xref: /netbsd-src/sys/ufs/lfs/lfs_vfsops.c (revision 6deb2c22d20de1d75d538e8a5c57b573926fd157)
1 /*	$NetBSD: lfs_vfsops.c,v 1.278 2009/09/13 05:17:37 tsutsui 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 1.278 2009/09/13 05:17:37 tsutsui 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 
96 #include <miscfs/specfs/specdev.h>
97 
98 #include <ufs/ufs/quota.h>
99 #include <ufs/ufs/inode.h>
100 #include <ufs/ufs/ufsmount.h>
101 #include <ufs/ufs/ufs_extern.h>
102 
103 #include <uvm/uvm.h>
104 #include <uvm/uvm_stat.h>
105 #include <uvm/uvm_pager.h>
106 #include <uvm/uvm_pdaemon.h>
107 
108 #include <ufs/lfs/lfs.h>
109 #include <ufs/lfs/lfs_extern.h>
110 
111 #include <miscfs/genfs/genfs.h>
112 #include <miscfs/genfs/genfs_node.h>
113 
114 MODULE(MODULE_CLASS_VFS, lfs, "ffs");
115 
116 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
117 static bool lfs_issequential_hole(const struct ufsmount *,
118     daddr_t, daddr_t);
119 
120 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *);
121 
122 void lfs_sysctl_setup(struct sysctllog *);
123 static struct sysctllog *lfs_sysctl_log;
124 
125 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc;
126 extern const struct vnodeopv_desc lfs_specop_opv_desc;
127 extern const struct vnodeopv_desc lfs_fifoop_opv_desc;
128 
129 pid_t lfs_writer_daemon = 0;
130 int lfs_do_flush = 0;
131 #ifdef LFS_KERNEL_RFW
132 int lfs_do_rfw = 0;
133 #endif
134 
135 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
136 	&lfs_vnodeop_opv_desc,
137 	&lfs_specop_opv_desc,
138 	&lfs_fifoop_opv_desc,
139 	NULL,
140 };
141 
142 struct vfsops lfs_vfsops = {
143 	MOUNT_LFS,
144 	sizeof (struct ufs_args),
145 	lfs_mount,
146 	ufs_start,
147 	lfs_unmount,
148 	ufs_root,
149 	ufs_quotactl,
150 	lfs_statvfs,
151 	lfs_sync,
152 	lfs_vget,
153 	lfs_fhtovp,
154 	lfs_vptofh,
155 	lfs_init,
156 	lfs_reinit,
157 	lfs_done,
158 	lfs_mountroot,
159 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
160 	vfs_stdextattrctl,
161 	(void *)eopnotsupp,	/* vfs_suspendctl */
162 	genfs_renamelock_enter,
163 	genfs_renamelock_exit,
164 	(void *)eopnotsupp,
165 	lfs_vnodeopv_descs,
166 	0,
167 	{ NULL, NULL },
168 };
169 
170 const struct genfs_ops lfs_genfsops = {
171 	.gop_size = lfs_gop_size,
172 	.gop_alloc = ufs_gop_alloc,
173 	.gop_write = lfs_gop_write,
174 	.gop_markupdate = ufs_gop_markupdate,
175 };
176 
177 static const struct ufs_ops lfs_ufsops = {
178 	.uo_itimes = NULL,
179 	.uo_update = lfs_update,
180 	.uo_truncate = lfs_truncate,
181 	.uo_valloc = lfs_valloc,
182 	.uo_vfree = lfs_vfree,
183 	.uo_balloc = lfs_balloc,
184 	.uo_unmark_vnode = lfs_unmark_vnode,
185 };
186 
187 struct shortlong {
188 	const char *sname;
189 	const char *lname;
190 };
191 
192 static int
193 sysctl_lfs_dostats(SYSCTLFN_ARGS)
194 {
195 	extern struct lfs_stats lfs_stats;
196 	extern int lfs_dostats;
197 	int error;
198 
199 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
200 	if (error || newp == NULL)
201 		return (error);
202 
203 	if (lfs_dostats == 0)
204 		memset(&lfs_stats, 0, sizeof(lfs_stats));
205 
206 	return (0);
207 }
208 
209 void
210 lfs_sysctl_setup(struct sysctllog *clog)
211 {
212 	int i;
213 	extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead,
214 		   lfs_fs_pagetrip, lfs_ignore_lazy_sync;
215 #ifdef DEBUG
216 	extern int lfs_debug_log_subsys[DLOG_MAX];
217 	struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */
218 		{ "rollforward", "Debug roll-forward code" },
219 		{ "alloc",	"Debug inode allocation and free list" },
220 		{ "avail",	"Debug space-available-now accounting" },
221 		{ "flush",	"Debug flush triggers" },
222 		{ "lockedlist",	"Debug locked list accounting" },
223 		{ "vnode_verbose", "Verbose per-vnode-written debugging" },
224 		{ "vnode",	"Debug vnode use during segment write" },
225 		{ "segment",	"Debug segment writing" },
226 		{ "seguse",	"Debug segment used-bytes accounting" },
227 		{ "cleaner",	"Debug cleaning routines" },
228 		{ "mount",	"Debug mount/unmount routines" },
229 		{ "pagecache",	"Debug UBC interactions" },
230 		{ "dirop",	"Debug directory-operation accounting" },
231 		{ "malloc",	"Debug private malloc accounting" },
232 	};
233 #endif /* DEBUG */
234 	struct shortlong stat_names[] = { /* Must match lfs.h! */
235 		{ "segsused",	    "Number of new segments allocated" },
236 		{ "psegwrites",	    "Number of partial-segment writes" },
237 		{ "psyncwrites",    "Number of synchronous partial-segment"
238 				    " writes" },
239 		{ "pcleanwrites",   "Number of partial-segment writes by the"
240 				    " cleaner" },
241 		{ "blocktot",       "Number of blocks written" },
242 		{ "cleanblocks",    "Number of blocks written by the cleaner" },
243 		{ "ncheckpoints",   "Number of checkpoints made" },
244 		{ "nwrites",        "Number of whole writes" },
245 		{ "nsync_writes",   "Number of synchronous writes" },
246 		{ "wait_exceeded",  "Number of times writer waited for"
247 				    " cleaner" },
248 		{ "write_exceeded", "Number of times writer invoked flush" },
249 		{ "flush_invoked",  "Number of times flush was invoked" },
250 		{ "vflush_invoked", "Number of time vflush was called" },
251 		{ "clean_inlocked", "Number of vnodes skipped for VI_XLOCK" },
252 		{ "clean_vnlocked", "Number of vnodes skipped for vget failure" },
253 		{ "segs_reclaimed", "Number of segments reclaimed" },
254 	};
255 
256 	sysctl_createv(&clog, 0, NULL, NULL,
257 		       CTLFLAG_PERMANENT,
258 		       CTLTYPE_NODE, "vfs", NULL,
259 		       NULL, 0, NULL, 0,
260 		       CTL_VFS, CTL_EOL);
261 	sysctl_createv(&clog, 0, NULL, NULL,
262 		       CTLFLAG_PERMANENT,
263 		       CTLTYPE_NODE, "lfs",
264 		       SYSCTL_DESCR("Log-structured file system"),
265 		       NULL, 0, NULL, 0,
266 		       CTL_VFS, 5, CTL_EOL);
267 	/*
268 	 * XXX the "5" above could be dynamic, thereby eliminating one
269 	 * more instance of the "number to vfs" mapping problem, but
270 	 * "5" is the order as taken from sys/mount.h
271 	 */
272 
273 	sysctl_createv(&clog, 0, NULL, NULL,
274 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
275 		       CTLTYPE_INT, "flushindir", NULL,
276 		       NULL, 0, &lfs_writeindir, 0,
277 		       CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
278 	sysctl_createv(&clog, 0, NULL, NULL,
279 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
280 		       CTLTYPE_INT, "clean_vnhead", NULL,
281 		       NULL, 0, &lfs_clean_vnhead, 0,
282 		       CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
283 	sysctl_createv(&clog, 0, NULL, NULL,
284 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
285 		       CTLTYPE_INT, "dostats",
286 		       SYSCTL_DESCR("Maintain statistics on LFS operations"),
287 		       sysctl_lfs_dostats, 0, &lfs_dostats, 0,
288 		       CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
289 	sysctl_createv(&clog, 0, NULL, NULL,
290 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
291 		       CTLTYPE_INT, "pagetrip",
292 		       SYSCTL_DESCR("How many dirty pages in fs triggers"
293 				    " a flush"),
294 		       NULL, 0, &lfs_fs_pagetrip, 0,
295 		       CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL);
296 	sysctl_createv(&clog, 0, NULL, NULL,
297 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
298 		       CTLTYPE_INT, "ignore_lazy_sync",
299 		       SYSCTL_DESCR("Lazy Sync is ignored entirely"),
300 		       NULL, 0, &lfs_ignore_lazy_sync, 0,
301 		       CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL);
302 #ifdef LFS_KERNEL_RFW
303 	sysctl_createv(&clog, 0, NULL, NULL,
304 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
305 		       CTLTYPE_INT, "rfw",
306 		       SYSCTL_DESCR("Use in-kernel roll-forward on mount"),
307 		       NULL, 0, &lfs_do_rfw, 0,
308 		       CTL_VFS, 5, LFS_DO_RFW, CTL_EOL);
309 #endif
310 
311 	sysctl_createv(&clog, 0, NULL, NULL,
312 		       CTLFLAG_PERMANENT,
313 		       CTLTYPE_NODE, "stats",
314 		       SYSCTL_DESCR("Debugging options"),
315 		       NULL, 0, NULL, 0,
316 		       CTL_VFS, 5, LFS_STATS, CTL_EOL);
317 	for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) {
318 		sysctl_createv(&clog, 0, NULL, NULL,
319 			       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
320 			       CTLTYPE_INT, stat_names[i].sname,
321 			       SYSCTL_DESCR(stat_names[i].lname),
322 			       NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]),
323 			       0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL);
324 	}
325 
326 #ifdef DEBUG
327 	sysctl_createv(&clog, 0, NULL, NULL,
328 		       CTLFLAG_PERMANENT,
329 		       CTLTYPE_NODE, "debug",
330 		       SYSCTL_DESCR("Debugging options"),
331 		       NULL, 0, NULL, 0,
332 		       CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL);
333 	for (i = 0; i < DLOG_MAX; i++) {
334 		sysctl_createv(&clog, 0, NULL, NULL,
335 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
336 			       CTLTYPE_INT, dlog_names[i].sname,
337 			       SYSCTL_DESCR(dlog_names[i].lname),
338 			       NULL, 0, &(lfs_debug_log_subsys[i]), 0,
339 			       CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL);
340 	}
341 #endif
342 }
343 
344 static int
345 lfs_modcmd(modcmd_t cmd, void *arg)
346 {
347 	int error;
348 
349 	switch (cmd) {
350 	case MODULE_CMD_INIT:
351 		error = vfs_attach(&lfs_vfsops);
352 		if (error != 0)
353 			break;
354 		lfs_sysctl_setup(lfs_sysctl_log);
355 		break;
356 	case MODULE_CMD_FINI:
357 		error = vfs_detach(&lfs_vfsops);
358 		if (error != 0)
359 			break;
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 	int fsflags;
389 	int loopcount;
390 
391 	lfs_writer_daemon = curproc->p_pid;
392 
393 	mutex_enter(&lfs_lock);
394 	for (;;) {
395 		mtsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", hz/10,
396 		    &lfs_lock);
397 
398 		/*
399 		 * Look through the list of LFSs to see if any of them
400 		 * have requested pageouts.
