xref: /netbsd-src/sys/ufs/lfs/lfs_vfsops.c (revision c38e7cc395b1472a774ff828e46123de44c628e9)
1 /*	$NetBSD: lfs_vfsops.c,v 1.361 2017/10/28 00:37:13 pgoyette 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 1.361 2017/10/28 00:37:13 pgoyette 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_bswap.h>
105 #include <ufs/lfs/ulfs_extern.h>
106 
107 #include <uvm/uvm.h>
108 #include <uvm/uvm_stat.h>
109 #include <uvm/uvm_pager.h>
110 #include <uvm/uvm_pdaemon.h>
111 
112 #include <ufs/lfs/lfs.h>
113 #include <ufs/lfs/lfs_accessors.h>
114 #include <ufs/lfs/lfs_kernel.h>
115 #include <ufs/lfs/lfs_extern.h>
116 
117 #include <miscfs/genfs/genfs.h>
118 #include <miscfs/genfs/genfs_node.h>
119 
120 MODULE(MODULE_CLASS_VFS, lfs, NULL);
121 
122 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
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 struct lwp * lfs_writer_daemon = NULL;
132 kcondvar_t lfs_writerd_cv;
133 
134 int lfs_do_flush = 0;
135 #ifdef LFS_KERNEL_RFW
136 int lfs_do_rfw = 0;
137 #endif
138 
139 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
140 	&lfs_vnodeop_opv_desc,
141 	&lfs_specop_opv_desc,
142 	&lfs_fifoop_opv_desc,
143 	NULL,
144 };
145 
146 struct vfsops lfs_vfsops = {
147 	.vfs_name = MOUNT_LFS,
148 	.vfs_min_mount_data = sizeof (struct ulfs_args),
149 	.vfs_mount = lfs_mount,
150 	.vfs_start = ulfs_start,
151 	.vfs_unmount = lfs_unmount,
152 	.vfs_root = ulfs_root,
153 	.vfs_quotactl = ulfs_quotactl,
154 	.vfs_statvfs = lfs_statvfs,
155 	.vfs_sync = lfs_sync,
156 	.vfs_vget = lfs_vget,
157 	.vfs_loadvnode = lfs_loadvnode,
158 	.vfs_newvnode = lfs_newvnode,
159 	.vfs_fhtovp = lfs_fhtovp,
160 	.vfs_vptofh = lfs_vptofh,
161 	.vfs_init = lfs_init,
162 	.vfs_reinit = lfs_reinit,
163 	.vfs_done = lfs_done,
164 	.vfs_mountroot = lfs_mountroot,
165 	.vfs_snapshot = (void *)eopnotsupp,
166 	.vfs_extattrctl = lfs_extattrctl,
167 	.vfs_suspendctl = genfs_suspendctl,
168 	.vfs_renamelock_enter = genfs_renamelock_enter,
169 	.vfs_renamelock_exit = genfs_renamelock_exit,
170 	.vfs_fsync = (void *)eopnotsupp,
171 	.vfs_opv_descs = lfs_vnodeopv_descs
172 };
173 
174 const struct genfs_ops lfs_genfsops = {
175 	.gop_size = lfs_gop_size,
176 	.gop_alloc = ulfs_gop_alloc,
177 	.gop_write = lfs_gop_write,
178 	.gop_markupdate = ulfs_gop_markupdate,
179 };
180 
181 struct shortlong {
182 	const char *sname;
183 	const char *lname;
184 };
185 
186 static int
187 sysctl_lfs_dostats(SYSCTLFN_ARGS)
188 {
189 	extern struct lfs_stats lfs_stats;
190 	extern int lfs_dostats;
191 	int error;
192 
193 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
194 	if (error || newp == NULL)
195 		return (error);
196 
197 	if (lfs_dostats == 0)
198 		memset(&lfs_stats, 0, sizeof(lfs_stats));
199 
200 	return (0);
201 }
202 
203 static void
204 lfs_sysctl_setup(struct sysctllog **clog)
205 {
206 	int i;
207 	extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead,
208 		   lfs_fs_pagetrip, lfs_ignore_lazy_sync;
209 #ifdef DEBUG
210 	extern int lfs_debug_log_subsys[DLOG_MAX];
211 	struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */
212 		{ "rollforward", "Debug roll-forward code" },
213 		{ "alloc",	"Debug inode allocation and free list" },
214 		{ "avail",	"Debug space-available-now accounting" },
215 		{ "flush",	"Debug flush triggers" },
216 		{ "lockedlist",	"Debug locked list accounting" },
217 		{ "vnode_verbose", "Verbose per-vnode-written debugging" },
218 		{ "vnode",	"Debug vnode use during segment write" },
219 		{ "segment",	"Debug segment writing" },
220 		{ "seguse",	"Debug segment used-bytes accounting" },
221 		{ "cleaner",	"Debug cleaning routines" },
222 		{ "mount",	"Debug mount/unmount routines" },
223 		{ "pagecache",	"Debug UBC interactions" },
224 		{ "dirop",	"Debug directory-operation accounting" },
225 		{ "malloc",	"Debug private malloc accounting" },
226 	};
227 #endif /* DEBUG */
228 	struct shortlong stat_names[] = { /* Must match lfs.h! */
229 		{ "segsused",	    "Number of new segments allocated" },
230 		{ "psegwrites",	    "Number of partial-segment writes" },
231 		{ "psyncwrites",    "Number of synchronous partial-segment"
232 				    " writes" },
233 		{ "pcleanwrites",   "Number of partial-segment writes by the"
234 				    " cleaner" },
235 		{ "blocktot",       "Number of blocks written" },
236 		{ "cleanblocks",    "Number of blocks written by the cleaner" },
237 		{ "ncheckpoints",   "Number of checkpoints made" },
238 		{ "nwrites",        "Number of whole writes" },
239 		{ "nsync_writes",   "Number of synchronous writes" },
240 		{ "wait_exceeded",  "Number of times writer waited for"
241 				    " cleaner" },
242 		{ "write_exceeded", "Number of times writer invoked flush" },
243 		{ "flush_invoked",  "Number of times flush was invoked" },
244 		{ "vflush_invoked", "Number of time vflush was called" },
245 		{ "clean_inlocked", "Number of vnodes skipped for being dead" },
246 		{ "clean_vnlocked", "Number of vnodes skipped for vget failure" },
247 		{ "segs_reclaimed", "Number of segments reclaimed" },
248 	};
249 
250 	sysctl_createv(clog, 0, NULL, NULL,
251 		       CTLFLAG_PERMANENT,
252 		       CTLTYPE_NODE, "lfs",
253 		       SYSCTL_DESCR("Log-structured file system"),
254 		       NULL, 0, NULL, 0,
255 		       CTL_VFS, 5, CTL_EOL);
256 	/*
257 	 * XXX the "5" above could be dynamic, thereby eliminating one
258 	 * more instance of the "number to vfs" mapping problem, but
259 	 * "5" is the order as taken from sys/mount.h
260 	 */
261 
262 	sysctl_createv(clog, 0, NULL, NULL,
263 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
264 		       CTLTYPE_INT, "flushindir", NULL,
265 		       NULL, 0, &lfs_writeindir, 0,
266 		       CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
267 	sysctl_createv(clog, 0, NULL, NULL,
268 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
269 		       CTLTYPE_INT, "clean_vnhead", NULL,
270 		       NULL, 0, &lfs_clean_vnhead, 0,
271 		       CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
272 	sysctl_createv(clog, 0, NULL, NULL,
273 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
274 		       CTLTYPE_INT, "dostats",
275 		       SYSCTL_DESCR("Maintain statistics on LFS operations"),
276 		       sysctl_lfs_dostats, 0, &lfs_dostats, 0,
277 		       CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
278 	sysctl_createv(clog, 0, NULL, NULL,
279 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
280 		       CTLTYPE_INT, "pagetrip",
281 		       SYSCTL_DESCR("How many dirty pages in fs triggers"
282 				    " a flush"),
283 		       NULL, 0, &lfs_fs_pagetrip, 0,
284 		       CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL);
285 	sysctl_createv(clog, 0, NULL, NULL,
286 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
287 		       CTLTYPE_INT, "ignore_lazy_sync",
288 		       SYSCTL_DESCR("Lazy Sync is ignored entirely"),
289 		       NULL, 0, &lfs_ignore_lazy_sync, 0,
290 		       CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL);
291 #ifdef LFS_KERNEL_RFW
292 	sysctl_createv(clog, 0, NULL, NULL,
293 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
294 		       CTLTYPE_INT, "rfw",
295 		       SYSCTL_DESCR("Use in-kernel roll-forward on mount"),
296 		       NULL, 0, &lfs_do_rfw, 0,
297 		       CTL_VFS, 5, LFS_DO_RFW, CTL_EOL);
298 #endif
299 
300 	sysctl_createv(clog, 0, NULL, NULL,
301 		       CTLFLAG_PERMANENT,
302 		       CTLTYPE_NODE, "stats",
303 		       SYSCTL_DESCR("Debugging options"),
304 		       NULL, 0, NULL, 0,
305 		       CTL_VFS, 5, LFS_STATS, CTL_EOL);
306 	for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) {
307 		sysctl_createv(clog, 0, NULL, NULL,
308 			       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
309 			       CTLTYPE_INT, stat_names[i].sname,
310 			       SYSCTL_DESCR(stat_names[i].lname),
311 			       NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]),
312 			       0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL);
313 	}
314 
315 #ifdef DEBUG
316 	sysctl_createv(clog, 0, NULL, NULL,
317 		       CTLFLAG_PERMANENT,
318 		       CTLTYPE_NODE, "debug",
319 		       SYSCTL_DESCR("Debugging options"),
320 		       NULL, 0, NULL, 0,
321 		       CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL);
322 	for (i = 0; i < DLOG_MAX; i++) {
323 		sysctl_createv(clog, 0, NULL, NULL,
324 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
325 			       CTLTYPE_INT, dlog_names[i].sname,
326 			       SYSCTL_DESCR(dlog_names[i].lname),
327 			       NULL, 0, &(lfs_debug_log_subsys[i]), 0,
328 			       CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL);
329 	}
330 #endif
331 }
332 
333 /* old cleaner syscall interface.  see VOP_FCNTL() */
334 static const struct syscall_package lfs_syscalls[] = {
335 	{ SYS_lfs_bmapv,	0, (sy_call_t *)sys_lfs_bmapv		},
336 	{ SYS_lfs_markv,	0, (sy_call_t *)sys_lfs_markv		},
337 	{ SYS___lfs_segwait50,	0, (sy_call_t *)sys___lfs_segwait50	},
338 	{ SYS_lfs_segclean,	0, (sy_call_t *)sys_lfs_segclean	},
339 	{ 0, 0, NULL },
340 };
341 
342 static int
343 lfs_modcmd(modcmd_t cmd, void *arg)
344 {
345 	int error;
346 
347 	switch (cmd) {
348 	case MODULE_CMD_INIT:
349 		error = syscall_establish(NULL, lfs_syscalls);
350 		if (error)
351 			return error;
352 		error = vfs_attach(&lfs_vfsops);
353 		if (error != 0) {
354 			syscall_disestablish(NULL, lfs_syscalls);
355 			break;
356 		}
357 		lfs_sysctl_setup(&lfs_sysctl_log);
358 		break;
359 	case MODULE_CMD_FINI:
360 		error = vfs_detach(&lfs_vfsops);
361 		if (error != 0)
362 			break;
363 		syscall_disestablish(NULL, lfs_syscalls);
364 		sysctl_teardown(&lfs_sysctl_log);
365 		break;
366 	default:
367 		error = ENOTTY;
368 		break;
369 	}
370 
371 	return (error);
372 }
373 
374 /*
375  * XXX Same structure as FFS inodes?  Should we share a common pool?
376  */
377 struct pool lfs_inode_pool;
378 struct pool lfs_dinode_pool;
379 struct pool lfs_inoext_pool;
380 struct pool lfs_lbnentry_pool;
381 
382 /*
383  * The writer daemon.  UVM keeps track of how many dirty pages we are holding
384  * in lfs_subsys_pages; the daemon flushes the filesystem when this value
385  * crosses the (user-defined) threshhold LFS_MAX_PAGES.
