xref: /netbsd-src/sys/ufs/lfs/lfs_vfsops.c (revision a5847cc334d9a7029f6352b847e9e8d71a0f9e0c)
1 /*	$NetBSD: lfs_vfsops.c,v 1.290 2011/07/11 08:27:40 hannken Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003, 2007, 2007
5  *     The NetBSD Foundation, Inc.
6  * All rights reserved.
7  *
8  * This code is derived from software contributed to The NetBSD Foundation
9  * by Konrad E. Schroder <perseant@hhhh.org>.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 /*-
33  * Copyright (c) 1989, 1991, 1993, 1994
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. Neither the name of the University nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  *
60  *	@(#)lfs_vfsops.c	8.20 (Berkeley) 6/10/95
61  */
62 
63 #include <sys/cdefs.h>
64 __KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.290 2011/07/11 08:27:40 hannken Exp $");
65 
66 #if defined(_KERNEL_OPT)
67 #include "opt_lfs.h"
68 #include "opt_quota.h"
69 #endif
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/namei.h>
74 #include <sys/proc.h>
75 #include <sys/kernel.h>
76 #include <sys/vnode.h>
77 #include <sys/mount.h>
78 #include <sys/kthread.h>
79 #include <sys/buf.h>
80 #include <sys/device.h>
81 #include <sys/mbuf.h>
82 #include <sys/file.h>
83 #include <sys/disklabel.h>
84 #include <sys/ioctl.h>
85 #include <sys/errno.h>
86 #include <sys/malloc.h>
87 #include <sys/pool.h>
88 #include <sys/socket.h>
89 #include <sys/syslog.h>
90 #include <uvm/uvm_extern.h>
91 #include <sys/sysctl.h>
92 #include <sys/conf.h>
93 #include <sys/kauth.h>
94 #include <sys/module.h>
95 #include <sys/syscallvar.h>
96 #include <sys/syscall.h>
97 #include <sys/syscallargs.h>
98 
99 #include <miscfs/specfs/specdev.h>
100 
101 #include <ufs/ufs/quota.h>
102 #include <ufs/ufs/inode.h>
103 #include <ufs/ufs/ufsmount.h>
104 #include <ufs/ufs/ufs_extern.h>
105 
106 #include <uvm/uvm.h>
107 #include <uvm/uvm_stat.h>
108 #include <uvm/uvm_pager.h>
109 #include <uvm/uvm_pdaemon.h>
110 
111 #include <ufs/lfs/lfs.h>
112 #include <ufs/lfs/lfs_extern.h>
113 
114 #include <miscfs/genfs/genfs.h>
115 #include <miscfs/genfs/genfs_node.h>
116 
117 MODULE(MODULE_CLASS_VFS, lfs, "ffs");
118 
119 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
120 static bool lfs_issequential_hole(const struct ufsmount *,
121     daddr_t, daddr_t);
122 
123 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *);
124 
125 static struct sysctllog *lfs_sysctl_log;
126 
127 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc;
128 extern const struct vnodeopv_desc lfs_specop_opv_desc;
129 extern const struct vnodeopv_desc lfs_fifoop_opv_desc;
130 
131 pid_t lfs_writer_daemon = 0;
132 int lfs_do_flush = 0;
133 #ifdef LFS_KERNEL_RFW
134 int lfs_do_rfw = 0;
135 #endif
136 
137 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
138 	&lfs_vnodeop_opv_desc,
139 	&lfs_specop_opv_desc,
140 	&lfs_fifoop_opv_desc,
141 	NULL,
142 };
143 
144 struct vfsops lfs_vfsops = {
145 	MOUNT_LFS,
146 	sizeof (struct ufs_args),
147 	lfs_mount,
148 	ufs_start,
149 	lfs_unmount,
150 	ufs_root,
151 	ufs_quotactl,
152 	lfs_statvfs,
153 	lfs_sync,
154 	lfs_vget,
155 	lfs_fhtovp,
156 	lfs_vptofh,
157 	lfs_init,
158 	lfs_reinit,
159 	lfs_done,
160 	lfs_mountroot,
161 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
162 	vfs_stdextattrctl,
163 	(void *)eopnotsupp,	/* vfs_suspendctl */
164 	genfs_renamelock_enter,
165 	genfs_renamelock_exit,
166 	(void *)eopnotsupp,
167 	lfs_vnodeopv_descs,
168 	0,
169 	{ NULL, NULL },
170 };
171 
172 const struct genfs_ops lfs_genfsops = {
173 	.gop_size = lfs_gop_size,
174 	.gop_alloc = ufs_gop_alloc,
175 	.gop_write = lfs_gop_write,
176 	.gop_markupdate = ufs_gop_markupdate,
177 };
178 
179 static const struct ufs_ops lfs_ufsops = {
180 	.uo_itimes = NULL,
181 	.uo_update = lfs_update,
182 	.uo_truncate = lfs_truncate,
183 	.uo_valloc = lfs_valloc,
184 	.uo_vfree = lfs_vfree,
185 	.uo_balloc = lfs_balloc,
186 	.uo_unmark_vnode = lfs_unmark_vnode,
187 };
188 
189 struct shortlong {
190 	const char *sname;
191 	const char *lname;
192 };
193 
194 static int
195 sysctl_lfs_dostats(SYSCTLFN_ARGS)
196 {
197 	extern struct lfs_stats lfs_stats;
198 	extern int lfs_dostats;
199 	int error;
200 
201 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
202 	if (error || newp == NULL)
203 		return (error);
204 
205 	if (lfs_dostats == 0)
206 		memset(&lfs_stats, 0, sizeof(lfs_stats));
207 
208 	return (0);
209 }
210 
211 static void
212 lfs_sysctl_setup(struct sysctllog **clog)
213 {
214 	int i;
215 	extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead,
216 		   lfs_fs_pagetrip, lfs_ignore_lazy_sync;
217 #ifdef DEBUG
218 	extern int lfs_debug_log_subsys[DLOG_MAX];
219 	struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */
220 		{ "rollforward", "Debug roll-forward code" },
221 		{ "alloc",	"Debug inode allocation and free list" },
222 		{ "avail",	"Debug space-available-now accounting" },
223 		{ "flush",	"Debug flush triggers" },
224 		{ "lockedlist",	"Debug locked list accounting" },
225 		{ "vnode_verbose", "Verbose per-vnode-written debugging" },
226 		{ "vnode",	"Debug vnode use during segment write" },
227 		{ "segment",	"Debug segment writing" },
228 		{ "seguse",	"Debug segment used-bytes accounting" },
229 		{ "cleaner",	"Debug cleaning routines" },
230 		{ "mount",	"Debug mount/unmount routines" },
231 		{ "pagecache",	"Debug UBC interactions" },
232 		{ "dirop",	"Debug directory-operation accounting" },
233 		{ "malloc",	"Debug private malloc accounting" },
234 	};
235 #endif /* DEBUG */
236 	struct shortlong stat_names[] = { /* Must match lfs.h! */
237 		{ "segsused",	    "Number of new segments allocated" },
238 		{ "psegwrites",	    "Number of partial-segment writes" },
239 		{ "psyncwrites",    "Number of synchronous partial-segment"
240 				    " writes" },
241 		{ "pcleanwrites",   "Number of partial-segment writes by the"
242 				    " cleaner" },
243 		{ "blocktot",       "Number of blocks written" },
244 		{ "cleanblocks",    "Number of blocks written by the cleaner" },
245 		{ "ncheckpoints",   "Number of checkpoints made" },
246 		{ "nwrites",        "Number of whole writes" },
247 		{ "nsync_writes",   "Number of synchronous writes" },
248 		{ "wait_exceeded",  "Number of times writer waited for"
249 				    " cleaner" },
250 		{ "write_exceeded", "Number of times writer invoked flush" },
251 		{ "flush_invoked",  "Number of times flush was invoked" },
252 		{ "vflush_invoked", "Number of time vflush was called" },
253 		{ "clean_inlocked", "Number of vnodes skipped for VI_XLOCK" },
254 		{ "clean_vnlocked", "Number of vnodes skipped for vget failure" },
255 		{ "segs_reclaimed", "Number of segments reclaimed" },
256 	};
257 
258 	sysctl_createv(clog, 0, NULL, NULL,
259 		       CTLFLAG_PERMANENT,
260 		       CTLTYPE_NODE, "vfs", NULL,
261 		       NULL, 0, NULL, 0,
262 		       CTL_VFS, CTL_EOL);
263 	sysctl_createv(clog, 0, NULL, NULL,
264 		       CTLFLAG_PERMANENT,
265 		       CTLTYPE_NODE, "lfs",
266 		       SYSCTL_DESCR("Log-structured file system"),
267 		       NULL, 0, NULL, 0,
268 		       CTL_VFS, 5, CTL_EOL);
269 	/*
270 	 * XXX the "5" above could be dynamic, thereby eliminating one
271 	 * more instance of the "number to vfs" mapping problem, but
272 	 * "5" is the order as taken from sys/mount.h
273 	 */
274 
275 	sysctl_createv(clog, 0, NULL, NULL,
276 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
277 		       CTLTYPE_INT, "flushindir", NULL,
278 		       NULL, 0, &lfs_writeindir, 0,
279 		       CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
280 	sysctl_createv(clog, 0, NULL, NULL,
281 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
282 		       CTLTYPE_INT, "clean_vnhead", NULL,
283 		       NULL, 0, &lfs_clean_vnhead, 0,
284 		       CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
285 	sysctl_createv(clog, 0, NULL, NULL,
286 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
287 		       CTLTYPE_INT, "dostats",