401 		 */
402 		mutex_enter(&mountlist_lock);
403 		for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
404 		     mp = nmp) {
405 			if (vfs_busy(mp, &nmp)) {
406 				continue;
407 			}
408 			if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS,
409 			    sizeof(mp->mnt_stat.f_fstypename)) == 0) {
410 				fs = VFSTOUFS(mp)->um_lfs;
411 				mutex_enter(&lfs_lock);
412 				fsflags = 0;
413 				if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) ||
414 				     lfs_dirvcount > LFS_MAX_DIROP) &&
415 				    fs->lfs_dirops == 0)
416 					fsflags |= SEGM_CKP;
417 				if (fs->lfs_pdflush) {
418 					DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n"));
419 					fs->lfs_pdflush = 0;
420 					lfs_flush_fs(fs, fsflags);
421 					mutex_exit(&lfs_lock);
422 				} else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) {
423 					DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n"));
424 					mutex_exit(&lfs_lock);
425 					lfs_writer_enter(fs, "wrdirop");
426 					lfs_flush_pchain(fs);
427 					lfs_writer_leave(fs);
428 				} else
429 					mutex_exit(&lfs_lock);
430 			}
431 			vfs_unbusy(mp, false, &nmp);
432 		}
433 		mutex_exit(&mountlist_lock);
434 
435 		/*
436 		 * If global state wants a flush, flush everything.
437 		 */
438 		mutex_enter(&lfs_lock);
439 		loopcount = 0;
440 		if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
441 			locked_queue_bytes > LFS_MAX_BYTES ||
442 			lfs_subsys_pages > LFS_MAX_PAGES) {
443 
444 			if (lfs_do_flush) {
445 				DLOG((DLOG_FLUSH, "daemon: lfs_do_flush\n"));
446 			}
447 			if (locked_queue_count > LFS_MAX_BUFS) {
448 				DLOG((DLOG_FLUSH, "daemon: lqc = %d, max %d\n",
449 				      locked_queue_count, LFS_MAX_BUFS));
450 			}
451 			if (locked_queue_bytes > LFS_MAX_BYTES) {
452 				DLOG((DLOG_FLUSH, "daemon: lqb = %ld, max %ld\n",
453 				      locked_queue_bytes, LFS_MAX_BYTES));
454 			}
455 			if (lfs_subsys_pages > LFS_MAX_PAGES) {
456 				DLOG((DLOG_FLUSH, "daemon: lssp = %d, max %d\n",
457 				      lfs_subsys_pages, LFS_MAX_PAGES));
458 			}
459 
460 			lfs_flush(NULL, SEGM_WRITERD, 0);
461 			lfs_do_flush = 0;
462 		}
463 	}
464 	/* NOTREACHED */
465 }
466 
467 /*
468  * Initialize the filesystem, most work done by ufs_init.
469  */
470 void
471 lfs_init(void)
472 {
473 
474 	malloc_type_attach(M_SEGMENT);
475 	pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
476 	    "lfsinopl", &pool_allocator_nointr, IPL_NONE);
477 	pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0,
478 	    "lfsdinopl", &pool_allocator_nointr, IPL_NONE);
479 	pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
480 	    "lfsinoextpl", &pool_allocator_nointr, IPL_NONE);
481 	pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0,
482 	    "lfslbnpool", &pool_allocator_nointr, IPL_NONE);
483 	ufs_init();
484 
485 #ifdef DEBUG
486 	memset(lfs_log, 0, sizeof(lfs_log));
487 #endif
488 	mutex_init(&lfs_lock, MUTEX_DEFAULT, IPL_NONE);
489 	cv_init(&locked_queue_cv, "lfsbuf");
490 	cv_init(&lfs_writing_cv, "lfsflush");
491 }
492 
493 void
494 lfs_reinit(void)
495 {
496 	ufs_reinit();
497 }
498 
499 void
500 lfs_done(void)
501 {
502 	ufs_done();
503 	mutex_destroy(&lfs_lock);
504 	cv_destroy(&locked_queue_cv);
505 	cv_destroy(&lfs_writing_cv);
506 	pool_destroy(&lfs_inode_pool);
507 	pool_destroy(&lfs_dinode_pool);
508 	pool_destroy(&lfs_inoext_pool);
509 	pool_destroy(&lfs_lbnentry_pool);
510 	malloc_type_detach(M_SEGMENT);
511 }
512 
513 /*
514  * Called by main() when ufs is going to be mounted as root.
515  */
516 int
517 lfs_mountroot(void)
518 {
519 	extern struct vnode *rootvp;
520 	struct mount *mp;
521 	struct lwp *l = curlwp;
522 	int error;
523 
524 	if (device_class(root_device) != DV_DISK)
525 		return (ENODEV);
526 
527 	if (rootdev == NODEV)
528 		return (ENODEV);
529 	if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
530 		vrele(rootvp);
531 		return (error);
532 	}
533 	if ((error = lfs_mountfs(rootvp, mp, l))) {
534 		vfs_unbusy(mp, false, NULL);
535 		vfs_destroy(mp);
536 		return (error);
537 	}
538 	mutex_enter(&mountlist_lock);
539 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
540 	mutex_exit(&mountlist_lock);
541 	(void)lfs_statvfs(mp, &mp->mnt_stat);
542 	vfs_unbusy(mp, false, NULL);
543 	setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp));
544 	return (0);
545 }
546 
547 /*
548  * VFS Operations.
549  *
550  * mount system call
551  */
552 int
553 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
554 {
555 	struct lwp *l = curlwp;
556 	struct vnode *devvp;
557 	struct ufs_args *args = data;
558 	struct ufsmount *ump = NULL;
559 	struct lfs *fs = NULL;				/* LFS */
560 	int error = 0, update;
561 	mode_t accessmode;
562 
563 	if (*data_len < sizeof *args)
564 		return EINVAL;
565 
566 	if (mp->mnt_flag & MNT_GETARGS) {
567 		ump = VFSTOUFS(mp);
568 		if (ump == NULL)
569 			return EIO;
570 		args->fspec = NULL;
571 		*data_len = sizeof *args;
572 		return 0;
573 	}
574 
575 	update = mp->mnt_flag & MNT_UPDATE;
576 
577 	/* Check arguments */
578 	if (args->fspec != NULL) {
579 		/*
580 		 * Look up the name and verify that it's sane.
581 		 */
582 		error = namei_simple_user(args->fspec,
583 					NSM_FOLLOW_NOEMULROOT, &devvp);
584 		if (error != 0)
585 			return (error);
586 
587 		if (!update) {
588 			/*
589 			 * Be sure this is a valid block device
590 			 */
591 			if (devvp->v_type != VBLK)
592 				error = ENOTBLK;
593 			else if (bdevsw_lookup(devvp->v_rdev) == NULL)
594 				error = ENXIO;
595 		} else {
596 			/*
597 			 * Be sure we're still naming the same device
598 			 * used for our initial mount
599 			 */
600 			ump = VFSTOUFS(mp);
601 			if (devvp != ump->um_devvp)
602 				error = EINVAL;
603 		}
604 	} else {
605 		if (!update) {
606 			/* New mounts must have a filename for the device */
607 			return (EINVAL);
608 		} else {
609 			/* Use the extant mount */
610 			ump = VFSTOUFS(mp);
611 			devvp = ump->um_devvp;
612 			vref(devvp);
613 		}
614 	}
615 
616 
617 	/*
618 	 * If mount by non-root, then verify that user has necessary
619 	 * permissions on the device.