386  */
387 static void
388 lfs_writerd(void *arg)
389 {
390 	mount_iterator_t *iter;
391  	struct mount *mp;
392  	struct lfs *fs;
393 	struct vfsops *vfs = NULL;
394  	int fsflags;
395 	int lfsc;
396 	int wrote_something = 0;
397 
398 	mutex_enter(&lfs_lock);
399 	KASSERTMSG(lfs_writer_daemon == NULL, "more than one LFS writer daemon");
400 	lfs_writer_daemon = curlwp;
401 	mutex_exit(&lfs_lock);
402 
403 	/* Take an extra reference to the LFS vfsops. */
404 	vfs = vfs_getopsbyname(MOUNT_LFS);
405 
406  	mutex_enter(&lfs_lock);
407  	for (;;) {
408 		KASSERT(mutex_owned(&lfs_lock));
409 		if (wrote_something == 0)
410 			cv_timedwait(&lfs_writerd_cv, &lfs_lock, hz/10 + 1);
411 		KASSERT(mutex_owned(&lfs_lock));
412 		wrote_something = 0;
413 
414 		/*
415 		 * If global state wants a flush, flush everything.
416 		 */
417 		if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
418 			locked_queue_bytes > LFS_MAX_BYTES ||
419 			lfs_subsys_pages > LFS_MAX_PAGES) {
420 
421 			if (lfs_do_flush) {
422 				DLOG((DLOG_FLUSH, "lfs_writerd: lfs_do_flush\n"));
423 			}
424 			if (locked_queue_count > LFS_MAX_BUFS) {
425 				DLOG((DLOG_FLUSH, "lfs_writerd: lqc = %d, max %d\n",
426 				      locked_queue_count, LFS_MAX_BUFS));
427 			}
428 			if (locked_queue_bytes > LFS_MAX_BYTES) {
429 				DLOG((DLOG_FLUSH, "lfs_writerd: lqb = %ld, max %ld\n",
430 				      locked_queue_bytes, LFS_MAX_BYTES));
431 			}
432 			if (lfs_subsys_pages > LFS_MAX_PAGES) {
433 				DLOG((DLOG_FLUSH, "lfs_writerd: lssp = %d, max %d\n",
434 				      lfs_subsys_pages, LFS_MAX_PAGES));
435 			}
436 
437 			lfs_flush(NULL, SEGM_WRITERD, 0);
438 			lfs_do_flush = 0;
439 			KASSERT(mutex_owned(&lfs_lock));
440 			continue;
441 		}
442 		KASSERT(mutex_owned(&lfs_lock));
443 		mutex_exit(&lfs_lock);
444 
445  		/*
446  		 * Look through the list of LFSs to see if any of them
447  		 * have requested pageouts.
448  		 */
449  		mountlist_iterator_init(&iter);
450 		lfsc = 0;
451 		while ((mp = mountlist_iterator_next(iter)) != NULL) {
452 			KASSERT(!mutex_owned(&lfs_lock));
453  			if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS,
454  			    sizeof(mp->mnt_stat.f_fstypename)) == 0) {
455 				++lfsc;
456  				fs = VFSTOULFS(mp)->um_lfs;
457 				daddr_t ooffset = 0;
458 				fsflags = SEGM_SINGLE;
459 
460  				mutex_enter(&lfs_lock);
461 				ooffset = lfs_sb_getoffset(fs);
462 
463 				if (lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs) && fs->lfs_nowrap) {
464 					/* Don't try to write if we're suspended */
465 					mutex_exit(&lfs_lock);
466 					continue;
467 				}
468 				if (LFS_STARVED_FOR_SEGS(fs)) {
469 					mutex_exit(&lfs_lock);
470 
471 					DLOG((DLOG_FLUSH, "lfs_writerd: need cleaning before writing possible\n"));
472 					lfs_wakeup_cleaner(fs);
473 					continue;
474 				}
475 
476  				if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) ||
477  				     lfs_dirvcount > LFS_MAX_DIROP) &&
478 				    fs->lfs_dirops == 0) {
479 					fsflags &= ~SEGM_SINGLE;
480  					fsflags |= SEGM_CKP;
481 					DLOG((DLOG_FLUSH, "lfs_writerd: checkpoint\n"));
482 					lfs_flush_fs(fs, fsflags);
483 				} else if (fs->lfs_pdflush) {
484  					DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n"));
485  					lfs_flush_fs(fs, fsflags);
486  				} else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) {
487  					DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n"));
488  					mutex_exit(&lfs_lock);
489  					lfs_writer_enter(fs, "wrdirop");
490  					lfs_flush_pchain(fs);
491  					lfs_writer_leave(fs);
492 					mutex_enter(&lfs_lock);
493 				}
494 				if (lfs_sb_getoffset(fs) != ooffset)
495 					++wrote_something;
496 				mutex_exit(&lfs_lock);
497  			}
498 			KASSERT(!mutex_owned(&lfs_lock));
499  		}
500 		if (lfsc == 0) {
501 			mutex_enter(&lfs_lock);
502 			lfs_writer_daemon = NULL;
503 			mutex_exit(&lfs_lock);
504 			mountlist_iterator_destroy(iter);
505 			break;
506 		}
507  		mountlist_iterator_destroy(iter);
508 
509  		mutex_enter(&lfs_lock);
510  	}
511 	KASSERT(!mutex_owned(&lfs_lock));
512 
513 	/* Give up our extra reference so the module can be unloaded. */
514 	mutex_enter(&vfs_list_lock);
515 	if (vfs != NULL)
516 		vfs->vfs_refcount--;
517 	mutex_exit(&vfs_list_lock);
518 
519 	/* Done! */
520 	kthread_exit(0);
521 }
522 
523 /*
524  * Initialize the filesystem, most work done by ulfs_init.
525  */
526 void
527 lfs_init(void)
528 {
529 
530 	/*
531 	 * XXX: should we use separate pools for 32-bit and 64-bit
532 	 * dinodes?
533 	 */
534 	malloc_type_attach(M_SEGMENT);
535 	pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
536 	    "lfsinopl", &pool_allocator_nointr, IPL_NONE);
537 	pool_init(&lfs_dinode_pool, sizeof(union lfs_dinode), 0, 0, 0,
538 	    "lfsdinopl", &pool_allocator_nointr, IPL_NONE);
539 	pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
540 	    "lfsinoextpl", &pool_allocator_nointr, IPL_NONE);
541 	pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0,
542 	    "lfslbnpool", &pool_allocator_nointr, IPL_NONE);
543 	ulfs_init();
544 
545 #ifdef DEBUG
546 	memset(lfs_log, 0, sizeof(lfs_log));
547 #endif
548 	mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE);
549 	cv_init(&lfs_writerd_cv, "lfswrite");
550 	cv_init(&locked_queue_cv, "lfsbuf");
551 	cv_init(&lfs_writing_cv, "lfsflush");
552 }
553 
554 void
555 lfs_reinit(void)
556 {
557 	ulfs_reinit();
558 }
559 
560 void
561 lfs_done(void)
562 {
563 	ulfs_done();
564 	mutex_destroy(&lfs_lock);
565 	cv_destroy(&lfs_writerd_cv);
566 	cv_destroy(&locked_queue_cv);
567 	cv_destroy(&lfs_writing_cv);
568 	pool_destroy(&lfs_inode_pool);
569 	pool_destroy(&lfs_dinode_pool);
570 	pool_destroy(&lfs_inoext_pool);
571 	pool_destroy(&lfs_lbnentry_pool);
572 	malloc_type_detach(M_SEGMENT);
573 }
574 
575 /*
576  * Called by main() when ulfs is going to be mounted as root.
577  */
578 int
579 lfs_mountroot(void)
580 {
581 	extern struct vnode *rootvp;
582 	struct lfs *fs = NULL;				/* LFS */
583 	struct mount *mp;
584 	struct lwp *l = curlwp;
585 	struct ulfsmount *ump;
586 	int error;
587 
588 	if (device_class(root_device) != DV_DISK)
589 		return (ENODEV);
590 
591 	if (rootdev == NODEV)
592 		return (ENODEV);
593 	if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
594 		vrele(rootvp);
595 		return (error);
596 	}
597 	if ((error = lfs_mountfs(rootvp, mp, l))) {
598 		vfs_unbusy(mp);
599 		vfs_rele(mp);
600 		return (error);
601 	}
602 	mountlist_append(mp);
603 	ump = VFSTOULFS(mp);
604 	fs = ump->um_lfs;
605 	lfs_sb_setfsmnt(fs, mp->mnt_stat.f_mntonname);
606 	(void)lfs_statvfs(mp, &mp->mnt_stat);
607 	vfs_unbusy(mp);
608 	setrootfstime((time_t)lfs_sb_gettstamp(VFSTOULFS(mp)->um_lfs));
609 	return (0);
610 }
611 
612 /*
613  * VFS Operations.
614  *
615  * mount system call
616  */
617 int
618 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
619 {
620 	struct lwp *l = curlwp;
621 	struct vnode *devvp;
622 	struct ulfs_args *args = data;
623 	struct ulfsmount *ump = NULL;
624 	struct lfs *fs = NULL;				/* LFS */
625 	int error = 0, update;
626 	mode_t accessmode;
627 
628 	if (args == NULL)
629 		return EINVAL;
630 	if (*data_len < sizeof *args)
631 		return EINVAL;
632 
633 	if (mp->mnt_flag & MNT_GETARGS) {
634 		ump = VFSTOULFS(mp);
635 		if (ump == NULL)
636 			return EIO;
637 		args->fspec = NULL;
638 		*data_len = sizeof *args;
639 		return 0;
640 	}
641 
642 	update = mp->mnt_flag & MNT_UPDATE;
643 
644 	/* Check arguments */
645 	if (args->fspec != NULL) {
646 		/*
647 		 * Look up the name and verify that it's sane.
648 		 */
649 		error = namei_simple_user(args->fspec,
650 					NSM_FOLLOW_NOEMULROOT, &devvp);
651 		if (error != 0)
652 			return (error);
653 
654 		if (!update) {
655 			/*
656 			 * Be sure this is a valid block device
657 			 */
658 			if (devvp->v_type != VBLK)
659 				error = ENOTBLK;
660 			else if (bdevsw_lookup(devvp->v_rdev) == NULL)
661 				error = ENXIO;
662 		} else {
663 			/*
664 			 * Be sure we're still naming the same device
665 			 * used for our initial mount
666 			 *
667 			 * XXX dholland 20151010: if namei gives us a
668 			 * different vnode for the same device,
669 			 * wouldn't it be better to use it going
670 			 * forward rather than ignore it in favor of
671 			 * the old one?
672 			 */
673 			ump = VFSTOULFS(mp);
674 			fs = ump->um_lfs;
675 			if (devvp != fs->lfs_devvp) {
676 				if (devvp->v_rdev != fs->lfs_devvp->v_rdev)
677 					error = EINVAL;
678 				else {
679 					vrele(devvp);
680 					devvp = fs->lfs_devvp;
681 					vref(devvp);
682 				}
683 			}
684 		}
685 	} else {
686 		if (!update) {
687 			/* New mounts must have a filename for the device */
688 			return (EINVAL);
689 		} else {
690 			/* Use the extant mount */
691 			ump = VFSTOULFS(mp);
692 			fs = ump->um_lfs;
693 			devvp = fs->lfs_devvp;
694 			vref(devvp);
695 		}
696 	}
697 
698 
699 	/*
700 	 * If mount by non-root, then verify that user has necessary
701 	 * permissions on the device.
702 	 */
703 	if (error == 0) {
704 		accessmode = VREAD;
705 		if (update ?
706 		    (mp->mnt_iflag & IMNT_WANTRDWR) != 0 :
707 		    (mp->mnt_flag & MNT_RDONLY) == 0)
708 			accessmode |= VWRITE;
709 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
710 		error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT,
711 		    KAUTH_REQ_SYSTEM_MOUNT_DEVICE, mp, devvp,
712 		    KAUTH_ARG(accessmode));
713 		VOP_UNLOCK(devvp);
714 	}
715 
716 	if (error) {
717 		vrele(devvp);
718 		return (error);
719 	}
720 
721 	if (!update) {
722 		int flags;
723 
724 		if (mp->mnt_flag & MNT_RDONLY)
725 			flags = FREAD;
726 		else
727 			flags = FREAD|FWRITE;
728 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
729 		error = VOP_OPEN(devvp, flags, FSCRED);
730 		VOP_UNLOCK(devvp);
731 		if (error)
732 			goto fail;
733 		error = lfs_mountfs(devvp, mp, l);		/* LFS */
734 		if (error) {
735 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
736 			(void)VOP_CLOSE(devvp, flags, NOCRED);
737 			VOP_UNLOCK(devvp);
738 			goto fail;
739 		}
740 
741 		ump = VFSTOULFS(mp);
742 		fs = ump->um_lfs;
743 	} else {
744 		/*
745 		 * Update the mount.
746 		 */
747 
748 		/*
749 		 * The initial mount got a reference on this
750 		 * device, so drop the one obtained via
751 		 * namei(), above.