288 		       SYSCTL_DESCR("Maintain statistics on LFS operations"),
289 		       sysctl_lfs_dostats, 0, &lfs_dostats, 0,
290 		       CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
291 	sysctl_createv(clog, 0, NULL, NULL,
292 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
293 		       CTLTYPE_INT, "pagetrip",
294 		       SYSCTL_DESCR("How many dirty pages in fs triggers"
295 				    " a flush"),
296 		       NULL, 0, &lfs_fs_pagetrip, 0,
297 		       CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL);
298 	sysctl_createv(clog, 0, NULL, NULL,
299 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
300 		       CTLTYPE_INT, "ignore_lazy_sync",
301 		       SYSCTL_DESCR("Lazy Sync is ignored entirely"),
302 		       NULL, 0, &lfs_ignore_lazy_sync, 0,
303 		       CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL);
304 #ifdef LFS_KERNEL_RFW
305 	sysctl_createv(clog, 0, NULL, NULL,
306 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
307 		       CTLTYPE_INT, "rfw",
308 		       SYSCTL_DESCR("Use in-kernel roll-forward on mount"),
309 		       NULL, 0, &lfs_do_rfw, 0,
310 		       CTL_VFS, 5, LFS_DO_RFW, CTL_EOL);
311 #endif
312 
313 	sysctl_createv(clog, 0, NULL, NULL,
314 		       CTLFLAG_PERMANENT,
315 		       CTLTYPE_NODE, "stats",
316 		       SYSCTL_DESCR("Debugging options"),
317 		       NULL, 0, NULL, 0,
318 		       CTL_VFS, 5, LFS_STATS, CTL_EOL);
319 	for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) {
320 		sysctl_createv(clog, 0, NULL, NULL,
321 			       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
322 			       CTLTYPE_INT, stat_names[i].sname,
323 			       SYSCTL_DESCR(stat_names[i].lname),
324 			       NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]),
325 			       0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL);
326 	}
327 
328 #ifdef DEBUG
329 	sysctl_createv(clog, 0, NULL, NULL,
330 		       CTLFLAG_PERMANENT,
331 		       CTLTYPE_NODE, "debug",
332 		       SYSCTL_DESCR("Debugging options"),
333 		       NULL, 0, NULL, 0,
334 		       CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL);
335 	for (i = 0; i < DLOG_MAX; i++) {
336 		sysctl_createv(clog, 0, NULL, NULL,
337 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
338 			       CTLTYPE_INT, dlog_names[i].sname,
339 			       SYSCTL_DESCR(dlog_names[i].lname),
340 			       NULL, 0, &(lfs_debug_log_subsys[i]), 0,
341 			       CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL);
342 	}
343 #endif
344 }
345 
346 /* old cleaner syscall interface.  see VOP_FCNTL() */
347 static const struct syscall_package lfs_syscalls[] = {
348 	{ SYS_lfs_bmapv,	0, (sy_call_t *)sys_lfs_bmapv		},
349 	{ SYS_lfs_markv,	0, (sy_call_t *)sys_lfs_markv		},
350 	{ SYS_lfs_segclean,	0, (sy_call_t *)sys___lfs_segwait50	},
351 	{ 0, 0, NULL },
352 };
353 
354 static int
355 lfs_modcmd(modcmd_t cmd, void *arg)
356 {
357 	int error;
358 
359 	switch (cmd) {
360 	case MODULE_CMD_INIT:
361 		error = syscall_establish(NULL, lfs_syscalls);
362 		if (error)
363 			return error;
364 		error = vfs_attach(&lfs_vfsops);
365 		if (error != 0) {
366 			syscall_disestablish(NULL, lfs_syscalls);
367 			break;
368 		}
369 		lfs_sysctl_setup(&lfs_sysctl_log);
370 		break;
371 	case MODULE_CMD_FINI:
372 		error = vfs_detach(&lfs_vfsops);
373 		if (error != 0)
374 			break;
375 		syscall_disestablish(NULL, lfs_syscalls);
376 		sysctl_teardown(&lfs_sysctl_log);
377 		break;
378 	default:
379 		error = ENOTTY;
380 		break;
381 	}
382 
383 	return (error);
384 }
385 
386 /*
387  * XXX Same structure as FFS inodes?  Should we share a common pool?
388  */
389 struct pool lfs_inode_pool;
390 struct pool lfs_dinode_pool;
391 struct pool lfs_inoext_pool;
392 struct pool lfs_lbnentry_pool;
393 
394 /*
395  * The writer daemon.  UVM keeps track of how many dirty pages we are holding
396  * in lfs_subsys_pages; the daemon flushes the filesystem when this value
397  * crosses the (user-defined) threshhold LFS_MAX_PAGES.
398  */
399 static void
400 lfs_writerd(void *arg)
401 {
402 	struct mount *mp, *nmp;
403 	struct lfs *fs;
404 	int fsflags;
405 	int loopcount;
406 
407 	lfs_writer_daemon = curproc->p_pid;
408 
409 	mutex_enter(&lfs_lock);
410 	for (;;) {
411 		mtsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", hz/10,
412 		    &lfs_lock);
413 
414 		/*
415 		 * Look through the list of LFSs to see if any of them
416 		 * have requested pageouts.
417 		 */
418 		mutex_enter(&mountlist_lock);
419 		for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
420 		     mp = nmp) {
421 			if (vfs_busy(mp, &nmp)) {
422 				continue;
423 			}
424 			if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS,
425 			    sizeof(mp->mnt_stat.f_fstypename)) == 0) {
426 				fs = VFSTOUFS(mp)->um_lfs;
427 				mutex_enter(&lfs_lock);
428 				fsflags = 0;
429 				if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) ||
430 				     lfs_dirvcount > LFS_MAX_DIROP) &&
431 				    fs->lfs_dirops == 0)
432 					fsflags |= SEGM_CKP;
433 				if (fs->lfs_pdflush) {
434 					DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n"));
435 					fs->lfs_pdflush = 0;
436 					lfs_flush_fs(fs, fsflags);
437 					mutex_exit(&lfs_lock);
438 				} else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) {
439 					DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n"));
440 					mutex_exit(&lfs_lock);
441 					lfs_writer_enter(fs, "wrdirop");
442 					lfs_flush_pchain(fs);
443 					lfs_writer_leave(fs);
444 				} else
445 					mutex_exit(&lfs_lock);
446 			}
447 			vfs_unbusy(mp, false, &nmp);
448 		}
449 		mutex_exit(&mountlist_lock);
450 
451 		/*
452 		 * If global state wants a flush, flush everything.
453 		 */
454 		mutex_enter(&lfs_lock);
455 		loopcount = 0;
456 		if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
457 			locked_queue_bytes > LFS_MAX_BYTES ||
458 			lfs_subsys_pages > LFS_MAX_PAGES) {
459 
460 			if (lfs_do_flush) {
461 				DLOG((DLOG_FLUSH, "daemon: lfs_do_flush\n"));
462 			}
463 			if (locked_queue_count > LFS_MAX_BUFS) {
464 				DLOG((DLOG_FLUSH, "daemon: lqc = %d, max %d\n",
465 				      locked_queue_count, LFS_MAX_BUFS));
466 			}
467 			if (locked_queue_bytes > LFS_MAX_BYTES) {
468 				DLOG((DLOG_FLUSH, "daemon: lqb = %ld, max %ld\n",
469 				      locked_queue_bytes, LFS_MAX_BYTES));
470 			}
471 			if (lfs_subsys_pages > LFS_MAX_PAGES) {
472 				DLOG((DLOG_FLUSH, "daemon: lssp = %d, max %d\n",
473 				      lfs_subsys_pages, LFS_MAX_PAGES));
474 			}
475 
476 			lfs_flush(NULL, SEGM_WRITERD, 0);
477 			lfs_do_flush = 0;
478 		}
479 	}
480 	/* NOTREACHED */
481 }
482 
483 /*
484  * Initialize the filesystem, most work done by ufs_init.
485  */
486 void
487 lfs_init(void)
488 {
489 
490 	malloc_type_attach(M_SEGMENT);
491 	pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
492 	    "lfsinopl", &pool_allocator_nointr, IPL_NONE);
493 	pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0,
494 	    "lfsdinopl", &pool_allocator_nointr, IPL_NONE);
495 	pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
496 	    "lfsinoextpl", &pool_allocator_nointr, IPL_NONE);
497 	pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0,
498 	    "lfslbnpool", &pool_allocator_nointr, IPL_NONE);
499 	ufs_init();
500 
501 #ifdef DEBUG
502 	memset(lfs_log, 0, sizeof(lfs_log));
503 #endif
504 	mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE);
505 	cv_init(&locked_queue_cv, "lfsbuf");
506 	cv_init(&lfs_writing_cv, "lfsflush");
507 }
508 
509 void
510 lfs_reinit(void)
511 {
512 	ufs_reinit();
513 }
514 
515 void
516 lfs_done(void)
517 {
518 	ufs_done();
519 	mutex_destroy(&lfs_lock);
520 	cv_destroy(&locked_queue_cv);
521 	cv_destroy(&lfs_writing_cv);
522 	pool_destroy(&lfs_inode_pool);
523 	pool_destroy(&lfs_dinode_pool);
524 	pool_destroy(&lfs_inoext_pool);
525 	pool_destroy(&lfs_lbnentry_pool);
526 	malloc_type_detach(M_SEGMENT);
527 }
528 
529 /*
530  * Called by main() when ufs is going to be mounted as root.