620 	 */
621 	if (error == 0) {
622 		accessmode = VREAD;
623 		if (update ?
624 		    (mp->mnt_iflag & IMNT_WANTRDWR) != 0 :
625 		    (mp->mnt_flag & MNT_RDONLY) == 0)
626 			accessmode |= VWRITE;
627 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
628 		error = genfs_can_mount(devvp, accessmode, l->l_cred);
629 		VOP_UNLOCK(devvp, 0);
630 	}
631 
632 	if (error) {
633 		vrele(devvp);
634 		return (error);
635 	}
636 
637 	if (!update) {
638 		int flags;
639 
640 		if (mp->mnt_flag & MNT_RDONLY)
641 			flags = FREAD;
642 		else
643 			flags = FREAD|FWRITE;
644 		error = VOP_OPEN(devvp, flags, FSCRED);
645 		if (error)
646 			goto fail;
647 		error = lfs_mountfs(devvp, mp, l);		/* LFS */
648 		if (error) {
649 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
650 			(void)VOP_CLOSE(devvp, flags, NOCRED);
651 			VOP_UNLOCK(devvp, 0);
652 			goto fail;
653 		}
654 
655 		ump = VFSTOUFS(mp);
656 		fs = ump->um_lfs;
657 	} else {
658 		/*
659 		 * Update the mount.
660 		 */
661 
662 		/*
663 		 * The initial mount got a reference on this
664 		 * device, so drop the one obtained via
665 		 * namei(), above.
666 		 */
667 		vrele(devvp);
668 
669 		ump = VFSTOUFS(mp);
670 		fs = ump->um_lfs;
671 		if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
672 			/*
673 			 * Changing from read-only to read/write.
674 			 * Note in the superblocks that we're writing.
675 			 */
676 			fs->lfs_ronly = 0;
677 			if (fs->lfs_pflags & LFS_PF_CLEAN) {
678 				fs->lfs_pflags &= ~LFS_PF_CLEAN;
679 				lfs_writesuper(fs, fs->lfs_sboffs[0]);
680 				lfs_writesuper(fs, fs->lfs_sboffs[1]);
681 			}
682 		}
683 		if (args->fspec == NULL)
684 			return EINVAL;
685 	}
686 
687 	error = set_statvfs_info(path, UIO_USERSPACE, args->fspec,
688 	    UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l);
689 	if (error == 0)
690 		(void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname,
691 			      sizeof(fs->lfs_fsmnt));
692 	return error;
693 
694 fail:
695 	vrele(devvp);
696 	return (error);
697 }
698 
699 
700 /*
701  * Common code for mount and mountroot
702  * LFS specific
703  */
704 int
705 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l)
706 {
707 	struct dlfs *tdfs, *dfs, *adfs;
708 	struct lfs *fs;
709 	struct ufsmount *ump;
710 	struct vnode *vp;
711 	struct buf *bp, *abp;
712 	struct partinfo dpart;
713 	dev_t dev;
714 	int error, i, ronly, secsize, fsbsize;
715 	kauth_cred_t cred;
716 	CLEANERINFO *cip;
717 	SEGUSE *sup;
718 	daddr_t sb_addr;
719 
720 	cred = l ? l->l_cred : NOCRED;
721 
722 	/*
723 	 * Flush out any old buffers remaining from a previous use.
724 	 */
725 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
726 	error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0);
727 	VOP_UNLOCK(devvp, 0);
728 	if (error)
729 		return (error);
730 
731 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
732 	if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred) != 0)
733 		secsize = DEV_BSIZE;
734 	else
735 		secsize = dpart.disklab->d_secsize;
736 
737 	/* Don't free random space on error. */
738 	bp = NULL;
739 	abp = NULL;
740 	ump = NULL;
741 
742 	sb_addr = LFS_LABELPAD / secsize;
743 	while (1) {
744 		/* Read in the superblock. */
745 		error = bread(devvp, sb_addr, LFS_SBPAD, cred, 0, &bp);
746 		if (error)
747 			goto out;
748 		dfs = (struct dlfs *)bp->b_data;
749 
750 		/* Check the basics. */
751 		if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE ||
752 		    dfs->dlfs_version > LFS_VERSION ||
753 		    dfs->dlfs_bsize < sizeof(struct dlfs)) {
754 			DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n"));
755 			error = EINVAL;		/* XXX needs translation */
756 			goto out;
757 		}
758 		if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) {
759 			DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n",
760 			       dfs->dlfs_inodefmt));
761 			error = EINVAL;
762 			goto out;
763 		}
764 
765 		if (dfs->dlfs_version == 1)
766 			fsbsize = secsize;
767 		else {
768 			fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb +
769 				dfs->dlfs_fsbtodb);
770 			/*
771 			 * Could be, if the frag size is large enough, that we
772 			 * don't have the "real" primary superblock.  If that's
773 			 * the case, get the real one, and try again.
774 			 */
775 			if (sb_addr != dfs->dlfs_sboffs[0] <<
776 				       dfs->dlfs_fsbtodb) {
777 				DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr"
778 				      " 0x%llx is not right, trying 0x%llx\n",
779 				      (long long)sb_addr,
780 				      (long long)(dfs->dlfs_sboffs[0] <<
781 						  dfs->dlfs_fsbtodb)));
782 				sb_addr = dfs->dlfs_sboffs[0] <<
783 					  dfs->dlfs_fsbtodb;
784 				brelse(bp, 0);
785 				continue;
786 			}
787 		}
788 		break;
789 	}
790 
791 	/*
792 	 * Check the second superblock to see which is newer; then mount
793 	 * using the older of the two.	This is necessary to ensure that
794 	 * the filesystem is valid if it was not unmounted cleanly.
795 	 */
796 
797 	if (dfs->dlfs_sboffs[1] &&
798 	    dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
799 	{
800 		error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize),
801 			LFS_SBPAD, cred, 0, &abp);
802 		if (error)
803 			goto out;
804 		adfs = (struct dlfs *)abp->b_data;
805 
806 		if (dfs->dlfs_version == 1) {
807 			/* 1s resolution comparison */
808 			if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
809 				tdfs = adfs;
810 			else
811 				tdfs = dfs;
812 		} else {
813 			/* monotonic infinite-resolution comparison */
814 			if (adfs->dlfs_serial < dfs->dlfs_serial)
815 				tdfs = adfs;
816 			else
817 				tdfs = dfs;
818 		}
819 
820 		/* Check the basics. */
821 		if (tdfs->dlfs_magic != LFS_MAGIC ||
822 		    tdfs->dlfs_bsize > MAXBSIZE ||
823 		    tdfs->dlfs_version > LFS_VERSION ||
824 		    tdfs->dlfs_bsize < sizeof(struct dlfs)) {
825 			DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock"
826 			      " sanity failed\n"));
827 			error = EINVAL;		/* XXX needs translation */
828 			goto out;
829 		}
830 	} else {
831 		DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock"
832 		      " daddr=0x%x\n", dfs->dlfs_sboffs[1]));
833 		error = EINVAL;
834 		goto out;
835 	}
836 
837 	/* Allocate the mount structure, copy the superblock into it. */
838 	fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO);
839 	memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs));
840 
841 	/* Compatibility */
842 	if (fs->lfs_version < 2) {
843 		fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
844 		fs->lfs_ibsize = fs->lfs_bsize;
845 		fs->lfs_start = fs->lfs_sboffs[0];
846 		fs->lfs_tstamp = fs->lfs_otstamp;
847 		fs->lfs_fsbtodb = 0;
848 	}
849 	if (fs->lfs_resvseg == 0)
850 		fs->lfs_resvseg = MIN(fs->lfs_minfreeseg - 1, \
851 			MAX(MIN_RESV_SEGS, fs->lfs_minfreeseg / 2 + 1));
852 
853 	/*
854 	 * If we aren't going to be able to write meaningfully to this
855 	 * filesystem, and were not mounted readonly, bomb out now.