752 		 */
753 		vrele(devvp);
754 
755 		ump = VFSTOULFS(mp);
756 		fs = ump->um_lfs;
757 
758 		if (fs->lfs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
759 			/*
760 			 * Changing from read/write to read-only.
761 			 * XXX: shouldn't we sync here? or does vfs do that?
762 			 */
763 #ifdef LFS_QUOTA2
764 			/* XXX: quotas should remain on when readonly */
765 			if (fs->lfs_use_quota2) {
766 				error = lfsquota2_umount(mp, 0);
767 				if (error) {
768 					return error;
769 				}
770 			}
771 #endif
772 		}
773 
774 		if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
775 			/*
776 			 * Changing from read-only to read/write.
777 			 * Note in the superblocks that we're writing.
778 			 */
779 
780 			/* XXX: quotas should have been on even if readonly */
781 			if (fs->lfs_use_quota2) {
782 #ifdef LFS_QUOTA2
783 				error = lfs_quota2_mount(mp);
784 #else
785 				uprintf("%s: no kernel support for this "
786 					"filesystem's quotas\n",
787 					mp->mnt_stat.f_mntonname);
788 				if (mp->mnt_flag & MNT_FORCE) {
789 					uprintf("%s: mounting anyway; "
790 						"fsck afterwards\n",
791 						mp->mnt_stat.f_mntonname);
792 				} else {
793 					error = EINVAL;
794 				}
795 #endif
796 				if (error) {
797 					return error;
798 				}
799 			}
800 
801 			fs->lfs_ronly = 0;
802 			if (lfs_sb_getpflags(fs) & LFS_PF_CLEAN) {
803 				lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN);
804 				lfs_writesuper(fs, lfs_sb_getsboff(fs, 0));
805 				lfs_writesuper(fs, lfs_sb_getsboff(fs, 1));
806 			}
807 		}
808 		if (args->fspec == NULL)
809 			return EINVAL;
810 	}
811 
812 	error = set_statvfs_info(path, UIO_USERSPACE, args->fspec,
813 	    UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l);
814 	if (error == 0)
815 		lfs_sb_setfsmnt(fs, mp->mnt_stat.f_mntonname);
816 	return error;
817 
818 fail:
819 	vrele(devvp);
820 	return (error);
821 }
822 
823 /*
824  * Helper for mountfs. Note that the fs pointer may be a dummy one
825  * pointing into a superblock buffer. (Which is gross; see below.)
826  */
827 static int
828 lfs_checkmagic(struct lfs *fs)
829 {
830 	switch (fs->lfs_dlfs_u.u_32.dlfs_magic) {
831 	    case LFS_MAGIC:
832 		fs->lfs_is64 = false;
833 		fs->lfs_dobyteswap = false;
834 		break;
835 	    case LFS64_MAGIC:
836 		fs->lfs_is64 = true;
837 		fs->lfs_dobyteswap = false;
838 		break;
839 #ifdef LFS_EI
840 	    case LFS_MAGIC_SWAPPED:
841 		fs->lfs_is64 = false;
842 		fs->lfs_dobyteswap = true;
843 		break;
844 	    case LFS64_MAGIC_SWAPPED:
845 		fs->lfs_is64 = true;
846 		fs->lfs_dobyteswap = true;
847 		break;
848 #endif
849 	    default:
850 		/* XXX needs translation */
851 		return EINVAL;
852 	}
853 	return 0;
854 }
855 
856 /*
857  * Common code for mount and mountroot
858  * LFS specific
859  */
860 int
861 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l)
862 {
863 	static bool lfs_mounted_once = false;
864 	struct lfs *primarysb, *altsb, *thesb;
865 	struct buf *primarybuf, *altbuf;
866 	struct lfs *fs;
867 	struct ulfsmount *ump;
868 	struct vnode *vp;
869 	dev_t dev;
870 	int error, i, ronly, fsbsize;
871 	kauth_cred_t cred;
872 	CLEANERINFO *cip;
873 	SEGUSE *sup;
874 	daddr_t sb_addr;
875 
876 	cred = l ? l->l_cred : NOCRED;
877 
878 	/* The superblock is supposed to be 512 bytes. */
879 	__CTASSERT(sizeof(struct dlfs) == DEV_BSIZE);
880 
881 	/*
882 	 * Flush out any old buffers remaining from a previous use.
883 	 */
884 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
885 	error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0);
886 	VOP_UNLOCK(devvp);
887 	if (error)
888 		return (error);
889 
890 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
891 
892 	/* Don't free random space on error. */
893 	primarybuf = NULL;
894 	altbuf = NULL;
895 	ump = NULL;
896 
897 	sb_addr = LFS_LABELPAD / DEV_BSIZE;
898 	while (1) {
899 		/*
900 		 * Read in the superblock.
901 		 *
902 		 * Note that because LFS_SBPAD is substantially larger
903 		 * (8K) than the actual on-disk superblock (512 bytes)
904 		 * the buffer contains enough space to be used as a
905 		 * whole struct lfs (in-memory superblock) - we do this
906 		 * only so we can set and use the is64 and dobyteswap
907 		 * members. XXX this is gross and the logic here should
908 		 * be reworked.
909 		 */
910 		error = bread(devvp, sb_addr, LFS_SBPAD, 0, &primarybuf);
911 		if (error)
912 			goto out;
913 		primarysb = (struct lfs *)primarybuf->b_data;
914 
915 		/* Check the basics. */
916 		error = lfs_checkmagic(primarysb);
917 		if (error) {
918 			DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock wrong magic\n"));
919 			goto out;
920 		}
921 		if (lfs_sb_getbsize(primarysb) > MAXBSIZE ||
922 		    lfs_sb_getversion(primarysb) > LFS_VERSION ||
923 		    lfs_sb_getbsize(primarysb) < sizeof(struct dlfs)) {
924 			DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n"));
925 			/* XXX needs translation */
926 			error = EINVAL;
927 			goto out;
928 		}
929 		if (lfs_sb_getinodefmt(primarysb) > LFS_MAXINODEFMT) {
930 			DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n",
931 			       lfs_sb_getinodefmt(primarysb)));
932 			error = EINVAL;
933 			goto out;
934 		}
935 
936 		if (lfs_sb_getversion(primarysb) == 1)
937 			fsbsize = DEV_BSIZE;
938 		else {
939 			fsbsize = 1 << lfs_sb_getffshift(primarysb);
940 			/*
941 			 * Could be, if the frag size is large enough, that we
942 			 * don't have the "real" primary superblock.  If that's
943 			 * the case, get the real one, and try again.
944 			 */
945 			if (sb_addr != (lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT))) {
946 				DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr"
947 				      " 0x%llx is not right, trying 0x%llx\n",
948 				      (long long)sb_addr,
949 				      (long long)(lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT))));
950 				sb_addr = lfs_sb_getsboff(primarysb, 0) << (lfs_sb_getffshift(primarysb) - DEV_BSHIFT);
951 				brelse(primarybuf, BC_INVAL);
952 				continue;
953 			}
954 		}
955 		break;
956 	}
957 
958 	/*
959 	 * Check the second superblock to see which is newer; then mount
960 	 * using the older of the two.	This is necessary to ensure that
961 	 * the filesystem is valid if it was not unmounted cleanly.
962 	 */
963 
964 	if (lfs_sb_getsboff(primarysb, 1) &&
965 	    lfs_sb_getsboff(primarysb, 1) - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
966 	{
967 		error = bread(devvp, lfs_sb_getsboff(primarysb, 1) * (fsbsize / DEV_BSIZE),
968 			LFS_SBPAD, 0, &altbuf);
969 		if (error)
970 			goto out;
971 		altsb = (struct lfs *)altbuf->b_data;
972 
973 		/*
974 		 * Note: this used to do the sanity check only if the
975 		 * timestamp/serial comparison required use of altsb;
976 		 * this way is less tolerant, but if altsb is corrupted
977 		 * enough that the magic number, version, and blocksize
978 		 * are bogus, why would the timestamp or serial fields
979 		 * mean anything either? If this kind of thing happens,
980 		 * you need to fsck anyway.
981 		 */
982 
983 		error = lfs_checkmagic(altsb);
984 		if (error)
985 			goto out;
986 
987 		/* Check the basics. */
988 		if (lfs_sb_getbsize(altsb) > MAXBSIZE ||
989 		    lfs_sb_getversion(altsb) > LFS_VERSION ||
990 		    lfs_sb_getbsize(altsb) < sizeof(struct dlfs)) {
991 			DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock"
992 			      " sanity failed\n"));
993 			error = EINVAL;		/* XXX needs translation */
994 			goto out;
995 		}
996 
997 		if (lfs_sb_getversion(primarysb) == 1) {
998 			/* 1s resolution comparison */
999 			if (lfs_sb_gettstamp(altsb) < lfs_sb_gettstamp(primarysb))
1000 				thesb = altsb;
1001 			else
1002 				thesb = primarysb;
1003 		} else {
1004 			/* monotonic infinite-resolution comparison */
1005 			if (lfs_sb_getserial(altsb) < lfs_sb_getserial(primarysb))
1006 				thesb = altsb;
1007 			else
1008 				thesb = primarysb;
1009 		}
1010 	} else {
1011 		DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock location"
1012 		      " daddr=0x%x\n", lfs_sb_getsboff(primarysb, 1)));
1013 		error = EINVAL;
1014 		goto out;
1015 	}
1016 
1017 	/*
1018 	 * Allocate the mount structure, copy the superblock into it.
1019 	 * Note that the 32-bit and 64-bit superblocks are the same size.
1020 	 */
1021 	fs = kmem_zalloc(sizeof(struct lfs), KM_SLEEP);
1022 	memcpy(&fs->lfs_dlfs_u.u_32, &thesb->lfs_dlfs_u.u_32,
1023 	       sizeof(struct dlfs));
1024 	fs->lfs_is64 = thesb->lfs_is64;
1025 	fs->lfs_dobyteswap = thesb->lfs_dobyteswap;
1026 	fs->lfs_hasolddirfmt = false; /* set for real below */
1027 
1028 	/* Compatibility */
1029 	if (lfs_sb_getversion(fs) < 2) {
1030 		lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE);
1031 		lfs_sb_setibsize(fs, lfs_sb_getbsize(fs));
1032 		lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0));
1033 		lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs));
1034 		lfs_sb_setfsbtodb(fs, 0);
1035 	}
1036 	if (lfs_sb_getresvseg(fs) == 0)
1037 		lfs_sb_setresvseg(fs, MIN(lfs_sb_getminfreeseg(fs) - 1, \
1038 			MAX(MIN_RESV_SEGS, lfs_sb_getminfreeseg(fs) / 2 + 1)));
1039 
1040 	/*
1041 	 * If we aren't going to be able to write meaningfully to this
1042 	 * filesystem, and were not mounted readonly, bomb out now.