531  */
532 int
533 lfs_mountroot(void)
534 {
535 	extern struct vnode *rootvp;
536 	struct lfs *fs = NULL;				/* LFS */
537 	struct mount *mp;
538 	struct lwp *l = curlwp;
539 	struct ufsmount *ump;
540 	int error;
541 
542 	if (device_class(root_device) != DV_DISK)
543 		return (ENODEV);
544 
545 	if (rootdev == NODEV)
546 		return (ENODEV);
547 	if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
548 		vrele(rootvp);
549 		return (error);
550 	}
551 	if ((error = lfs_mountfs(rootvp, mp, l))) {
552 		vfs_unbusy(mp, false, NULL);
553 		vfs_destroy(mp);
554 		return (error);
555 	}
556 	mutex_enter(&mountlist_lock);
557 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
558 	mutex_exit(&mountlist_lock);
559 	ump = VFSTOUFS(mp);
560 	fs = ump->um_lfs;
561 	memset(fs->lfs_fsmnt, 0, sizeof(fs->lfs_fsmnt));
562 	(void)copystr(mp->mnt_stat.f_mntonname, fs->lfs_fsmnt, MNAMELEN - 1, 0);
563 	(void)lfs_statvfs(mp, &mp->mnt_stat);
564 	vfs_unbusy(mp, false, NULL);
565 	setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp));
566 	return (0);
567 }
568 
569 /*
570  * VFS Operations.
571  *
572  * mount system call
573  */
574 int
575 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
576 {
577 	struct lwp *l = curlwp;
578 	struct vnode *devvp;
579 	struct ufs_args *args = data;
580 	struct ufsmount *ump = NULL;
581 	struct lfs *fs = NULL;				/* LFS */
582 	int error = 0, update;
583 	mode_t accessmode;
584 
585 	if (*data_len < sizeof *args)
586 		return EINVAL;
587 
588 	if (mp->mnt_flag & MNT_GETARGS) {
589 		ump = VFSTOUFS(mp);
590 		if (ump == NULL)
591 			return EIO;
592 		args->fspec = NULL;
593 		*data_len = sizeof *args;
594 		return 0;
595 	}
596 
597 	update = mp->mnt_flag & MNT_UPDATE;
598 
599 	/* Check arguments */
600 	if (args->fspec != NULL) {
601 		/*
602 		 * Look up the name and verify that it's sane.
603 		 */
604 		error = namei_simple_user(args->fspec,
605 					NSM_FOLLOW_NOEMULROOT, &devvp);
606 		if (error != 0)
607 			return (error);
608 
609 		if (!update) {
610 			/*
611 			 * Be sure this is a valid block device
612 			 */
613 			if (devvp->v_type != VBLK)
614 				error = ENOTBLK;
615 			else if (bdevsw_lookup(devvp->v_rdev) == NULL)
616 				error = ENXIO;
617 		} else {
618 			/*
619 			 * Be sure we're still naming the same device
620 			 * used for our initial mount
621 			 */
622 			ump = VFSTOUFS(mp);
623 			if (devvp != ump->um_devvp) {
624 				if (devvp->v_rdev != ump->um_devvp->v_rdev)
625 					error = EINVAL;
626 				else {
627 					vrele(devvp);
628 					devvp = ump->um_devvp;
629 					vref(devvp);
630 				}
631 			}
632 		}
633 	} else {
634 		if (!update) {
635 			/* New mounts must have a filename for the device */
636 			return (EINVAL);
637 		} else {
638 			/* Use the extant mount */
639 			ump = VFSTOUFS(mp);
640 			devvp = ump->um_devvp;
641 			vref(devvp);
642 		}
643 	}
644 
645 
646 	/*
647 	 * If mount by non-root, then verify that user has necessary
648 	 * permissions on the device.
649 	 */
650 	if (error == 0) {
651 		accessmode = VREAD;
652 		if (update ?
653 		    (mp->mnt_iflag & IMNT_WANTRDWR) != 0 :
654 		    (mp->mnt_flag & MNT_RDONLY) == 0)
655 			accessmode |= VWRITE;
656 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
657 		error = genfs_can_mount(devvp, accessmode, l->l_cred);
658 		VOP_UNLOCK(devvp);
659 	}
660 
661 	if (error) {
662 		vrele(devvp);
663 		return (error);
664 	}
665 
666 	if (!update) {
667 		int flags;
668 
669 		if (mp->mnt_flag & MNT_RDONLY)
670 			flags = FREAD;
671 		else
672 			flags = FREAD|FWRITE;
673 		error = VOP_OPEN(devvp, flags, FSCRED);
674 		if (error)
675 			goto fail;
676 		error = lfs_mountfs(devvp, mp, l);		/* LFS */
677 		if (error) {
678 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
679 			(void)VOP_CLOSE(devvp, flags, NOCRED);
680 			VOP_UNLOCK(devvp);
681 			goto fail;
682 		}
683 
684 		ump = VFSTOUFS(mp);
685 		fs = ump->um_lfs;
686 	} else {
687 		/*
688 		 * Update the mount.
689 		 */
690 
691 		/*
692 		 * The initial mount got a reference on this
693 		 * device, so drop the one obtained via
694 		 * namei(), above.
695 		 */
696 		vrele(devvp);
697 
698 		ump = VFSTOUFS(mp);
699 		fs = ump->um_lfs;
700 		if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
701 			/*
702 			 * Changing from read-only to read/write.
703 			 * Note in the superblocks that we're writing.
704 			 */
705 			fs->lfs_ronly = 0;
706 			if (fs->lfs_pflags & LFS_PF_CLEAN) {
707 				fs->lfs_pflags &= ~LFS_PF_CLEAN;
708 				lfs_writesuper(fs, fs->lfs_sboffs[0]);
709 				lfs_writesuper(fs, fs->lfs_sboffs[1]);
710 			}
711 		}
712 		if (args->fspec == NULL)
713 			return EINVAL;
714 	}
715 
716 	error = set_statvfs_info(path, UIO_USERSPACE, args->fspec,
717 	    UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l);
718 	if (error == 0)
719 		(void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname,
720 			      sizeof(fs->lfs_fsmnt));
721 	return error;
722 
723 fail:
724 	vrele(devvp);
725 	return (error);
726 }
727 
728 
729 /*
730  * Common code for mount and mountroot
731  * LFS specific
732  */
733 int
734 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l)
735 {
736 	struct dlfs *tdfs, *dfs, *adfs;
737 	struct lfs *fs;
738 	struct ufsmount *ump;
739 	struct vnode *vp;
740 	struct buf *bp, *abp;
741 	dev_t dev;
742 	int error, i, ronly, fsbsize;
743 	kauth_cred_t cred;
744 	CLEANERINFO *cip;
745 	SEGUSE *sup;
746 	daddr_t sb_addr;
747 
748 	cred = l ? l->l_cred : NOCRED;
749 
750 	/*
751 	 * Flush out any old buffers remaining from a previous use.
752 	 */
753 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
754 	error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0);
755 	VOP_UNLOCK(devvp);
756 	if (error)
757 		return (error);
758 
759 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
760 
761 	/* Don't free random space on error. */
762 	bp = NULL;
763 	abp = NULL;
764 	ump = NULL;
765 
766 	sb_addr = LFS_LABELPAD / DEV_BSIZE;
767 	while (1) {
768 		/* Read in the superblock. */
769 		error = bread(devvp, sb_addr, LFS_SBPAD, cred, 0, &bp);
770 		if (error)
771 			goto out;
772 		dfs = (struct dlfs *)bp->b_data;
773 
774 		/* Check the basics. */
775 		if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE ||
776 		    dfs->dlfs_version > LFS_VERSION ||
777 		    dfs->dlfs_bsize < sizeof(struct dlfs)) {
778 			DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n"));
779 			error = EINVAL;		/* XXX needs translation */
780 			goto out;
781 		}
782 		if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) {
783 			DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n",
784 			       dfs->dlfs_inodefmt));
785 			error = EINVAL;
786 			goto out;
787 		}
788 
789 		if (dfs->dlfs_version == 1)
790 			fsbsize = DEV_BSIZE;
791 		else {
792 			fsbsize = 1 << dfs->dlfs_ffshift;
793 			/*
794 			 * Could be, if the frag size is large enough, that we
795 			 * don't have the "real" primary superblock.  If that's
796 			 * the case, get the real one, and try again.
797 			 */
798 			if (sb_addr != (dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT))) {
799 				DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr"
800 				      " 0x%llx is not right, trying 0x%llx\n",
801 				      (long long)sb_addr,
802 				      (long long)(dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT))));
803 				sb_addr = dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT);
804 				brelse(bp, 0);
805 				continue;
806 			}
807 		}
808 		break;
809 	}
810 
811 	/*
812 	 * Check the second superblock to see which is newer; then mount
813 	 * using the older of the two.	This is necessary to ensure that
814 	 * the filesystem is valid if it was not unmounted cleanly.
815 	 */
816 
817 	if (dfs->dlfs_sboffs[1] &&
818 	    dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
819 	{
820 		error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / DEV_BSIZE),
821 			LFS_SBPAD, cred, 0, &abp);
822 		if (error)
823 			goto out;
824 		adfs = (struct dlfs *)abp->b_data;
825 
826 		if (dfs->dlfs_version == 1) {
827 			/* 1s resolution comparison */
828 			if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
829 				tdfs = adfs;
830 			else
831 				tdfs = dfs;
832 		} else {
833 			/* monotonic infinite-resolution comparison */
834 			if (adfs->dlfs_serial < dfs->dlfs_serial)
835 				tdfs = adfs;
836 			else
837 				tdfs = dfs;
838 		}
839 
840 		/* Check the basics. */
841 		if (tdfs->dlfs_magic != LFS_MAGIC ||
842 		    tdfs->dlfs_bsize > MAXBSIZE ||
843 		    tdfs->dlfs_version > LFS_VERSION ||
844 		    tdfs->dlfs_bsize < sizeof(struct dlfs)) {
845 			DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock"
846 			      " sanity failed\n"));
847 			error = EINVAL;		/* XXX needs translation */
848 			goto out;
849 		}
850 	} else {
851 		DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock"
852 		      " daddr=0x%x\n", dfs->dlfs_sboffs[1]));
853 		error = EINVAL;
854 		goto out;
855 	}
856 
857 	/* Allocate the mount structure, copy the superblock into it. */
858 	fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO);
859 	memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs));
860 
861 	/* Compatibility */
862 	if (fs->lfs_version < 2) {
863 		fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
864 		fs->lfs_ibsize = fs->lfs_bsize;
865 		fs->lfs_start = fs->lfs_sboffs[0];
866 		fs->lfs_tstamp = fs->lfs_otstamp;
867 		fs->lfs_fsbtodb = 0;
868 	}
869 	if (fs->lfs_resvseg == 0)
870 		fs->lfs_resvseg = MIN(fs->lfs_minfreeseg - 1, \
871 			MAX(MIN_RESV_SEGS, fs->lfs_minfreeseg / 2 + 1));
872 
873 	/*
874 	 * If we aren't going to be able to write meaningfully to this
875 	 * filesystem, and were not mounted readonly, bomb out now.