856 	 */
857 	if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) {
858 		DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write,"
859 		      " we need BUFPAGES >= %lld\n",
860 		      (long long)((bufmem_hiwater / bufmem_lowater) *
861 				  LFS_INVERSE_MAX_BYTES(
862 					  fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT)));
863 		free(fs, M_UFSMNT);
864 		error = EFBIG; /* XXX needs translation */
865 		goto out;
866 	}
867 
868 	/* Before rolling forward, lock so vget will sleep for other procs */
869 	if (l != NULL) {
870 		fs->lfs_flags = LFS_NOTYET;
871 		fs->lfs_rfpid = l->l_proc->p_pid;
872 	}
873 
874 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
875 	ump->um_lfs = fs;
876 	ump->um_ops = &lfs_ufsops;
877 	ump->um_fstype = UFS1;
878 	if (sizeof(struct lfs) < LFS_SBPAD) {			/* XXX why? */
879 		brelse(bp, BC_INVAL);
880 		brelse(abp, BC_INVAL);
881 	} else {
882 		brelse(bp, 0);
883 		brelse(abp, 0);
884 	}
885 	bp = NULL;
886 	abp = NULL;
887 
888 
889 	/* Set up the I/O information */
890 	fs->lfs_devbsize = secsize;
891 	fs->lfs_iocount = 0;
892 	fs->lfs_diropwait = 0;
893 	fs->lfs_activesb = 0;
894 	fs->lfs_uinodes = 0;
895 	fs->lfs_ravail = 0;
896 	fs->lfs_favail = 0;
897 	fs->lfs_sbactive = 0;
898 
899 	/* Set up the ifile and lock aflags */
900 	fs->lfs_doifile = 0;
901 	fs->lfs_writer = 0;
902 	fs->lfs_dirops = 0;
903 	fs->lfs_nadirop = 0;
904 	fs->lfs_seglock = 0;
905 	fs->lfs_pdflush = 0;
906 	fs->lfs_sleepers = 0;
907 	fs->lfs_pages = 0;
908 	rw_init(&fs->lfs_fraglock);
909 	rw_init(&fs->lfs_iflock);
910 	cv_init(&fs->lfs_stopcv, "lfsstop");
911 
912 	/* Set the file system readonly/modify bits. */
913 	fs->lfs_ronly = ronly;
914 	if (ronly == 0)
915 		fs->lfs_fmod = 1;
916 
917 	/* Initialize the mount structure. */
918 	dev = devvp->v_rdev;
919 	mp->mnt_data = ump;
920 	mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
921 	mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS);
922 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
923 	mp->mnt_stat.f_namemax = LFS_MAXNAMLEN;
924 	mp->mnt_stat.f_iosize = fs->lfs_bsize;
925 	mp->mnt_flag |= MNT_LOCAL;
926 	mp->mnt_fs_bshift = fs->lfs_bshift;
927 	ump->um_flags = 0;
928 	ump->um_mountp = mp;
929 	ump->um_dev = dev;
930 	ump->um_devvp = devvp;
931 	ump->um_bptrtodb = fs->lfs_fsbtodb;
932 	ump->um_seqinc = fragstofsb(fs, fs->lfs_frag);
933 	ump->um_nindir = fs->lfs_nindir;
934 	ump->um_lognindir = ffs(fs->lfs_nindir) - 1;
935 	for (i = 0; i < MAXQUOTAS; i++)
936 		ump->um_quotas[i] = NULLVP;
937 	ump->um_maxsymlinklen = fs->lfs_maxsymlinklen;
938 	ump->um_dirblksiz = DIRBLKSIZ;
939 	ump->um_maxfilesize = fs->lfs_maxfilesize;
940 	if (ump->um_maxsymlinklen > 0)
941 		mp->mnt_iflag |= IMNT_DTYPE;
942 	devvp->v_specmountpoint = mp;
943 
944 	/* Set up reserved memory for pageout */
945 	lfs_setup_resblks(fs);
946 	/* Set up vdirop tailq */
947 	TAILQ_INIT(&fs->lfs_dchainhd);
948 	/* and paging tailq */
949 	TAILQ_INIT(&fs->lfs_pchainhd);
950 	/* and delayed segment accounting for truncation list */
951 	LIST_INIT(&fs->lfs_segdhd);
952 
953 	/*
954 	 * We use the ifile vnode for almost every operation.  Instead of
955 	 * retrieving it from the hash table each time we retrieve it here,
956 	 * artificially increment the reference count and keep a pointer
957 	 * to it in the incore copy of the superblock.
958 	 */
959 	if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
960 		DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error));
961 		goto out;
962 	}
963 	fs->lfs_ivnode = vp;
964 	VREF(vp);
965 
966 	/* Set up inode bitmap and order free list */
967 	lfs_order_freelist(fs);
968 
969 	/* Set up segment usage flags for the autocleaner. */
970 	fs->lfs_nactive = 0;
971 	fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *),
972 						M_SEGMENT, M_WAITOK);
973 	fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
974 						 M_SEGMENT, M_WAITOK);
975 	fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
976 						 M_SEGMENT, M_WAITOK);
977 	memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t));
978 	for (i = 0; i < fs->lfs_nseg; i++) {
979 		int changed;
980 
981 		LFS_SEGENTRY(sup, fs, i, bp);
982 		changed = 0;
983 		if (!ronly) {
984 			if (sup->su_nbytes == 0 &&
985 			    !(sup->su_flags & SEGUSE_EMPTY)) {
986 				sup->su_flags |= SEGUSE_EMPTY;
987 				++changed;
988 			} else if (!(sup->su_nbytes == 0) &&
989 				   (sup->su_flags & SEGUSE_EMPTY)) {
990 				sup->su_flags &= ~SEGUSE_EMPTY;
991 				++changed;
992 			}
993 			if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) {
994 				sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL);
995 				++changed;
996 			}
997 		}
998 		fs->lfs_suflags[0][i] = sup->su_flags;
999 		if (changed)
1000 			LFS_WRITESEGENTRY(sup, fs, i, bp);
1001 		else
1002 			brelse(bp, 0);
1003 	}
1004 
1005 #ifdef LFS_KERNEL_RFW
1006 	lfs_roll_forward(fs, mp, l);
1007 #endif
1008 
1009 	/* If writing, sb is not clean; record in case of immediate crash */
1010 	if (!fs->lfs_ronly) {
1011 		fs->lfs_pflags &= ~LFS_PF_CLEAN;
1012 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
1013 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
1014 	}
1015 
1016 	/* Allow vget now that roll-forward is complete */
1017 	fs->lfs_flags &= ~(LFS_NOTYET);
1018 	wakeup(&fs->lfs_flags);
1019 
1020 	/*
1021 	 * Initialize the ifile cleaner info with information from
1022 	 * the superblock.
1023 	 */
1024 	LFS_CLEANERINFO(cip, fs, bp);
1025 	cip->clean = fs->lfs_nclean;
1026 	cip->dirty = fs->lfs_nseg - fs->lfs_nclean;
1027 	cip->avail = fs->lfs_avail;
1028 	cip->bfree = fs->lfs_bfree;
1029 	(void) LFS_BWRITE_LOG(bp); /* Ifile */
1030 
1031 	/*
1032 	 * Mark the current segment as ACTIVE, since we're going to
1033 	 * be writing to it.
1034 	 */
1035 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);
1036 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1037 	fs->lfs_nactive++;
1038 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);  /* Ifile */
1039 
1040 	/* Now that roll-forward is done, unlock the Ifile */
1041 	vput(vp);
1042 
1043 	/* Start the pagedaemon-anticipating daemon */
1044 	if (lfs_writer_daemon == 0 && kthread_create(PRI_BIO, 0, NULL,
1045 	    lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1046 		panic("fork lfs_writer");
1047 
1048 	printf("WARNING: the log-structured file system is experimental\n"
1049 	    "WARNING: it may cause system crashes and/or corrupt data\n");
1050 
1051 	return (0);
1052 
1053 out:
1054 	if (bp)
1055 		brelse(bp, 0);
1056 	if (abp)
1057 		brelse(abp, 0);
1058 	if (ump) {
1059 		free(ump->um_lfs, M_UFSMNT);
1060 		free(ump, M_UFSMNT);
1061 		mp->mnt_data = NULL;
1062 	}
1063 
1064 	return (error);
1065 }
1066 
1067 /*
1068  * unmount system call
1069  */
1070 int
1071 lfs_unmount(struct mount *mp, int mntflags)
1072 {
1073 	struct lwp *l = curlwp;
1074 	struct ufsmount *ump;
1075 	struct lfs *fs;
1076 	int error, flags, ronly;
1077 	vnode_t *vp;
1078 
1079 	flags = 0;
1080 	if (mntflags & MNT_FORCE)
1081 		flags |= FORCECLOSE;
1082 
1083 	ump = VFSTOUFS(mp);
1084 	fs = ump->um_lfs;
1085 
1086 	/* Two checkpoints */
1087 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1088 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1089 
1090 	/* wake up the cleaner so it can die */
1091 	lfs_wakeup_cleaner(fs);
1092 	mutex_enter(&lfs_lock);
1093 	while (fs->lfs_sleepers)
1094 		mtsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1095 			&lfs_lock);
1096 	mutex_exit(&lfs_lock);
1097 
1098 #ifdef QUOTA
1099 	if (mp->mnt_flag & MNT_QUOTA) {
1100 		int i;
1101 		error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags);
1102 		if (error)
1103 			return (error);
1104 		for (i = 0; i < MAXQUOTAS; i++) {
1105 			if (ump->um_quotas[i] == NULLVP)
1106 				continue;
1107 			quotaoff(l, mp, i);
1108 		}
1109 		/*
1110 		 * Here we fall through to vflush again to ensure
1111 		 * that we have gotten rid of all the system vnodes.
