xref: /netbsd-src/sys/ufs/lfs/lfs_vfsops.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: lfs_vfsops.c,v 1.321 2014/04/16 18:55:19 maxv 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 1.321 2014/04/16 18:55:19 maxv 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/lfs/ulfs_quotacommon.h>
102 #include <ufs/lfs/ulfs_inode.h>
103 #include <ufs/lfs/ulfsmount.h>
104 #include <ufs/lfs/ulfs_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_kernel.h>
113 #include <ufs/lfs/lfs_extern.h>
114 
115 #include <miscfs/genfs/genfs.h>
116 #include <miscfs/genfs/genfs_node.h>
117 
118 MODULE(MODULE_CLASS_VFS, lfs, NULL);
119 
120 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
121 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *);
122 static int lfs_extattrctl(struct mount *, int, struct vnode *, int,
123 			  const char *);
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 lwpid_t lfs_writer_lid = 0;
133 int lfs_do_flush = 0;
134 #ifdef LFS_KERNEL_RFW
135 int lfs_do_rfw = 0;
136 #endif
137 
138 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
139 	&lfs_vnodeop_opv_desc,
140 	&lfs_specop_opv_desc,
141 	&lfs_fifoop_opv_desc,
142 	NULL,
143 };
144 
145 struct vfsops lfs_vfsops = {
146 	.vfs_name = MOUNT_LFS,
147 	.vfs_min_mount_data = sizeof (struct ulfs_args),
148 	.vfs_mount = lfs_mount,
149 	.vfs_start = ulfs_start,
150 	.vfs_unmount = lfs_unmount,
151 	.vfs_root = ulfs_root,
152 	.vfs_quotactl = ulfs_quotactl,
153 	.vfs_statvfs = lfs_statvfs,
154 	.vfs_sync = lfs_sync,
155 	.vfs_vget = lfs_vget,
156 	.vfs_fhtovp = lfs_fhtovp,
157 	.vfs_vptofh = lfs_vptofh,
158 	.vfs_init = lfs_init,
159 	.vfs_reinit = lfs_reinit,
160 	.vfs_done = lfs_done,
161 	.vfs_mountroot = lfs_mountroot,
162 	.vfs_snapshot = (void *)eopnotsupp,
163 	.vfs_extattrctl = lfs_extattrctl,
164 	.vfs_suspendctl = (void *)eopnotsupp,
165 	.vfs_renamelock_enter = genfs_renamelock_enter,
166 	.vfs_renamelock_exit = genfs_renamelock_exit,
167 	.vfs_fsync = (void *)eopnotsupp,
168 	.vfs_opv_descs = lfs_vnodeopv_descs
169 };
170 
171 const struct genfs_ops lfs_genfsops = {
172 	.gop_size = lfs_gop_size,
173 	.gop_alloc = ulfs_gop_alloc,
174 	.gop_write = lfs_gop_write,
175 	.gop_markupdate = ulfs_gop_markupdate,
176 };
177 
178 struct shortlong {
179 	const char *sname;
180 	const char *lname;
181 };
182 
183 static int
184 sysctl_lfs_dostats(SYSCTLFN_ARGS)
185 {
186 	extern struct lfs_stats lfs_stats;
187 	extern int lfs_dostats;
188 	int error;
189 
190 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
191 	if (error || newp == NULL)
192 		return (error);
193 
194 	if (lfs_dostats == 0)
195 		memset(&lfs_stats, 0, sizeof(lfs_stats));
196 
197 	return (0);
198 }
199 
200 static void
201 lfs_sysctl_setup(struct sysctllog **clog)
202 {
203 	int i;
204 	extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead,
205 		   lfs_fs_pagetrip, lfs_ignore_lazy_sync;
206 #ifdef DEBUG
207 	extern int lfs_debug_log_subsys[DLOG_MAX];
208 	struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */
209 		{ "rollforward", "Debug roll-forward code" },
210 		{ "alloc",	"Debug inode allocation and free list" },
211 		{ "avail",	"Debug space-available-now accounting" },
212 		{ "flush",	"Debug flush triggers" },
213 		{ "lockedlist",	"Debug locked list accounting" },
214 		{ "vnode_verbose", "Verbose per-vnode-written debugging" },
215 		{ "vnode",	"Debug vnode use during segment write" },
216 		{ "segment",	"Debug segment writing" },
217 		{ "seguse",	"Debug segment used-bytes accounting" },
218 		{ "cleaner",	"Debug cleaning routines" },
219 		{ "mount",	"Debug mount/unmount routines" },
220 		{ "pagecache",	"Debug UBC interactions" },
221 		{ "dirop",	"Debug directory-operation accounting" },
222 		{ "malloc",	"Debug private malloc accounting" },
223 	};
224 #endif /* DEBUG */
225 	struct shortlong stat_names[] = { /* Must match lfs.h! */
226 		{ "segsused",	    "Number of new segments allocated" },
227 		{ "psegwrites",	    "Number of partial-segment writes" },
228 		{ "psyncwrites",    "Number of synchronous partial-segment"
229 				    " writes" },
230 		{ "pcleanwrites",   "Number of partial-segment writes by the"
231 				    " cleaner" },
232 		{ "blocktot",       "Number of blocks written" },
233 		{ "cleanblocks",    "Number of blocks written by the cleaner" },
234 		{ "ncheckpoints",   "Number of checkpoints made" },
235 		{ "nwrites",        "Number of whole writes" },
236 		{ "nsync_writes",   "Number of synchronous writes" },
237 		{ "wait_exceeded",  "Number of times writer waited for"
238 				    " cleaner" },
239 		{ "write_exceeded", "Number of times writer invoked flush" },
240 		{ "flush_invoked",  "Number of times flush was invoked" },
241 		{ "vflush_invoked", "Number of time vflush was called" },
242 		{ "clean_inlocked", "Number of vnodes skipped for being dead" },
243 		{ "clean_vnlocked", "Number of vnodes skipped for vget failure" },
244 		{ "segs_reclaimed", "Number of segments reclaimed" },
245 	};
246 
247 	sysctl_createv(clog, 0, NULL, NULL,
248 		       CTLFLAG_PERMANENT,
249 		       CTLTYPE_NODE, "lfs",
250 		       SYSCTL_DESCR("Log-structured file system"),
251 		       NULL, 0, NULL, 0,
252 		       CTL_VFS, 5, CTL_EOL);
253 	/*
254 	 * XXX the "5" above could be dynamic, thereby eliminating one
255 	 * more instance of the "number to vfs" mapping problem, but
256 	 * "5" is the order as taken from sys/mount.h
257 	 */
258 
259 	sysctl_createv(clog, 0, NULL, NULL,
260 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
261 		       CTLTYPE_INT, "flushindir", NULL,
262 		       NULL, 0, &lfs_writeindir, 0,
263 		       CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
264 	sysctl_createv(clog, 0, NULL, NULL,
265 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
266 		       CTLTYPE_INT, "clean_vnhead", NULL,
267 		       NULL, 0, &lfs_clean_vnhead, 0,
268 		       CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
269 	sysctl_createv(clog, 0, NULL, NULL,
270 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
271 		       CTLTYPE_INT, "dostats",
272 		       SYSCTL_DESCR("Maintain statistics on LFS operations"),
273 		       sysctl_lfs_dostats, 0, &lfs_dostats, 0,
274 		       CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
275 	sysctl_createv(clog, 0, NULL, NULL,
276 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
277 		       CTLTYPE_INT, "pagetrip",
278 		       SYSCTL_DESCR("How many dirty pages in fs triggers"
279 				    " a flush"),
280 		       NULL, 0, &lfs_fs_pagetrip, 0,
281 		       CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL);
282 	sysctl_createv(clog, 0, NULL, NULL,
283 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
284 		       CTLTYPE_INT, "ignore_lazy_sync",
285 		       SYSCTL_DESCR("Lazy Sync is ignored entirely"),
286 		       NULL, 0, &lfs_ignore_lazy_sync, 0,
287 		       CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL);
288 #ifdef LFS_KERNEL_RFW
289 	sysctl_createv(clog, 0, NULL, NULL,
290 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
291 		       CTLTYPE_INT, "rfw",
292 		       SYSCTL_DESCR("Use in-kernel roll-forward on mount"),
293 		       NULL, 0, &lfs_do_rfw, 0,
294 		       CTL_VFS, 5, LFS_DO_RFW, CTL_EOL);
295 #endif
296 
297 	sysctl_createv(clog, 0, NULL, NULL,
298 		       CTLFLAG_PERMANENT,
299 		       CTLTYPE_NODE, "stats",
300 		       SYSCTL_DESCR("Debugging options"),
301 		       NULL, 0, NULL, 0,
302 		       CTL_VFS, 5, LFS_STATS, CTL_EOL);
303 	for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) {
304 		sysctl_createv(clog, 0, NULL, NULL,
305 			       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
306 			       CTLTYPE_INT, stat_names[i].sname,
307 			       SYSCTL_DESCR(stat_names[i].lname),
308 			       NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]),
309 			       0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL);
310 	}
311 
312 #ifdef DEBUG
313 	sysctl_createv(clog, 0, NULL, NULL,
314 		       CTLFLAG_PERMANENT,
315 		       CTLTYPE_NODE, "debug",
316 		       SYSCTL_DESCR("Debugging options"),
317 		       NULL, 0, NULL, 0,
318 		       CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL);
319 	for (i = 0; i < DLOG_MAX; i++) {
320 		sysctl_createv(clog, 0, NULL, NULL,
321 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
322 			       CTLTYPE_INT, dlog_names[i].sname,
323 			       SYSCTL_DESCR(dlog_names[i].lname),
324 			       NULL, 0, &(lfs_debug_log_subsys[i]), 0,
325 			       CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL);
326 	}
327 #endif
328 }
329 
330 /* old cleaner syscall interface.  see VOP_FCNTL() */
331 static const struct syscall_package lfs_syscalls[] = {
332 	{ SYS_lfs_bmapv,	0, (sy_call_t *)sys_lfs_bmapv		},
333 	{ SYS_lfs_markv,	0, (sy_call_t *)sys_lfs_markv		},
334 	{ SYS_lfs_segclean,	0, (sy_call_t *)sys___lfs_segwait50	},
335 	{ 0, 0, NULL },
336 };
337 
338 static int
339 lfs_modcmd(modcmd_t cmd, void *arg)
340 {
341 	int error;
342 
343 	switch (cmd) {
344 	case MODULE_CMD_INIT:
345 		error = syscall_establish(NULL, lfs_syscalls);
346 		if (error)
347 			return error;
348 		error = vfs_attach(&lfs_vfsops);
349 		if (error != 0) {
350 			syscall_disestablish(NULL, lfs_syscalls);
351 			break;
352 		}
353 		lfs_sysctl_setup(&lfs_sysctl_log);
354 		break;
355 	case MODULE_CMD_FINI:
356 		error = vfs_detach(&lfs_vfsops);
357 		if (error != 0)
358 			break;
359 		syscall_disestablish(NULL, lfs_syscalls);
360 		sysctl_teardown(&lfs_sysctl_log);
361 		break;
362 	default:
363 		error = ENOTTY;
364 		break;
365 	}
366 
367 	return (error);
368 }
369 
370 /*
371  * XXX Same structure as FFS inodes?  Should we share a common pool?
