirtied in our absence.
2292 	 */
2293 #ifdef DEBUG
2294 	debug_n_again = 0;
2295 #endif
2296 	do {
2297 		busypg = NULL;
2298 		KASSERT(mutex_owned(vp->v_interlock));
2299 		if (check_dirty(fs, vp, startoffset, endoffset, blkeof,
2300 				ap->a_flags, 0, &busypg) < 0) {
2301 			mutex_exit(vp->v_interlock);
2302 			/* XXX why? --ks */
2303 			mutex_enter(vp->v_interlock);
2304 			write_and_wait(fs, vp, busypg, seglocked, NULL);
2305 			if (!seglocked) {
2306 				mutex_exit(vp->v_interlock);
2307 				lfs_release_finfo(fs);
2308 				lfs_segunlock(fs);
2309 				mutex_enter(vp->v_interlock);
2310 			}
2311 			sp->vp = NULL;
2312 			goto get_seglock;
2313 		}
2314 
2315 		busypg = NULL;
2316 		KASSERT(!mutex_owned(&uvm_pageqlock));
2317 		oreclaim = (ap->a_flags & PGO_RECLAIM);
2318 		ap->a_flags &= ~PGO_RECLAIM;
2319 		error = genfs_do_putpages(vp, startoffset, endoffset,
2320 					   ap->a_flags, &busypg);
2321 		ap->a_flags |= oreclaim;
2322 
2323 		if (error == EDEADLK || error == EAGAIN) {
2324 			DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned"
2325 			      " %d ino %d off %x (seg %d)\n", error,
2326 			      ip->i_number, fs->lfs_offset,
2327 			      lfs_dtosn(fs, fs->lfs_offset)));
2328 
2329 			if (oreclaim) {
2330 				mutex_enter(vp->v_interlock);
2331 				write_and_wait(fs, vp, busypg, seglocked, "again");
2332 				mutex_exit(vp->v_interlock);
2333 			} else {
2334 				if ((sp->seg_flags & SEGM_SINGLE) &&
2335 				    fs->lfs_curseg != fs->lfs_startseg)
2336 					donewriting = 1;
2337 			}
2338 		} else if (error) {
2339 			DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned"
2340 			      " %d ino %d off %x (seg %d)\n", error,
2341 			      (int)ip->i_number, fs->lfs_offset,
2342 			      lfs_dtosn(fs, fs->lfs_offset)));
2343 		}
2344 		/* genfs_do_putpages loses the interlock */
2345 #ifdef DEBUG
2346 		++debug_n_again;
2347 #endif
2348 		if (oreclaim && error == EAGAIN) {
2349 			DLOG((DLOG_PAGE, "vp %p ino %d vi_flags %x a_flags %x avoiding vclean panic\n",
2350 			      vp, (int)ip->i_number, vp->v_iflag, ap->a_flags));
2351 			mutex_enter(vp->v_interlock);
2352 		}
2353 		if (error == EDEADLK)
2354 			mutex_enter(vp->v_interlock);
2355 	} while (error == EDEADLK || (oreclaim && error == EAGAIN));
2356 #ifdef DEBUG
2357 	if (debug_n_again > TOOMANY)
2358 		DLOG((DLOG_PAGE, "lfs_putpages: again: looping, n = %d\n", debug_n_again));
2359 #endif
2360 
2361 	KASSERT(sp != NULL && sp->vp == vp);
2362 	if (!seglocked && !donewriting) {
2363 		sp->vp = NULL;
2364 
2365 		/* Write indirect blocks as well */
2366 		lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir);
2367 		lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir);
2368 		lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir);
2369 
2370 		KASSERT(sp->vp == NULL);
2371 		sp->vp = vp;
2372 	}
2373 
2374 	/*
2375 	 * Blocks are now gathered into a segment waiting to be written.
2376 	 * All that's left to do is update metadata, and write them.
2377 	 */
2378 	lfs_updatemeta(sp);
2379 	KASSERT(sp->vp == vp);
2380 	sp->vp = NULL;
2381 
2382 	/*
2383 	 * If we were called from lfs_writefile, we don't need to clean up
2384 	 * the FIP or unlock the segment lock.	We're done.
2385 	 */
2386 	if (seglocked) {
2387 		KASSERT(!mutex_owned(vp->v_interlock));
2388 		return error;
2389 	}
2390 
2391 	/* Clean up FIP and send it to disk. */
2392 	lfs_release_finfo(fs);
2393 	lfs_writeseg(fs, fs->lfs_sp);
2394 
2395 	/*
2396 	 * Remove us from paging queue if we wrote all our pages.