1112 		 */
1113 	}
1114 #endif
1115 	if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1116 		return (error);
1117 	if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0)
1118 		return (error);
1119 	vp = fs->lfs_ivnode;
1120 	mutex_enter(&vp->v_interlock);
1121 	if (LIST_FIRST(&vp->v_dirtyblkhd))
1122 		panic("lfs_unmount: still dirty blocks on ifile vnode");
1123 	mutex_exit(&vp->v_interlock);
1124 
1125 	/* Explicitly write the superblock, to update serial and pflags */
1126 	fs->lfs_pflags |= LFS_PF_CLEAN;
1127 	lfs_writesuper(fs, fs->lfs_sboffs[0]);
1128 	lfs_writesuper(fs, fs->lfs_sboffs[1]);
1129 	mutex_enter(&lfs_lock);
1130 	while (fs->lfs_iocount)
1131 		mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0,
1132 			&lfs_lock);
1133 	mutex_exit(&lfs_lock);
1134 
1135 	/* Finish with the Ifile, now that we're done with it */
1136 	vgone(fs->lfs_ivnode);
1137 
1138 	ronly = !fs->lfs_ronly;
1139 	if (ump->um_devvp->v_type != VBAD)
1140 		ump->um_devvp->v_specmountpoint = NULL;
1141 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1142 	error = VOP_CLOSE(ump->um_devvp,
1143 	    ronly ? FREAD : FREAD|FWRITE, NOCRED);
1144 	vput(ump->um_devvp);
1145 
1146 	/* Complain about page leakage */
1147 	if (fs->lfs_pages > 0)
1148 		printf("lfs_unmount: still claim %d pages (%d in subsystem)\n",
1149 			fs->lfs_pages, lfs_subsys_pages);
1150 
1151 	/* Free per-mount data structures */
1152 	free(fs->lfs_ino_bitmap, M_SEGMENT);
1153 	free(fs->lfs_suflags[0], M_SEGMENT);
1154 	free(fs->lfs_suflags[1], M_SEGMENT);
1155 	free(fs->lfs_suflags, M_SEGMENT);
1156 	lfs_free_resblks(fs);
1157 	cv_destroy(&fs->lfs_stopcv);
1158 	rw_destroy(&fs->lfs_fraglock);
1159 	rw_destroy(&fs->lfs_iflock);
1160 	free(fs, M_UFSMNT);
1161 	free(ump, M_UFSMNT);
1162 
1163 	mp->mnt_data = NULL;
1164 	mp->mnt_flag &= ~MNT_LOCAL;
1165 	return (error);
1166 }
1167 
1168 /*
1169  * Get file system statistics.
1170  *
1171  * NB: We don't lock to access the superblock here, because it's not
1172  * really that important if we get it wrong.
1173  */
1174 int
1175 lfs_statvfs(struct mount *mp, struct statvfs *sbp)
1176 {
1177 	struct lfs *fs;
1178 	struct ufsmount *ump;
1179 
1180 	ump = VFSTOUFS(mp);
1181 	fs = ump->um_lfs;
1182 	if (fs->lfs_magic != LFS_MAGIC)
1183 		panic("lfs_statvfs: magic");
1184 
1185 	sbp->f_bsize = fs->lfs_bsize;
1186 	sbp->f_frsize = fs->lfs_fsize;
1187 	sbp->f_iosize = fs->lfs_bsize;
1188 	sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks);
1189 
1190 	sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs));
1191 	KASSERT(sbp->f_bfree <= fs->lfs_dsize);
1192 #if 0
1193 	if (sbp->f_bfree < 0)
1194 		sbp->f_bfree = 0;
1195 #endif
1196 
1197 	sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs));
1198 	if (sbp->f_bfree > sbp->f_bresvd)
1199 		sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
1200 	else
1201 		sbp->f_bavail = 0;
1202 
1203 	sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs);
1204 	sbp->f_ffree = sbp->f_files - fs->lfs_nfiles;
1205 	sbp->f_favail = sbp->f_ffree;
1206 	sbp->f_fresvd = 0;
1207 	copy_statvfs_info(sbp, mp);
1208 	return (0);
1209 }
1210 
1211 /*
1212  * Go through the disk queues to initiate sandbagged IO;
1213  * go through the inodes to write those that have been modified;
1214  * initiate the writing of the super block if it has been modified.
1215  *
1216  * Note: we are always called with the filesystem marked `MPBUSY'.
1217  */
1218 int
1219 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
1220 {
1221 	int error;
1222 	struct lfs *fs;
1223 
1224 	fs = VFSTOUFS(mp)->um_lfs;
1225 	if (fs->lfs_ronly)
1226 		return 0;
1227 
1228 	/* Snapshots should not hose the syncer */
1229 	/*
1230 	 * XXX Sync can block here anyway, since we don't have a very
1231 	 * XXX good idea of how much data is pending.  If it's more
1232 	 * XXX than a segment and lfs_nextseg is close to the end of
1233 	 * XXX the log, we'll likely block.
1234 	 */
1235 	mutex_enter(&lfs_lock);
1236 	if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) {
1237 		mutex_exit(&lfs_lock);
1238 		return 0;
1239 	}
1240 	mutex_exit(&lfs_lock);
1241 
1242 	lfs_writer_enter(fs, "lfs_dirops");
1243 
1244 	/* All syncs must be checkpoints until roll-forward is implemented. */
1245 	DLOG((DLOG_FLUSH, "lfs_sync at 0x%x\n", fs->lfs_offset));
1246 	error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1247 	lfs_writer_leave(fs);
1248 #ifdef QUOTA
1249 	qsync(mp);
1250 #endif
1251 	return (error);
1252 }
1253 
1254 /*
1255  * Look up an LFS dinode number to find its incore vnode.  If not already
1256  * in core, read it in from the specified device.  Return the inode locked.
1257  * Detection and handling of mount points must be done by the calling routine.
1258  */
1259 int
1260 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1261 {
1262 	struct lfs *fs;
1263 	struct ufs1_dinode *dip;
1264 	struct inode *ip;
1265 	struct buf *bp;
1266 	struct ifile *ifp;
1267 	struct vnode *vp;
1268 	struct ufsmount *ump;
1269 	daddr_t daddr;
1270 	dev_t dev;
1271 	int error, retries;
1272 	struct timespec ts;
1273 
1274 	memset(&ts, 0, sizeof ts);	/* XXX gcc */
1275 
1276 	ump = VFSTOUFS(mp);
1277 	dev = ump->um_dev;
1278 	fs = ump->um_lfs;
1279 
1280 	/*
1281 	 * If the filesystem is not completely mounted yet, suspend
1282 	 * any access requests (wait for roll-forward to complete).
1283 	 */
1284 	mutex_enter(&lfs_lock);
1285 	while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1286 		mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0,
1287 			&lfs_lock);
1288 	mutex_exit(&lfs_lock);
1289 
1290 retry:
1291 	if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
1292 		return (0);
1293 
1294 	if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) {
1295 		*vpp = NULL;
1296 		 return (error);
1297 	}
1298 
1299 	mutex_enter(&ufs_hashlock);
1300 	if (ufs_ihashget(dev, ino, 0) != NULL) {
1301 		mutex_exit(&ufs_hashlock);
1302 		ungetnewvnode(vp);
1303 		goto retry;
1304 	}
1305 
1306 	/* Translate the inode number to a disk address. */
1307 	if (ino == LFS_IFILE_INUM)
1308 		daddr = fs->lfs_idaddr;
1309 	else {
1310 		/* XXX bounds-check this too */
1311 		LFS_IENTRY(ifp, fs, ino, bp);
1312 		daddr = ifp->if_daddr;
1313 		if (fs->lfs_version > 1) {
1314 			ts.tv_sec = ifp->if_atime_sec;
1315 			ts.tv_nsec = ifp->if_atime_nsec;
1316 		}
1317 
1318 		brelse(bp, 0);
1319 		if (daddr == LFS_UNUSED_DADDR) {
1320 			*vpp = NULLVP;
1321 			mutex_exit(&ufs_hashlock);
1322 			ungetnewvnode(vp);
1323 			return (ENOENT);
1324 		}
1325 	}
1326 
1327 	/* Allocate/init new vnode/inode. */
1328 	lfs_vcreate(mp, ino, vp);
1329 
1330 	/*
1331 	 * Put it onto its hash chain and lock it so that other requests for
1332 	 * this inode will block if they arrive while we are sleeping waiting
1333 	 * for old data structures to be purged or for the contents of the
1334 	 * disk portion of this inode to be read.
1335 	 */
1336 	ip = VTOI(vp);
1337 	ufs_ihashins(ip);
1338 	mutex_exit(&ufs_hashlock);
1339 
1340 	/*
1341 	 * XXX
1342 	 * This may not need to be here, logically it should go down with
1343 	 * the i_devvp initialization.
1344 	 * Ask Kirk.
1345 	 */
1346 	ip->i_lfs = ump->um_lfs;
1347 
1348 	/* Read in the disk contents for the inode, copy into the inode. */
1349 	retries = 0;
1350     again:
1351 	error = bread(ump->um_devvp, fsbtodb(fs, daddr),
1352 		(fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize),
1353 		NOCRED, 0, &bp);
1354 	if (error) {
1355 		/*
1356 		 * The inode does not contain anything useful, so it would
1357 		 * be misleading to leave it on its hash chain. With mode
1358 		 * still zero, it will be unlinked and returned to the free
1359 		 * list by vput().