876 	 */
877 	if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) {
878 		DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write,"
879 		      " we need BUFPAGES >= %lld\n",
880 		      (long long)((bufmem_hiwater / bufmem_lowater) *
881 				  LFS_INVERSE_MAX_BYTES(
882 					  fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT)));
883 		free(fs, M_UFSMNT);
884 		error = EFBIG; /* XXX needs translation */
885 		goto out;
886 	}
887 
888 	/* Before rolling forward, lock so vget will sleep for other procs */
889 	if (l != NULL) {
890 		fs->lfs_flags = LFS_NOTYET;
891 		fs->lfs_rfpid = l->l_proc->p_pid;
892 	}
893 
894 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
895 	ump->um_lfs = fs;
896 	ump->um_ops = &lfs_ufsops;
897 	ump->um_fstype = UFS1;
898 	if (sizeof(struct lfs) < LFS_SBPAD) {			/* XXX why? */
899 		brelse(bp, BC_INVAL);
900 		brelse(abp, BC_INVAL);
901 	} else {
902 		brelse(bp, 0);
903 		brelse(abp, 0);
904 	}
905 	bp = NULL;
906 	abp = NULL;
907 
908 
909 	/* Set up the I/O information */
910 	fs->lfs_devbsize = DEV_BSIZE;
911 	fs->lfs_iocount = 0;
912 	fs->lfs_diropwait = 0;
913 	fs->lfs_activesb = 0;
914 	fs->lfs_uinodes = 0;
915 	fs->lfs_ravail = 0;
916 	fs->lfs_favail = 0;
917 	fs->lfs_sbactive = 0;
918 
919 	/* Set up the ifile and lock aflags */
920 	fs->lfs_doifile = 0;
921 	fs->lfs_writer = 0;
922 	fs->lfs_dirops = 0;
923 	fs->lfs_nadirop = 0;
924 	fs->lfs_seglock = 0;
925 	fs->lfs_pdflush = 0;
926 	fs->lfs_sleepers = 0;
927 	fs->lfs_pages = 0;
928 	rw_init(&fs->lfs_fraglock);
929 	rw_init(&fs->lfs_iflock);
930 	cv_init(&fs->lfs_stopcv, "lfsstop");
931 
932 	/* Set the file system readonly/modify bits. */
933 	fs->lfs_ronly = ronly;
934 	if (ronly == 0)
935 		fs->lfs_fmod = 1;
936 
937 	/* Initialize the mount structure. */
938 	dev = devvp->v_rdev;
939 	mp->mnt_data = ump;
940 	mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
941 	mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS);
942 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
943 	mp->mnt_stat.f_namemax = LFS_MAXNAMLEN;
944 	mp->mnt_stat.f_iosize = fs->lfs_bsize;
945 	mp->mnt_flag |= MNT_LOCAL;
946 	mp->mnt_fs_bshift = fs->lfs_bshift;
947 	ump->um_flags = 0;
948 	ump->um_mountp = mp;
949 	ump->um_dev = dev;
950 	ump->um_devvp = devvp;
951 	ump->um_bptrtodb = fs->lfs_ffshift - DEV_BSHIFT;
952 	ump->um_seqinc = fs->lfs_frag;
953 	ump->um_nindir = fs->lfs_nindir;
954 	ump->um_lognindir = ffs(fs->lfs_nindir) - 1;
955 	for (i = 0; i < MAXQUOTAS; i++)
956 		ump->um_quotas[i] = NULLVP;
957 	ump->um_maxsymlinklen = fs->lfs_maxsymlinklen;
958 	ump->um_dirblksiz = DIRBLKSIZ;
959 	ump->um_maxfilesize = fs->lfs_maxfilesize;
960 	if (ump->um_maxsymlinklen > 0)
961 		mp->mnt_iflag |= IMNT_DTYPE;
962 	devvp->v_specmountpoint = mp;
963 
964 	/* Set up reserved memory for pageout */
965 	lfs_setup_resblks(fs);
966 	/* Set up vdirop tailq */
967 	TAILQ_INIT(&fs->lfs_dchainhd);
968 	/* and paging tailq */
969 	TAILQ_INIT(&fs->lfs_pchainhd);
970 	/* and delayed segment accounting for truncation list */
971 	LIST_INIT(&fs->lfs_segdhd);
972 
973 	/*
974 	 * We use the ifile vnode for almost every operation.  Instead of
975 	 * retrieving it from the hash table each time we retrieve it here,
976 	 * artificially increment the reference count and keep a pointer
977 	 * to it in the incore copy of the superblock.
978 	 */
979 	if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
980 		DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error));
981 		goto out;
982 	}
983 	fs->lfs_ivnode = vp;
984 	vref(vp);
985 
986 	/* Set up inode bitmap and order free list */
987 	lfs_order_freelist(fs);
988 
989 	/* Set up segment usage flags for the autocleaner. */
990 	fs->lfs_nactive = 0;
991 	fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *),
992 						M_SEGMENT, M_WAITOK);
993 	fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
994 						 M_SEGMENT, M_WAITOK);
995 	fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
996 						 M_SEGMENT, M_WAITOK);
997 	memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t));
998 	for (i = 0; i < fs->lfs_nseg; i++) {
999 		int changed;
1000 
1001 		LFS_SEGENTRY(sup, fs, i, bp);
1002 		changed = 0;
1003 		if (!ronly) {
1004 			if (sup->su_nbytes == 0 &&
1005 			    !(sup->su_flags & SEGUSE_EMPTY)) {
1006 				sup->su_flags |= SEGUSE_EMPTY;
1007 				++changed;
1008 			} else if (!(sup->su_nbytes == 0) &&
1009 				   (sup->su_flags & SEGUSE_EMPTY)) {
1010 				sup->su_flags &= ~SEGUSE_EMPTY;
1011 				++changed;
1012 			}
1013 			if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) {
1014 				sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL);
1015 				++changed;
1016 			}
1017 		}
1018 		fs->lfs_suflags[0][i] = sup->su_flags;
1019 		if (changed)
1020 			LFS_WRITESEGENTRY(sup, fs, i, bp);
1021 		else
1022 			brelse(bp, 0);
1023 	}
1024 
1025 #ifdef LFS_KERNEL_RFW
1026 	lfs_roll_forward(fs, mp, l);
1027 #endif
1028 
1029 	/* If writing, sb is not clean; record in case of immediate crash */
1030 	if (!fs->lfs_ronly) {
1031 		fs->lfs_pflags &= ~LFS_PF_CLEAN;
1032 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
1033 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
1034 	}
1035 
1036 	/* Allow vget now that roll-forward is complete */
1037 	fs->lfs_flags &= ~(LFS_NOTYET);
1038 	wakeup(&fs->lfs_flags);
1039 
1040 	/*
1041 	 * Initialize the ifile cleaner info with information from
1042 	 * the superblock.
1043 	 */
1044 	LFS_CLEANERINFO(cip, fs, bp);
1045 	cip->clean = fs->lfs_nclean;
1046 	cip->dirty = fs->lfs_nseg - fs->lfs_nclean;
1047 	cip->avail = fs->lfs_avail;
1048 	cip->bfree = fs->lfs_bfree;
1049 	(void) LFS_BWRITE_LOG(bp); /* Ifile */
1050 
1051 	/*
1052 	 * Mark the current segment as ACTIVE, since we're going to
1053 	 * be writing to it.
1054 	 */
1055 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);
1056 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1057 	fs->lfs_nactive++;
1058 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);  /* Ifile */
1059 
1060 	/* Now that roll-forward is done, unlock the Ifile */
1061 	vput(vp);
1062 
1063 	/* Start the pagedaemon-anticipating daemon */
1064 	if (lfs_writer_daemon == 0 && kthread_create(PRI_BIO, 0, NULL,
1065 	    lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1066 		panic("fork lfs_writer");
1067 	/*
1068 	 * XXX: Get extra reference to LFS vfsops.  This prevents unload,
1069 	 * but also prevents kernel panic due to text being unloaded
1070 	 * from below lfs_writerd.  When lfs_writerd can exit, remove
1071 	 * this!!!