1043 	 */
1044 	if (lfs_fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) {
1045 		DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write,"
1046 		      " we need BUFPAGES >= %lld\n",
1047 		      (long long)((bufmem_hiwater / bufmem_lowater) *
1048 				  LFS_INVERSE_MAX_BYTES(
1049 					  lfs_fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT)));
1050 		kmem_free(fs, sizeof(struct lfs));
1051 		error = EFBIG; /* XXX needs translation */
1052 		goto out;
1053 	}
1054 
1055 	/* Before rolling forward, lock so vget will sleep for other procs */
1056 	if (l != NULL) {
1057 		fs->lfs_flags = LFS_NOTYET;
1058 		fs->lfs_rfpid = l->l_proc->p_pid;
1059 	}
1060 
1061 	ump = kmem_zalloc(sizeof(*ump), KM_SLEEP);
1062 	ump->um_lfs = fs;
1063 	ump->um_fstype = fs->lfs_is64 ? ULFS2 : ULFS1;
1064 	/* ump->um_cleaner_thread = NULL; */
1065 	brelse(primarybuf, BC_INVAL);
1066 	brelse(altbuf, BC_INVAL);
1067 	primarybuf = NULL;
1068 	altbuf = NULL;
1069 
1070 
1071 	/* Set up the I/O information */
1072 	fs->lfs_devbsize = DEV_BSIZE;
1073 	fs->lfs_iocount = 0;
1074 	fs->lfs_diropwait = 0;
1075 	fs->lfs_activesb = 0;
1076 	lfs_sb_setuinodes(fs, 0);
1077 	fs->lfs_ravail = 0;
1078 	fs->lfs_favail = 0;
1079 	fs->lfs_sbactive = 0;
1080 
1081 	/* Set up the ifile and lock aflags */
1082 	fs->lfs_doifile = 0;
1083 	fs->lfs_writer = 0;
1084 	fs->lfs_dirops = 0;
1085 	fs->lfs_nadirop = 0;
1086 	fs->lfs_seglock = 0;
1087 	fs->lfs_pdflush = 0;
1088 	fs->lfs_sleepers = 0;
1089 	fs->lfs_pages = 0;
1090 	rw_init(&fs->lfs_fraglock);
1091 	rw_init(&fs->lfs_iflock);
1092 	cv_init(&fs->lfs_sleeperscv, "lfs_slp");
1093 	cv_init(&fs->lfs_diropscv, "lfs_dirop");
1094 	cv_init(&fs->lfs_stopcv, "lfsstop");
1095 	cv_init(&fs->lfs_nextsegsleep, "segment");
1096 
1097 	/* Initialize values for all LFS mounts */
1098 	if (!lfs_mounted_once) {
1099 		cv_init(&lfs_allclean_wakeup, "segment");
1100 		lfs_mounted_once = true;
1101 	}
1102 
1103 	/* Set the file system readonly/modify bits. */
1104 	fs->lfs_ronly = ronly;
1105 	if (ronly == 0)
1106 		fs->lfs_fmod = 1;
1107 
1108 	/* Device we're using */
1109 	dev = devvp->v_rdev;
1110 	fs->lfs_dev = dev;
1111 	fs->lfs_devvp = devvp;
1112 
1113 	/* ulfs-level information */
1114 	fs->um_flags = 0;
1115 	fs->um_bptrtodb = lfs_sb_getffshift(fs) - DEV_BSHIFT;
1116 	fs->um_seqinc = lfs_sb_getfrag(fs);
1117 	fs->um_nindir = lfs_sb_getnindir(fs);
1118 	fs->um_lognindir = ffs(lfs_sb_getnindir(fs)) - 1;
1119 	fs->um_maxsymlinklen = lfs_sb_getmaxsymlinklen(fs);
1120 	fs->um_dirblksiz = LFS_DIRBLKSIZ;
1121 	fs->um_maxfilesize = lfs_sb_getmaxfilesize(fs);
1122 
1123 	/* quota stuff */
1124 	/* XXX: these need to come from the on-disk superblock to be used */
1125 	fs->lfs_use_quota2 = 0;
1126 	fs->lfs_quota_magic = 0;
1127 	fs->lfs_quota_flags = 0;
1128 	fs->lfs_quotaino[0] = 0;
1129 	fs->lfs_quotaino[1] = 0;
1130 
1131 	/* Initialize the mount structure. */
1132 	mp->mnt_data = ump;
1133 	mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
1134 	mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS);
1135 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
1136 	mp->mnt_stat.f_namemax = LFS_MAXNAMLEN;
1137 	mp->mnt_stat.f_iosize = lfs_sb_getbsize(fs);
1138 	mp->mnt_flag |= MNT_LOCAL;
1139 	mp->mnt_fs_bshift = lfs_sb_getbshift(fs);
1140 	if (fs->um_maxsymlinklen > 0)
1141 		mp->mnt_iflag |= IMNT_DTYPE;
1142 	else
1143 		fs->lfs_hasolddirfmt = true;
1144 
1145 	ump->um_mountp = mp;
1146 	for (i = 0; i < ULFS_MAXQUOTAS; i++)
1147 		ump->um_quotas[i] = NULLVP;
1148 	spec_node_setmountedfs(devvp, mp);
1149 
1150 	/* Set up reserved memory for pageout */
1151 	lfs_setup_resblks(fs);
1152 	/* Set up vdirop tailq */
1153 	TAILQ_INIT(&fs->lfs_dchainhd);
1154 	/* and paging tailq */
1155 	TAILQ_INIT(&fs->lfs_pchainhd);
1156 	/* and delayed segment accounting for truncation list */
1157 	LIST_INIT(&fs->lfs_segdhd);
1158 
1159 	/*
1160 	 * We use the ifile vnode for almost every operation.  Instead of
1161 	 * retrieving it from the hash table each time we retrieve it here,
1162 	 * artificially increment the reference count and keep a pointer
1163 	 * to it in the incore copy of the superblock.
1164 	 */
1165 	if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
1166 		DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error));
1167 		goto out;
1168 	}
1169 	fs->lfs_ivnode = vp;
1170 	vref(vp);
1171 
1172 	/* Set up inode bitmap and order free list */
1173 	lfs_order_freelist(fs);
1174 
1175 	/* Set up segment usage flags for the autocleaner. */
1176 	fs->lfs_nactive = 0;
1177 	fs->lfs_suflags = malloc(2 * sizeof(u_int32_t *),
1178 				 M_SEGMENT, M_WAITOK);
1179 	fs->lfs_suflags[0] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t),
1180 				    M_SEGMENT, M_WAITOK);
1181 	fs->lfs_suflags[1] = malloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t),
1182 				    M_SEGMENT, M_WAITOK);
1183 	memset(fs->lfs_suflags[1], 0, lfs_sb_getnseg(fs) * sizeof(u_int32_t));
1184 	for (i = 0; i < lfs_sb_getnseg(fs); i++) {
1185 		int changed;
1186 		struct buf *bp;
1187 
1188 		LFS_SEGENTRY(sup, fs, i, bp);
1189 		changed = 0;
1190 		if (!ronly) {
1191 			if (sup->su_nbytes == 0 &&
1192 			    !(sup->su_flags & SEGUSE_EMPTY)) {
1193 				sup->su_flags |= SEGUSE_EMPTY;
1194 				++changed;
1195 			} else if (!(sup->su_nbytes == 0) &&
1196 				   (sup->su_flags & SEGUSE_EMPTY)) {
1197 				sup->su_flags &= ~SEGUSE_EMPTY;
1198 				++changed;
1199 			}
1200 			if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) {
1201 				sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL);
1202 				++changed;
1203 			}
1204 		}
1205 		fs->lfs_suflags[0][i] = sup->su_flags;
1206 		if (changed)
1207 			LFS_WRITESEGENTRY(sup, fs, i, bp);
1208 		else
1209 			brelse(bp, 0);
1210 	}
1211 
1212 	/*
1213 	 * XXX: if the fs has quotas, quotas should be on even if
1214 	 * readonly. Otherwise you can't query the quota info!
1215 	 * However, that's not how the quota2 code got written and I
1216 	 * don't know if it'll behave itself if enabled while
1217 	 * readonly, so for now use the same enable logic as ffs.
1218 	 *
1219 	 * XXX: also, if you use the -f behavior allowed here (and
1220 	 * equivalently above for remount) it will corrupt the fs. It
1221 	 * ought not to allow that. It should allow mounting readonly
1222 	 * if there are quotas and the kernel doesn't have the quota
1223 	 * code, but only readonly.
1224 	 *
1225 	 * XXX: and if you use the -f behavior allowed here it will
1226 	 * likely crash at unmount time (or remount time) because we
1227 	 * think quotas are active.
1228 	 *
1229 	 * Although none of this applies until there's a way to set
1230 	 * lfs_use_quota2 and have quotas in the fs at all.
1231 	 */
1232 	if (!ronly && fs->lfs_use_quota2) {
1233 #ifdef LFS_QUOTA2
1234 		error = lfs_quota2_mount(mp);
1235 #else
1236 		uprintf("%s: no kernel support for this filesystem's quotas\n",
1237 			mp->mnt_stat.f_mntonname);
1238 		if (mp->mnt_flag & MNT_FORCE) {
1239 			uprintf("%s: mounting anyway; fsck afterwards\n",
1240 				mp->mnt_stat.f_mntonname);
1241 		} else {
1242 			error = EINVAL;
1243 		}
1244 #endif
1245 		if (error) {
1246 			/* XXX XXX must clean up the stuff immediately above */
1247 			printf("lfs_mountfs: sorry, leaking some memory\n");
1248 			goto out;
1249 		}
1250 	}
1251 
1252 #ifdef LFS_KERNEL_RFW
1253 	lfs_roll_forward(fs, mp, l);
1254 #endif
1255 
1256 	/* If writing, sb is not clean; record in case of immediate crash */
1257 	if (!fs->lfs_ronly) {
1258 		lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) & ~LFS_PF_CLEAN);
1259 		lfs_writesuper(fs, lfs_sb_getsboff(fs, 0));
1260 		lfs_writesuper(fs, lfs_sb_getsboff(fs, 1));
1261 	}
1262 
1263 	/* Allow vget now that roll-forward is complete */
1264 	fs->lfs_flags &= ~(LFS_NOTYET);
1265 	wakeup(&fs->lfs_flags);
1266 
1267 	/*
1268 	 * Initialize the ifile cleaner info with information from
1269 	 * the superblock.
1270 	 */
1271 	{
1272 		struct buf *bp;
1273 
1274 		LFS_CLEANERINFO(cip, fs, bp);
1275 		lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs));
1276 		lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs));
1277 		lfs_ci_setavail(fs, cip, lfs_sb_getavail(fs));
1278 		lfs_ci_setbfree(fs, cip, lfs_sb_getbfree(fs));
1279 		(void) LFS_BWRITE_LOG(bp); /* Ifile */
1280 	}
1281 
1282 	/*
1283 	 * Mark the current segment as ACTIVE, since we're going to
1284 	 * be writing to it.
1285 	 */
1286 	{
1287 		struct buf *bp;
1288 
1289 		LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp);
1290 		sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1291 		fs->lfs_nactive++;
1292 		LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)), bp);  /* Ifile */
1293 	}
1294 
1295 	/* Now that roll-forward is done, unlock the Ifile */
1296 	vput(vp);
1297 
1298 	/* Start the pagedaemon-anticipating daemon */
1299 	mutex_enter(&lfs_lock);
1300 	if (lfs_writer_daemon == NULL &&
1301 	    kthread_create(PRI_BIO, 0, NULL,
1302 	    lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1303 		panic("fork lfs_writer");
1304 	mutex_exit(&lfs_lock);
1305 
1306 	printf("WARNING: the log-structured file system is experimental\n"
1307 	    "WARNING: it may cause system crashes and/or corrupt data\n");
1308 
1309 	return (0);
1310 
1311 out:
1312 	if (primarybuf)
1313 		brelse(primarybuf, BC_INVAL);
1314 	if (altbuf)
1315 		brelse(altbuf, BC_INVAL);
1316 	if (ump) {
1317 		kmem_free(ump->um_lfs, sizeof(struct lfs));
1318 		kmem_free(ump, sizeof(*ump));
1319 		mp->mnt_data = NULL;
1320 	}
1321 
1322 	return (error);
1323 }
1324 
1325 /*
1326  * unmount system call
1327  */
1328 int
1329 lfs_unmount(struct mount *mp, int mntflags)
1330 {
1331 	struct lwp *l = curlwp;
1332 	struct ulfsmount *ump;
1333 	struct lfs *fs;
1334 	int error, flags, ronly;
1335 	vnode_t *vp;
1336 
1337 	flags = 0;
1338 	if (mntflags & MNT_FORCE)
1339 		flags |= FORCECLOSE;
1340 
1341 	ump = VFSTOULFS(mp);
1342 	fs = ump->um_lfs;
1343 
1344 	/* Two checkpoints */
1345 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1346 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1347 
1348 	/* wake up the cleaner so it can die */
1349 	/* XXX: shouldn't this be *after* the error cases below? */
1350 	lfs_wakeup_cleaner(fs);
1351 	mutex_enter(&lfs_lock);
1352 	while (fs->lfs_sleepers)
1353 		cv_wait(&fs->lfs_sleeperscv, &lfs_lock);
1354 	mutex_exit(&lfs_lock);
1355 
1356 #ifdef LFS_EXTATTR
1357 	if (ump->um_fstype == ULFS1) {
1358 		if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) {
1359 			ulfs_extattr_stop(mp, curlwp);
1360 		}
1361 		if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED) {
1362 			ulfs_extattr_uepm_destroy(&ump->um_extattr);
1363 			mp->mnt_flag &= ~MNT_EXTATTR;
1364 		}
1365 	}
1366 #endif
1367 #ifdef LFS_QUOTA
1368         if ((error = lfsquota1_umount(mp, flags)) != 0)
1369 		return (error);
1370 #endif
1371 #ifdef LFS_QUOTA2
1372         if ((error = lfsquota2_umount(mp, flags)) != 0)
1373 		return (error);
1374 #endif
1375 	if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1376 		return (error);
1377 	if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0)
1378 		return (error);
1379 	vp = fs->lfs_ivnode;
1380 	mutex_enter(vp->v_interlock);
1381 	if (LIST_FIRST(&vp->v_dirtyblkhd))
1382 		panic("lfs_unmount: still dirty blocks on ifile vnode");
1383 	mutex_exit(vp->v_interlock);
1384 
1385 	/* Explicitly write the superblock, to update serial and pflags */
1386 	lfs_sb_setpflags(fs, lfs_sb_getpflags(fs) | LFS_PF_CLEAN);
1387 	lfs_writesuper(fs, lfs_sb_getsboff(fs, 0));
1388 	lfs_writesuper(fs, lfs_sb_getsboff(fs, 1));
1389 	mutex_enter(&lfs_lock);
1390 	while (fs->lfs_iocount)
1391 		mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0,
1392 			&lfs_lock);
1393 	mutex_exit(&lfs_lock);
1394 
1395 	/* Finish with the Ifile, now that we're done with it */
1396 	vgone(fs->lfs_ivnode);
1397 
1398 	ronly = !fs->lfs_ronly;
1399 	if (fs->lfs_devvp->v_type != VBAD)
1400 		spec_node_setmountedfs(fs->lfs_devvp, NULL);
1401 	vn_lock(fs->lfs_devvp, LK_EXCLUSIVE | LK_RETRY);
1402 	error = VOP_CLOSE(fs->lfs_devvp,
1403 	    ronly ? FREAD : FREAD|FWRITE, NOCRED);
1404 	vput(fs->lfs_devvp);
1405 
1406 	/* Complain about page leakage */
1407 	if (fs->lfs_pages > 0)
1408 		printf("lfs_unmount: still claim %d pages (%d in subsystem)\n",
1409 			fs->lfs_pages, lfs_subsys_pages);
1410 
1411 	/* Free per-mount data structures */
1412 	free(fs->lfs_ino_bitmap, M_SEGMENT);
1413 	free(fs->lfs_suflags[0], M_SEGMENT);
1414 	free(fs->lfs_suflags[1], M_SEGMENT);
1415 	free(fs->lfs_suflags, M_SEGMENT);
1416 	lfs_free_resblks(fs);
1417 	cv_destroy(&fs->lfs_sleeperscv);
1418 	cv_destroy(&fs->lfs_diropscv);
1419 	cv_destroy(&fs->lfs_stopcv);
1420 	cv_destroy(&fs->lfs_nextsegsleep);
1421 
1422 	rw_destroy(&fs->lfs_fraglock);
1423 	rw_destroy(&fs->lfs_iflock);
1424 
1425 	kmem_free(fs, sizeof(struct lfs));
1426 	kmem_free(ump, sizeof(*ump));
1427 
1428 	mp->mnt_data = NULL;
1429 	mp->mnt_flag &= ~MNT_LOCAL;
1430 	return (error);
1431 }
1432 
1433 /*
1434  * Get file system statistics.