372  */
373 struct pool lfs_inode_pool;
374 struct pool lfs_dinode_pool;
375 struct pool lfs_inoext_pool;
376 struct pool lfs_lbnentry_pool;
377 
378 /*
379  * The writer daemon.  UVM keeps track of how many dirty pages we are holding
380  * in lfs_subsys_pages; the daemon flushes the filesystem when this value
381  * crosses the (user-defined) threshhold LFS_MAX_PAGES.
382  */
383 static void
384 lfs_writerd(void *arg)
385 {
386  	struct mount *mp, *nmp;
387  	struct lfs *fs;
388 	struct vfsops *vfs = NULL;
389  	int fsflags;
390 	int skipc;
391 	int lfsc;
392 	int wrote_something = 0;
393 
394 	mutex_enter(&lfs_lock);
395  	lfs_writer_daemon = curproc->p_pid;
396 	lfs_writer_lid = curlwp->l_lid;
397 	mutex_exit(&lfs_lock);
398 
399 	/* Take an extra reference to the LFS vfsops. */
400 	vfs = vfs_getopsbyname(MOUNT_LFS);
401 
402  	mutex_enter(&lfs_lock);
403  	for (;;) {
404 		KASSERT(mutex_owned(&lfs_lock));
405 		if (wrote_something == 0)
406 			mtsleep(&lfs_writer_daemon, PVM, "lfswriter", hz/10 + 1,
407 				&lfs_lock);
408 
409 		KASSERT(mutex_owned(&lfs_lock));
410 		wrote_something = 0;
411 
412 		/*
413 		 * If global state wants a flush, flush everything.
414 		 */
415 		if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
416 			locked_queue_bytes > LFS_MAX_BYTES ||
417 			lfs_subsys_pages > LFS_MAX_PAGES) {
418 
419 			if (lfs_do_flush) {
420 				DLOG((DLOG_FLUSH, "lfs_writerd: lfs_do_flush\n"));
421 			}
422 			if (locked_queue_count > LFS_MAX_BUFS) {
423 				DLOG((DLOG_FLUSH, "lfs_writerd: lqc = %d, max %d\n",
424 				      locked_queue_count, LFS_MAX_BUFS));
425 			}
426 			if (locked_queue_bytes > LFS_MAX_BYTES) {
427 				DLOG((DLOG_FLUSH, "lfs_writerd: lqb = %ld, max %ld\n",
428 				      locked_queue_bytes, LFS_MAX_BYTES));
429 			}
430 			if (lfs_subsys_pages > LFS_MAX_PAGES) {
431 				DLOG((DLOG_FLUSH, "lfs_writerd: lssp = %d, max %d\n",
432 				      lfs_subsys_pages, LFS_MAX_PAGES));
433 			}
434 
435 			lfs_flush(NULL, SEGM_WRITERD, 0);
436 			lfs_do_flush = 0;
437 			KASSERT(mutex_owned(&lfs_lock));
438 			continue;
439 		}
440 		KASSERT(mutex_owned(&lfs_lock));
441 		mutex_exit(&lfs_lock);
442 
443  		/*
444  		 * Look through the list of LFSs to see if any of them
445  		 * have requested pageouts.
446  		 */
447  		mutex_enter(&mountlist_lock);
448 		lfsc = 0;
449 		skipc = 0;
450  		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
451  			if (vfs_busy(mp, &nmp)) {
452 				++skipc;
453  				continue;
454  			}
455 			KASSERT(!mutex_owned(&lfs_lock));
456  			if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS,
457  			    sizeof(mp->mnt_stat.f_fstypename)) == 0) {
458 				++lfsc;
459  				fs = VFSTOULFS(mp)->um_lfs;
460 				int32_t ooffset = 0;
461 				fsflags = SEGM_SINGLE;
462 
463  				mutex_enter(&lfs_lock);
464 				ooffset = fs->lfs_offset;
465 
466 				if (fs->lfs_nextseg < fs->lfs_curseg && fs->lfs_nowrap) {
467 					/* Don't try to write if we're suspended */
468 					mutex_exit(&lfs_lock);
469 					vfs_unbusy(mp, false, &nmp);
470 					continue;
471 				}
472 				if (LFS_STARVED_FOR_SEGS(fs)) {
473 					mutex_exit(&lfs_lock);
474 
475 					DLOG((DLOG_FLUSH, "lfs_writerd: need cleaning before writing possible\n"));
476 					lfs_wakeup_cleaner(fs);
477 					vfs_unbusy(mp, false, &nmp);
478 					continue;
479 				}
480 
481  				if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) ||
482  				     lfs_dirvcount > LFS_MAX_DIROP) &&
483 				    fs->lfs_dirops == 0) {
484 					fsflags &= ~SEGM_SINGLE;
485  					fsflags |= SEGM_CKP;
486 					DLOG((DLOG_FLUSH, "lfs_writerd: checkpoint\n"));
487 					lfs_flush_fs(fs, fsflags);
488 				} else if (fs->lfs_pdflush) {
489  					DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n"));
490  					lfs_flush_fs(fs, fsflags);
491  				} else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) {
492  					DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n"));
493  					mutex_exit(&lfs_lock);
494  					lfs_writer_enter(fs, "wrdirop");
495  					lfs_flush_pchain(fs);
496  					lfs_writer_leave(fs);
497 					mutex_enter(&lfs_lock);
498 				}
499 				if (fs->lfs_offset != ooffset)
500 					++wrote_something;
501 				mutex_exit(&lfs_lock);
502  			}
503 			KASSERT(!mutex_owned(&lfs_lock));
504  			vfs_unbusy(mp, false, &nmp);
505  		}
506 		if (lfsc + skipc == 0) {
507 			mutex_enter(&lfs_lock);
508 			lfs_writer_daemon = 0;
509 			lfs_writer_lid = 0;
510 			mutex_exit(&lfs_lock);
511 			mutex_exit(&mountlist_lock);
512 			break;
513 		}
514  		mutex_exit(&mountlist_lock);
515 
516  		mutex_enter(&lfs_lock);
517  	}
518 	KASSERT(!mutex_owned(&lfs_lock));
519 	KASSERT(!mutex_owned(&mountlist_lock));
520 
521 	/* Give up our extra reference so the module can be unloaded. */
522 	mutex_enter(&vfs_list_lock);
523 	if (vfs != NULL)
524 		vfs->vfs_refcount--;
525 	mutex_exit(&vfs_list_lock);
526 
527 	/* Done! */
528 	kthread_exit(0);
529 }
530 
531 /*
532  * Initialize the filesystem, most work done by ulfs_init.
533  */
534 void
535 lfs_init(void)
536 {
537 
538 	malloc_type_attach(M_SEGMENT);
539 	pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
540 	    "lfsinopl", &pool_allocator_nointr, IPL_NONE);
541 	pool_init(&lfs_dinode_pool, sizeof(struct ulfs1_dinode), 0, 0, 0,
542 	    "lfsdinopl", &pool_allocator_nointr, IPL_NONE);
543 	pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
544 	    "lfsinoextpl", &pool_allocator_nointr, IPL_NONE);
545 	pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0,
546 	    "lfslbnpool", &pool_allocator_nointr, IPL_NONE);
547 	ulfs_init();
548 
549 #ifdef DEBUG
550 	memset(lfs_log, 0, sizeof(lfs_log));
551 #endif
552 	mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE);
553 	cv_init(&locked_queue_cv, "lfsbuf");
554 	cv_init(&lfs_writing_cv, "lfsflush");
555 }
556 
557 void
558 lfs_reinit(void)
559 {
560 	ulfs_reinit();
561 }
562 
563 void
564 lfs_done(void)
565 {
566 	ulfs_done();
567 	mutex_destroy(&lfs_lock);
568 	cv_destroy(&locked_queue_cv);
569 	cv_destroy(&lfs_writing_cv);
570 	pool_destroy(&lfs_inode_pool);
571 	pool_destroy(&lfs_dinode_pool);
572 	pool_destroy(&lfs_inoext_pool);
573 	pool_destroy(&lfs_lbnentry_pool);
574 	malloc_type_detach(M_SEGMENT);
575 }
576 
577 /*
578  * Called by main() when ulfs is going to be mounted as root.
579  */
580 int
581 lfs_mountroot(void)
582 {
583 	extern struct vnode *rootvp;
584 	struct lfs *fs = NULL;				/* LFS */
585 	struct mount *mp;
586 	struct lwp *l = curlwp;
587 	struct ulfsmount *ump;
588 	int error;
589 
590 	if (device_class(root_device) != DV_DISK)
591 		return (ENODEV);
592 
593 	if (rootdev == NODEV)
594 		return (ENODEV);
595 	if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
596 		vrele(rootvp);
597 		return (error);
598 	}
599 	if ((error = lfs_mountfs(rootvp, mp, l))) {
600 		vfs_unbusy(mp, false, NULL);
601 		vfs_destroy(mp);
602 		return (error);
603 	}
604 	mountlist_append(mp);
605 	ump = VFSTOULFS(mp);
606 	fs = ump->um_lfs;
607 	memset(fs->lfs_fsmnt, 0, sizeof(fs->lfs_fsmnt));
608 	(void)copystr(mp->mnt_stat.f_mntonname, fs->lfs_fsmnt, MNAMELEN - 1, 0);
609 	(void)lfs_statvfs(mp, &mp->mnt_stat);
610 	vfs_unbusy(mp, false, NULL);
611 	setrootfstime((time_t)(VFSTOULFS(mp)->um_lfs->lfs_tstamp));
612 	return (0);
613 }
614 
615 /*
616  * VFS Operations.
617  *
618  * mount system call
619  */
620 int
621 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
622 {
623 	struct lwp *l = curlwp;
624 	struct vnode *devvp;
625 	struct ulfs_args *args = data;
626 	struct ulfsmount *ump = NULL;
627 	struct lfs *fs = NULL;				/* LFS */
628 	int error = 0, update;
629 	mode_t accessmode;
630 
631 	if (args == NULL)
632 		return EINVAL;
633 	if (*data_len < sizeof *args)
634 		return EINVAL;
635 
636 	if (mp->mnt_flag & MNT_GETARGS) {
637 		ump = VFSTOULFS(mp);
638 		if (ump == NULL)
639 			return EIO;
640 		args->fspec = NULL;
641 		*data_len = sizeof *args;
642 		return 0;
643 	}
644 
645 	update = mp->mnt_flag & MNT_UPDATE;
646 
647 	/* Check arguments */
648 	if (args->fspec != NULL) {
649 		/*
650 		 * Look up the name and verify that it's sane.