1360 		 */
1361 		vput(vp);
1362 		brelse(bp, 0);
1363 		*vpp = NULL;
1364 		return (error);
1365 	}
1366 
1367 	dip = lfs_ifind(fs, ino, bp);
1368 	if (dip == NULL) {
1369 		/* Assume write has not completed yet; try again */
1370 		brelse(bp, BC_INVAL);
1371 		++retries;
1372 		if (retries > LFS_IFIND_RETRIES) {
1373 #ifdef DEBUG
1374 			/* If the seglock is held look at the bpp to see
1375 			   what is there anyway */
1376 			mutex_enter(&lfs_lock);
1377 			if (fs->lfs_seglock > 0) {
1378 				struct buf **bpp;
1379 				struct ufs1_dinode *dp;
1380 				int i;
1381 
1382 				for (bpp = fs->lfs_sp->bpp;
1383 				     bpp != fs->lfs_sp->cbpp; ++bpp) {
1384 					if ((*bpp)->b_vp == fs->lfs_ivnode &&
1385 					    bpp != fs->lfs_sp->bpp) {
1386 						/* Inode block */
1387 						printf("lfs_vget: block 0x%" PRIx64 ": ",
1388 						       (*bpp)->b_blkno);
1389 						dp = (struct ufs1_dinode *)(*bpp)->b_data;
1390 						for (i = 0; i < INOPB(fs); i++)
1391 							if (dp[i].di_u.inumber)
1392 								printf("%d ", dp[i].di_u.inumber);
1393 						printf("\n");
1394 					}
1395 				}
1396 			}
1397 			mutex_exit(&lfs_lock);
1398 #endif /* DEBUG */
1399 			panic("lfs_vget: dinode not found");
1400 		}
1401 		mutex_enter(&lfs_lock);
1402 		if (fs->lfs_iocount) {
1403 			DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino));
1404 			(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
1405 				      "lfs ifind", 1, &lfs_lock);
1406 		} else
1407 			retries = LFS_IFIND_RETRIES;
1408 		mutex_exit(&lfs_lock);
1409 		goto again;
1410 	}
1411 	*ip->i_din.ffs1_din = *dip;
1412 	brelse(bp, 0);
1413 
1414 	if (fs->lfs_version > 1) {
1415 		ip->i_ffs1_atime = ts.tv_sec;
1416 		ip->i_ffs1_atimensec = ts.tv_nsec;
1417 	}
1418 
1419 	lfs_vinit(mp, &vp);
1420 
1421 	*vpp = vp;
1422 
1423 	KASSERT(VOP_ISLOCKED(vp));
1424 
1425 	return (0);
1426 }
1427 
1428 /*
1429  * File handle to vnode
1430  */
1431 int
1432 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1433 {
1434 	struct lfid lfh;
1435 	struct buf *bp;
1436 	IFILE *ifp;
1437 	int32_t daddr;
1438 	struct lfs *fs;
1439 	vnode_t *vp;
1440 
1441 	if (fhp->fid_len != sizeof(struct lfid))
1442 		return EINVAL;
1443 
1444 	memcpy(&lfh, fhp, sizeof(lfh));
1445 	if (lfh.lfid_ino < LFS_IFILE_INUM)
1446 		return ESTALE;
1447 
1448 	fs = VFSTOUFS(mp)->um_lfs;
1449 	if (lfh.lfid_ident != fs->lfs_ident)
1450 		return ESTALE;
1451 
1452 	if (lfh.lfid_ino >
1453 	    ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) -
1454 	     fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb)
1455 		return ESTALE;
1456 
1457 	mutex_enter(&ufs_ihash_lock);
1458 	vp = ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfh.lfid_ino);
1459 	mutex_exit(&ufs_ihash_lock);
1460 	if (vp == NULL) {
1461 		LFS_IENTRY(ifp, fs, lfh.lfid_ino, bp);
1462 		daddr = ifp->if_daddr;
1463 		brelse(bp, 0);
1464 		if (daddr == LFS_UNUSED_DADDR)
1465 			return ESTALE;
1466 	}
1467 
1468 	return (ufs_fhtovp(mp, &lfh.lfid_ufid, vpp));
1469 }
1470 
1471 /*
1472  * Vnode pointer to File handle
1473  */
1474 /* ARGSUSED */
1475 int
1476 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size)
1477 {
1478 	struct inode *ip;
1479 	struct lfid lfh;
1480 
1481 	if (*fh_size < sizeof(struct lfid)) {
1482 		*fh_size = sizeof(struct lfid);
1483 		return E2BIG;
1484 	}
1485 	*fh_size = sizeof(struct lfid);
1486 	ip = VTOI(vp);
1487 	memset(&lfh, 0, sizeof(lfh));
1488 	lfh.lfid_len = sizeof(struct lfid);
1489 	lfh.lfid_ino = ip->i_number;
1490 	lfh.lfid_gen = ip->i_gen;
1491 	lfh.lfid_ident = ip->i_lfs->lfs_ident;
1492 	memcpy(fhp, &lfh, sizeof(lfh));
1493 	return (0);
1494 }
1495 
1496 /*
1497  * ufs_bmaparray callback function for writing.
1498  *
1499  * Since blocks will be written to the new segment anyway,
1500  * we don't care about current daddr of them.
1501  */
1502 static bool
1503 lfs_issequential_hole(const struct ufsmount *ump,
1504     daddr_t daddr0, daddr_t daddr1)
1505 {
1506 	daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */
1507 	daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */
1508 
1509 	KASSERT(daddr0 == UNWRITTEN ||
1510 	    (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR));
1511 	KASSERT(daddr1 == UNWRITTEN ||
1512 	    (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR));
1513 
1514 	/* NOTE: all we want to know here is 'hole or not'. */
1515 	/* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */
1516 
1517 	/*
1518 	 * treat UNWRITTENs and all resident blocks as 'contiguous'
1519 	 */
1520 	if (daddr0 != 0 && daddr1 != 0)
1521 		return true;
1522 
1523 	/*
1524 	 * both are in hole?
1525 	 */
1526 	if (daddr0 == 0 && daddr1 == 0)
1527 		return true; /* all holes are 'contiguous' for us. */
1528 
1529 	return false;
1530 }
1531 
1532 /*
1533  * lfs_gop_write functions exactly like genfs_gop_write, except that
1534  * (1) it requires the seglock to be held by its caller, and sp->fip
1535  *     to be properly initialized (it will return without re-initializing
1536  *     sp->fip, and without calling lfs_writeseg).
1537  * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1538  *     to determine how large a block it can write at once (though it does
1539  *     still use VOP_BMAP to find holes in the file);
1540  * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1541  *     (leaving lfs_writeseg to deal with the cluster blocks, so we might
1542  *     now have clusters of clusters, ick.)
1543  */
1544 static int
1545 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages,
1546     int flags)
1547 {
1548 	int i, error, run, haveeof = 0;
1549 	int fs_bshift;
1550 	vaddr_t kva;
1551 	off_t eof, offset, startoffset = 0;
1552 	size_t bytes, iobytes, skipbytes;
1553 	daddr_t lbn, blkno;
1554 	struct vm_page *pg;
1555 	struct buf *mbp, *bp;
1556 	struct vnode *devvp = VTOI(vp)->i_devvp;
1557 	struct inode *ip = VTOI(vp);
1558 	struct lfs *fs = ip->i_lfs;
1559 	struct segment *sp = fs->lfs_sp;
1560 	UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1561 
1562 	ASSERT_SEGLOCK(fs);
1563 
1564 	/* The Ifile lives in the buffer cache */
1565 	KASSERT(vp != fs->lfs_ivnode);
1566 
1567 	/*
1568 	 * We don't want to fill the disk before the cleaner has a chance
1569 	 * to make room for us.  If we're in danger of doing that, fail
1570 	 * with EAGAIN.  The caller will have to notice this, unlock
1571 	 * so the cleaner can run, relock and try again.
1572 	 *
1573 	 * We must write everything, however, if our vnode is being
1574 	 * reclaimed.
1575 	 */
1576 	if (LFS_STARVED_FOR_SEGS(fs) && vp != fs->lfs_flushvp)
1577 		goto tryagain;
1578 
1579 	/*
1580 	 * Sometimes things slip past the filters in lfs_putpages,
1581 	 * and the pagedaemon tries to write pages---problem is
1582 	 * that the pagedaemon never acquires the segment lock.
1583 	 *
1584 	 * Alternatively, pages that were clean when we called
1585 	 * genfs_putpages may have become dirty in the meantime.  In this
1586 	 * case the segment header is not properly set up for blocks
1587 	 * to be added to it.
1588 	 *
1589 	 * Unbusy and unclean the pages, and put them on the ACTIVE
1590 	 * queue under the hypothesis that they couldn't have got here
1591 	 * unless they were modified *quite* recently.
1592 	 *
1593 	 * XXXUBC that last statement is an oversimplification of course.
1594 	 */
1595 	if (!LFS_SEGLOCK_HELD(fs) ||
1596 	    (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) ||
1597 	    (pgs[0]->offset & fs->lfs_bmask) != 0) {
1598 		goto tryagain;
1599 	}
1600 
1601 	UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1602 	    vp, pgs, npages, flags);
1603 
1604 	GOP_SIZE(vp, vp->v_size, &eof, 0);
1605 	haveeof = 1;
1606 
1607 	if (vp->v_type == VREG)
1608 		fs_bshift = vp->v_mount->mnt_fs_bshift;
1609 	else
1610 		fs_bshift = DEV_BSHIFT;
1611 	error = 0;
1612 	pg = pgs[0];
1613 	startoffset = pg->offset;
1614 	KASSERT(eof >= 0);
1615 
1616 	if (startoffset >= eof) {
1617 		goto tryagain;
1618 	} else
1619 		bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1620 	skipbytes = 0;
1621 
1622 	KASSERT(bytes != 0);
1623 
1624 	/* Swap PG_DELWRI for PG_PAGEOUT */
1625 	for (i = 0; i < npages; i++) {
1626 		if (pgs[i]->flags & PG_DELWRI) {
1627 			KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
1628 			pgs[i]->flags &= ~PG_DELWRI;
1629 			pgs[i]->flags |= PG_PAGEOUT;
1630 			uvm_pageout_start(1);
1631 			mutex_enter(&uvm_pageqlock);
1632 			uvm_pageunwire(pgs[i]);
1633 			mutex_exit(&uvm_pageqlock);
1634 		}
1635 	}
1636 
1637 	/*
1638 	 * Check to make sure we're starting on a block boundary.