1072 	 */
1073 	vfs_getopsbyname(MOUNT_LFS);
1074 
1075 	printf("WARNING: the log-structured file system is experimental\n"
1076 	    "WARNING: it may cause system crashes and/or corrupt data\n");
1077 
1078 	return (0);
1079 
1080 out:
1081 	if (bp)
1082 		brelse(bp, 0);
1083 	if (abp)
1084 		brelse(abp, 0);
1085 	if (ump) {
1086 		free(ump->um_lfs, M_UFSMNT);
1087 		free(ump, M_UFSMNT);
1088 		mp->mnt_data = NULL;
1089 	}
1090 
1091 	return (error);
1092 }
1093 
1094 /*
1095  * unmount system call
1096  */
1097 int
1098 lfs_unmount(struct mount *mp, int mntflags)
1099 {
1100 	struct lwp *l = curlwp;
1101 	struct ufsmount *ump;
1102 	struct lfs *fs;
1103 	int error, flags, ronly;
1104 	vnode_t *vp;
1105 
1106 	flags = 0;
1107 	if (mntflags & MNT_FORCE)
1108 		flags |= FORCECLOSE;
1109 
1110 	ump = VFSTOUFS(mp);
1111 	fs = ump->um_lfs;
1112 
1113 	/* Two checkpoints */
1114 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1115 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1116 
1117 	/* wake up the cleaner so it can die */
1118 	lfs_wakeup_cleaner(fs);
1119 	mutex_enter(&lfs_lock);
1120 	while (fs->lfs_sleepers)
1121 		mtsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1122 			&lfs_lock);
1123 	mutex_exit(&lfs_lock);
1124 
1125 #ifdef QUOTA
1126         if ((error = quota1_umount(mp, flags)) != 0)
1127 		return (error);
1128 #endif
1129 	if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1130 		return (error);
1131 	if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0)
1132 		return (error);
1133 	vp = fs->lfs_ivnode;
1134 	mutex_enter(vp->v_interlock);
1135 	if (LIST_FIRST(&vp->v_dirtyblkhd))
1136 		panic("lfs_unmount: still dirty blocks on ifile vnode");
1137 	mutex_exit(vp->v_interlock);
1138 
1139 	/* Explicitly write the superblock, to update serial and pflags */
1140 	fs->lfs_pflags |= LFS_PF_CLEAN;
1141 	lfs_writesuper(fs, fs->lfs_sboffs[0]);
1142 	lfs_writesuper(fs, fs->lfs_sboffs[1]);
1143 	mutex_enter(&lfs_lock);
1144 	while (fs->lfs_iocount)
1145 		mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0,
1146 			&lfs_lock);
1147 	mutex_exit(&lfs_lock);
1148 
1149 	/* Finish with the Ifile, now that we're done with it */
1150 	vgone(fs->lfs_ivnode);
1151 
1152 	ronly = !fs->lfs_ronly;
1153 	if (ump->um_devvp->v_type != VBAD)
1154 		ump->um_devvp->v_specmountpoint = NULL;
1155 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1156 	error = VOP_CLOSE(ump->um_devvp,
1157 	    ronly ? FREAD : FREAD|FWRITE, NOCRED);
1158 	vput(ump->um_devvp);
1159 
1160 	/* Complain about page leakage */
1161 	if (fs->lfs_pages > 0)
1162 		printf("lfs_unmount: still claim %d pages (%d in subsystem)\n",
1163 			fs->lfs_pages, lfs_subsys_pages);
1164 
1165 	/* Free per-mount data structures */
1166 	free(fs->lfs_ino_bitmap, M_SEGMENT);
1167 	free(fs->lfs_suflags[0], M_SEGMENT);
1168 	free(fs->lfs_suflags[1], M_SEGMENT);
1169 	free(fs->lfs_suflags, M_SEGMENT);
1170 	lfs_free_resblks(fs);
1171 	cv_destroy(&fs->lfs_stopcv);
1172 	rw_destroy(&fs->lfs_fraglock);
1173 	rw_destroy(&fs->lfs_iflock);
1174 	free(fs, M_UFSMNT);
1175 	free(ump, M_UFSMNT);
1176 
1177 	mp->mnt_data = NULL;
1178 	mp->mnt_flag &= ~MNT_LOCAL;
1179 	return (error);
1180 }
1181 
1182 /*
1183  * Get file system statistics.
1184  *
1185  * NB: We don't lock to access the superblock here, because it's not
1186  * really that important if we get it wrong.
1187  */
1188 int
1189 lfs_statvfs(struct mount *mp, struct statvfs *sbp)
1190 {
1191 	struct lfs *fs;
1192 	struct ufsmount *ump;
1193 
1194 	ump = VFSTOUFS(mp);
1195 	fs = ump->um_lfs;
1196 	if (fs->lfs_magic != LFS_MAGIC)
1197 		panic("lfs_statvfs: magic");
1198 
1199 	sbp->f_bsize = fs->lfs_bsize;
1200 	sbp->f_frsize = fs->lfs_fsize;
1201 	sbp->f_iosize = fs->lfs_bsize;
1202 	sbp->f_blocks = LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks;
1203 
1204 	sbp->f_bfree = LFS_EST_BFREE(fs);
1205 	KASSERT(sbp->f_bfree <= fs->lfs_dsize);
1206 #if 0
1207 	if (sbp->f_bfree < 0)
1208 		sbp->f_bfree = 0;
1209 #endif
1210 
1211 	sbp->f_bresvd = LFS_EST_RSVD(fs);
1212 	if (sbp->f_bfree > sbp->f_bresvd)
1213 		sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
1214 	else
1215 		sbp->f_bavail = 0;
1216 
1217 	sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs);
1218 	sbp->f_ffree = sbp->f_files - fs->lfs_nfiles;
1219 	sbp->f_favail = sbp->f_ffree;
1220 	sbp->f_fresvd = 0;
1221 	copy_statvfs_info(sbp, mp);
1222 	return (0);
1223 }
1224 
1225 /*
1226  * Go through the disk queues to initiate sandbagged IO;
1227  * go through the inodes to write those that have been modified;
1228  * initiate the writing of the super block if it has been modified.
1229  *
1230  * Note: we are always called with the filesystem marked `MPBUSY'.
1231  */
1232 int
1233 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
1234 {
1235 	int error;
1236 	struct lfs *fs;
1237 
1238 	fs = VFSTOUFS(mp)->um_lfs;
1239 	if (fs->lfs_ronly)
1240 		return 0;
1241 
1242 	/* Snapshots should not hose the syncer */
1243 	/*
1244 	 * XXX Sync can block here anyway, since we don't have a very
1245 	 * XXX good idea of how much data is pending.  If it's more
1246 	 * XXX than a segment and lfs_nextseg is close to the end of
1247 	 * XXX the log, we'll likely block.
1248 	 */
1249 	mutex_enter(&lfs_lock);
1250 	if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) {
1251 		mutex_exit(&lfs_lock);
1252 		return 0;
1253 	}
1254 	mutex_exit(&lfs_lock);
1255 
1256 	lfs_writer_enter(fs, "lfs_dirops");
1257 
1258 	/* All syncs must be checkpoints until roll-forward is implemented. */
1259 	DLOG((DLOG_FLUSH, "lfs_sync at 0x%x\n", fs->lfs_offset));
1260 	error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1261 	lfs_writer_leave(fs);
1262 #ifdef QUOTA
1263 	qsync(mp);
1264 #endif
1265 	return (error);
1266 }
1267 
1268 /*
1269  * Look up an LFS dinode number to find its incore vnode.  If not already
1270  * in core, read it in from the specified device.  Return the inode locked.
1271  * Detection and handling of mount points must be done by the calling routine.
1272  */
1273 int
1274 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1275 {
1276 	struct lfs *fs;
1277 	struct ufs1_dinode *dip;
1278 	struct inode *ip;
1279 	struct buf *bp;
1280 	struct ifile *ifp;
1281 	struct vnode *vp;
1282 	struct ufsmount *ump;
1283 	daddr_t daddr;
1284 	dev_t dev;
1285 	int error, retries;
1286 	struct timespec ts;
1287 
1288 	memset(&ts, 0, sizeof ts);	/* XXX gcc */
1289 
1290 	ump = VFSTOUFS(mp);
1291 	dev = ump->um_dev;
1292 	fs = ump->um_lfs;
1293 
1294 	/*
1295 	 * If the filesystem is not completely mounted yet, suspend
1296 	 * any access requests (wait for roll-forward to complete).
1297 	 */
1298 	mutex_enter(&lfs_lock);
1299 	while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1300 		mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0,
1301 			&lfs_lock);
1302 	mutex_exit(&lfs_lock);
1303 
1304 retry:
1305 	if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
1306 		return (0);
1307 
1308 	error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, NULL, &vp);
1309 	if (error) {
1310 		*vpp = NULL;
1311 		 return (error);
1312 	}
1313 
1314 	mutex_enter(&ufs_hashlock);
1315 	if (ufs_ihashget(dev, ino, 0) != NULL) {
1316 		mutex_exit(&ufs_hashlock);
1317 		ungetnewvnode(vp);
1318 		goto retry;
1319 	}
1320 
1321 	/* Translate the inode number to a disk address. */
1322 	if (ino == LFS_IFILE_INUM)
1323 		daddr = fs->lfs_idaddr;
1324 	else {
1325 		/* XXX bounds-check this too */
1326 		LFS_IENTRY(ifp, fs, ino, bp);
1327 		daddr = ifp->if_daddr;
1328 		if (fs->lfs_version > 1) {
1329 			ts.tv_sec = ifp->if_atime_sec;
1330 			ts.tv_nsec = ifp->if_atime_nsec;
1331 		}
1332 
1333 		brelse(bp, 0);
1334 		if (daddr == LFS_UNUSED_DADDR) {
1335 			*vpp = NULLVP;
1336 			mutex_exit(&ufs_hashlock);
1337 			ungetnewvnode(vp);
1338 			return (ENOENT);
1339 		}
1340 	}
1341 
1342 	/* Allocate/init new vnode/inode. */
1343 	lfs_vcreate(mp, ino, vp);
1344 
1345 	/*
1346 	 * Put it onto its hash chain and lock it so that other requests for
1347 	 * this inode will block if they arrive while we are sleeping waiting
1348 	 * for old data structures to be purged or for the contents of the
1349 	 * disk portion of this inode to be read.
1350 	 */
1351 	ip = VTOI(vp);
1352 	ufs_ihashins(ip);
1353 	mutex_exit(&ufs_hashlock);
1354 
1355 	/*
1356 	 * XXX
1357 	 * This may not need to be here, logically it should go down with
1358 	 * the i_devvp initialization.
1359 	 * Ask Kirk.
1360 	 */
1361 	ip->i_lfs = ump->um_lfs;
1362 
1363 	/* Read in the disk contents for the inode, copy into the inode. */
1364 	retries = 0;
1365     again:
1366 	error = bread(ump->um_devvp, fsbtodb(fs, daddr),
1367 		(fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize),
1368 		NOCRED, 0, &bp);
1369 	if (error) {
1370 		/*
1371 		 * The inode does not contain anything useful, so it would
1372 		 * be misleading to leave it on its hash chain. With mode
1373 		 * still zero, it will be unlinked and returned to the free
1374 		 * list by vput().