651 		 */
652 		error = namei_simple_user(args->fspec,
653 					NSM_FOLLOW_NOEMULROOT, &devvp);
654 		if (error != 0)
655 			return (error);
656 
657 		if (!update) {
658 			/*
659 			 * Be sure this is a valid block device
660 			 */
661 			if (devvp->v_type != VBLK)
662 				error = ENOTBLK;
663 			else if (bdevsw_lookup(devvp->v_rdev) == NULL)
664 				error = ENXIO;
665 		} else {
666 			/*
667 			 * Be sure we're still naming the same device
668 			 * used for our initial mount
669 			 */
670 			ump = VFSTOULFS(mp);
671 			if (devvp != ump->um_devvp) {
672 				if (devvp->v_rdev != ump->um_devvp->v_rdev)
673 					error = EINVAL;
674 				else {
675 					vrele(devvp);
676 					devvp = ump->um_devvp;
677 					vref(devvp);
678 				}
679 			}
680 		}
681 	} else {
682 		if (!update) {
683 			/* New mounts must have a filename for the device */
684 			return (EINVAL);
685 		} else {
686 			/* Use the extant mount */
687 			ump = VFSTOULFS(mp);
688 			devvp = ump->um_devvp;
689 			vref(devvp);
690 		}
691 	}
692 
693 
694 	/*
695 	 * If mount by non-root, then verify that user has necessary
696 	 * permissions on the device.
697 	 */
698 	if (error == 0) {
699 		accessmode = VREAD;
700 		if (update ?
701 		    (mp->mnt_iflag & IMNT_WANTRDWR) != 0 :
702 		    (mp->mnt_flag & MNT_RDONLY) == 0)
703 			accessmode |= VWRITE;
704 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
705 		error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT,
706 		    KAUTH_REQ_SYSTEM_MOUNT_DEVICE, mp, devvp,
707 		    KAUTH_ARG(accessmode));
708 		VOP_UNLOCK(devvp);
709 	}
710 
711 	if (error) {
712 		vrele(devvp);
713 		return (error);
714 	}
715 
716 	if (!update) {
717 		int flags;
718 
719 		if (mp->mnt_flag & MNT_RDONLY)
720 			flags = FREAD;
721 		else
722 			flags = FREAD|FWRITE;
723 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
724 		error = VOP_OPEN(devvp, flags, FSCRED);
725 		VOP_UNLOCK(devvp);
726 		if (error)
727 			goto fail;
728 		error = lfs_mountfs(devvp, mp, l);		/* LFS */
729 		if (error) {
730 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
731 			(void)VOP_CLOSE(devvp, flags, NOCRED);
732 			VOP_UNLOCK(devvp);
733 			goto fail;
734 		}
735 
736 		ump = VFSTOULFS(mp);
737 		fs = ump->um_lfs;
738 	} else {
739 		/*
740 		 * Update the mount.
741 		 */
742 
743 		/*
744 		 * The initial mount got a reference on this
745 		 * device, so drop the one obtained via
746 		 * namei(), above.
747 		 */
748 		vrele(devvp);
749 
750 		ump = VFSTOULFS(mp);
751 		fs = ump->um_lfs;
752 
753 		if (fs->lfs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
754 			/*
755 			 * Changing from read/write to read-only.
756 			 * XXX: shouldn't we sync here? or does vfs do that?
757 			 */
758 #ifdef LFS_QUOTA2
759 			/* XXX: quotas should remain on when readonly */
760 			if (fs->lfs_use_quota2) {
761 				error = lfsquota2_umount(mp, 0);
762 				if (error) {
763 					return error;
764 				}
765 			}
766 #endif
767 		}
768 
769 		if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
770 			/*
771 			 * Changing from read-only to read/write.
772 			 * Note in the superblocks that we're writing.
773 			 */
774 
775 			/* XXX: quotas should have been on even if readonly */
776 			if (fs->lfs_use_quota2) {
777 #ifdef LFS_QUOTA2
778 				error = lfs_quota2_mount(mp);
779 #else
780 				uprintf("%s: no kernel support for this "
781 					"filesystem's quotas\n",
782 					mp->mnt_stat.f_mntonname);
783 				if (mp->mnt_flag & MNT_FORCE) {
784 					uprintf("%s: mounting anyway; "
785 						"fsck afterwards\n",
786 						mp->mnt_stat.f_mntonname);
787 				} else {
788 					error = EINVAL;
789 				}
790 #endif
791 				if (error) {
792 					return error;
793 				}
794 			}
795 
796 			fs->lfs_ronly = 0;
797 			if (fs->lfs_pflags & LFS_PF_CLEAN) {
798 				fs->lfs_pflags &= ~LFS_PF_CLEAN;
799 				lfs_writesuper(fs, fs->lfs_sboffs[0]);
800 				lfs_writesuper(fs, fs->lfs_sboffs[1]);
801 			}
802 		}
803 		if (args->fspec == NULL)
804 			return EINVAL;
805 	}
806 
807 	error = set_statvfs_info(path, UIO_USERSPACE, args->fspec,
808 	    UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l);
809 	if (error == 0)
810 		(void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname,
811 			      sizeof(fs->lfs_fsmnt));
812 	return error;
813 
814 fail:
815 	vrele(devvp);
816 	return (error);
817 }
818 
819 
820 /*
821  * Common code for mount and mountroot
822  * LFS specific
823  */
824 int
825 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l)
826 {
827 	struct dlfs *tdfs, *dfs, *adfs;
828 	struct lfs *fs;
829 	struct ulfsmount *ump;
830 	struct vnode *vp;
831 	struct buf *bp, *abp;
832 	dev_t dev;
833 	int error, i, ronly, fsbsize;
834 	kauth_cred_t cred;
835 	CLEANERINFO *cip;
836 	SEGUSE *sup;
837 	daddr_t sb_addr;
838 
839 	cred = l ? l->l_cred : NOCRED;
840 
841 	/*
842 	 * Flush out any old buffers remaining from a previous use.
843 	 */
844 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
845 	error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0);
846 	VOP_UNLOCK(devvp);
847 	if (error)
848 		return (error);
849 
850 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
851 
852 	/* Don't free random space on error. */
853 	bp = NULL;
854 	abp = NULL;
855 	ump = NULL;
856 
857 	sb_addr = LFS_LABELPAD / DEV_BSIZE;
858 	while (1) {
859 		/* Read in the superblock. */
860 		error = bread(devvp, sb_addr, LFS_SBPAD, cred, 0, &bp);
861 		if (error)
862 			goto out;
863 		dfs = (struct dlfs *)bp->b_data;
864 
865 		/* Check the basics. */
866 		if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE ||
867 		    dfs->dlfs_version > LFS_VERSION ||
868 		    dfs->dlfs_bsize < sizeof(struct dlfs)) {
869 			DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n"));
870 			error = EINVAL;		/* XXX needs translation */
871 			goto out;
872 		}
873 		if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) {
874 			DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n",
875 			       dfs->dlfs_inodefmt));
876 			error = EINVAL;
877 			goto out;
878 		}
879 
880 		if (dfs->dlfs_version == 1)
881 			fsbsize = DEV_BSIZE;
882 		else {
883 			fsbsize = 1 << dfs->dlfs_ffshift;
884 			/*
885 			 * Could be, if the frag size is large enough, that we
886 			 * don't have the "real" primary superblock.  If that's
887 			 * the case, get the real one, and try again.
888 			 */
889 			if (sb_addr != (dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT))) {
890 				DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr"
891 				      " 0x%llx is not right, trying 0x%llx\n",
892 				      (long long)sb_addr,
893 				      (long long)(dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT))));
894 				sb_addr = dfs->dlfs_sboffs[0] << (dfs->dlfs_ffshift - DEV_BSHIFT);
895 				brelse(bp, 0);
896 				continue;
897 			}
898 		}
899 		break;
900 	}
901 
902 	/*
903 	 * Check the second superblock to see which is newer; then mount
904 	 * using the older of the two.	This is necessary to ensure that
905 	 * the filesystem is valid if it was not unmounted cleanly.
906 	 */
907 
908 	if (dfs->dlfs_sboffs[1] &&
909 	    dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
910 	{
911 		error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / DEV_BSIZE),
912 			LFS_SBPAD, cred, 0, &abp);
913 		if (error)
914 			goto out;
915 		adfs = (struct dlfs *)abp->b_data;
916 
917 		if (dfs->dlfs_version == 1) {
918 			/* 1s resolution comparison */
919 			if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
920 				tdfs = adfs;
921 			else
922 				tdfs = dfs;
923 		} else {
924 			/* monotonic infinite-resolution comparison */
925 			if (adfs->dlfs_serial < dfs->dlfs_serial)
926 				tdfs = adfs;
927 			else
928 				tdfs = dfs;
929 		}
930 
931 		/* Check the basics. */
932 		if (tdfs->dlfs_magic != LFS_MAGIC ||
933 		    tdfs->dlfs_bsize > MAXBSIZE ||
934 		    tdfs->dlfs_version > LFS_VERSION ||
935 		    tdfs->dlfs_bsize < sizeof(struct dlfs)) {
936 			DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock"
937 			      " sanity failed\n"));
938 			error = EINVAL;		/* XXX needs translation */
939 			goto out;
940 		}
941 	} else {
942 		DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock"
943 		      " daddr=0x%x\n", dfs->dlfs_sboffs[1]));
944 		error = EINVAL;
945 		goto out;
946 	}
947 
948 	/* Allocate the mount structure, copy the superblock into it. */
949 	fs = kmem_zalloc(sizeof(struct lfs), KM_SLEEP);
950 	memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs));
951 
952 	/* Compatibility */
953 	if (fs->lfs_version < 2) {
954 		fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
955 		fs->lfs_ibsize = fs->lfs_bsize;
956 		fs->lfs_start = fs->lfs_sboffs[0];
957 		fs->lfs_tstamp = fs->lfs_otstamp;
958 		fs->lfs_fsbtodb = 0;
959 	}
960 	if (fs->lfs_resvseg == 0)
961 		fs->lfs_resvseg = MIN(fs->lfs_minfreeseg - 1, \
962 			MAX(MIN_RESV_SEGS, fs->lfs_minfreeseg / 2 + 1));
963 
964 	/*
965 	 * If we aren't going to be able to write meaningfully to this
966 	 * filesystem, and were not mounted readonly, bomb out now.