1435  *
1436  * NB: We don't lock to access the superblock here, because it's not
1437  * really that important if we get it wrong.
1438  */
1439 int
1440 lfs_statvfs(struct mount *mp, struct statvfs *sbp)
1441 {
1442 	struct lfs *fs;
1443 	struct ulfsmount *ump;
1444 
1445 	ump = VFSTOULFS(mp);
1446 	fs = ump->um_lfs;
1447 
1448 	sbp->f_bsize = lfs_sb_getbsize(fs);
1449 	sbp->f_frsize = lfs_sb_getfsize(fs);
1450 	sbp->f_iosize = lfs_sb_getbsize(fs);
1451 	sbp->f_blocks = LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks;
1452 
1453 	sbp->f_bfree = LFS_EST_BFREE(fs);
1454 	/*
1455 	 * XXX this should be lfs_sb_getsize (measured in frags)
1456 	 * rather than dsize (measured in diskblocks). However,
1457 	 * getsize needs a format version check (for version 1 it
1458 	 * needs to be blockstofrags'd) so for the moment I'm going to
1459 	 * leave this...  it won't fire wrongly as frags are at least
1460 	 * as big as diskblocks.
1461 	 */
1462 	KASSERT(sbp->f_bfree <= lfs_sb_getdsize(fs));
1463 #if 0
1464 	if (sbp->f_bfree < 0)
1465 		sbp->f_bfree = 0;
1466 #endif
1467 
1468 	sbp->f_bresvd = LFS_EST_RSVD(fs);
1469 	if (sbp->f_bfree > sbp->f_bresvd)
1470 		sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
1471 	else
1472 		sbp->f_bavail = 0;
1473 
1474 	/* XXX: huh? - dholland 20150728 */
1475 	sbp->f_files = lfs_sb_getbfree(fs) / lfs_btofsb(fs, lfs_sb_getibsize(fs))
1476 	    * LFS_INOPB(fs);
1477 	sbp->f_ffree = sbp->f_files - lfs_sb_getnfiles(fs);
1478 	sbp->f_favail = sbp->f_ffree;
1479 	sbp->f_fresvd = 0;
1480 	copy_statvfs_info(sbp, mp);
1481 	return (0);
1482 }
1483 
1484 /*
1485  * Go through the disk queues to initiate sandbagged IO;
1486  * go through the inodes to write those that have been modified;
1487  * initiate the writing of the super block if it has been modified.
1488  *
1489  * Note: we are always called with the filesystem marked `MPBUSY'.
1490  */
1491 int
1492 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
1493 {
1494 	int error;
1495 	struct lfs *fs;
1496 
1497 	fs = VFSTOULFS(mp)->um_lfs;
1498 	if (fs->lfs_ronly)
1499 		return 0;
1500 
1501 	/* Snapshots should not hose the syncer */
1502 	/*
1503 	 * XXX Sync can block here anyway, since we don't have a very
1504 	 * XXX good idea of how much data is pending.  If it's more
1505 	 * XXX than a segment and lfs_nextseg is close to the end of
1506 	 * XXX the log, we'll likely block.
1507 	 */
1508 	mutex_enter(&lfs_lock);
1509 	if (fs->lfs_nowrap && lfs_sb_getnextseg(fs) < lfs_sb_getcurseg(fs)) {
1510 		mutex_exit(&lfs_lock);
1511 		return 0;
1512 	}
1513 	mutex_exit(&lfs_lock);
1514 
1515 	lfs_writer_enter(fs, "lfs_dirops");
1516 
1517 	/* All syncs must be checkpoints until roll-forward is implemented. */
1518 	DLOG((DLOG_FLUSH, "lfs_sync at 0x%jx\n",
1519 	      (uintmax_t)lfs_sb_getoffset(fs)));
1520 	error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1521 	lfs_writer_leave(fs);
1522 #ifdef LFS_QUOTA
1523 	lfs_qsync(mp);
1524 #endif
1525 	return (error);
1526 }
1527 
1528 /*
1529  * Look up an LFS dinode number to find its incore vnode.  If not already
1530  * in core, read it in from the specified device.  Return the inode locked.
1531  * Detection and handling of mount points must be done by the calling routine.
1532  */
1533 int
1534 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1535 {
1536 	int error;
1537 
1538 	error = vcache_get(mp, &ino, sizeof(ino), vpp);
1539 	if (error)
1540 		return error;
1541 	error = vn_lock(*vpp, LK_EXCLUSIVE);
1542 	if (error) {
1543 		vrele(*vpp);
1544 		*vpp = NULL;
1545 		return error;
1546 	}
1547 
1548 	return 0;
1549 }
1550 
1551 /*
1552  * Create a new vnode/inode pair and initialize what fields we can.
1553  */
1554 static void
1555 lfs_init_vnode(struct ulfsmount *ump, ino_t ino, struct vnode *vp)
1556 {
1557 	struct lfs *fs = ump->um_lfs;
1558 	struct inode *ip;
1559 	union lfs_dinode *dp;
1560 
1561 	ASSERT_NO_SEGLOCK(fs);
1562 
1563 	/* Initialize the inode. */
1564 	ip = pool_get(&lfs_inode_pool, PR_WAITOK);
1565 	memset(ip, 0, sizeof(*ip));
1566 	dp = pool_get(&lfs_dinode_pool, PR_WAITOK);
1567 	memset(dp, 0, sizeof(*dp));
1568 	ip->inode_ext.lfs = pool_get(&lfs_inoext_pool, PR_WAITOK);
1569 	memset(ip->inode_ext.lfs, 0, sizeof(*ip->inode_ext.lfs));
1570 	ip->i_din = dp;
1571 	ip->i_ump = ump;
1572 	ip->i_vnode = vp;
1573 	ip->i_dev = fs->lfs_dev;
1574 	lfs_dino_setinumber(fs, dp, ino);
1575 	ip->i_number = ino;
1576 	ip->i_lfs = fs;
1577 	ip->i_lfs_effnblks = 0;
1578 	SPLAY_INIT(&ip->i_lfs_lbtree);
1579 	ip->i_lfs_nbtree = 0;
1580 	LIST_INIT(&ip->i_lfs_segdhd);
1581 
1582 	vp->v_tag = VT_LFS;
1583 	vp->v_op = lfs_vnodeop_p;
1584 	vp->v_data = ip;
1585 }
1586 
1587 /*
1588  * Undo lfs_init_vnode().
1589  */
1590 static void
1591 lfs_deinit_vnode(struct ulfsmount *ump, struct vnode *vp)
1592 {
1593 	struct inode *ip = VTOI(vp);
1594 
1595 	pool_put(&lfs_inoext_pool, ip->inode_ext.lfs);
1596 	pool_put(&lfs_dinode_pool, ip->i_din);
1597 	pool_put(&lfs_inode_pool, ip);
1598 	vp->v_data = NULL;
1599 }
1600 
1601 /*
1602  * Read an inode from disk and initialize this vnode / inode pair.
1603  * Caller assures no other thread will try to load this inode.
1604  */
1605 int
1606 lfs_loadvnode(struct mount *mp, struct vnode *vp,
1607     const void *key, size_t key_len, const void **new_key)
1608 {
1609 	struct lfs *fs;
1610 	union lfs_dinode *dip;
1611 	struct inode *ip;
1612 	struct buf *bp;
1613 	IFILE *ifp;
1614 	struct ulfsmount *ump;
1615 	ino_t ino;
1616 	daddr_t daddr;
1617 	int error, retries;
1618 	struct timespec ts;
1619 
1620 	KASSERT(key_len == sizeof(ino));
1621 	memcpy(&ino, key, key_len);
1622 
1623 	memset(&ts, 0, sizeof ts);	/* XXX gcc */
1624 
1625 	ump = VFSTOULFS(mp);
1626 	fs = ump->um_lfs;
1627 
1628 	/*
1629 	 * If the filesystem is not completely mounted yet, suspend
1630 	 * any access requests (wait for roll-forward to complete).
