xref: /netbsd-src/sys/ufs/lfs/lfs_vfsops.c (revision 274254cdae52594c1aa480a736aef78313d15c9c)
1 /*	$NetBSD: lfs_vfsops.c,v 1.272 2009/04/04 10:33:59 ad 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 1.272 2009/04/04 10:33:59 ad 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 nameidata nd;
557 	struct vnode *devvp;
558 	struct ufs_args *args = data;
559 	struct ufsmount *ump = NULL;
560 	struct lfs *fs = NULL;				/* LFS */
561 	int error = 0, update;
562 	mode_t accessmode;
563 
564 	if (*data_len < sizeof *args)
565 		return EINVAL;
566 
567 	if (mp->mnt_flag & MNT_GETARGS) {
568 		ump = VFSTOUFS(mp);
569 		if (ump == NULL)
570 			return EIO;
571 		args->fspec = NULL;
572 		*data_len = sizeof *args;
573 		return 0;
574 	}
575 
576 	update = mp->mnt_flag & MNT_UPDATE;
577 
578 	/* Check arguments */
579 	if (args->fspec != NULL) {
580 		/*
581 		 * Look up the name and verify that it's sane.
582 		 */
583 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, args->fspec);
584 		if ((error = namei(&nd)) != 0)
585 			return (error);
586 		devvp = nd.ni_vp;
587 
588 		if (!update) {
589 			/*
590 			 * Be sure this is a valid block device
591 			 */
592 			if (devvp->v_type != VBLK)
593 				error = ENOTBLK;
594 			else if (bdevsw_lookup(devvp->v_rdev) == NULL)
595 				error = ENXIO;
596 		} else {
597 			/*
598 			 * Be sure we're still naming the same device
599 			 * used for our initial mount
600 			 */
601 			ump = VFSTOUFS(mp);
602 			if (devvp != ump->um_devvp)
603 				error = EINVAL;
604 		}
605 	} else {
606 		if (!update) {
607 			/* New mounts must have a filename for the device */
608 			return (EINVAL);
609 		} else {
610 			/* Use the extant mount */
611 			ump = VFSTOUFS(mp);
612 			devvp = ump->um_devvp;
613 			vref(devvp);
614 		}
615 	}
616 
617 
618 	/*
619 	 * If mount by non-root, then verify that user has necessary
620 	 * permissions on the device.
621 	 */
622 	if (error == 0 && kauth_authorize_generic(l->l_cred,
623 	    KAUTH_GENERIC_ISSUSER, NULL) != 0) {
624 		accessmode = VREAD;
625 		if (update ?
626 		    (mp->mnt_iflag & IMNT_WANTRDWR) != 0 :
627 		    (mp->mnt_flag & MNT_RDONLY) == 0)
628 			accessmode |= VWRITE;
629 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
630 		error = VOP_ACCESS(devvp, accessmode, l->l_cred);
631 		VOP_UNLOCK(devvp, 0);
632 	}
633 
634 	if (error) {
635 		vrele(devvp);
636 		return (error);
637 	}
638 
639 	if (!update) {
640 		int flags;
641 
642 		if (mp->mnt_flag & MNT_RDONLY)
643 			flags = FREAD;
644 		else
645 			flags = FREAD|FWRITE;
646 		error = VOP_OPEN(devvp, flags, FSCRED);
647 		if (error)
648 			goto fail;
649 		error = lfs_mountfs(devvp, mp, l);		/* LFS */
650 		if (error) {
651 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
652 			(void)VOP_CLOSE(devvp, flags, NOCRED);
653 			VOP_UNLOCK(devvp, 0);
654 			goto fail;
655 		}
656 
657 		ump = VFSTOUFS(mp);
658 		fs = ump->um_lfs;
659 	} else {
660 		/*
661 		 * Update the mount.
662 		 */
663 
664 		/*
665 		 * The initial mount got a reference on this
666 		 * device, so drop the one obtained via
667 		 * namei(), above.
668 		 */
669 		vrele(devvp);
670 
671 		ump = VFSTOUFS(mp);
672 		fs = ump->um_lfs;
673 		if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
674 			/*
675 			 * Changing from read-only to read/write.