967 	 */
968 	if (lfs_fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) {
969 		DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write,"
970 		      " we need BUFPAGES >= %lld\n",
971 		      (long long)((bufmem_hiwater / bufmem_lowater) *
972 				  LFS_INVERSE_MAX_BYTES(
973 					  lfs_fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT)));
974 		kmem_free(fs, sizeof(struct lfs));
975 		error = EFBIG; /* XXX needs translation */
976 		goto out;
977 	}
978 
979 	/* Before rolling forward, lock so vget will sleep for other procs */
980 	if (l != NULL) {
981 		fs->lfs_flags = LFS_NOTYET;
982 		fs->lfs_rfpid = l->l_proc->p_pid;
983 	}
984 
985 	ump = kmem_zalloc(sizeof(*ump), KM_SLEEP);
986 	ump->um_lfs = fs;
987 	ump->um_fstype = ULFS1;
988 	if (sizeof(struct lfs) < LFS_SBPAD) {			/* XXX why? */
989 		brelse(bp, BC_INVAL);
990 		brelse(abp, BC_INVAL);
991 	} else {
992 		brelse(bp, 0);
993 		brelse(abp, 0);
994 	}
995 	bp = NULL;
996 	abp = NULL;
997 
998 
999 	/* Set up the I/O information */
1000 	fs->lfs_devbsize = DEV_BSIZE;
1001 	fs->lfs_iocount = 0;
1002 	fs->lfs_diropwait = 0;
1003 	fs->lfs_activesb = 0;
1004 	fs->lfs_uinodes = 0;
1005 	fs->lfs_ravail = 0;
1006 	fs->lfs_favail = 0;
1007 	fs->lfs_sbactive = 0;
1008 
1009 	/* Set up the ifile and lock aflags */
1010 	fs->lfs_doifile = 0;
1011 	fs->lfs_writer = 0;
1012 	fs->lfs_dirops = 0;
1013 	fs->lfs_nadirop = 0;
1014 	fs->lfs_seglock = 0;
1015 	fs->lfs_pdflush = 0;
1016 	fs->lfs_sleepers = 0;
1017 	fs->lfs_pages = 0;
1018 	rw_init(&fs->lfs_fraglock);
1019 	rw_init(&fs->lfs_iflock);
1020 	cv_init(&fs->lfs_stopcv, "lfsstop");
1021 
1022 	/* Set the file system readonly/modify bits. */
1023 	fs->lfs_ronly = ronly;
1024 	if (ronly == 0)
1025 		fs->lfs_fmod = 1;
1026 
1027 	/* ulfs-level information */
1028 	fs->um_flags = 0;
1029 	fs->um_bptrtodb = fs->lfs_ffshift - DEV_BSHIFT;
1030 	fs->um_seqinc = fs->lfs_frag;
1031 	fs->um_nindir = fs->lfs_nindir;
1032 	fs->um_lognindir = ffs(fs->lfs_nindir) - 1;
1033 	fs->um_maxsymlinklen = fs->lfs_maxsymlinklen;
1034 	fs->um_dirblksiz = LFS_DIRBLKSIZ;
1035 	fs->um_maxfilesize = fs->lfs_maxfilesize;
1036 
1037 	/* quota stuff */
1038 	/* XXX: these need to come from the on-disk superblock to be used */
1039 	fs->lfs_use_quota2 = 0;
1040 	fs->lfs_quota_magic = 0;
1041 	fs->lfs_quota_flags = 0;
1042 	fs->lfs_quotaino[0] = 0;
1043 	fs->lfs_quotaino[1] = 0;
1044 
1045 	/* Initialize the mount structure. */
1046 	dev = devvp->v_rdev;
1047 	mp->mnt_data = ump;
1048 	mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
1049 	mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS);
1050 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
1051 	mp->mnt_stat.f_namemax = LFS_MAXNAMLEN;
1052 	mp->mnt_stat.f_iosize = fs->lfs_bsize;
1053 	mp->mnt_flag |= MNT_LOCAL;
1054 	mp->mnt_fs_bshift = fs->lfs_bshift;
1055 	if (fs->um_maxsymlinklen > 0)
1056 		mp->mnt_iflag |= IMNT_DTYPE;
1057 
1058 	ump->um_mountp = mp;
1059 	ump->um_dev = dev;
1060 	ump->um_devvp = devvp;
1061 	for (i = 0; i < ULFS_MAXQUOTAS; i++)
1062 		ump->um_quotas[i] = NULLVP;
1063 	spec_node_setmountedfs(devvp, mp);
1064 
1065 	/* Set up reserved memory for pageout */
1066 	lfs_setup_resblks(fs);
1067 	/* Set up vdirop tailq */
1068 	TAILQ_INIT(&fs->lfs_dchainhd);
1069 	/* and paging tailq */
1070 	TAILQ_INIT(&fs->lfs_pchainhd);
1071 	/* and delayed segment accounting for truncation list */
1072 	LIST_INIT(&fs->lfs_segdhd);
1073 
1074 	/*
1075 	 * We use the ifile vnode for almost every operation.  Instead of
1076 	 * retrieving it from the hash table each time we retrieve it here,
1077 	 * artificially increment the reference count and keep a pointer
1078 	 * to it in the incore copy of the superblock.
1079 	 */
1080 	if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
1081 		DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error));
1082 		goto out;
1083 	}
1084 	fs->lfs_ivnode = vp;
1085 	vref(vp);
1086 
1087 	/* Set up inode bitmap and order free list */
1088 	lfs_order_freelist(fs);
1089 
1090 	/* Set up segment usage flags for the autocleaner. */
1091 	fs->lfs_nactive = 0;
1092 	fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *),
1093 						M_SEGMENT, M_WAITOK);
1094 	fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1095 						 M_SEGMENT, M_WAITOK);
1096 	fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1097 						 M_SEGMENT, M_WAITOK);
1098 	memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t));
1099 	for (i = 0; i < fs->lfs_nseg; i++) {
1100 		int changed;
1101 
1102 		LFS_SEGENTRY(sup, fs, i, bp);
1103 		changed = 0;
1104 		if (!ronly) {
1105 			if (sup->su_nbytes == 0 &&
1106 			    !(sup->su_flags & SEGUSE_EMPTY)) {
1107 				sup->su_flags |= SEGUSE_EMPTY;
1108 				++changed;
1109 			} else if (!(sup->su_nbytes == 0) &&
1110 				   (sup->su_flags & SEGUSE_EMPTY)) {
1111 				sup->su_flags &= ~SEGUSE_EMPTY;
1112 				++changed;
1113 			}
1114 			if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) {
1115 				sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL);
1116 				++changed;
1117 			}
1118 		}
1119 		fs->lfs_suflags[0][i] = sup->su_flags;
1120 		if (changed)
1121 			LFS_WRITESEGENTRY(sup, fs, i, bp);
1122 		else
1123 			brelse(bp, 0);
1124 	}
1125 
1126 	/*
1127 	 * XXX: if the fs has quotas, quotas should be on even if
1128 	 * readonly. Otherwise you can't query the quota info!
1129 	 * However, that's not how the quota2 code got written and I
1130 	 * don't know if it'll behave itself if enabled while
1131 	 * readonly, so for now use the same enable logic as ffs.
1132 	 *
1133 	 * XXX: also, if you use the -f behavior allowed here (and
1134 	 * equivalently above for remount) it will corrupt the fs. It
1135 	 * ought not to allow that. It should allow mounting readonly
1136 	 * if there are quotas and the kernel doesn't have the quota
1137 	 * code, but only readonly.
1138 	 *
1139 	 * XXX: and if you use the -f behavior allowed here it will
1140 	 * likely crash at unmount time (or remount time) because we
1141 	 * think quotas are active.
1142 	 *
1143 	 * Although none of this applies until there's a way to set
1144 	 * lfs_use_quota2 and have quotas in the fs at all.
1145 	 */
1146 	if (!ronly && fs->lfs_use_quota2) {
1147 #ifdef LFS_QUOTA2
1148 		error = lfs_quota2_mount(mp);
1149 #else
1150 		uprintf("%s: no kernel support for this filesystem's quotas\n",
1151 			mp->mnt_stat.f_mntonname);
1152 		if (mp->mnt_flag & MNT_FORCE) {
1153 			uprintf("%s: mounting anyway; fsck afterwards\n",
1154 				mp->mnt_stat.f_mntonname);
1155 		} else {
1156 			error = EINVAL;
1157 		}
1158 #endif
1159 		if (error) {
1160 			/* XXX XXX must clean up the stuff immediately above */
1161 			printf("lfs_mountfs: sorry, leaking some memory\n");
1162 			goto out;
1163 		}
1164 	}
1165 
1166 #ifdef LFS_EXTATTR
1167 	/*
1168 	 * Initialize file-backed extended attributes for ULFS1 file
1169 	 * systems.
1170 	 *
1171 	 * XXX: why is this limited to ULFS1?
1172 	 */
1173 	if (ump->um_fstype == ULFS1) {
1174 		ulfs_extattr_uepm_init(&ump->um_extattr);
1175 	}
1176 #endif
1177 
1178 #ifdef LFS_KERNEL_RFW
1179 	lfs_roll_forward(fs, mp, l);
1180 #endif
1181 
1182 	/* If writing, sb is not clean; record in case of immediate crash */
1183 	if (!fs->lfs_ronly) {
1184 		fs->lfs_pflags &= ~LFS_PF_CLEAN;
1185 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
1186 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
1187 	}
1188 
1189 	/* Allow vget now that roll-forward is complete */
1190 	fs->lfs_flags &= ~(LFS_NOTYET);
1191 	wakeup(&fs->lfs_flags);
1192 
1193 	/*
1194 	 * Initialize the ifile cleaner info with information from
1195 	 * the superblock.
1196 	 */
1197 	LFS_CLEANERINFO(cip, fs, bp);
1198 	cip->clean = fs->lfs_nclean;
1199 	cip->dirty = fs->lfs_nseg - fs->lfs_nclean;
1200 	cip->avail = fs->lfs_avail;
1201 	cip->bfree = fs->lfs_bfree;
1202 	(void) LFS_BWRITE_LOG(bp); /* Ifile */
1203 
1204 	/*
1205 	 * Mark the current segment as ACTIVE, since we're going to
1206 	 * be writing to it.