2397 	 */
2398 	if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) {
2399 		mutex_enter(&lfs_lock);
2400 		if (ip->i_flags & IN_PAGING) {
2401 			ip->i_flags &= ~IN_PAGING;
2402 			TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain);
2403 		}
2404 		mutex_exit(&lfs_lock);
2405 	}
2406 
2407 	/*
2408 	 * XXX - with the malloc/copy writeseg, the pages are freed by now
2409 	 * even if we don't wait (e.g. if we hold a nested lock).  This
2410 	 * will not be true if we stop using malloc/copy.
2411 	 */
2412 	KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT);
2413 	lfs_segunlock(fs);
2414 
2415 	/*
2416 	 * Wait for v_numoutput to drop to zero.  The seglock should
2417 	 * take care of this, but there is a slight possibility that
2418 	 * aiodoned might not have got around to our buffers yet.
2419 	 */
2420 	if (sync) {
2421 		mutex_enter(vp->v_interlock);
2422 		while (vp->v_numoutput > 0) {
2423 			DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on"
2424 			      " num %d\n", ip->i_number, vp->v_numoutput));
2425 			cv_wait(&vp->v_cv, vp->v_interlock);
2426 		}
2427 		mutex_exit(vp->v_interlock);
2428 	}
2429 	KASSERT(!mutex_owned(vp->v_interlock));
2430 	return error;
2431 }
2432 
2433 /*
2434  * Return the last logical file offset that should be written for this file
2435  * if we're doing a write that ends at "size".	If writing, we need to know
2436  * about sizes on disk, i.e. fragments if there are any; if reading, we need
2437  * to know about entire blocks.
2438  */
2439 void
2440 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags)
2441 {
2442 	struct inode *ip = VTOI(vp);
2443 	struct lfs *fs = ip->i_lfs;
2444 	daddr_t olbn, nlbn;
2445 
2446 	olbn = lfs_lblkno(fs, ip->i_size);
2447 	nlbn = lfs_lblkno(fs, size);
2448 	if (!(flags & GOP_SIZE_MEM) && nlbn < ULFS_NDADDR && olbn <= nlbn) {
2449 		*eobp = lfs_fragroundup(fs, size);
2450 	} else {
2451 		*eobp = lfs_blkroundup(fs, size);
2452 	}
2453 }
2454 
2455 #ifdef DEBUG
2456 void lfs_dump_vop(void *);
2457 
2458 void
2459 lfs_dump_vop(void *v)
2460 {
2461 	struct vop_putpages_args /* {
2462 		struct vnode *a_vp;
2463 		voff_t a_offlo;
2464 		voff_t a_offhi;
2465 		int a_flags;
2466 	} */ *ap = v;
2467 
2468 #ifdef DDB
2469 	vfs_vnode_print(ap->a_vp, 0, printf);
2470 #endif
2471 	lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din);
2472 }
2473 #endif
2474 
2475 int
2476 lfs_mmap(void *v)
2477 {
2478 	struct vop_mmap_args /* {
2479 		const struct vnodeop_desc *a_desc;
2480 		struct vnode *a_vp;
2481 		vm_prot_t a_prot;
2482 		kauth_cred_t a_cred;
2483 	} */ *ap = v;
2484 
2485 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM)
2486 		return EOPNOTSUPP;
2487 	return ulfs_mmap(v);
2488 }
2489 
2490 static int
2491 lfs_openextattr(void *v)
2492 {
2493 	struct vop_openextattr_args /* {
2494 		struct vnode *a_vp;
2495 		kauth_cred_t a_cred;
2496 		struct proc *a_p;
2497 	} */ *ap = v;
2498 	struct inode *ip = VTOI(ap->a_vp);
2499 	struct ulfsmount *ump = ip->i_ump;
2500 	//struct lfs *fs = ip->i_lfs;
2501 
2502 	/* Not supported for ULFS1 file systems. */
2503 	if (ump->um_fstype == ULFS1)
2504 		return (EOPNOTSUPP);
2505 
2506 	/* XXX Not implemented for ULFS2 file systems. */
2507 	return (EOPNOTSUPP);
2508 }
2509 
2510 static int
2511 lfs_closeextattr(void *v)
2512 {
2513 	struct vop_closeextattr_args /* {
2514 		struct vnode *a_vp;
2515 		int a_commit;
2516 		kauth_cred_t a_cred;
2517 		struct proc *a_p;
2518 	} */ *ap = v;
2519 	struct inode *ip = VTOI(ap->a_vp);