1375 		 */
1376 		vput(vp);
1377 		brelse(bp, 0);
1378 		*vpp = NULL;
1379 		return (error);
1380 	}
1381 
1382 	dip = lfs_ifind(fs, ino, bp);
1383 	if (dip == NULL) {
1384 		/* Assume write has not completed yet; try again */
1385 		brelse(bp, BC_INVAL);
1386 		++retries;
1387 		if (retries > LFS_IFIND_RETRIES) {
1388 #ifdef DEBUG
1389 			/* If the seglock is held look at the bpp to see
1390 			   what is there anyway */
1391 			mutex_enter(&lfs_lock);
1392 			if (fs->lfs_seglock > 0) {
1393 				struct buf **bpp;
1394 				struct ufs1_dinode *dp;
1395 				int i;
1396 
1397 				for (bpp = fs->lfs_sp->bpp;
1398 				     bpp != fs->lfs_sp->cbpp; ++bpp) {
1399 					if ((*bpp)->b_vp == fs->lfs_ivnode &&
1400 					    bpp != fs->lfs_sp->bpp) {
1401 						/* Inode block */
1402 						printf("lfs_vget: block 0x%" PRIx64 ": ",
1403 						       (*bpp)->b_blkno);
1404 						dp = (struct ufs1_dinode *)(*bpp)->b_data;
1405 						for (i = 0; i < INOPB(fs); i++)
1406 							if (dp[i].di_u.inumber)
1407 								printf("%d ", dp[i].di_u.inumber);
1408 						printf("\n");
1409 					}
1410 				}
1411 			}
1412 			mutex_exit(&lfs_lock);
1413 #endif /* DEBUG */
1414 			panic("lfs_vget: dinode not found");
1415 		}
1416 		mutex_enter(&lfs_lock);
1417 		if (fs->lfs_iocount) {
1418 			DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino));
1419 			(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
1420 				      "lfs ifind", 1, &lfs_lock);
1421 		} else
1422 			retries = LFS_IFIND_RETRIES;
1423 		mutex_exit(&lfs_lock);
1424 		goto again;
1425 	}
1426 	*ip->i_din.ffs1_din = *dip;
1427 	brelse(bp, 0);
1428 
1429 	if (fs->lfs_version > 1) {
1430 		ip->i_ffs1_atime = ts.tv_sec;
1431 		ip->i_ffs1_atimensec = ts.tv_nsec;
1432 	}
1433 
1434 	lfs_vinit(mp, &vp);
1435 
1436 	*vpp = vp;
1437 
1438 	KASSERT(VOP_ISLOCKED(vp));
1439 
1440 	return (0);
1441 }
1442 
1443 /*
1444  * File handle to vnode
1445  */
1446 int
1447 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1448 {
1449 	struct lfid lfh;
1450 	struct buf *bp;
1451 	IFILE *ifp;
1452 	int32_t daddr;
1453 	struct lfs *fs;
1454 	vnode_t *vp;
1455 
1456 	if (fhp->fid_len != sizeof(struct lfid))
1457 		return EINVAL;
1458 
1459 	memcpy(&lfh, fhp, sizeof(lfh));
1460 	if (lfh.lfid_ino < LFS_IFILE_INUM)
1461 		return ESTALE;
1462 
1463 	fs = VFSTOUFS(mp)->um_lfs;
1464 	if (lfh.lfid_ident != fs->lfs_ident)
1465 		return ESTALE;
1466 
1467 	if (lfh.lfid_ino >
1468 	    ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) -
1469 	     fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb)
1470 		return ESTALE;
1471 
1472 	mutex_enter(&ufs_ihash_lock);
1473 	vp = ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfh.lfid_ino);
1474 	mutex_exit(&ufs_ihash_lock);
1475 	if (vp == NULL) {
1476 		LFS_IENTRY(ifp, fs, lfh.lfid_ino, bp);
1477 		daddr = ifp->if_daddr;
1478 		brelse(bp, 0);
1479 		if (daddr == LFS_UNUSED_DADDR)
1480 			return ESTALE;
1481 	}
1482 
1483 	return (ufs_fhtovp(mp, &lfh.lfid_ufid, vpp));
1484 }
1485 
1486 /*
1487  * Vnode pointer to File handle
1488  */
1489 /* ARGSUSED */
1490 int
1491 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size)
1492 {
1493 	struct inode *ip;
1494 	struct lfid lfh;
1495 
1496 	if (*fh_size < sizeof(struct lfid)) {
1497 		*fh_size = sizeof(struct lfid);
1498 		return E2BIG;
1499 	}
1500 	*fh_size = sizeof(struct lfid);
1501 	ip = VTOI(vp);
1502 	memset(&lfh, 0, sizeof(lfh));
1503 	lfh.lfid_len = sizeof(struct lfid);
1504 	lfh.lfid_ino = ip->i_number;
1505 	lfh.lfid_gen = ip->i_gen;
1506 	lfh.lfid_ident = ip->i_lfs->lfs_ident;
1507 	memcpy(fhp, &lfh, sizeof(lfh));
1508 	return (0);
1509 }
1510 
1511 /*
1512  * ufs_bmaparray callback function for writing.
1513  *
1514  * Since blocks will be written to the new segment anyway,
1515  * we don't care about current daddr of them.
1516  */
1517 static bool
1518 lfs_issequential_hole(const struct ufsmount *ump,
1519     daddr_t daddr0, daddr_t daddr1)
1520 {
1521 	daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */
1522 	daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */
1523 
1524 	KASSERT(daddr0 == UNWRITTEN ||
1525 	    (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR));
1526 	KASSERT(daddr1 == UNWRITTEN ||
1527 	    (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR));
1528 
1529 	/* NOTE: all we want to know here is 'hole or not'. */
1530 	/* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */
1531 
1532 	/*
1533 	 * treat UNWRITTENs and all resident blocks as 'contiguous'
1534 	 */
1535 	if (daddr0 != 0 && daddr1 != 0)
1536 		return true;
1537 
1538 	/*
1539 	 * both are in hole?
1540 	 */
1541 	if (daddr0 == 0 && daddr1 == 0)
1542 		return true; /* all holes are 'contiguous' for us. */
1543 
1544 	return false;
1545 }
1546 
1547 /*
1548  * lfs_gop_write functions exactly like genfs_gop_write, except that
1549  * (1) it requires the seglock to be held by its caller, and sp->fip
1550  *     to be properly initialized (it will return without re-initializing
1551  *     sp->fip, and without calling lfs_writeseg).
1552  * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1553  *     to determine how large a block it can write at once (though it does
1554  *     still use VOP_BMAP to find holes in the file);
1555  * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1556  *     (leaving lfs_writeseg to deal with the cluster blocks, so we might
1557  *     now have clusters of clusters, ick.)
1558  */
1559 static int
1560 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages,
1561     int flags)
1562 {
1563 	int i, error, run, haveeof = 0;
1564 	int fs_bshift;
1565 	vaddr_t kva;
1566 	off_t eof, offset, startoffset = 0;
1567 	size_t bytes, iobytes, skipbytes;
1568 	bool async = (flags & PGO_SYNCIO) == 0;
1569 	daddr_t lbn, blkno;
1570 	struct vm_page *pg;
1571 	struct buf *mbp, *bp;
1572 	struct vnode *devvp = VTOI(vp)->i_devvp;
1573 	struct inode *ip = VTOI(vp);
1574 	struct lfs *fs = ip->i_lfs;
1575 	struct segment *sp = fs->lfs_sp;
1576 	UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1577 
1578 	ASSERT_SEGLOCK(fs);
1579 
1580 	/* The Ifile lives in the buffer cache */
1581 	KASSERT(vp != fs->lfs_ivnode);
1582 
1583 	/*
1584 	 * We don't want to fill the disk before the cleaner has a chance
1585 	 * to make room for us.  If we're in danger of doing that, fail
1586 	 * with EAGAIN.  The caller will have to notice this, unlock
1587 	 * so the cleaner can run, relock and try again.
1588 	 *
1589 	 * We must write everything, however, if our vnode is being
1590 	 * reclaimed.
1591 	 */
1592 	if (LFS_STARVED_FOR_SEGS(fs) && vp != fs->lfs_flushvp)
1593 		goto tryagain;
1594 
1595 	/*
1596 	 * Sometimes things slip past the filters in lfs_putpages,
1597 	 * and the pagedaemon tries to write pages---problem is
1598 	 * that the pagedaemon never acquires the segment lock.
1599 	 *
1600 	 * Alternatively, pages that were clean when we called
1601 	 * genfs_putpages may have become dirty in the meantime.  In this
1602 	 * case the segment header is not properly set up for blocks
1603 	 * to be added to it.
1604 	 *
1605 	 * Unbusy and unclean the pages, and put them on the ACTIVE
1606 	 * queue under the hypothesis that they couldn't have got here
1607 	 * unless they were modified *quite* recently.
1608 	 *
1609 	 * XXXUBC that last statement is an oversimplification of course.