1207 	 */
1208 	LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, fs->lfs_offset), bp);
1209 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1210 	fs->lfs_nactive++;
1211 	LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, fs->lfs_offset), bp);  /* Ifile */
1212 
1213 	/* Now that roll-forward is done, unlock the Ifile */
1214 	vput(vp);
1215 
1216 	/* Start the pagedaemon-anticipating daemon */
1217 	mutex_enter(&lfs_lock);
1218 	if (lfs_writer_daemon == 0 && lfs_writer_lid == 0 &&
1219 	    kthread_create(PRI_BIO, 0, NULL,
1220 	    lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1221 		panic("fork lfs_writer");
1222 	mutex_exit(&lfs_lock);
1223 
1224 	printf("WARNING: the log-structured file system is experimental\n"
1225 	    "WARNING: it may cause system crashes and/or corrupt data\n");
1226 
1227 	return (0);
1228 
1229 out:
1230 	if (bp)
1231 		brelse(bp, 0);
1232 	if (abp)
1233 		brelse(abp, 0);
1234 	if (ump) {
1235 		kmem_free(ump->um_lfs, sizeof(struct lfs));
1236 		kmem_free(ump, sizeof(*ump));
1237 		mp->mnt_data = NULL;
1238 	}
1239 
1240 	return (error);
1241 }
1242 
1243 /*
1244  * unmount system call
1245  */
1246 int
1247 lfs_unmount(struct mount *mp, int mntflags)
1248 {
1249 	struct lwp *l = curlwp;
1250 	struct ulfsmount *ump;
1251 	struct lfs *fs;
1252 	int error, flags, ronly;
1253 	vnode_t *vp;
1254 
1255 	flags = 0;
1256 	if (mntflags & MNT_FORCE)
1257 		flags |= FORCECLOSE;
1258 
1259 	ump = VFSTOULFS(mp);
1260 	fs = ump->um_lfs;
1261 
1262 	/* Two checkpoints */
1263 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1264 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1265 
1266 	/* wake up the cleaner so it can die */
1267 	/* XXX: shouldn't this be *after* the error cases below? */
1268 	lfs_wakeup_cleaner(fs);
1269 	mutex_enter(&lfs_lock);
1270 	while (fs->lfs_sleepers)
1271 		mtsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1272 			&lfs_lock);
1273 	mutex_exit(&lfs_lock);
1274 
1275 #ifdef LFS_EXTATTR
1276 	if (ump->um_fstype == ULFS1) {
1277 		if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) {
1278 			ulfs_extattr_stop(mp, curlwp);
1279 		}
1280 		if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED) {
1281 			ulfs_extattr_uepm_destroy(&ump->um_extattr);
1282 		}
1283 	}
1284 #endif
1285 #ifdef LFS_QUOTA
1286         if ((error = lfsquota1_umount(mp, flags)) != 0)
1287 		return (error);
1288 #endif
1289 #ifdef LFS_QUOTA2
1290         if ((error = lfsquota2_umount(mp, flags)) != 0)
1291 		return (error);
1292 #endif
1293 	if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1294 		return (error);
1295 	if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0)
1296 		return (error);
1297 	vp = fs->lfs_ivnode;
1298 	mutex_enter(vp->v_interlock);
1299 	if (LIST_FIRST(&vp->v_dirtyblkhd))
1300 		panic("lfs_unmount: still dirty blocks on ifile vnode");
1301 	mutex_exit(vp->v_interlock);
1302 
1303 	/* Explicitly write the superblock, to update serial and pflags */
1304 	fs->lfs_pflags |= LFS_PF_CLEAN;
1305 	lfs_writesuper(fs, fs->lfs_sboffs[0]);
1306 	lfs_writesuper(fs, fs->lfs_sboffs[1]);
1307 	mutex_enter(&lfs_lock);
1308 	while (fs->lfs_iocount)
1309 		mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0,
1310 			&lfs_lock);
1311 	mutex_exit(&lfs_lock);
1312 
1313 	/* Finish with the Ifile, now that we're done with it */
1314 	vgone(fs->lfs_ivnode);
1315 
1316 	ronly = !fs->lfs_ronly;
1317 	if (ump->um_devvp->v_type != VBAD)
1318 		spec_node_setmountedfs(ump->um_devvp, NULL);
1319 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1320 	error = VOP_CLOSE(ump->um_devvp,
1321 	    ronly ? FREAD : FREAD|FWRITE, NOCRED);
1322 	vput(ump->um_devvp);
1323 
1324 	/* Complain about page leakage */
1325 	if (fs->lfs_pages > 0)
1326 		printf("lfs_unmount: still claim %d pages (%d in subsystem)\n",
1327 			fs->lfs_pages, lfs_subsys_pages);
1328 
1329 	/* Free per-mount data structures */
1330 	free(fs->lfs_ino_bitmap, M_SEGMENT);
1331 	free(fs->lfs_suflags[0], M_SEGMENT);
1332 	free(fs->lfs_suflags[1], M_SEGMENT);
1333 	free(fs->lfs_suflags, M_SEGMENT);
1334 	lfs_free_resblks(fs);
1335 	cv_destroy(&fs->lfs_stopcv);
1336 	rw_destroy(&fs->lfs_fraglock);
1337 	rw_destroy(&fs->lfs_iflock);
1338 
1339 	kmem_free(fs, sizeof(struct lfs));
1340 	kmem_free(ump, sizeof(*ump));
1341 
1342 	mp->mnt_data = NULL;
1343 	mp->mnt_flag &= ~MNT_LOCAL;
1344 	return (error);
1345 }
1346 
1347 /*
1348  * Get file system statistics.
1349  *
1350  * NB: We don't lock to access the superblock here, because it's not
1351  * really that important if we get it wrong.
1352  */
1353 int
1354 lfs_statvfs(struct mount *mp, struct statvfs *sbp)
1355 {
1356 	struct lfs *fs;
1357 	struct ulfsmount *ump;
1358 
1359 	ump = VFSTOULFS(mp);
1360 	fs = ump->um_lfs;
1361 	if (fs->lfs_magic != LFS_MAGIC)
1362 		panic("lfs_statvfs: magic");
1363 
1364 	sbp->f_bsize = fs->lfs_bsize;
1365 	sbp->f_frsize = fs->lfs_fsize;
1366 	sbp->f_iosize = fs->lfs_bsize;
1367 	sbp->f_blocks = LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks;
1368 
1369 	sbp->f_bfree = LFS_EST_BFREE(fs);
1370 	KASSERT(sbp->f_bfree <= fs->lfs_dsize);
1371 #if 0
1372 	if (sbp->f_bfree < 0)
1373 		sbp->f_bfree = 0;
1374 #endif
1375 
1376 	sbp->f_bresvd = LFS_EST_RSVD(fs);
1377 	if (sbp->f_bfree > sbp->f_bresvd)
1378 		sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
1379 	else
1380 		sbp->f_bavail = 0;
1381 
1382 	sbp->f_files = fs->lfs_bfree / lfs_btofsb(fs, fs->lfs_ibsize)
1383 	    * LFS_INOPB(fs);
1384 	sbp->f_ffree = sbp->f_files - fs->lfs_nfiles;
1385 	sbp->f_favail = sbp->f_ffree;
1386 	sbp->f_fresvd = 0;
1387 	copy_statvfs_info(sbp, mp);
1388 	return (0);
1389 }
1390 
1391 /*
1392  * Go through the disk queues to initiate sandbagged IO;
1393  * go through the inodes to write those that have been modified;
1394  * initiate the writing of the super block if it has been modified.
1395  *
1396  * Note: we are always called with the filesystem marked `MPBUSY'.
1397  */
1398 int
1399 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
1400 {
1401 	int error;
1402 	struct lfs *fs;
1403 
1404 	fs = VFSTOULFS(mp)->um_lfs;
1405 	if (fs->lfs_ronly)
1406 		return 0;
1407 
1408 	/* Snapshots should not hose the syncer */
1409 	/*
1410 	 * XXX Sync can block here anyway, since we don't have a very
1411 	 * XXX good idea of how much data is pending.  If it's more
1412 	 * XXX than a segment and lfs_nextseg is close to the end of
1413 	 * XXX the log, we'll likely block.
1414 	 */
1415 	mutex_enter(&lfs_lock);
1416 	if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) {
1417 		mutex_exit(&lfs_lock);
1418 		return 0;
1419 	}
1420 	mutex_exit(&lfs_lock);
1421 
1422 	lfs_writer_enter(fs, "lfs_dirops");
1423 
1424 	/* All syncs must be checkpoints until roll-forward is implemented. */
1425 	DLOG((DLOG_FLUSH, "lfs_sync at 0x%x\n", fs->lfs_offset));
1426 	error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1427 	lfs_writer_leave(fs);
1428 #ifdef LFS_QUOTA
1429 	lfs_qsync(mp);
1430 #endif
1431 	return (error);
1432 }
1433 
1434 /*
1435  * Look up an LFS dinode number to find its incore vnode.  If not already
1436  * in core, read it in from the specified device.  Return the inode locked.
1437  * Detection and handling of mount points must be done by the calling routine.
1438  */
1439 int
1440 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1441 {
1442 	struct lfs *fs;
1443 	struct ulfs1_dinode *dip;
1444 	struct inode *ip;
1445 	struct buf *bp;
1446 	struct ifile *ifp;
1447 	struct vnode *vp;
1448 	struct ulfsmount *ump;
1449 	daddr_t daddr;
1450 	dev_t dev;
1451 	int error, retries;
1452 	struct timespec ts;
1453 
1454 	memset(&ts, 0, sizeof ts);	/* XXX gcc */
1455 
1456 	ump = VFSTOULFS(mp);
1457 	dev = ump->um_dev;
1458 	fs = ump->um_lfs;
1459 
1460 	/*
1461 	 * If the filesystem is not completely mounted yet, suspend
1462 	 * any access requests (wait for roll-forward to complete).
1463 	 */
1464 	mutex_enter(&lfs_lock);
1465 	while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1466 		mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0,
1467 			&lfs_lock);
1468 	mutex_exit(&lfs_lock);
1469 
1470 retry:
1471 	if ((*vpp = ulfs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
1472 		return (0);
1473 
1474 	error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, NULL, &vp);
1475 	if (error) {
1476 		*vpp = NULL;
1477 		 return (error);
1478 	}
1479 
1480 	mutex_enter(&ulfs_hashlock);
1481 	if (ulfs_ihashget(dev, ino, 0) != NULL) {
1482 		mutex_exit(&ulfs_hashlock);
1483 		ungetnewvnode(vp);
1484 		goto retry;
1485 	}
1486 
1487 	/* Translate the inode number to a disk address. */
1488 	if (ino == LFS_IFILE_INUM)
1489 		daddr = fs->lfs_idaddr;
1490 	else {
1491 		/* XXX bounds-check this too */
1492 		LFS_IENTRY(ifp, fs, ino, bp);
1493 		daddr = ifp->if_daddr;
1494 		if (fs->lfs_version > 1) {
1495 			ts.tv_sec = ifp->if_atime_sec;
1496 			ts.tv_nsec = ifp->if_atime_nsec;
1497 		}
1498 
1499 		brelse(bp, 0);
1500 		if (daddr == LFS_UNUSED_DADDR) {
1501 			*vpp = NULLVP;
1502 			mutex_exit(&ulfs_hashlock);
1503 			ungetnewvnode(vp);
1504 			return (ENOENT);
1505 		}
1506 	}
1507 
1508 	/* Allocate/init new vnode/inode. */
1509 	lfs_vcreate(mp, ino, vp);
1510 
1511 	/*
1512 	 * Put it onto its hash chain and lock it so that other requests for
1513 	 * this inode will block if they arrive while we are sleeping waiting
1514 	 * for old data structures to be purged or for the contents of the
1515 	 * disk portion of this inode to be read.
1516 	 */
1517 	ip = VTOI(vp);
1518 	ulfs_ihashins(ip);
1519 	mutex_exit(&ulfs_hashlock);
1520 
1521 	/*
1522 	 * XXX
1523 	 * This may not need to be here, logically it should go down with
1524 	 * the i_devvp initialization.
1525 	 * Ask Kirk.