1631 	 */
1632 	mutex_enter(&lfs_lock);
1633 	while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1634 		mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0,
1635 			&lfs_lock);
1636 	mutex_exit(&lfs_lock);
1637 
1638 	/* Translate the inode number to a disk address. */
1639 	if (ino == LFS_IFILE_INUM)
1640 		daddr = lfs_sb_getidaddr(fs);
1641 	else {
1642 		/* XXX bounds-check this too */
1643 		LFS_IENTRY(ifp, fs, ino, bp);
1644 		daddr = lfs_if_getdaddr(fs, ifp);
1645 		if (lfs_sb_getversion(fs) > 1) {
1646 			ts.tv_sec = lfs_if_getatime_sec(fs, ifp);
1647 			ts.tv_nsec = lfs_if_getatime_nsec(fs, ifp);
1648 		}
1649 
1650 		brelse(bp, 0);
1651 		if (daddr == LFS_UNUSED_DADDR)
1652 			return (ENOENT);
1653 	}
1654 
1655 	/* Allocate/init new vnode/inode. */
1656 	lfs_init_vnode(ump, ino, vp);
1657 	ip = VTOI(vp);
1658 
1659 	/* If the cleaner supplied the inode, use it. */
1660 	if (curlwp == fs->lfs_cleaner_thread && fs->lfs_cleaner_hint != NULL &&
1661 	    fs->lfs_cleaner_hint->bi_lbn == LFS_UNUSED_LBN) {
1662 		dip = fs->lfs_cleaner_hint->bi_bp;
1663 		if (fs->lfs_is64) {
1664 			error = copyin(dip, &ip->i_din->u_64,
1665 				       sizeof(struct lfs64_dinode));
1666 		} else {
1667 			error = copyin(dip, &ip->i_din->u_32,
1668 				       sizeof(struct lfs32_dinode));
1669 		}
1670 		if (error) {
1671 			lfs_deinit_vnode(ump, vp);
1672 			return error;
1673 		}
1674 		KASSERT(ip->i_number == ino);
1675 		goto out;
1676 	}
1677 
1678 	/* Read in the disk contents for the inode, copy into the inode. */
1679 	retries = 0;
1680 again:
1681 	error = bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr),
1682 		(lfs_sb_getversion(fs) == 1 ? lfs_sb_getbsize(fs) : lfs_sb_getibsize(fs)),
1683 		0, &bp);
1684 	if (error) {
1685 		lfs_deinit_vnode(ump, vp);
1686 		return error;
1687 	}
1688 
1689 	dip = lfs_ifind(fs, ino, bp);
1690 	if (dip == NULL) {
1691 		/* Assume write has not completed yet; try again */
1692 		brelse(bp, BC_INVAL);
1693 		++retries;
1694 		if (retries <= LFS_IFIND_RETRIES) {
1695 			mutex_enter(&lfs_lock);
1696 			if (fs->lfs_iocount) {
1697 				DLOG((DLOG_VNODE,
1698 				    "%s: dinode %d not found, retrying...\n",
1699 				    __func__, ino));
1700 				(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
1701 					      "lfs ifind", 1, &lfs_lock);
1702 			} else
1703 				retries = LFS_IFIND_RETRIES;
1704 			mutex_exit(&lfs_lock);
1705 			goto again;
1706 		}
1707 #ifdef DEBUG
1708 		/* If the seglock is held look at the bpp to see
1709 		   what is there anyway */
1710 		mutex_enter(&lfs_lock);
1711 		if (fs->lfs_seglock > 0) {
1712 			struct buf **bpp;
1713 			union lfs_dinode *dp;
1714 			int i;
1715 
1716 			for (bpp = fs->lfs_sp->bpp;
1717 			     bpp != fs->lfs_sp->cbpp; ++bpp) {
1718 				if ((*bpp)->b_vp == fs->lfs_ivnode &&
1719 				    bpp != fs->lfs_sp->bpp) {
1720 					/* Inode block */
1721 					printf("%s: block 0x%" PRIx64 ": ",
1722 					       __func__, (*bpp)->b_blkno);
1723 					for (i = 0; i < LFS_INOPB(fs); i++) {
1724 						dp = DINO_IN_BLOCK(fs,
1725 						    (*bpp)->b_data, i);
1726 						if (lfs_dino_getinumber(fs, dp))
1727 							printf("%ju ",
1728 							    (uintmax_t)lfs_dino_getinumber(fs, dp));
1729 					}
1730 					printf("\n");
1731 				}
1732 			}
1733 		}
1734 		mutex_exit(&lfs_lock);
1735 #endif /* DEBUG */
1736 		panic("lfs_loadvnode: dinode not found");
1737 	}
1738 	lfs_copy_dinode(fs, ip->i_din, dip);
1739 	brelse(bp, 0);
1740 
1741 out:
1742 	if (lfs_sb_getversion(fs) > 1) {
1743 		lfs_dino_setatime(fs, ip->i_din, ts.tv_sec);
1744 		lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec);
1745 	}
1746 
1747 	lfs_vinit(mp, &vp);
1748 
1749 	*new_key = &ip->i_number;
1750 	return 0;
1751 }
1752 
1753 /*
1754  * Create a new inode and initialize this vnode / inode pair.
1755  */
1756 int
1757 lfs_newvnode(struct mount *mp, struct vnode *dvp, struct vnode *vp,
1758     struct vattr *vap, kauth_cred_t cred,
1759     size_t *key_len, const void **new_key)
1760 {
1761 	ino_t ino;
1762 	struct inode *ip;
1763 	struct ulfsmount *ump;
1764 	struct lfs *fs;
1765 	int error, mode, gen;
1766 
1767 	KASSERT(dvp != NULL || vap->va_fileid > 0);
1768 	KASSERT(dvp != NULL && dvp->v_mount == mp);
1769 	KASSERT(vap->va_type != VNON);
1770 
1771 	*key_len = sizeof(ino);
1772 	ump = VFSTOULFS(mp);
1773 	fs = ump->um_lfs;
1774 	mode = MAKEIMODE(vap->va_type, vap->va_mode);
1775 
1776 	/*
1777 	 * Allocate fresh inode.  With "dvp == NULL" take the inode number
1778 	 * and version from "vap".
1779 	*/
1780 	if (dvp == NULL) {
1781 		ino = vap->va_fileid;
1782 		gen = vap->va_gen;
1783 		error = lfs_valloc_fixed(fs, ino, gen);
1784 	} else {
1785 		error = lfs_valloc(dvp, mode, cred, &ino, &gen);
1786 	}
1787 	if (error)
1788 		return error;
1789 
1790 	/* Attach inode to vnode. */
1791 	lfs_init_vnode(ump, ino, vp);
1792 	ip = VTOI(vp);
1793 
1794 	mutex_enter(&lfs_lock);
1795 	LFS_SET_UINO(ip, IN_CHANGE);
1796 	mutex_exit(&lfs_lock);
1797 
1798 	/* Note no blocks yet */
1799 	ip->i_lfs_hiblk = -1;
1800 
1801 	/* Set a new generation number for this inode. */
1802 	ip->i_gen = gen;
1803 	lfs_dino_setgen(fs, ip->i_din, gen);
1804 
1805 	memset(ip->i_lfs_fragsize, 0,
1806 	    ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize));
1807 
1808 	/* Set uid / gid. */
1809 	if (cred == NOCRED || cred == FSCRED) {
1810 		ip->i_gid = 0;
1811 		ip->i_uid = 0;
1812 	} else {
1813 		ip->i_gid = VTOI(dvp)->i_gid;
1814 		ip->i_uid = kauth_cred_geteuid(cred);
1815 	}
1816 	DIP_ASSIGN(ip, gid, ip->i_gid);
1817 	DIP_ASSIGN(ip, uid, ip->i_uid);
1818 
1819 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
1820 	error = lfs_chkiq(ip, 1, cred, 0);
1821 	if (error) {
1822 		lfs_vfree(dvp, ino, mode);
1823 		lfs_deinit_vnode(ump, vp);
1824 
1825 		return error;
1826 	}
1827 #endif
1828 
1829 	/* Set type and finalize. */
1830 	ip->i_flags = 0;
1831 	DIP_ASSIGN(ip, flags, 0);
1832 	ip->i_mode = mode;
1833 	DIP_ASSIGN(ip, mode, mode);
1834 	if (vap->va_rdev != VNOVAL) {
1835 		/*
1836 		 * Want to be able to use this to make badblock
1837 		 * inodes, so don't truncate the dev number.
1838 		 */
1839 		// XXX clean this up
1840 		if (ump->um_fstype == ULFS1)
1841 			ip->i_din->u_32.di_rdev = ulfs_rw32(vap->va_rdev,
1842 			    ULFS_MPNEEDSWAP(fs));
1843 		else
1844 			ip->i_din->u_64.di_rdev = ulfs_rw64(vap->va_rdev,
1845 			    ULFS_MPNEEDSWAP(fs));
1846 	}
1847 	lfs_vinit(mp, &vp);
1848 
1849 	*new_key = &ip->i_number;
1850 	return 0;
1851 }
1852 
1853 /*
1854  * File handle to vnode
1855  */
1856 int
1857 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1858 {
1859 	struct lfid lfh;
1860 	struct lfs *fs;
1861 
1862 	if (fhp->fid_len != sizeof(struct lfid))
1863 		return EINVAL;
1864 
1865 	memcpy(&lfh, fhp, sizeof(lfh));
1866 	if (lfh.lfid_ino < LFS_IFILE_INUM)
1867 		return ESTALE;
1868 
1869 	fs = VFSTOULFS(mp)->um_lfs;
1870 	if (lfh.lfid_ident != lfs_sb_getident(fs))
1871 		return ESTALE;
1872 
1873 	if (lfh.lfid_ino >
1874 	    ((lfs_dino_getsize(fs, VTOI(fs->lfs_ivnode)->i_din) >> lfs_sb_getbshift(fs)) -
1875 	     lfs_sb_getcleansz(fs) - lfs_sb_getsegtabsz(fs)) * lfs_sb_getifpb(fs))
1876 		return ESTALE;
1877 
1878 	return (ulfs_fhtovp(mp, &lfh.lfid_ufid, vpp));
1879 }
1880 
1881 /*
1882  * Vnode pointer to File handle
1883  */
1884 /* ARGSUSED */
1885 int
1886 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size)
1887 {
1888 	struct inode *ip;
1889 	struct lfid lfh;
1890 
1891 	if (*fh_size < sizeof(struct lfid)) {
1892 		*fh_size = sizeof(struct lfid);
1893 		return E2BIG;
1894 	}
1895 	*fh_size = sizeof(struct lfid);
1896 	ip = VTOI(vp);
1897 	memset(&lfh, 0, sizeof(lfh));
1898 	lfh.lfid_len = sizeof(struct lfid);
1899 	lfh.lfid_ino = ip->i_number;
1900 	lfh.lfid_gen = ip->i_gen;
1901 	lfh.lfid_ident = lfs_sb_getident(ip->i_lfs);
1902 	memcpy(fhp, &lfh, sizeof(lfh));
1903 	return (0);
1904 }
1905 
1906 /*
1907  * ulfs_bmaparray callback function for writing.
1908  *
1909  * Since blocks will be written to the new segment anyway,
1910  * we don't care about current daddr of them.
1911  */
1912 static bool
1913 lfs_issequential_hole(const struct lfs *fs,
1914     daddr_t daddr0, daddr_t daddr1)
1915 {
1916 	(void)fs; /* not used */
1917 
1918 	KASSERT(daddr0 == UNWRITTEN ||
1919 	    (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR(fs)));
1920 	KASSERT(daddr1 == UNWRITTEN ||
1921 	    (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR(fs)));
1922 
1923 	/* NOTE: all we want to know here is 'hole or not'. */
1924 	/* NOTE: UNASSIGNED is converted to 0 by ulfs_bmaparray. */
1925 
1926 	/*
1927 	 * treat UNWRITTENs and all resident blocks as 'contiguous'
1928 	 */
1929 	if (daddr0 != 0 && daddr1 != 0)
1930 		return true;
1931 
1932 	/*
1933 	 * both are in hole?
1934 	 */
1935 	if (daddr0 == 0 && daddr1 == 0)
1936 		return true; /* all holes are 'contiguous' for us. */
1937 
1938 	return false;
1939 }
1940 
1941 /*
1942  * lfs_gop_write functions exactly like genfs_gop_write, except that
1943  * (1) it requires the seglock to be held by its caller, and sp->fip
1944  *     to be properly initialized (it will return without re-initializing
1945  *     sp->fip, and without calling lfs_writeseg).
1946  * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1947  *     to determine how large a block it can write at once (though it does
1948  *     still use VOP_BMAP to find holes in the file);
1949  * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1950  *     (leaving lfs_writeseg to deal with the cluster blocks, so we might
1951  *     now have clusters of clusters, ick.)
1952  */
1953 static int
1954 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages,
1955     int flags)
1956 {
1957 	int i, error, run, haveeof = 0;
1958 	int fs_bshift;
1959 	vaddr_t kva;
1960 	off_t eof, offset, startoffset = 0;
1961 	size_t bytes, iobytes, skipbytes;
1962 	bool async = (flags & PGO_SYNCIO) == 0;
1963 	daddr_t lbn, blkno;
1964 	struct vm_page *pg;
1965 	struct buf *mbp, *bp;
1966 	struct vnode *devvp = VTOI(vp)->i_devvp;
1967 	struct inode *ip = VTOI(vp);
1968 	struct lfs *fs = ip->i_lfs;
1969 	struct segment *sp = fs->lfs_sp;
1970 	SEGSUM *ssp;
1971 	UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1972 	const char * failreason = NULL;
1973 
1974 	ASSERT_SEGLOCK(fs);
1975 
1976 	/* The Ifile lives in the buffer cache */
1977 	KASSERT(vp != fs->lfs_ivnode);
1978 
1979 	/*
1980 	 * We don't want to fill the disk before the cleaner has a chance
1981 	 * to make room for us.  If we're in danger of doing that, fail
1982 	 * with EAGAIN.  The caller will have to notice this, unlock
1983 	 * so the cleaner can run, relock and try again.
1984 	 *
1985 	 * We must write everything, however, if our vnode is being
1986 	 * reclaimed.
1987 	 */
1988 	mutex_enter(vp->v_interlock);
1989 	if (LFS_STARVED_FOR_SEGS(fs) && vdead_check(vp, VDEAD_NOWAIT) == 0) {
1990 		mutex_exit(vp->v_interlock);
1991 		failreason = "Starved for segs and not flushing vp";
1992  		goto tryagain;
1993 	}
1994 	mutex_exit(vp->v_interlock);
1995 
1996 	/*
1997 	 * Sometimes things slip past the filters in lfs_putpages,
1998 	 * and the pagedaemon tries to write pages---problem is
1999 	 * that the pagedaemon never acquires the segment lock.