1639 	 * We'll check later to make sure we always write entire
1640 	 * blocks (or fragments).
1641 	 */
1642 	if (startoffset & fs->lfs_bmask)
1643 		printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n",
1644 		       startoffset, fs->lfs_bmask,
1645 		       startoffset & fs->lfs_bmask);
1646 	KASSERT((startoffset & fs->lfs_bmask) == 0);
1647 	if (bytes & fs->lfs_ffmask) {
1648 		printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
1649 		panic("lfs_gop_write: non-integer blocks");
1650 	}
1651 
1652 	/*
1653 	 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK.
1654 	 * If we would, write what we have and try again.  If we don't
1655 	 * have anything to write, we'll have to sleep.
1656 	 */
1657 	if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
1658 				      (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ?
1659 				       UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) {
1660 		DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n"));
1661 #if 0
1662 		      " with nfinfo=%d at offset 0x%x\n",
1663 		      (int)((SEGSUM *)(sp->segsum))->ss_nfinfo,
1664 		      (unsigned)fs->lfs_offset));
1665 #endif
1666 		lfs_updatemeta(sp);
1667 		lfs_release_finfo(fs);
1668 		(void) lfs_writeseg(fs, sp);
1669 
1670 		lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
1671 
1672 		/*
1673 		 * Having given up all of the pager_map we were holding,
1674 		 * we can now wait for aiodoned to reclaim it for us
1675 		 * without fear of deadlock.
1676 		 */
1677 		kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
1678 				     UVMPAGER_MAPIN_WAITOK);
1679 	}
1680 
1681 	mutex_enter(&vp->v_interlock);
1682 	vp->v_numoutput += 2; /* one for biodone, one for aiodone */
1683 	mutex_exit(&vp->v_interlock);
1684 
1685 	mbp = getiobuf(NULL, true);
1686 	UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
1687 	    vp, mbp, vp->v_numoutput, bytes);
1688 	mbp->b_bufsize = npages << PAGE_SHIFT;
1689 	mbp->b_data = (void *)kva;
1690 	mbp->b_resid = mbp->b_bcount = bytes;
1691 	mbp->b_cflags = BC_BUSY|BC_AGE;
1692 	mbp->b_iodone = uvm_aio_biodone;
1693 
1694 	bp = NULL;
1695 	for (offset = startoffset;
1696 	    bytes > 0;
1697 	    offset += iobytes, bytes -= iobytes) {
1698 		lbn = offset >> fs_bshift;
1699 		error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
1700 		    lfs_issequential_hole);
1701 		if (error) {
1702 			UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d",
1703 			    error,0,0,0);
1704 			skipbytes += bytes;
1705 			bytes = 0;
1706 			break;
1707 		}
1708 
1709 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
1710 		    bytes);
1711 		if (blkno == (daddr_t)-1) {
1712 			skipbytes += iobytes;
1713 			continue;
1714 		}
1715 
1716 		/*
1717 		 * Discover how much we can really pack into this buffer.
1718 		 */
1719 		/* If no room in the current segment, finish it up */
1720 		if (sp->sum_bytes_left < sizeof(int32_t) ||
1721 		    sp->seg_bytes_left < (1 << fs->lfs_bshift)) {
1722 			int vers;
1723 
1724 			lfs_updatemeta(sp);
1725 			vers = sp->fip->fi_version;
1726 			lfs_release_finfo(fs);
1727 			(void) lfs_writeseg(fs, sp);
1728 
1729 			lfs_acquire_finfo(fs, ip->i_number, vers);
1730 		}
1731 		/* Check both for space in segment and space in segsum */
1732 		iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
1733 					<< fs_bshift);
1734 		iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
1735 				       << fs_bshift);
1736 		KASSERT(iobytes > 0);
1737 
1738 		/* if it's really one i/o, don't make a second buf */
1739 		if (offset == startoffset && iobytes == bytes) {
1740 			bp = mbp;
1741 			/* correct overcount if there is no second buffer */
1742 			mutex_enter(&vp->v_interlock);
1743 			--vp->v_numoutput;
1744 			mutex_exit(&vp->v_interlock);
1745 		} else {
1746 			bp = getiobuf(NULL, true);
1747 			UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
1748 			    vp, bp, vp->v_numoutput, 0);
1749 			nestiobuf_setup(mbp, bp, offset - pg->offset, iobytes);
1750 			/*
1751 			 * LFS doesn't like async I/O here, dies with
1752 			 * and assert in lfs_bwrite().  Is that assert
1753 			 * valid?  I retained non-async behaviour when
1754 			 * converted this to use nestiobuf --pooka
1755 			 */
1756 			bp->b_flags &= ~B_ASYNC;
1757 			/*
1758 			 * LFS uses VOP_BWRITE instead of VOP_STRATEGY.
1759 			 * Therefore biodone doesn't get called for
1760 			 * the buffer.  Therefore decrement the output
1761 			 * counter that nestiobuf_setup() incremented.
1762 			 */
1763 			mutex_enter(&vp->v_interlock);
1764 			vp->v_numoutput--;
1765 			mutex_exit(&vp->v_interlock);
1766 		}
1767 
1768 		/* XXX This is silly ... is this necessary? */
1769 		mutex_enter(&bufcache_lock);
1770 		mutex_enter(&vp->v_interlock);
1771 		bgetvp(vp, bp);
1772 		mutex_exit(&vp->v_interlock);
1773 		mutex_exit(&bufcache_lock);
1774 
1775 		bp->b_lblkno = lblkno(fs, offset);
1776 		bp->b_private = mbp;
1777 		if (devvp->v_type == VBLK) {
1778 			bp->b_dev = devvp->v_rdev;
1779 		}
1780 		VOP_BWRITE(bp);
1781 		while (lfs_gatherblock(sp, bp, NULL))
1782 			continue;
1783 	}
1784 
1785 	nestiobuf_done(mbp, skipbytes, error);
1786 	if (skipbytes) {
1787 		UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
1788 	}
1789 	UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
1790 	return (0);
1791 
1792     tryagain:
1793 	/*
1794 	 * We can't write the pages, for whatever reason.
1795 	 * Clean up after ourselves, and make the caller try again.
1796 	 */
1797 	mutex_enter(&vp->v_interlock);
1798 
1799 	/* Tell why we're here, if we know */
1800 	if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) {
1801 		DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n"));
1802 	} else if ((pgs[0]->offset & fs->lfs_bmask) != 0) {
1803 		DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n"));
1804 	} else if (haveeof && startoffset >= eof) {
1805 		DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64
1806 		      " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number,
1807 		      pgs[0]->offset, eof, npages));
1808 	} else if (LFS_STARVED_FOR_SEGS(fs)) {
1809 		DLOG((DLOG_PAGE, "lfs_gop_write: avail too low\n"));
1810 	} else {
1811 		DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n"));
1812 	}
1813 
1814 	mutex_enter(&uvm_pageqlock);
1815 	for (i = 0; i < npages; i++) {
1816 		pg = pgs[i];
1817 
1818 		if (pg->flags & PG_PAGEOUT)
1819 			uvm_pageout_done(1);
1820 		if (pg->flags & PG_DELWRI) {
1821 			uvm_pageunwire(pg);
1822 		}
1823 		uvm_pageactivate(pg);
1824 		pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
1825 		DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg,
1826 			vp, pg->offset));
1827 		DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags));
1828 		DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags));
1829 		DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon));
1830 		DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject));
1831 		DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i,
1832 		      pg->wire_count));
1833 		DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i,
1834 		      pg->loan_count));
1835 	}
1836 	/* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
1837 	uvm_page_unbusy(pgs, npages);
1838 	mutex_exit(&uvm_pageqlock);
1839 	mutex_exit(&vp->v_interlock);
1840 	return EAGAIN;
1841 }
1842 
1843 /*
1844  * finish vnode/inode initialization.
1845  * used by lfs_vget and lfs_fastvget.
1846  */
1847 void
1848 lfs_vinit(struct mount *mp, struct vnode **vpp)
1849 {
1850 	struct vnode *vp = *vpp;
1851 	struct inode *ip = VTOI(vp);
1852 	struct ufsmount *ump = VFSTOUFS(mp);
1853 	struct lfs *fs = ump->um_lfs;
1854 	int i;
1855 
1856 	ip->i_mode = ip->i_ffs1_mode;
1857 	ip->i_nlink = ip->i_ffs1_nlink;
1858 	ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size;
1859 	ip->i_flags = ip->i_ffs1_flags;
1860 	ip->i_gen = ip->i_ffs1_gen;
1861 	ip->i_uid = ip->i_ffs1_uid;
1862 	ip->i_gid = ip->i_ffs1_gid;
1863 
1864 	ip->i_lfs_effnblks = ip->i_ffs1_blocks;
1865 	ip->i_lfs_odnlink = ip->i_ffs1_nlink;
1866 
1867 	/*
1868 	 * Initialize the vnode from the inode, check for aliases.  In all
1869 	 * cases re-init ip, the underlying vnode/inode may have changed.