1526 	 */
1527 	ip->i_lfs = ump->um_lfs;
1528 
1529 	/* Read in the disk contents for the inode, copy into the inode. */
1530 	retries = 0;
1531     again:
1532 	error = bread(ump->um_devvp, LFS_FSBTODB(fs, daddr),
1533 		(fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize),
1534 		NOCRED, 0, &bp);
1535 	if (error) {
1536 		/*
1537 		 * The inode does not contain anything useful, so it would
1538 		 * be misleading to leave it on its hash chain. With mode
1539 		 * still zero, it will be unlinked and returned to the free
1540 		 * list by vput().
1541 		 */
1542 		vput(vp);
1543 		*vpp = NULL;
1544 		return (error);
1545 	}
1546 
1547 	dip = lfs_ifind(fs, ino, bp);
1548 	if (dip == NULL) {
1549 		/* Assume write has not completed yet; try again */
1550 		brelse(bp, BC_INVAL);
1551 		++retries;
1552 		if (retries > LFS_IFIND_RETRIES) {
1553 #ifdef DEBUG
1554 			/* If the seglock is held look at the bpp to see
1555 			   what is there anyway */
1556 			mutex_enter(&lfs_lock);
1557 			if (fs->lfs_seglock > 0) {
1558 				struct buf **bpp;
1559 				struct ulfs1_dinode *dp;
1560 				int i;
1561 
1562 				for (bpp = fs->lfs_sp->bpp;
1563 				     bpp != fs->lfs_sp->cbpp; ++bpp) {
1564 					if ((*bpp)->b_vp == fs->lfs_ivnode &&
1565 					    bpp != fs->lfs_sp->bpp) {
1566 						/* Inode block */
1567 						printf("lfs_vget: block 0x%" PRIx64 ": ",
1568 						       (*bpp)->b_blkno);
1569 						dp = (struct ulfs1_dinode *)(*bpp)->b_data;
1570 						for (i = 0; i < LFS_INOPB(fs); i++)
1571 							if (dp[i].di_inumber)
1572 								printf("%d ", dp[i].di_inumber);
1573 						printf("\n");
1574 					}
1575 				}
1576 			}
1577 			mutex_exit(&lfs_lock);
1578 #endif /* DEBUG */
1579 			panic("lfs_vget: dinode not found");
1580 		}
1581 		mutex_enter(&lfs_lock);
1582 		if (fs->lfs_iocount) {
1583 			DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino));
1584 			(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
1585 				      "lfs ifind", 1, &lfs_lock);
1586 		} else
1587 			retries = LFS_IFIND_RETRIES;
1588 		mutex_exit(&lfs_lock);
1589 		goto again;
1590 	}
1591 	*ip->i_din.ffs1_din = *dip;
1592 	brelse(bp, 0);
1593 
1594 	if (fs->lfs_version > 1) {
1595 		ip->i_ffs1_atime = ts.tv_sec;
1596 		ip->i_ffs1_atimensec = ts.tv_nsec;
1597 	}
1598 
1599 	lfs_vinit(mp, &vp);
1600 
1601 	*vpp = vp;
1602 
1603 	KASSERT(VOP_ISLOCKED(vp));
1604 
1605 	return (0);
1606 }
1607 
1608 /*
1609  * File handle to vnode
1610  */
1611 int
1612 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1613 {
1614 	struct lfid lfh;
1615 	struct buf *bp;
1616 	IFILE *ifp;
1617 	int32_t daddr;
1618 	struct lfs *fs;
1619 	vnode_t *vp;
1620 
1621 	if (fhp->fid_len != sizeof(struct lfid))
1622 		return EINVAL;
1623 
1624 	memcpy(&lfh, fhp, sizeof(lfh));
1625 	if (lfh.lfid_ino < LFS_IFILE_INUM)
1626 		return ESTALE;
1627 
1628 	fs = VFSTOULFS(mp)->um_lfs;
1629 	if (lfh.lfid_ident != fs->lfs_ident)
1630 		return ESTALE;
1631 
1632 	if (lfh.lfid_ino >
1633 	    ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) -
1634 	     fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb)
1635 		return ESTALE;
1636 
1637 	mutex_enter(&ulfs_ihash_lock);
1638 	vp = ulfs_ihashlookup(VFSTOULFS(mp)->um_dev, lfh.lfid_ino);
1639 	mutex_exit(&ulfs_ihash_lock);
1640 	if (vp == NULL) {
1641 		LFS_IENTRY(ifp, fs, lfh.lfid_ino, bp);
1642 		daddr = ifp->if_daddr;
1643 		brelse(bp, 0);
1644 		if (daddr == LFS_UNUSED_DADDR)
1645 			return ESTALE;
1646 	}
1647 
1648 	return (ulfs_fhtovp(mp, &lfh.lfid_ufid, vpp));
1649 }
1650 
1651 /*
1652  * Vnode pointer to File handle
1653  */
1654 /* ARGSUSED */
1655 int
1656 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size)
1657 {
1658 	struct inode *ip;
1659 	struct lfid lfh;
1660 
1661 	if (*fh_size < sizeof(struct lfid)) {
1662 		*fh_size = sizeof(struct lfid);
1663 		return E2BIG;
1664 	}
1665 	*fh_size = sizeof(struct lfid);
1666 	ip = VTOI(vp);
1667 	memset(&lfh, 0, sizeof(lfh));
1668 	lfh.lfid_len = sizeof(struct lfid);
1669 	lfh.lfid_ino = ip->i_number;
1670 	lfh.lfid_gen = ip->i_gen;
1671 	lfh.lfid_ident = ip->i_lfs->lfs_ident;
1672 	memcpy(fhp, &lfh, sizeof(lfh));
1673 	return (0);
1674 }
1675 
1676 /*
1677  * ulfs_bmaparray callback function for writing.
1678  *
1679  * Since blocks will be written to the new segment anyway,
1680  * we don't care about current daddr of them.
1681  */
1682 static bool
1683 lfs_issequential_hole(const struct lfs *fs,
1684     daddr_t daddr0, daddr_t daddr1)
1685 {
1686 	(void)fs; /* not used */
1687 
1688 	daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */
1689 	daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */
1690 
1691 	KASSERT(daddr0 == UNWRITTEN ||
1692 	    (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR));
1693 	KASSERT(daddr1 == UNWRITTEN ||
1694 	    (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR));
1695 
1696 	/* NOTE: all we want to know here is 'hole or not'. */
1697 	/* NOTE: UNASSIGNED is converted to 0 by ulfs_bmaparray. */
1698 
1699 	/*
1700 	 * treat UNWRITTENs and all resident blocks as 'contiguous'
1701 	 */
1702 	if (daddr0 != 0 && daddr1 != 0)
1703 		return true;
1704 
1705 	/*
1706 	 * both are in hole?
1707 	 */
1708 	if (daddr0 == 0 && daddr1 == 0)
1709 		return true; /* all holes are 'contiguous' for us. */
1710 
1711 	return false;
1712 }
1713 
1714 /*
1715  * lfs_gop_write functions exactly like genfs_gop_write, except that
1716  * (1) it requires the seglock to be held by its caller, and sp->fip
1717  *     to be properly initialized (it will return without re-initializing
1718  *     sp->fip, and without calling lfs_writeseg).
1719  * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1720  *     to determine how large a block it can write at once (though it does
1721  *     still use VOP_BMAP to find holes in the file);
1722  * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1723  *     (leaving lfs_writeseg to deal with the cluster blocks, so we might
1724  *     now have clusters of clusters, ick.)
1725  */
1726 static int
1727 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages,
1728     int flags)
1729 {
1730 	int i, error, run, haveeof = 0;
1731 	int fs_bshift;
1732 	vaddr_t kva;
1733 	off_t eof, offset, startoffset = 0;
1734 	size_t bytes, iobytes, skipbytes;
1735 	bool async = (flags & PGO_SYNCIO) == 0;
1736 	daddr_t lbn, blkno;
1737 	struct vm_page *pg;
1738 	struct buf *mbp, *bp;
1739 	struct vnode *devvp = VTOI(vp)->i_devvp;
1740 	struct inode *ip = VTOI(vp);
1741 	struct lfs *fs = ip->i_lfs;
1742 	struct segment *sp = fs->lfs_sp;
1743 	UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1744 	const char * failreason = NULL;
1745 
1746 	ASSERT_SEGLOCK(fs);
1747 
1748 	/* The Ifile lives in the buffer cache */
1749 	KASSERT(vp != fs->lfs_ivnode);
1750 
1751 	/*
1752 	 * We don't want to fill the disk before the cleaner has a chance
1753 	 * to make room for us.  If we're in danger of doing that, fail
1754 	 * with EAGAIN.  The caller will have to notice this, unlock
1755 	 * so the cleaner can run, relock and try again.
1756 	 *
1757 	 * We must write everything, however, if our vnode is being
1758 	 * reclaimed.
1759 	 */
1760 	mutex_enter(vp->v_interlock);
1761 	if (LFS_STARVED_FOR_SEGS(fs) && vdead_check(vp, VDEAD_NOWAIT) == 0) {
1762 		mutex_exit(vp->v_interlock);
1763 		failreason = "Starved for segs and not flushing vp";
1764  		goto tryagain;
1765 	}
1766 	mutex_exit(vp->v_interlock);
1767 
1768 	/*
1769 	 * Sometimes things slip past the filters in lfs_putpages,
1770 	 * and the pagedaemon tries to write pages---problem is
1771 	 * that the pagedaemon never acquires the segment lock.
1772 	 *
1773 	 * Alternatively, pages that were clean when we called
1774 	 * genfs_putpages may have become dirty in the meantime.  In this
1775 	 * case the segment header is not properly set up for blocks
1776 	 * to be added to it.
1777 	 *
1778 	 * Unbusy and unclean the pages, and put them on the ACTIVE
1779 	 * queue under the hypothesis that they couldn't have got here
1780 	 * unless they were modified *quite* recently.
1781 	 *
1782 	 * XXXUBC that last statement is an oversimplification of course.