2000 	 *
2001 	 * Alternatively, pages that were clean when we called
2002 	 * genfs_putpages may have become dirty in the meantime.  In this
2003 	 * case the segment header is not properly set up for blocks
2004 	 * to be added to it.
2005 	 *
2006 	 * Unbusy and unclean the pages, and put them on the ACTIVE
2007 	 * queue under the hypothesis that they couldn't have got here
2008 	 * unless they were modified *quite* recently.
2009 	 *
2010 	 * XXXUBC that last statement is an oversimplification of course.
2011 	 */
2012 	if (!LFS_SEGLOCK_HELD(fs)) {
2013 		failreason = "Seglock not held";
2014 		goto tryagain;
2015 	}
2016 	if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) {
2017 		failreason = "Inode with no_gop_write";
2018 		goto tryagain;
2019 	}
2020 	if ((pgs[0]->offset & lfs_sb_getbmask(fs)) != 0) {
2021 		failreason = "Bad page offset";
2022 		goto tryagain;
2023 	}
2024 
2025 	UVMHIST_LOG(ubchist, "vp %#jx pgs %#jx npages %jd flags 0x%jx",
2026 	    (uintptr_t)vp, (uintptr_t)pgs, npages, flags);
2027 
2028 	GOP_SIZE(vp, vp->v_size, &eof, 0);
2029 	haveeof = 1;
2030 
2031 	if (vp->v_type == VREG)
2032 		fs_bshift = vp->v_mount->mnt_fs_bshift;
2033 	else
2034 		fs_bshift = DEV_BSHIFT;
2035 	error = 0;
2036 	pg = pgs[0];
2037 	startoffset = pg->offset;
2038 	KASSERT(eof >= 0);
2039 
2040 	if (startoffset >= eof) {
2041 		failreason = "Offset beyond EOF";
2042 		goto tryagain;
2043 	} else
2044 		bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
2045 	skipbytes = 0;
2046 
2047 	KASSERT(bytes != 0);
2048 
2049 	/* Swap PG_DELWRI for PG_PAGEOUT */
2050 	for (i = 0; i < npages; i++) {
2051 		if (pgs[i]->flags & PG_DELWRI) {
2052 			KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
2053 			pgs[i]->flags &= ~PG_DELWRI;
2054 			pgs[i]->flags |= PG_PAGEOUT;
2055 			uvm_pageout_start(1);
2056 			mutex_enter(vp->v_interlock);
2057 			mutex_enter(&uvm_pageqlock);
2058 			uvm_pageunwire(pgs[i]);
2059 			mutex_exit(&uvm_pageqlock);
2060 			mutex_exit(vp->v_interlock);
2061 		}
2062 	}
2063 
2064 	/*
2065 	 * Check to make sure we're starting on a block boundary.
2066 	 * We'll check later to make sure we always write entire
2067 	 * blocks (or fragments).
2068 	 */
2069 	if (startoffset & lfs_sb_getbmask(fs))
2070 		printf("%" PRId64 " & %" PRIu64 " = %" PRId64 "\n",
2071 		       startoffset, lfs_sb_getbmask(fs),
2072 		       startoffset & lfs_sb_getbmask(fs));
2073 	KASSERT((startoffset & lfs_sb_getbmask(fs)) == 0);
2074 	if (bytes & lfs_sb_getffmask(fs)) {
2075 		printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
2076 		panic("lfs_gop_write: non-integer blocks");
2077 	}
2078 
2079 	/*
2080 	 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK.
2081 	 * If we would, write what we have and try again.  If we don't
2082 	 * have anything to write, we'll have to sleep.
2083 	 */
2084 	ssp = (SEGSUM *)sp->segsum;
2085 	if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
2086 				      (lfs_ss_getnfinfo(fs, ssp) < 1 ?
2087 				       UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) {
2088 		DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n"));
2089 #if 0
2090 		      " with nfinfo=%d at offset 0x%jx\n",
2091 		      (int)lfs_ss_getnfinfo(fs, ssp),
2092 		      (uintmax_t)lfs_sb_getoffset(fs)));
2093 #endif
2094 		lfs_updatemeta(sp);
2095 		lfs_release_finfo(fs);
2096 		(void) lfs_writeseg(fs, sp);
2097 
2098 		lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
2099 
2100 		/*
2101 		 * Having given up all of the pager_map we were holding,
2102 		 * we can now wait for aiodoned to reclaim it for us
2103 		 * without fear of deadlock.
2104 		 */
2105 		kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
2106 				     UVMPAGER_MAPIN_WAITOK);
2107 	}
2108 
2109 	mbp = getiobuf(NULL, true);
2110 	UVMHIST_LOG(ubchist, "vp %#jx mbp %#jx num now %jd bytes 0x%jx",
2111 	    (uintptr_t)vp, (uintptr_t)mbp, vp->v_numoutput, bytes);
2112 	mbp->b_bufsize = npages << PAGE_SHIFT;
2113 	mbp->b_data = (void *)kva;
2114 	mbp->b_resid = mbp->b_bcount = bytes;
2115 	mbp->b_cflags = BC_BUSY|BC_AGE;
2116 	mbp->b_iodone = uvm_aio_biodone;
2117 
2118 	bp = NULL;
2119 	for (offset = startoffset;
2120 	    bytes > 0;
2121 	    offset += iobytes, bytes -= iobytes) {
2122 		lbn = offset >> fs_bshift;
2123 		error = ulfs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
2124 		    lfs_issequential_hole);
2125 		if (error) {
2126 			UVMHIST_LOG(ubchist, "ulfs_bmaparray() -> %jd",
2127 			    error,0,0,0);
2128 			skipbytes += bytes;
2129 			bytes = 0;
2130 			break;
2131 		}
2132 
2133 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
2134 		    bytes);
2135 		if (blkno == (daddr_t)-1) {
2136 			skipbytes += iobytes;
2137 			continue;
2138 		}
2139 
2140 		/*
2141 		 * Discover how much we can really pack into this buffer.
2142 		 */
2143 		/* If no room in the current segment, finish it up */
2144 		if (sp->sum_bytes_left < sizeof(int32_t) ||
2145 		    sp->seg_bytes_left < (1 << lfs_sb_getbshift(fs))) {
2146 			int vers;
2147 
2148 			lfs_updatemeta(sp);
2149 			vers = lfs_fi_getversion(fs, sp->fip);
2150 			lfs_release_finfo(fs);
2151 			(void) lfs_writeseg(fs, sp);
2152 
2153 			lfs_acquire_finfo(fs, ip->i_number, vers);
2154 		}
2155 		/* Check both for space in segment and space in segsum */
2156 		iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
2157 					<< fs_bshift);
2158 		iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
2159 				       << fs_bshift);
2160 		KASSERT(iobytes > 0);
2161 
2162 		/* if it's really one i/o, don't make a second buf */
2163 		if (offset == startoffset && iobytes == bytes) {
2164 			bp = mbp;
2165 			/*
2166 			 * All the LFS output is done by the segwriter.  It
2167 			 * will increment numoutput by one for all the bufs it
2168 			 * recieves.  However this buffer needs one extra to
2169 			 * account for aiodone.
2170 			 */
2171 			mutex_enter(vp->v_interlock);
2172 			vp->v_numoutput++;
2173 			mutex_exit(vp->v_interlock);
2174 		} else {
2175 			bp = getiobuf(NULL, true);
2176 			UVMHIST_LOG(ubchist, "vp %#jx bp %#jx num now %jd",
2177 			    (uintptr_t)vp, (uintptr_t)bp, vp->v_numoutput, 0);
2178 			nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes);
2179 			/*
2180 			 * LFS doesn't like async I/O here, dies with
2181 			 * an assert in lfs_bwrite().  Is that assert
2182 			 * valid?  I retained non-async behaviour when
2183 			 * converted this to use nestiobuf --pooka
2184 			 */
2185 			bp->b_flags &= ~B_ASYNC;
2186 		}
2187 
2188 		/* XXX This is silly ... is this necessary? */
2189 		mutex_enter(&bufcache_lock);
2190 		mutex_enter(vp->v_interlock);
2191 		bgetvp(vp, bp);
2192 		mutex_exit(vp->v_interlock);
2193 		mutex_exit(&bufcache_lock);
2194 
2195 		bp->b_lblkno = lfs_lblkno(fs, offset);
2196 		bp->b_private = mbp;
2197 		if (devvp->v_type == VBLK) {
2198 			bp->b_dev = devvp->v_rdev;
2199 		}
2200 		VOP_BWRITE(bp->b_vp, bp);
2201 		while (lfs_gatherblock(sp, bp, NULL))
2202 			continue;
2203 	}
2204 
2205 	nestiobuf_done(mbp, skipbytes, error);
2206 	if (skipbytes) {
2207 		UVMHIST_LOG(ubchist, "skipbytes %jd", skipbytes, 0,0,0);
2208 	}
2209 	UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
2210 
2211 	if (!async) {
2212 		/* Start a segment write. */
2213 		UVMHIST_LOG(ubchist, "flushing", 0,0,0,0);
2214 		mutex_enter(&lfs_lock);
2215 		lfs_flush(fs, 0, 1);
2216 		mutex_exit(&lfs_lock);
2217 	}
2218 
2219 	if ((sp->seg_flags & SEGM_SINGLE) && lfs_sb_getcurseg(fs) != fs->lfs_startseg)
2220 		return EAGAIN;
2221 
2222 	return (0);
2223 
2224     tryagain:
2225 	/*
2226 	 * We can't write the pages, for whatever reason.
2227 	 * Clean up after ourselves, and make the caller try again.
2228 	 */
2229 	mutex_enter(vp->v_interlock);
2230 
2231 	/* Tell why we're here, if we know */
2232 	if (failreason != NULL) {
2233 		DLOG((DLOG_PAGE, "lfs_gop_write: %s\n", failreason));
2234 	}
2235 	if (haveeof && startoffset >= eof) {
2236  		DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64
2237  		      " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number,
2238  		      pgs[0]->offset, eof, npages));
2239 	}
2240 
2241 	mutex_enter(&uvm_pageqlock);
2242 	for (i = 0; i < npages; i++) {
2243 		pg = pgs[i];
2244 
2245 		if (pg->flags & PG_PAGEOUT)
2246 			uvm_pageout_done(1);
2247 		if (pg->flags & PG_DELWRI) {
2248 			uvm_pageunwire(pg);
2249 		}
2250 		uvm_pageactivate(pg);
2251 		pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
2252 		DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg,
2253 			vp, pg->offset));
2254 		DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags));
2255 		DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags));
2256 		DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon));
2257 		DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject));
2258 		DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i,
2259 		      pg->wire_count));
2260 		DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i,
2261 		      pg->loan_count));
2262 	}
2263 	/* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
2264 	uvm_page_unbusy(pgs, npages);
2265 	mutex_exit(&uvm_pageqlock);
2266 	mutex_exit(vp->v_interlock);
2267 	return EAGAIN;
2268 }
2269 
2270 /*
2271  * finish vnode/inode initialization.
2272  * used by lfs_vget.
2273  */
2274 void
2275 lfs_vinit(struct mount *mp, struct vnode **vpp)
2276 {
2277 	struct vnode *vp = *vpp;
2278 	struct inode *ip = VTOI(vp);
2279 	struct ulfsmount *ump = VFSTOULFS(mp);
2280 	struct lfs *fs = ump->um_lfs;
2281 	int i;
2282 
2283 	ip->i_mode = lfs_dino_getmode(fs, ip->i_din);
2284 	ip->i_nlink = lfs_dino_getnlink(fs, ip->i_din);
2285 	ip->i_lfs_osize = ip->i_size = lfs_dino_getsize(fs, ip->i_din);
2286 	ip->i_flags = lfs_dino_getflags(fs, ip->i_din);
2287 	ip->i_gen = lfs_dino_getgen(fs, ip->i_din);
2288 	ip->i_uid = lfs_dino_getuid(fs, ip->i_din);
2289 	ip->i_gid = lfs_dino_getgid(fs, ip->i_din);
2290 
2291 	ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din);
2292 	ip->i_lfs_odnlink = lfs_dino_getnlink(fs, ip->i_din);
2293 
2294 	/*
2295 	 * Initialize the vnode from the inode, check for aliases.  In all
2296 	 * cases re-init ip, the underlying vnode/inode may have changed.