1870 	 */
1871 	ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
1872 	ip = VTOI(vp);
1873 
1874 	memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize));
1875 	if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) {
1876 #ifdef DEBUG
1877 		for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift;
1878 		    i < NDADDR; i++) {
1879 			if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
1880 			    i == 0)
1881 				continue;
1882 			if (ip->i_ffs1_db[i] != 0) {
1883 inconsistent:
1884 				lfs_dump_dinode(ip->i_din.ffs1_din);
1885 				panic("inconsistent inode");
1886 			}
1887 		}
1888 		for ( ; i < NDADDR + NIADDR; i++) {
1889 			if (ip->i_ffs1_ib[i - NDADDR] != 0) {
1890 				goto inconsistent;
1891 			}
1892 		}
1893 #endif /* DEBUG */
1894 		for (i = 0; i < NDADDR; i++)
1895 			if (ip->i_ffs1_db[i] != 0)
1896 				ip->i_lfs_fragsize[i] = blksize(fs, ip, i);
1897 	}
1898 
1899 #ifdef DIAGNOSTIC
1900 	if (vp->v_type == VNON) {
1901 # ifdef DEBUG
1902 		lfs_dump_dinode(ip->i_din.ffs1_din);
1903 # endif
1904 		panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n",
1905 		      (unsigned long long)ip->i_number,
1906 		      (ip->i_mode & IFMT) >> 12);
1907 	}
1908 #endif /* DIAGNOSTIC */
1909 
1910 	/*
1911 	 * Finish inode initialization now that aliasing has been resolved.
1912 	 */
1913 
1914 	ip->i_devvp = ump->um_devvp;
1915 	VREF(ip->i_devvp);
1916 	genfs_node_init(vp, &lfs_genfsops);
1917 	uvm_vnp_setsize(vp, ip->i_size);
1918 
1919 	/* Initialize hiblk from file size */
1920 	ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1;
1921 
1922 	*vpp = vp;
1923 }
1924 
1925 /*
1926  * Resize the filesystem to contain the specified number of segments.
1927  */
1928 int
1929 lfs_resize_fs(struct lfs *fs, int newnsegs)
1930 {
1931 	SEGUSE *sup;
1932 	struct buf *bp, *obp;
1933 	daddr_t olast, nlast, ilast, noff, start, end;
1934 	struct vnode *ivp;
1935 	struct inode *ip;
1936 	int error, badnews, inc, oldnsegs;
1937 	int sbbytes, csbbytes, gain, cgain;
1938 	int i;
1939 
1940 	/* Only support v2 and up */
1941 	if (fs->lfs_version < 2)
1942 		return EOPNOTSUPP;
1943 
1944 	/* If we're doing nothing, do it fast */
1945 	oldnsegs = fs->lfs_nseg;
1946 	if (newnsegs == oldnsegs)
1947 		return 0;
1948 
1949 	/* We always have to have two superblocks */
1950 	if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1]))
1951 		return EFBIG;
1952 
1953 	ivp = fs->lfs_ivnode;
1954 	ip = VTOI(ivp);
1955 	error = 0;
1956 
1957 	/* Take the segment lock so no one else calls lfs_newseg() */
1958 	lfs_seglock(fs, SEGM_PROT);
1959 
1960 	/*
1961 	 * Make sure the segments we're going to be losing, if any,
1962 	 * are in fact empty.  We hold the seglock, so their status
1963 	 * cannot change underneath us.  Count the superblocks we lose,
1964 	 * while we're at it.
1965 	 */
1966 	sbbytes = csbbytes = 0;
1967 	cgain = 0;
1968 	for (i = newnsegs; i < oldnsegs; i++) {
1969 		LFS_SEGENTRY(sup, fs, i, bp);
1970 		badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL);
1971 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
1972 			sbbytes += LFS_SBPAD;
1973 		if (!(sup->su_flags & SEGUSE_DIRTY)) {
1974 			++cgain;
1975 			if (sup->su_flags & SEGUSE_SUPERBLOCK)
1976 				csbbytes += LFS_SBPAD;
1977 		}
1978 		brelse(bp, 0);
1979 		if (badnews) {
1980 			error = EBUSY;
1981 			goto out;
1982 		}
1983 	}
1984 
1985 	/* Note old and new segment table endpoints, and old ifile size */
1986 	olast = fs->lfs_cleansz + fs->lfs_segtabsz;
1987 	nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz;
1988 	ilast = ivp->v_size >> fs->lfs_bshift;
1989 	noff = nlast - olast;
1990 
1991 	/*
1992 	 * Make sure no one can use the Ifile while we change it around.
1993 	 * Even after taking the iflock we need to make sure no one still
1994 	 * is holding Ifile buffers, so we get each one, to drain them.
1995 	 * (XXX this could be done better.)
1996 	 */
1997 	rw_enter(&fs->lfs_iflock, RW_WRITER);
1998 	vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY);
1999 	for (i = 0; i < ilast; i++) {
2000 		bread(ivp, i, fs->lfs_bsize, NOCRED, 0, &bp);
2001 		brelse(bp, 0);
2002 	}
2003 
2004 	/* Allocate new Ifile blocks */
2005 	for (i = ilast; i < ilast + noff; i++) {
2006 		if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0,
2007 			       &bp) != 0)
2008 			panic("balloc extending ifile");
2009 		memset(bp->b_data, 0, fs->lfs_bsize);
2010 		VOP_BWRITE(bp);
2011 	}
2012 
2013 	/* Register new ifile size */
2014 	ip->i_size += noff * fs->lfs_bsize;
2015 	ip->i_ffs1_size = ip->i_size;
2016 	uvm_vnp_setsize(ivp, ip->i_size);
2017 
2018 	/* Copy the inode table to its new position */
2019 	if (noff != 0) {
2020 		if (noff < 0) {
2021 			start = nlast;
2022 			end = ilast + noff;
2023 			inc = 1;
2024 		} else {
2025 			start = ilast + noff - 1;
2026 			end = nlast - 1;
2027 			inc = -1;
2028 		}
2029 		for (i = start; i != end; i += inc) {
2030 			if (bread(ivp, i, fs->lfs_bsize, NOCRED,
2031 			    B_MODIFY, &bp) != 0)
2032 				panic("resize: bread dst blk failed");
2033 			if (bread(ivp, i - noff, fs->lfs_bsize,
2034 			    NOCRED, 0, &obp))
2035 				panic("resize: bread src blk failed");
2036 			memcpy(bp->b_data, obp->b_data, fs->lfs_bsize);
2037 			VOP_BWRITE(bp);
2038 			brelse(obp, 0);
2039 		}
2040 	}
2041 
2042 	/* If we are expanding, write the new empty SEGUSE entries */
2043 	if (newnsegs > oldnsegs) {
2044 		for (i = oldnsegs; i < newnsegs; i++) {
2045 			if ((error = bread(ivp, i / fs->lfs_sepb +
2046 					   fs->lfs_cleansz, fs->lfs_bsize,
2047 					   NOCRED, B_MODIFY, &bp)) != 0)
2048 				panic("lfs: ifile read: %d", error);
2049 			while ((i + 1) % fs->lfs_sepb && i < newnsegs) {
2050 				sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb];
2051 				memset(sup, 0, sizeof(*sup));
2052 				i++;
2053 			}
2054 			VOP_BWRITE(bp);
2055 		}
2056 	}
2057 
2058 	/* Zero out unused superblock offsets */
2059 	for (i = 2; i < LFS_MAXNUMSB; i++)
2060 		if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs)
2061 			fs->lfs_sboffs[i] = 0x0;
2062 
2063 	/*
2064 	 * Correct superblock entries that depend on fs size.
2065 	 * The computations of these are as follows:
2066 	 *
2067 	 * size  = segtod(fs, nseg)
2068 	 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD)
2069 	 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used
2070 	 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD)
2071 	 *         + (segtod(fs, 1) - (offset - curseg))
2072 	 *	   - segtod(fs, minfreeseg - (minfreeseg / 2))
2073 	 *
2074 	 * XXX - we should probably adjust minfreeseg as well.
2075 	 */
2076 	gain = (newnsegs - oldnsegs);
2077 	fs->lfs_nseg = newnsegs;
2078 	fs->lfs_segtabsz = nlast - fs->lfs_cleansz;
2079 	fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize);
2080 	fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes);
2081 	fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes)
2082 		       - gain * btofsb(fs, fs->lfs_bsize / 2);
2083 	if (gain > 0) {
2084 		fs->lfs_nclean += gain;
2085 		fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize);
2086 	} else {
2087 		fs->lfs_nclean -= cgain;
2088 		fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) -
2089 				 btofsb(fs, csbbytes);
2090 	}
2091 
2092 	/* Resize segment flag cache */
2093 	fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0],
2094 						  fs->lfs_nseg * sizeof(u_int32_t),
2095 						  M_SEGMENT, M_WAITOK);
2096 	fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[1],
2097 						  fs->lfs_nseg * sizeof(u_int32_t),
2098 						  M_SEGMENT, M_WAITOK);
2099 	for (i = oldnsegs; i < newnsegs; i++)
2100 		fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0;
2101 
2102 	/* Truncate Ifile if necessary */
2103 	if (noff < 0)
2104 		lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0,
2105 		    NOCRED);
2106 
2107 	/* Update cleaner info so the cleaner can die */
2108 	bread(ivp, 0, fs->lfs_bsize, NOCRED, B_MODIFY, &bp);
2109 	((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean;
2110 	((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean;
2111 	VOP_BWRITE(bp);
2112 
2113 	/* Let Ifile accesses proceed */
2114 	VOP_UNLOCK(ivp, 0);
2115 	rw_exit(&fs->lfs_iflock);
2116 
2117     out:
2118 	lfs_segunlock(fs);
2119 	return error;
2120 }
2121