1783 	 */
1784 	if (!LFS_SEGLOCK_HELD(fs)) {
1785 		failreason = "Seglock not held";
1786 		goto tryagain;
1787 	}
1788 	if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) {
1789 		failreason = "Inode with no_gop_write";
1790 		goto tryagain;
1791 	}
1792 	if ((pgs[0]->offset & fs->lfs_bmask) != 0) {
1793 		failreason = "Bad page offset";
1794 		goto tryagain;
1795 	}
1796 
1797 	UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1798 	    vp, pgs, npages, flags);
1799 
1800 	GOP_SIZE(vp, vp->v_size, &eof, 0);
1801 	haveeof = 1;
1802 
1803 	if (vp->v_type == VREG)
1804 		fs_bshift = vp->v_mount->mnt_fs_bshift;
1805 	else
1806 		fs_bshift = DEV_BSHIFT;
1807 	error = 0;
1808 	pg = pgs[0];
1809 	startoffset = pg->offset;
1810 	KASSERT(eof >= 0);
1811 
1812 	if (startoffset >= eof) {
1813 		failreason = "Offset beyond EOF";
1814 		goto tryagain;
1815 	} else
1816 		bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1817 	skipbytes = 0;
1818 
1819 	KASSERT(bytes != 0);
1820 
1821 	/* Swap PG_DELWRI for PG_PAGEOUT */
1822 	for (i = 0; i < npages; i++) {
1823 		if (pgs[i]->flags & PG_DELWRI) {
1824 			KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
1825 			pgs[i]->flags &= ~PG_DELWRI;
1826 			pgs[i]->flags |= PG_PAGEOUT;
1827 			uvm_pageout_start(1);
1828 			mutex_enter(vp->v_interlock);
1829 			mutex_enter(&uvm_pageqlock);
1830 			uvm_pageunwire(pgs[i]);
1831 			mutex_exit(&uvm_pageqlock);
1832 			mutex_exit(vp->v_interlock);
1833 		}
1834 	}
1835 
1836 	/*
1837 	 * Check to make sure we're starting on a block boundary.
1838 	 * We'll check later to make sure we always write entire
1839 	 * blocks (or fragments).
1840 	 */
1841 	if (startoffset & fs->lfs_bmask)
1842 		printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n",
1843 		       startoffset, fs->lfs_bmask,
1844 		       startoffset & fs->lfs_bmask);
1845 	KASSERT((startoffset & fs->lfs_bmask) == 0);
1846 	if (bytes & fs->lfs_ffmask) {
1847 		printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
1848 		panic("lfs_gop_write: non-integer blocks");
1849 	}
1850 
1851 	/*
1852 	 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK.
1853 	 * If we would, write what we have and try again.  If we don't
1854 	 * have anything to write, we'll have to sleep.
1855 	 */
1856 	if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
1857 				      (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ?
1858 				       UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) {
1859 		DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n"));
1860 #if 0
1861 		      " with nfinfo=%d at offset 0x%x\n",
1862 		      (int)((SEGSUM *)(sp->segsum))->ss_nfinfo,
1863 		      (unsigned)fs->lfs_offset));
1864 #endif
1865 		lfs_updatemeta(sp);
1866 		lfs_release_finfo(fs);
1867 		(void) lfs_writeseg(fs, sp);
1868 
1869 		lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
1870 
1871 		/*
1872 		 * Having given up all of the pager_map we were holding,
1873 		 * we can now wait for aiodoned to reclaim it for us
1874 		 * without fear of deadlock.
1875 		 */
1876 		kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
1877 				     UVMPAGER_MAPIN_WAITOK);
1878 	}
1879 
1880 	mbp = getiobuf(NULL, true);
1881 	UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
1882 	    vp, mbp, vp->v_numoutput, bytes);
1883 	mbp->b_bufsize = npages << PAGE_SHIFT;
1884 	mbp->b_data = (void *)kva;
1885 	mbp->b_resid = mbp->b_bcount = bytes;
1886 	mbp->b_cflags = BC_BUSY|BC_AGE;
1887 	mbp->b_iodone = uvm_aio_biodone;
1888 
1889 	bp = NULL;
1890 	for (offset = startoffset;
1891 	    bytes > 0;
1892 	    offset += iobytes, bytes -= iobytes) {
1893 		lbn = offset >> fs_bshift;
1894 		error = ulfs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
1895 		    lfs_issequential_hole);
1896 		if (error) {
1897 			UVMHIST_LOG(ubchist, "ulfs_bmaparray() -> %d",
1898 			    error,0,0,0);
1899 			skipbytes += bytes;
1900 			bytes = 0;
1901 			break;
1902 		}
1903 
1904 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
1905 		    bytes);
1906 		if (blkno == (daddr_t)-1) {
1907 			skipbytes += iobytes;
1908 			continue;
1909 		}
1910 
1911 		/*
1912 		 * Discover how much we can really pack into this buffer.
1913 		 */
1914 		/* If no room in the current segment, finish it up */
1915 		if (sp->sum_bytes_left < sizeof(int32_t) ||
1916 		    sp->seg_bytes_left < (1 << fs->lfs_bshift)) {
1917 			int vers;
1918 
1919 			lfs_updatemeta(sp);
1920 			vers = sp->fip->fi_version;
1921 			lfs_release_finfo(fs);
1922 			(void) lfs_writeseg(fs, sp);
1923 
1924 			lfs_acquire_finfo(fs, ip->i_number, vers);
1925 		}
1926 		/* Check both for space in segment and space in segsum */
1927 		iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
1928 					<< fs_bshift);
1929 		iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
1930 				       << fs_bshift);
1931 		KASSERT(iobytes > 0);
1932 
1933 		/* if it's really one i/o, don't make a second buf */
1934 		if (offset == startoffset && iobytes == bytes) {
1935 			bp = mbp;
1936 			/*
1937 			 * All the LFS output is done by the segwriter.  It
1938 			 * will increment numoutput by one for all the bufs it
1939 			 * recieves.  However this buffer needs one extra to
1940 			 * account for aiodone.
1941 			 */
1942 			mutex_enter(vp->v_interlock);
1943 			vp->v_numoutput++;
1944 			mutex_exit(vp->v_interlock);
1945 		} else {
1946 			bp = getiobuf(NULL, true);
1947 			UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
1948 			    vp, bp, vp->v_numoutput, 0);
1949 			nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes);
1950 			/*
1951 			 * LFS doesn't like async I/O here, dies with
1952 			 * an assert in lfs_bwrite().  Is that assert
1953 			 * valid?  I retained non-async behaviour when
1954 			 * converted this to use nestiobuf --pooka
1955 			 */
1956 			bp->b_flags &= ~B_ASYNC;
1957 		}
1958 
1959 		/* XXX This is silly ... is this necessary? */
1960 		mutex_enter(&bufcache_lock);
1961 		mutex_enter(vp->v_interlock);
1962 		bgetvp(vp, bp);
1963 		mutex_exit(vp->v_interlock);
1964 		mutex_exit(&bufcache_lock);
1965 
1966 		bp->b_lblkno = lfs_lblkno(fs, offset);
1967 		bp->b_private = mbp;
1968 		if (devvp->v_type == VBLK) {
1969 			bp->b_dev = devvp->v_rdev;
1970 		}
1971 		VOP_BWRITE(bp->b_vp, bp);
1972 		while (lfs_gatherblock(sp, bp, NULL))
1973 			continue;
1974 	}
1975 
1976 	nestiobuf_done(mbp, skipbytes, error);
1977 	if (skipbytes) {
1978 		UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
1979 	}
1980 	UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
1981 
1982 	if (!async) {
1983 		/* Start a segment write. */
1984 		UVMHIST_LOG(ubchist, "flushing", 0,0,0,0);
1985 		mutex_enter(&lfs_lock);
1986 		lfs_flush(fs, 0, 1);
1987 		mutex_exit(&lfs_lock);
1988 	}
1989 
1990 	if ((sp->seg_flags & SEGM_SINGLE) && fs->lfs_curseg != fs->lfs_startseg)
1991 		return EAGAIN;
1992 
1993 	return (0);
1994 
1995     tryagain:
1996 	/*
1997 	 * We can't write the pages, for whatever reason.
1998 	 * Clean up after ourselves, and make the caller try again.
1999 	 */
2000 	mutex_enter(vp->v_interlock);
2001 
2002 	/* Tell why we're here, if we know */
2003 	if (failreason != NULL) {
2004 		DLOG((DLOG_PAGE, "lfs_gop_write: %s\n", failreason));
2005 	}
2006 	if (haveeof && startoffset >= eof) {
2007  		DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64
2008  		      " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number,
2009  		      pgs[0]->offset, eof, npages));
2010 	}
2011 
2012 	mutex_enter(&uvm_pageqlock);
2013 	for (i = 0; i < npages; i++) {
2014 		pg = pgs[i];
2015 
2016 		if (pg->flags & PG_PAGEOUT)
2017 			uvm_pageout_done(1);
2018 		if (pg->flags & PG_DELWRI) {
2019 			uvm_pageunwire(pg);
2020 		}
2021 		uvm_pageactivate(pg);
2022 		pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
2023 		DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg,
2024 			vp, pg->offset));
2025 		DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags));
2026 		DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags));
2027 		DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon));
2028 		DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject));
2029 		DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i,
2030 		      pg->wire_count));
2031 		DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i,
2032 		      pg->loan_count));
2033 	}
2034 	/* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
2035 	uvm_page_unbusy(pgs, npages);
2036 	mutex_exit(&uvm_pageqlock);
2037 	mutex_exit(vp->v_interlock);
2038 	return EAGAIN;
2039 }
2040 
2041 /*
2042  * finish vnode/inode initialization.
2043  * used by lfs_vget and lfs_fastvget.
2044  */
2045 void
2046 lfs_vinit(struct mount *mp, struct vnode **vpp)
2047 {
2048 	struct vnode *vp = *vpp;
2049 	struct inode *ip = VTOI(vp);
2050 	struct ulfsmount *ump = VFSTOULFS(mp);
2051 	struct lfs *fs = ump->um_lfs;
2052 	int i;
2053 
2054 	ip->i_mode = ip->i_ffs1_mode;
2055 	ip->i_nlink = ip->i_ffs1_nlink;
2056 	ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size;
2057 	ip->i_flags = ip->i_ffs1_flags;
2058 	ip->i_gen = ip->i_ffs1_gen;
2059 	ip->i_uid = ip->i_ffs1_uid;
2060 	ip->i_gid = ip->i_ffs1_gid;
2061 
2062 	ip->i_lfs_effnblks = ip->i_ffs1_blocks;
2063 	ip->i_lfs_odnlink = ip->i_ffs1_nlink;
2064 
2065 	/*
2066 	 * Initialize the vnode from the inode, check for aliases.  In all
2067 	 * cases re-init ip, the underlying vnode/inode may have changed.
2068 	 */
2069 	ulfs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
2070 	ip = VTOI(vp);
2071 
2072 	memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize));
2073 	if (vp->v_type != VLNK || ip->i_size >= ip->i_lfs->um_maxsymlinklen) {
2074 #ifdef DEBUG
2075 		for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift;
2076 		    i < ULFS_NDADDR; i++) {
2077 			if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
2078 			    i == 0)
2079 				continue;
2080 			if (ip->i_ffs1_db[i] != 0) {
2081 				lfs_dump_dinode(ip->i_din.ffs1_din);
2082 				panic("inconsistent inode (direct)");
2083 			}
2084 		}
2085 		for ( ; i < ULFS_NDADDR + ULFS_NIADDR; i++) {
2086 			if (ip->i_ffs1_ib[i - ULFS_NDADDR] != 0) {
2087 				lfs_dump_dinode(ip->i_din.ffs1_din);
2088 				panic("inconsistent inode (indirect)");
2089 			}
2090 		}
2091 #endif /* DEBUG */
2092 		for (i = 0; i < ULFS_NDADDR; i++)
2093 			if (ip->i_ffs1_db[i] != 0)
2094 				ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i);
2095 	}
2096 
2097 #ifdef DIAGNOSTIC
2098 	if (vp->v_type == VNON) {
2099 # ifdef DEBUG
2100 		lfs_dump_dinode(ip->i_din.ffs1_din);
2101 # endif
2102 		panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n",
2103 		      (unsigned long long)ip->i_number,
2104 		      (ip->i_mode & LFS_IFMT) >> 12);
2105 	}
2106 #endif /* DIAGNOSTIC */
2107 
2108 	/*
2109 	 * Finish inode initialization now that aliasing has been resolved.