1610 	 */
1611 	if (!LFS_SEGLOCK_HELD(fs) ||
1612 	    (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) ||
1613 	    (pgs[0]->offset & fs->lfs_bmask) != 0) {
1614 		goto tryagain;
1615 	}
1616 
1617 	UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1618 	    vp, pgs, npages, flags);
1619 
1620 	GOP_SIZE(vp, vp->v_size, &eof, 0);
1621 	haveeof = 1;
1622 
1623 	if (vp->v_type == VREG)
1624 		fs_bshift = vp->v_mount->mnt_fs_bshift;
1625 	else
1626 		fs_bshift = DEV_BSHIFT;
1627 	error = 0;
1628 	pg = pgs[0];
1629 	startoffset = pg->offset;
1630 	KASSERT(eof >= 0);
1631 
1632 	if (startoffset >= eof) {
1633 		goto tryagain;
1634 	} else
1635 		bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1636 	skipbytes = 0;
1637 
1638 	KASSERT(bytes != 0);
1639 
1640 	/* Swap PG_DELWRI for PG_PAGEOUT */
1641 	for (i = 0; i < npages; i++) {
1642 		if (pgs[i]->flags & PG_DELWRI) {
1643 			KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
1644 			pgs[i]->flags &= ~PG_DELWRI;
1645 			pgs[i]->flags |= PG_PAGEOUT;
1646 			uvm_pageout_start(1);
1647 			mutex_enter(&uvm_pageqlock);
1648 			uvm_pageunwire(pgs[i]);
1649 			mutex_exit(&uvm_pageqlock);
1650 		}
1651 	}
1652 
1653 	/*
1654 	 * Check to make sure we're starting on a block boundary.
1655 	 * We'll check later to make sure we always write entire
1656 	 * blocks (or fragments).
1657 	 */
1658 	if (startoffset & fs->lfs_bmask)
1659 		printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n",
1660 		       startoffset, fs->lfs_bmask,
1661 		       startoffset & fs->lfs_bmask);
1662 	KASSERT((startoffset & fs->lfs_bmask) == 0);
1663 	if (bytes & fs->lfs_ffmask) {
1664 		printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
1665 		panic("lfs_gop_write: non-integer blocks");
1666 	}
1667 
1668 	/*
1669 	 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK.
1670 	 * If we would, write what we have and try again.  If we don't
1671 	 * have anything to write, we'll have to sleep.
1672 	 */
1673 	if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
1674 				      (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ?
1675 				       UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) {
1676 		DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n"));
1677 #if 0
1678 		      " with nfinfo=%d at offset 0x%x\n",
1679 		      (int)((SEGSUM *)(sp->segsum))->ss_nfinfo,
1680 		      (unsigned)fs->lfs_offset));
1681 #endif
1682 		lfs_updatemeta(sp);
1683 		lfs_release_finfo(fs);
1684 		(void) lfs_writeseg(fs, sp);
1685 
1686 		lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
1687 
1688 		/*
1689 		 * Having given up all of the pager_map we were holding,
1690 		 * we can now wait for aiodoned to reclaim it for us
1691 		 * without fear of deadlock.
1692 		 */
1693 		kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
1694 				     UVMPAGER_MAPIN_WAITOK);
1695 	}
1696 
1697 	mbp = getiobuf(NULL, true);
1698 	UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
1699 	    vp, mbp, vp->v_numoutput, bytes);
1700 	mbp->b_bufsize = npages << PAGE_SHIFT;
1701 	mbp->b_data = (void *)kva;
1702 	mbp->b_resid = mbp->b_bcount = bytes;
1703 	mbp->b_cflags = BC_BUSY|BC_AGE;
1704 	mbp->b_iodone = uvm_aio_biodone;
1705 
1706 	bp = NULL;
1707 	for (offset = startoffset;
1708 	    bytes > 0;
1709 	    offset += iobytes, bytes -= iobytes) {
1710 		lbn = offset >> fs_bshift;
1711 		error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
1712 		    lfs_issequential_hole);
1713 		if (error) {
1714 			UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d",
1715 			    error,0,0,0);
1716 			skipbytes += bytes;
1717 			bytes = 0;
1718 			break;
1719 		}
1720 
1721 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
1722 		    bytes);
1723 		if (blkno == (daddr_t)-1) {
1724 			skipbytes += iobytes;
1725 			continue;
1726 		}
1727 
1728 		/*
1729 		 * Discover how much we can really pack into this buffer.
1730 		 */
1731 		/* If no room in the current segment, finish it up */
1732 		if (sp->sum_bytes_left < sizeof(int32_t) ||
1733 		    sp->seg_bytes_left < (1 << fs->lfs_bshift)) {
1734 			int vers;
1735 
1736 			lfs_updatemeta(sp);
1737 			vers = sp->fip->fi_version;
1738 			lfs_release_finfo(fs);
1739 			(void) lfs_writeseg(fs, sp);
1740 
1741 			lfs_acquire_finfo(fs, ip->i_number, vers);
1742 		}
1743 		/* Check both for space in segment and space in segsum */
1744 		iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
1745 					<< fs_bshift);
1746 		iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
1747 				       << fs_bshift);
1748 		KASSERT(iobytes > 0);
1749 
1750 		/* if it's really one i/o, don't make a second buf */
1751 		if (offset == startoffset && iobytes == bytes) {
1752 			bp = mbp;
1753 			/*
1754 			 * All the LFS output is done by the segwriter.  It
1755 			 * will increment numoutput by one for all the bufs it
1756 			 * recieves.  However this buffer needs one extra to
1757 			 * account for aiodone.
1758 			 */
1759 			mutex_enter(vp->v_interlock);
1760 			vp->v_numoutput++;
1761 			mutex_exit(vp->v_interlock);
1762 		} else {
1763 			bp = getiobuf(NULL, true);
1764 			UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
1765 			    vp, bp, vp->v_numoutput, 0);
1766 			nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes);
1767 			/*
1768 			 * LFS doesn't like async I/O here, dies with
1769 			 * and assert in lfs_bwrite().  Is that assert
1770 			 * valid?  I retained non-async behaviour when
1771 			 * converted this to use nestiobuf --pooka
1772 			 */
1773 			bp->b_flags &= ~B_ASYNC;
1774 		}
1775 
1776 		/* XXX This is silly ... is this necessary? */
1777 		mutex_enter(&bufcache_lock);
1778 		mutex_enter(vp->v_interlock);
1779 		bgetvp(vp, bp);
1780 		mutex_exit(vp->v_interlock);
1781 		mutex_exit(&bufcache_lock);
1782 
1783 		bp->b_lblkno = lblkno(fs, offset);
1784 		bp->b_private = mbp;
1785 		if (devvp->v_type == VBLK) {
1786 			bp->b_dev = devvp->v_rdev;
1787 		}
1788 		VOP_BWRITE(bp->b_vp, bp);
1789 		while (lfs_gatherblock(sp, bp, NULL))
1790 			continue;
1791 	}
1792 
1793 	nestiobuf_done(mbp, skipbytes, error);
1794 	if (skipbytes) {
1795 		UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
1796 	}
1797 	UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
1798 
1799 	if (!async) {
1800 		/* Start a segment write. */
1801 		UVMHIST_LOG(ubchist, "flushing", 0,0,0,0);
1802 		mutex_enter(&lfs_lock);
1803 		lfs_flush(fs, 0, 1);
1804 		mutex_exit(&lfs_lock);
1805 	}
1806 	return (0);
1807 
1808     tryagain:
1809 	/*
1810 	 * We can't write the pages, for whatever reason.
1811 	 * Clean up after ourselves, and make the caller try again.
1812 	 */
1813 	mutex_enter(vp->v_interlock);
1814 
1815 	/* Tell why we're here, if we know */
1816 	if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) {
1817 		DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n"));
1818 	} else if ((pgs[0]->offset & fs->lfs_bmask) != 0) {
1819 		DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n"));
1820 	} else if (haveeof && startoffset >= eof) {
1821 		DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64
1822 		      " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number,
1823 		      pgs[0]->offset, eof, npages));
1824 	} else if (LFS_STARVED_FOR_SEGS(fs)) {
1825 		DLOG((DLOG_PAGE, "lfs_gop_write: avail too low\n"));
1826 	} else {
1827 		DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n"));
1828 	}
1829 
1830 	mutex_enter(&uvm_pageqlock);
1831 	for (i = 0; i < npages; i++) {
1832 		pg = pgs[i];
1833 
1834 		if (pg->flags & PG_PAGEOUT)
1835 			uvm_pageout_done(1);
1836 		if (pg->flags & PG_DELWRI) {
1837 			uvm_pageunwire(pg);
1838 		}
1839 		uvm_pageactivate(pg);
1840 		pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
1841 		DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg,
1842 			vp, pg->offset));
1843 		DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags));
1844 		DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags));
1845 		DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon));
1846 		DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject));
1847 		DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i,
1848 		      pg->wire_count));
1849 		DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i,
1850 		      pg->loan_count));
1851 	}
1852 	/* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
1853 	uvm_page_unbusy(pgs, npages);
1854 	mutex_exit(&uvm_pageqlock);
1855 	mutex_exit(vp->v_interlock);
1856 	return EAGAIN;
1857 }
1858 
1859 /*
1860  * finish vnode/inode initialization.
1861  * used by lfs_vget and lfs_fastvget.
1862  */
1863 void
1864 lfs_vinit(struct mount *mp, struct vnode **vpp)
1865 {
1866 	struct vnode *vp = *vpp;
1867 	struct inode *ip = VTOI(vp);
1868 	struct ufsmount *ump = VFSTOUFS(mp);
1869 	struct lfs *fs = ump->um_lfs;
1870 	int i;
1871 
1872 	ip->i_mode = ip->i_ffs1_mode;
1873 	ip->i_nlink = ip->i_ffs1_nlink;
1874 	ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size;
1875 	ip->i_flags = ip->i_ffs1_flags;
1876 	ip->i_gen = ip->i_ffs1_gen;
1877 	ip->i_uid = ip->i_ffs1_uid;
1878 	ip->i_gid = ip->i_ffs1_gid;
1879 
1880 	ip->i_lfs_effnblks = ip->i_ffs1_blocks;
1881 	ip->i_lfs_odnlink = ip->i_ffs1_nlink;
1882 
1883 	/*
1884 	 * Initialize the vnode from the inode, check for aliases.  In all
1885 	 * cases re-init ip, the underlying vnode/inode may have changed.