2520 	struct ulfsmount *ump = ip->i_ump;
2521 	//struct lfs *fs = ip->i_lfs;
2522 
2523 	/* Not supported for ULFS1 file systems. */
2524 	if (ump->um_fstype == ULFS1)
2525 		return (EOPNOTSUPP);
2526 
2527 	/* XXX Not implemented for ULFS2 file systems. */
2528 	return (EOPNOTSUPP);
2529 }
2530 
2531 static int
2532 lfs_getextattr(void *v)
2533 {
2534 	struct vop_getextattr_args /* {
2535 		struct vnode *a_vp;
2536 		int a_attrnamespace;
2537 		const char *a_name;
2538 		struct uio *a_uio;
2539 		size_t *a_size;
2540 		kauth_cred_t a_cred;
2541 		struct proc *a_p;
2542 	} */ *ap = v;
2543 	struct vnode *vp = ap->a_vp;
2544 	struct inode *ip = VTOI(vp);
2545 	struct ulfsmount *ump = ip->i_ump;
2546 	//struct lfs *fs = ip->i_lfs;
2547 	int error;
2548 
2549 	if (ump->um_fstype == ULFS1) {
2550 #ifdef LFS_EXTATTR
2551 		fstrans_start(vp->v_mount, FSTRANS_SHARED);
2552 		error = ulfs_getextattr(ap);
2553 		fstrans_done(vp->v_mount);
2554 #else
2555 		error = EOPNOTSUPP;
2556 #endif
2557 		return error;
2558 	}
2559 
2560 	/* XXX Not implemented for ULFS2 file systems. */
2561 	return (EOPNOTSUPP);
2562 }
2563 
2564 static int
2565 lfs_setextattr(void *v)
2566 {
2567 	struct vop_setextattr_args /* {
2568 		struct vnode *a_vp;
2569 		int a_attrnamespace;
2570 		const char *a_name;
2571 		struct uio *a_uio;
2572 		kauth_cred_t a_cred;
2573 		struct proc *a_p;
2574 	} */ *ap = v;
2575 	struct vnode *vp = ap->a_vp;
2576 	struct inode *ip = VTOI(vp);
2577 	struct ulfsmount *ump = ip->i_ump;
2578 	//struct lfs *fs = ip->i_lfs;
2579 	int error;
2580 
2581 	if (ump->um_fstype == ULFS1) {
2582 #ifdef LFS_EXTATTR
2583 		fstrans_start(vp->v_mount, FSTRANS_SHARED);
2584 		error = ulfs_setextattr(ap);
2585 		fstrans_done(vp->v_mount);
2586 #else
2587 		error = EOPNOTSUPP;
2588 #endif
2589 		return error;
2590 	}
2591 
2592 	/* XXX Not implemented for ULFS2 file systems. */
2593 	return (EOPNOTSUPP);
2594 }
2595 
2596 static int
2597 lfs_listextattr(void *v)
2598 {
2599 	struct vop_listextattr_args /* {
2600 		struct vnode *a_vp;
2601 		int a_attrnamespace;
2602 		struct uio *a_uio;
2603 		size_t *a_size;
2604 		kauth_cred_t a_cred;
2605 		struct proc *a_p;
2606 	} */ *ap = v;
2607 	struct vnode *vp = ap->a_vp;
2608 	struct inode *ip = VTOI(vp);
2609 	struct ulfsmount *ump = ip->i_ump;
2610 	//struct lfs *fs = ip->i_lfs;
2611 	int error;
2612 
2613 	if (ump->um_fstype == ULFS1) {
2614 #ifdef LFS_EXTATTR
2615 		fstrans_start(vp->v_mount, FSTRANS_SHARED);
2616 		error = ulfs_listextattr(ap);
2617 		fstrans_done(vp->v_mount);
2618 #else
2619 		error = EOPNOTSUPP;
2620 #endif
2621 		return error;
2622 	}
2623 
2624 	/* XXX Not implemented for ULFS2 file systems. */
2625 	return (EOPNOTSUPP);
2626 }
2627 
2628 static int
2629 lfs_deleteextattr(void *v)
2630 {
2631 	struct vop_deleteextattr_args /* {
2632 		struct vnode *a_vp;
2633 		int a_attrnamespace;
2634 		kauth_cred_t a_cred;
2635 		struct proc *a_p;
2636 	} */ *ap = v;
2637 	struct vnode *vp = ap->a_vp;
2638 	struct inode *ip = VTOI(vp);
2639 	struct ulfsmount *ump = ip->i_ump;
2640 	//struct fs *fs = ip->i_lfs;
2641 	int error;
2642 
2643 	if (ump->um_fstype == ULFS1) {
2644 #ifdef LFS_EXTATTR
2645 		fstrans_start(vp->v_mount, FSTRANS_SHARED);
2646 		error = ulfs_deleteextattr(ap);
2647 		fstrans_done(vp->v_mount);
2648 #else
2649 		error = EOPNOTSUPP;
2650 #endif
2651 		return error;
2652 	}
2653 
2654 	/* XXX Not implemented for ULFS2 file systems. */
2655 	return (EOPNOTSUPP);
2656 }
2657