2297 	 */
2298 	ulfs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
2299 	ip = VTOI(vp);
2300 
2301 	memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize));
2302 	if (vp->v_type != VLNK || ip->i_size >= ip->i_lfs->um_maxsymlinklen) {
2303 #ifdef DEBUG
2304 		for (i = (ip->i_size + lfs_sb_getbsize(fs) - 1) >> lfs_sb_getbshift(fs);
2305 		    i < ULFS_NDADDR; i++) {
2306 			if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
2307 			    i == 0)
2308 				continue;
2309 			if (lfs_dino_getdb(fs, ip->i_din, i) != 0) {
2310 				lfs_dump_dinode(fs, ip->i_din);
2311 				panic("inconsistent inode (direct)");
2312 			}
2313 		}
2314 		for ( ; i < ULFS_NDADDR + ULFS_NIADDR; i++) {
2315 			if (lfs_dino_getib(fs, ip->i_din, i - ULFS_NDADDR) != 0) {
2316 				lfs_dump_dinode(fs, ip->i_din);
2317 				panic("inconsistent inode (indirect)");
2318 			}
2319 		}
2320 #endif /* DEBUG */
2321 		for (i = 0; i < ULFS_NDADDR; i++)
2322 			if (lfs_dino_getdb(fs, ip->i_din, i) != 0)
2323 				ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i);
2324 	}
2325 
2326 	KASSERTMSG((vp->v_type != VNON),
2327 	    "lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n",
2328 	    (unsigned long long)ip->i_number,
2329 	    (ip->i_mode & LFS_IFMT) >> 12);
2330 
2331 	/*
2332 	 * Finish inode initialization now that aliasing has been resolved.
2333 	 */
2334 
2335 	ip->i_devvp = fs->lfs_devvp;
2336 	vref(ip->i_devvp);
2337 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
2338 	ulfsquota_init(ip);
2339 #endif
2340 	genfs_node_init(vp, &lfs_genfsops);
2341 	uvm_vnp_setsize(vp, ip->i_size);
2342 
2343 	/* Initialize hiblk from file size */
2344 	ip->i_lfs_hiblk = lfs_lblkno(ip->i_lfs, ip->i_size + lfs_sb_getbsize(ip->i_lfs) - 1) - 1;
2345 
2346 	*vpp = vp;
2347 }
2348 
2349 /*
2350  * Resize the filesystem to contain the specified number of segments.
2351  */
2352 int
2353 lfs_resize_fs(struct lfs *fs, int newnsegs)
2354 {
2355 	SEGUSE *sup;
2356 	CLEANERINFO *cip;
2357 	struct buf *bp, *obp;
2358 	daddr_t olast, nlast, ilast, noff, start, end;
2359 	struct vnode *ivp;
2360 	struct inode *ip;
2361 	int error, badnews, inc, oldnsegs;
2362 	int sbbytes, csbbytes, gain, cgain;
2363 	int i;
2364 
2365 	/* Only support v2 and up */
2366 	if (lfs_sb_getversion(fs) < 2)
2367 		return EOPNOTSUPP;
2368 
2369 	/* If we're doing nothing, do it fast */
2370 	oldnsegs = lfs_sb_getnseg(fs);
2371 	if (newnsegs == oldnsegs)
2372 		return 0;
2373 
2374 	/* We always have to have two superblocks */
2375 	if (newnsegs <= lfs_dtosn(fs, lfs_sb_getsboff(fs, 1)))
2376 		/* XXX this error code is rather nonsense */
2377 		return EFBIG;
2378 
2379 	ivp = fs->lfs_ivnode;
2380 	ip = VTOI(ivp);
2381 	error = 0;
2382 
2383 	/* Take the segment lock so no one else calls lfs_newseg() */
2384 	lfs_seglock(fs, SEGM_PROT);
2385 
2386 	/*
2387 	 * Make sure the segments we're going to be losing, if any,
2388 	 * are in fact empty.  We hold the seglock, so their status
2389 	 * cannot change underneath us.  Count the superblocks we lose,
2390 	 * while we're at it.
2391 	 */
2392 	sbbytes = csbbytes = 0;
2393 	cgain = 0;
2394 	for (i = newnsegs; i < oldnsegs; i++) {
2395 		LFS_SEGENTRY(sup, fs, i, bp);
2396 		badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL);
2397 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
2398 			sbbytes += LFS_SBPAD;
2399 		if (!(sup->su_flags & SEGUSE_DIRTY)) {
2400 			++cgain;
2401 			if (sup->su_flags & SEGUSE_SUPERBLOCK)
2402 				csbbytes += LFS_SBPAD;
2403 		}
2404 		brelse(bp, 0);
2405 		if (badnews) {
2406 			error = EBUSY;
2407 			goto out;
2408 		}
2409 	}
2410 
2411 	/* Note old and new segment table endpoints, and old ifile size */
2412 	olast = lfs_sb_getcleansz(fs) + lfs_sb_getsegtabsz(fs);
2413 	nlast = howmany(newnsegs, lfs_sb_getsepb(fs)) + lfs_sb_getcleansz(fs);
2414 	ilast = ivp->v_size >> lfs_sb_getbshift(fs);
2415 	noff = nlast - olast;
2416 
2417 	/*
2418 	 * Make sure no one can use the Ifile while we change it around.
2419 	 * Even after taking the iflock we need to make sure no one still
2420 	 * is holding Ifile buffers, so we get each one, to drain them.
2421 	 * (XXX this could be done better.)
2422 	 */
2423 	rw_enter(&fs->lfs_iflock, RW_WRITER);
2424 	for (i = 0; i < ilast; i++) {
2425 		/* XXX what to do if bread fails? */
2426 		bread(ivp, i, lfs_sb_getbsize(fs), 0, &bp);
2427 		brelse(bp, 0);
2428 	}
2429 
2430 	/* Allocate new Ifile blocks */
2431 	for (i = ilast; i < ilast + noff; i++) {
2432 		if (lfs_balloc(ivp, i * lfs_sb_getbsize(fs), lfs_sb_getbsize(fs), NOCRED, 0,
2433 			       &bp) != 0)
2434 			panic("balloc extending ifile");
2435 		memset(bp->b_data, 0, lfs_sb_getbsize(fs));
2436 		VOP_BWRITE(bp->b_vp, bp);
2437 	}
2438 
2439 	/* Register new ifile size */
2440 	ip->i_size += noff * lfs_sb_getbsize(fs);
2441 	lfs_dino_setsize(fs, ip->i_din, ip->i_size);
2442 	uvm_vnp_setsize(ivp, ip->i_size);
2443 
2444 	/* Copy the inode table to its new position */
2445 	if (noff != 0) {
2446 		if (noff < 0) {
2447 			start = nlast;
2448 			end = ilast + noff;
2449 			inc = 1;
2450 		} else {
2451 			start = ilast + noff - 1;
2452 			end = nlast - 1;
2453 			inc = -1;
2454 		}
2455 		for (i = start; i != end; i += inc) {
2456 			if (bread(ivp, i, lfs_sb_getbsize(fs),
2457 			    B_MODIFY, &bp) != 0)
2458 				panic("resize: bread dst blk failed");
2459 			if (bread(ivp, i - noff, lfs_sb_getbsize(fs),
2460 			    0, &obp))
2461 				panic("resize: bread src blk failed");
2462 			memcpy(bp->b_data, obp->b_data, lfs_sb_getbsize(fs));
2463 			VOP_BWRITE(bp->b_vp, bp);
2464 			brelse(obp, 0);
2465 		}
2466 	}
2467 
2468 	/* If we are expanding, write the new empty SEGUSE entries */
2469 	if (newnsegs > oldnsegs) {
2470 		for (i = oldnsegs; i < newnsegs; i++) {
2471 			if ((error = bread(ivp, i / lfs_sb_getsepb(fs) +
2472 					   lfs_sb_getcleansz(fs), lfs_sb_getbsize(fs),
2473 					   B_MODIFY, &bp)) != 0)
2474 				panic("lfs: ifile read: %d", error);
2475 			while ((i + 1) % lfs_sb_getsepb(fs) && i < newnsegs) {
2476 				sup = &((SEGUSE *)bp->b_data)[i % lfs_sb_getsepb(fs)];
2477 				memset(sup, 0, sizeof(*sup));
2478 				i++;
2479 			}
2480 			VOP_BWRITE(bp->b_vp, bp);
2481 		}
2482 	}
2483 
2484 	/* Zero out unused superblock offsets */
2485 	for (i = 2; i < LFS_MAXNUMSB; i++)
2486 		if (lfs_dtosn(fs, lfs_sb_getsboff(fs, i)) >= newnsegs)
2487 			lfs_sb_setsboff(fs, i, 0x0);
2488 
2489 	/*
2490 	 * Correct superblock entries that depend on fs size.
2491 	 * The computations of these are as follows:
2492 	 *
2493 	 * size  = lfs_segtod(fs, nseg)
2494 	 * dsize = lfs_segtod(fs, nseg - minfreeseg) - lfs_btofsb(#super * LFS_SBPAD)
2495 	 * bfree = dsize - lfs_btofsb(fs, bsize * nseg / 2) - blocks_actually_used
2496 	 * avail = lfs_segtod(fs, nclean) - lfs_btofsb(#clean_super * LFS_SBPAD)
2497 	 *         + (lfs_segtod(fs, 1) - (offset - curseg))
2498 	 *	   - lfs_segtod(fs, minfreeseg - (minfreeseg / 2))
2499 	 *
2500 	 * XXX - we should probably adjust minfreeseg as well.
2501 	 */
2502 	gain = (newnsegs - oldnsegs);
2503 	lfs_sb_setnseg(fs, newnsegs);
2504 	lfs_sb_setsegtabsz(fs, nlast - lfs_sb_getcleansz(fs));
2505 	lfs_sb_addsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)));
2506 	lfs_sb_adddsize(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes));
2507 	lfs_sb_addbfree(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)) - lfs_btofsb(fs, sbbytes)
2508 		       - gain * lfs_btofsb(fs, lfs_sb_getbsize(fs) / 2));
2509 	if (gain > 0) {
2510 		lfs_sb_addnclean(fs, gain);
2511 		lfs_sb_addavail(fs, gain * lfs_btofsb(fs, lfs_sb_getssize(fs)));
2512 	} else {
2513 		lfs_sb_subnclean(fs, cgain);
2514 		lfs_sb_subavail(fs, cgain * lfs_btofsb(fs, lfs_sb_getssize(fs)) -
2515 				 lfs_btofsb(fs, csbbytes));
2516 	}
2517 
2518 	/* Resize segment flag cache */
2519 	fs->lfs_suflags[0] = realloc(fs->lfs_suflags[0],
2520 	    lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK);
2521 	fs->lfs_suflags[1] = realloc(fs->lfs_suflags[1],
2522 	    lfs_sb_getnseg(fs) * sizeof(u_int32_t), M_SEGMENT, M_WAITOK);
2523 	for (i = oldnsegs; i < newnsegs; i++)
2524 		fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0;
2525 
2526 	/* Truncate Ifile if necessary */
2527 	if (noff < 0)
2528 		lfs_truncate(ivp, ivp->v_size + (noff << lfs_sb_getbshift(fs)), 0,
2529 		    NOCRED);
2530 
2531 	/* Update cleaner info so the cleaner can die */
2532 	/* XXX what to do if bread fails? */
2533 	bread(ivp, 0, lfs_sb_getbsize(fs), B_MODIFY, &bp);
2534 	cip = bp->b_data;
2535 	lfs_ci_setclean(fs, cip, lfs_sb_getnclean(fs));
2536 	lfs_ci_setdirty(fs, cip, lfs_sb_getnseg(fs) - lfs_sb_getnclean(fs));
2537 	VOP_BWRITE(bp->b_vp, bp);
2538 
2539 	/* Let Ifile accesses proceed */
2540 	rw_exit(&fs->lfs_iflock);
2541 
2542     out:
2543 	lfs_segunlock(fs);
2544 	return error;
2545 }
2546 
2547 /*
2548  * Extended attribute dispatch
2549  */
2550 int
2551 lfs_extattrctl(struct mount *mp, int cmd, struct vnode *vp,
2552 	       int attrnamespace, const char *attrname)
2553 {
2554 #ifdef LFS_EXTATTR
2555 	struct ulfsmount *ump;
2556 
2557 	ump = VFSTOULFS(mp);
2558 	if (ump->um_fstype == ULFS1) {
2559 		return ulfs_extattrctl(mp, cmd, vp, attrnamespace, attrname);
2560 	}
2561 #endif
2562 	return vfs_stdextattrctl(mp, cmd, vp, attrnamespace, attrname);
2563 }
2564