1886 	 */
1887 	ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
1888 	ip = VTOI(vp);
1889 
1890 	memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize));
1891 	if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) {
1892 #ifdef DEBUG
1893 		for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift;
1894 		    i < NDADDR; i++) {
1895 			if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
1896 			    i == 0)
1897 				continue;
1898 			if (ip->i_ffs1_db[i] != 0) {
1899 inconsistent:
1900 				lfs_dump_dinode(ip->i_din.ffs1_din);
1901 				panic("inconsistent inode");
1902 			}
1903 		}
1904 		for ( ; i < NDADDR + NIADDR; i++) {
1905 			if (ip->i_ffs1_ib[i - NDADDR] != 0) {
1906 				goto inconsistent;
1907 			}
1908 		}
1909 #endif /* DEBUG */
1910 		for (i = 0; i < NDADDR; i++)
1911 			if (ip->i_ffs1_db[i] != 0)
1912 				ip->i_lfs_fragsize[i] = blksize(fs, ip, i);
1913 	}
1914 
1915 #ifdef DIAGNOSTIC
1916 	if (vp->v_type == VNON) {
1917 # ifdef DEBUG
1918 		lfs_dump_dinode(ip->i_din.ffs1_din);
1919 # endif
1920 		panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n",
1921 		      (unsigned long long)ip->i_number,
1922 		      (ip->i_mode & IFMT) >> 12);
1923 	}
1924 #endif /* DIAGNOSTIC */
1925 
1926 	/*
1927 	 * Finish inode initialization now that aliasing has been resolved.
1928 	 */
1929 
1930 	ip->i_devvp = ump->um_devvp;
1931 	vref(ip->i_devvp);
1932 	genfs_node_init(vp, &lfs_genfsops);
1933 	uvm_vnp_setsize(vp, ip->i_size);
1934 
1935 	/* Initialize hiblk from file size */
1936 	ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1;
1937 
1938 	*vpp = vp;
1939 }
1940 
1941 /*
1942  * Resize the filesystem to contain the specified number of segments.
1943  */
1944 int
1945 lfs_resize_fs(struct lfs *fs, int newnsegs)
1946 {
1947 	SEGUSE *sup;
1948 	struct buf *bp, *obp;
1949 	daddr_t olast, nlast, ilast, noff, start, end;
1950 	struct vnode *ivp;
1951 	struct inode *ip;
1952 	int error, badnews, inc, oldnsegs;
1953 	int sbbytes, csbbytes, gain, cgain;
1954 	int i;
1955 
1956 	/* Only support v2 and up */
1957 	if (fs->lfs_version < 2)
1958 		return EOPNOTSUPP;
1959 
1960 	/* If we're doing nothing, do it fast */
1961 	oldnsegs = fs->lfs_nseg;
1962 	if (newnsegs == oldnsegs)
1963 		return 0;
1964 
1965 	/* We always have to have two superblocks */
1966 	if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1]))
1967 		return EFBIG;
1968 
1969 	ivp = fs->lfs_ivnode;
1970 	ip = VTOI(ivp);
1971 	error = 0;
1972 
1973 	/* Take the segment lock so no one else calls lfs_newseg() */
1974 	lfs_seglock(fs, SEGM_PROT);
1975 
1976 	/*
1977 	 * Make sure the segments we're going to be losing, if any,
1978 	 * are in fact empty.  We hold the seglock, so their status
1979 	 * cannot change underneath us.  Count the superblocks we lose,
1980 	 * while we're at it.
1981 	 */
1982 	sbbytes = csbbytes = 0;
1983 	cgain = 0;
1984 	for (i = newnsegs; i < oldnsegs; i++) {
1985 		LFS_SEGENTRY(sup, fs, i, bp);
1986 		badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL);
1987 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
1988 			sbbytes += LFS_SBPAD;
1989 		if (!(sup->su_flags & SEGUSE_DIRTY)) {
1990 			++cgain;
1991 			if (sup->su_flags & SEGUSE_SUPERBLOCK)
1992 				csbbytes += LFS_SBPAD;
1993 		}
1994 		brelse(bp, 0);
1995 		if (badnews) {
1996 			error = EBUSY;
1997 			goto out;
1998 		}
1999 	}
2000 
2001 	/* Note old and new segment table endpoints, and old ifile size */
2002 	olast = fs->lfs_cleansz + fs->lfs_segtabsz;
2003 	nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz;
2004 	ilast = ivp->v_size >> fs->lfs_bshift;
2005 	noff = nlast - olast;
2006 
2007 	/*
2008 	 * Make sure no one can use the Ifile while we change it around.
2009 	 * Even after taking the iflock we need to make sure no one still
2010 	 * is holding Ifile buffers, so we get each one, to drain them.
2011 	 * (XXX this could be done better.)
2012 	 */
2013 	rw_enter(&fs->lfs_iflock, RW_WRITER);
2014 	vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY);
2015 	for (i = 0; i < ilast; i++) {
2016 		bread(ivp, i, fs->lfs_bsize, NOCRED, 0, &bp);
2017 		brelse(bp, 0);
2018 	}
2019 
2020 	/* Allocate new Ifile blocks */
2021 	for (i = ilast; i < ilast + noff; i++) {
2022 		if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0,
2023 			       &bp) != 0)
2024 			panic("balloc extending ifile");
2025 		memset(bp->b_data, 0, fs->lfs_bsize);
2026 		VOP_BWRITE(bp->b_vp, bp);
2027 	}
2028 
2029 	/* Register new ifile size */
2030 	ip->i_size += noff * fs->lfs_bsize;
2031 	ip->i_ffs1_size = ip->i_size;
2032 	uvm_vnp_setsize(ivp, ip->i_size);
2033 
2034 	/* Copy the inode table to its new position */
2035 	if (noff != 0) {
2036 		if (noff < 0) {
2037 			start = nlast;
2038 			end = ilast + noff;
2039 			inc = 1;
2040 		} else {
2041 			start = ilast + noff - 1;
2042 			end = nlast - 1;
2043 			inc = -1;
2044 		}
2045 		for (i = start; i != end; i += inc) {
2046 			if (bread(ivp, i, fs->lfs_bsize, NOCRED,
2047 			    B_MODIFY, &bp) != 0)
2048 				panic("resize: bread dst blk failed");
2049 			if (bread(ivp, i - noff, fs->lfs_bsize,
2050 			    NOCRED, 0, &obp))
2051 				panic("resize: bread src blk failed");
2052 			memcpy(bp->b_data, obp->b_data, fs->lfs_bsize);
2053 			VOP_BWRITE(bp->b_vp, bp);
2054 			brelse(obp, 0);
2055 		}
2056 	}
2057 
2058 	/* If we are expanding, write the new empty SEGUSE entries */
2059 	if (newnsegs > oldnsegs) {
2060 		for (i = oldnsegs; i < newnsegs; i++) {
2061 			if ((error = bread(ivp, i / fs->lfs_sepb +
2062 					   fs->lfs_cleansz, fs->lfs_bsize,
2063 					   NOCRED, B_MODIFY, &bp)) != 0)
2064 				panic("lfs: ifile read: %d", error);
2065 			while ((i + 1) % fs->lfs_sepb && i < newnsegs) {
2066 				sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb];
2067 				memset(sup, 0, sizeof(*sup));
2068 				i++;
2069 			}
2070 			VOP_BWRITE(bp->b_vp, bp);
2071 		}
2072 	}
2073 
2074 	/* Zero out unused superblock offsets */
2075 	for (i = 2; i < LFS_MAXNUMSB; i++)
2076 		if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs)
2077 			fs->lfs_sboffs[i] = 0x0;
2078 
2079 	/*
2080 	 * Correct superblock entries that depend on fs size.
2081 	 * The computations of these are as follows:
2082 	 *
2083 	 * size  = segtod(fs, nseg)
2084 	 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD)
2085 	 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used
2086 	 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD)
2087 	 *         + (segtod(fs, 1) - (offset - curseg))
2088 	 *	   - segtod(fs, minfreeseg - (minfreeseg / 2))
2089 	 *
2090 	 * XXX - we should probably adjust minfreeseg as well.
2091 	 */
2092 	gain = (newnsegs - oldnsegs);
2093 	fs->lfs_nseg = newnsegs;
2094 	fs->lfs_segtabsz = nlast - fs->lfs_cleansz;
2095 	fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize);
2096 	fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes);
2097 	fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes)
2098 		       - gain * btofsb(fs, fs->lfs_bsize / 2);
2099 	if (gain > 0) {
2100 		fs->lfs_nclean += gain;
2101 		fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize);
2102 	} else {
2103 		fs->lfs_nclean -= cgain;
2104 		fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) -
2105 				 btofsb(fs, csbbytes);
2106 	}
2107 
2108 	/* Resize segment flag cache */
2109 	fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0],
2110 						  fs->lfs_nseg * sizeof(u_int32_t),
2111 						  M_SEGMENT, M_WAITOK);
2112 	fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[1],
2113 						  fs->lfs_nseg * sizeof(u_int32_t),
2114 						  M_SEGMENT, M_WAITOK);
2115 	for (i = oldnsegs; i < newnsegs; i++)
2116 		fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0;
2117 
2118 	/* Truncate Ifile if necessary */
2119 	if (noff < 0)
2120 		lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0,
2121 		    NOCRED);
2122 
2123 	/* Update cleaner info so the cleaner can die */
2124 	bread(ivp, 0, fs->lfs_bsize, NOCRED, B_MODIFY, &bp);
2125 	((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean;
2126 	((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean;
2127 	VOP_BWRITE(bp->b_vp, bp);
2128 
2129 	/* Let Ifile accesses proceed */
2130 	VOP_UNLOCK(ivp);
2131 	rw_exit(&fs->lfs_iflock);
2132 
2133     out:
2134 	lfs_segunlock(fs);
2135 	return error;
2136 }
2137