2110 	 */
2111 
2112 	ip->i_devvp = ump->um_devvp;
2113 	vref(ip->i_devvp);
2114 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
2115 	ulfsquota_init(ip);
2116 #endif
2117 	genfs_node_init(vp, &lfs_genfsops);
2118 	uvm_vnp_setsize(vp, ip->i_size);
2119 
2120 	/* Initialize hiblk from file size */
2121 	ip->i_lfs_hiblk = lfs_lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1;
2122 
2123 	*vpp = vp;
2124 }
2125 
2126 /*
2127  * Resize the filesystem to contain the specified number of segments.
2128  */
2129 int
2130 lfs_resize_fs(struct lfs *fs, int newnsegs)
2131 {
2132 	SEGUSE *sup;
2133 	struct buf *bp, *obp;
2134 	daddr_t olast, nlast, ilast, noff, start, end;
2135 	struct vnode *ivp;
2136 	struct inode *ip;
2137 	int error, badnews, inc, oldnsegs;
2138 	int sbbytes, csbbytes, gain, cgain;
2139 	int i;
2140 
2141 	/* Only support v2 and up */
2142 	if (fs->lfs_version < 2)
2143 		return EOPNOTSUPP;
2144 
2145 	/* If we're doing nothing, do it fast */
2146 	oldnsegs = fs->lfs_nseg;
2147 	if (newnsegs == oldnsegs)
2148 		return 0;
2149 
2150 	/* We always have to have two superblocks */
2151 	if (newnsegs <= lfs_dtosn(fs, fs->lfs_sboffs[1]))
2152 		return EFBIG;
2153 
2154 	ivp = fs->lfs_ivnode;
2155 	ip = VTOI(ivp);
2156 	error = 0;
2157 
2158 	/* Take the segment lock so no one else calls lfs_newseg() */
2159 	lfs_seglock(fs, SEGM_PROT);
2160 
2161 	/*
2162 	 * Make sure the segments we're going to be losing, if any,
2163 	 * are in fact empty.  We hold the seglock, so their status
2164 	 * cannot change underneath us.  Count the superblocks we lose,
2165 	 * while we're at it.
2166 	 */
2167 	sbbytes = csbbytes = 0;
2168 	cgain = 0;
2169 	for (i = newnsegs; i < oldnsegs; i++) {
2170 		LFS_SEGENTRY(sup, fs, i, bp);
2171 		badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL);
2172 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
2173 			sbbytes += LFS_SBPAD;
2174 		if (!(sup->su_flags & SEGUSE_DIRTY)) {
2175 			++cgain;
2176 			if (sup->su_flags & SEGUSE_SUPERBLOCK)
2177 				csbbytes += LFS_SBPAD;
2178 		}
2179 		brelse(bp, 0);
2180 		if (badnews) {
2181 			error = EBUSY;
2182 			goto out;
2183 		}
2184 	}
2185 
2186 	/* Note old and new segment table endpoints, and old ifile size */
2187 	olast = fs->lfs_cleansz + fs->lfs_segtabsz;
2188 	nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz;
2189 	ilast = ivp->v_size >> fs->lfs_bshift;
2190 	noff = nlast - olast;
2191 
2192 	/*
2193 	 * Make sure no one can use the Ifile while we change it around.
2194 	 * Even after taking the iflock we need to make sure no one still
2195 	 * is holding Ifile buffers, so we get each one, to drain them.
2196 	 * (XXX this could be done better.)
2197 	 */
2198 	rw_enter(&fs->lfs_iflock, RW_WRITER);
2199 	for (i = 0; i < ilast; i++) {
2200 		/* XXX what to do if bread fails? */
2201 		bread(ivp, i, fs->lfs_bsize, NOCRED, 0, &bp);
2202 		brelse(bp, 0);
2203 	}
2204 
2205 	/* Allocate new Ifile blocks */
2206 	for (i = ilast; i < ilast + noff; i++) {
2207 		if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0,
2208 			       &bp) != 0)
2209 			panic("balloc extending ifile");
2210 		memset(bp->b_data, 0, fs->lfs_bsize);
2211 		VOP_BWRITE(bp->b_vp, bp);
2212 	}
2213 
2214 	/* Register new ifile size */
2215 	ip->i_size += noff * fs->lfs_bsize;
2216 	ip->i_ffs1_size = ip->i_size;
2217 	uvm_vnp_setsize(ivp, ip->i_size);
2218 
2219 	/* Copy the inode table to its new position */
2220 	if (noff != 0) {
2221 		if (noff < 0) {
2222 			start = nlast;
2223 			end = ilast + noff;
2224 			inc = 1;
2225 		} else {
2226 			start = ilast + noff - 1;
2227 			end = nlast - 1;
2228 			inc = -1;
2229 		}
2230 		for (i = start; i != end; i += inc) {
2231 			if (bread(ivp, i, fs->lfs_bsize, NOCRED,
2232 			    B_MODIFY, &bp) != 0)
2233 				panic("resize: bread dst blk failed");
2234 			if (bread(ivp, i - noff, fs->lfs_bsize,
2235 			    NOCRED, 0, &obp))
2236 				panic("resize: bread src blk failed");
2237 			memcpy(bp->b_data, obp->b_data, fs->lfs_bsize);
2238 			VOP_BWRITE(bp->b_vp, bp);
2239 			brelse(obp, 0);
2240 		}
2241 	}
2242 
2243 	/* If we are expanding, write the new empty SEGUSE entries */
2244 	if (newnsegs > oldnsegs) {
2245 		for (i = oldnsegs; i < newnsegs; i++) {
2246 			if ((error = bread(ivp, i / fs->lfs_sepb +
2247 					   fs->lfs_cleansz, fs->lfs_bsize,
2248 					   NOCRED, B_MODIFY, &bp)) != 0)
2249 				panic("lfs: ifile read: %d", error);
2250 			while ((i + 1) % fs->lfs_sepb && i < newnsegs) {
2251 				sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb];
2252 				memset(sup, 0, sizeof(*sup));
2253 				i++;
2254 			}
2255 			VOP_BWRITE(bp->b_vp, bp);
2256 		}
2257 	}
2258 
2259 	/* Zero out unused superblock offsets */
2260 	for (i = 2; i < LFS_MAXNUMSB; i++)
2261 		if (lfs_dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs)
2262 			fs->lfs_sboffs[i] = 0x0;
2263 
2264 	/*
2265 	 * Correct superblock entries that depend on fs size.
2266 	 * The computations of these are as follows:
2267 	 *
2268 	 * size  = lfs_segtod(fs, nseg)
2269 	 * dsize = lfs_segtod(fs, nseg - minfreeseg) - lfs_btofsb(#super * LFS_SBPAD)
2270 	 * bfree = dsize - lfs_btofsb(fs, bsize * nseg / 2) - blocks_actually_used
2271 	 * avail = lfs_segtod(fs, nclean) - lfs_btofsb(#clean_super * LFS_SBPAD)
2272 	 *         + (lfs_segtod(fs, 1) - (offset - curseg))
2273 	 *	   - lfs_segtod(fs, minfreeseg - (minfreeseg / 2))
2274 	 *
2275 	 * XXX - we should probably adjust minfreeseg as well.
2276 	 */
2277 	gain = (newnsegs - oldnsegs);
2278 	fs->lfs_nseg = newnsegs;
2279 	fs->lfs_segtabsz = nlast - fs->lfs_cleansz;
2280 	fs->lfs_size += gain * lfs_btofsb(fs, fs->lfs_ssize);
2281 	fs->lfs_dsize += gain * lfs_btofsb(fs, fs->lfs_ssize) - lfs_btofsb(fs, sbbytes);
2282 	fs->lfs_bfree += gain * lfs_btofsb(fs, fs->lfs_ssize) - lfs_btofsb(fs, sbbytes)
2283 		       - gain * lfs_btofsb(fs, fs->lfs_bsize / 2);
2284 	if (gain > 0) {
2285 		fs->lfs_nclean += gain;
2286 		fs->lfs_avail += gain * lfs_btofsb(fs, fs->lfs_ssize);
2287 	} else {
2288 		fs->lfs_nclean -= cgain;
2289 		fs->lfs_avail -= cgain * lfs_btofsb(fs, fs->lfs_ssize) -
2290 				 lfs_btofsb(fs, csbbytes);
2291 	}
2292 
2293 	/* Resize segment flag cache */
2294 	fs->lfs_suflags[0] = realloc(fs->lfs_suflags[0],
2295 	    fs->lfs_nseg * sizeof(u_int32_t), M_SEGMENT, M_WAITOK);
2296 	fs->lfs_suflags[1] = realloc(fs->lfs_suflags[1],
2297 	    fs->lfs_nseg * sizeof(u_int32_t), M_SEGMENT, M_WAITOK);
2298 	for (i = oldnsegs; i < newnsegs; i++)
2299 		fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0;
2300 
2301 	/* Truncate Ifile if necessary */
2302 	if (noff < 0)
2303 		lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0,
2304 		    NOCRED);
2305 
2306 	/* Update cleaner info so the cleaner can die */
2307 	/* XXX what to do if bread fails? */
2308 	bread(ivp, 0, fs->lfs_bsize, NOCRED, B_MODIFY, &bp);
2309 	((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean;
2310 	((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean;
2311 	VOP_BWRITE(bp->b_vp, bp);
2312 
2313 	/* Let Ifile accesses proceed */
2314 	rw_exit(&fs->lfs_iflock);
2315 
2316     out:
2317 	lfs_segunlock(fs);
2318 	return error;
2319 }
2320 
2321 /*
2322  * Extended attribute dispatch
2323  */
2324 static int
2325 lfs_extattrctl(struct mount *mp, int cmd, struct vnode *vp,
2326 	       int attrnamespace, const char *attrname)
2327 {
2328 #ifdef LFS_EXTATTR
2329 	struct ulfsmount *ump;
2330 
2331 	ump = VFSTOULFS(mp);
2332 	if (ump->um_fstype == ULFS1) {
2333 		return ulfs_extattrctl(mp, cmd, vp, attrnamespace, attrname);
2334 	}
2335 #endif
2336 	return vfs_stdextattrctl(mp, cmd, vp, attrnamespace, attrname);
2337 }
2338