676 			 * Note in the superblocks that we're writing.
677 			 */
678 			fs->lfs_ronly = 0;
679 			if (fs->lfs_pflags & LFS_PF_CLEAN) {
680 				fs->lfs_pflags &= ~LFS_PF_CLEAN;
681 				lfs_writesuper(fs, fs->lfs_sboffs[0]);
682 				lfs_writesuper(fs, fs->lfs_sboffs[1]);
683 			}
684 		}
685 		if (args->fspec == NULL)
686 			return EINVAL;
687 	}
688 
689 	error = set_statvfs_info(path, UIO_USERSPACE, args->fspec,
690 	    UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l);
691 	if (error == 0)
692 		(void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname,
693 			      sizeof(fs->lfs_fsmnt));
694 	return error;
695 
696 fail:
697 	vrele(devvp);
698 	return (error);
699 }
700 
701 
702 /*
703  * Common code for mount and mountroot
704  * LFS specific
705  */
706 int
707 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l)
708 {
709 	struct dlfs *tdfs, *dfs, *adfs;
710 	struct lfs *fs;
711 	struct ufsmount *ump;
712 	struct vnode *vp;
713 	struct buf *bp, *abp;
714 	struct partinfo dpart;
715 	dev_t dev;
716 	int error, i, ronly, secsize, fsbsize;
717 	kauth_cred_t cred;
718 	CLEANERINFO *cip;
719 	SEGUSE *sup;
720 	daddr_t sb_addr;
721 
722 	cred = l ? l->l_cred : NOCRED;
723 
724 	/*
725 	 * Flush out any old buffers remaining from a previous use.
726 	 */
727 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
728 	error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0);
729 	VOP_UNLOCK(devvp, 0);
730 	if (error)
731 		return (error);
732 
733 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
734 	if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred) != 0)
735 		secsize = DEV_BSIZE;
736 	else
737 		secsize = dpart.disklab->d_secsize;
738 
739 	/* Don't free random space on error. */
740 	bp = NULL;
741 	abp = NULL;
742 	ump = NULL;
743 
744 	sb_addr = LFS_LABELPAD / secsize;
745 	while (1) {
746 		/* Read in the superblock. */
747 		error = bread(devvp, sb_addr, LFS_SBPAD, cred, 0, &bp);
748 		if (error)
749 			goto out;
750 		dfs = (struct dlfs *)bp->b_data;
751 
752 		/* Check the basics. */
753 		if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE ||
754 		    dfs->dlfs_version > LFS_VERSION ||
755 		    dfs->dlfs_bsize < sizeof(struct dlfs)) {
756 			DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n"));
757 			error = EINVAL;		/* XXX needs translation */
758 			goto out;
759 		}
760 		if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) {
761 			DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n",
762 			       dfs->dlfs_inodefmt));
763 			error = EINVAL;
764 			goto out;
765 		}
766 
767 		if (dfs->dlfs_version == 1)
768 			fsbsize = secsize;
769 		else {
770 			fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb +
771 				dfs->dlfs_fsbtodb);
772 			/*
773 			 * Could be, if the frag size is large enough, that we
774 			 * don't have the "real" primary superblock.  If that's
775 			 * the case, get the real one, and try again.
776 			 */
777 			if (sb_addr != dfs->dlfs_sboffs[0] <<
778 				       dfs->dlfs_fsbtodb) {
779 				DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr"
780 				      " 0x%llx is not right, trying 0x%llx\n",
781 				      (long long)sb_addr,
782 				      (long long)(dfs->dlfs_sboffs[0] <<
783 						  dfs->dlfs_fsbtodb)));
784 				sb_addr = dfs->dlfs_sboffs[0] <<
785 					  dfs->dlfs_fsbtodb;
786 				brelse(bp, 0);
787 				continue;
788 			}
789 		}
790 		break;
791 	}
792 
793 	/*
794 	 * Check the second superblock to see which is newer; then mount
795 	 * using the older of the two.	This is necessary to ensure that
796 	 * the filesystem is valid if it was not unmounted cleanly.
797 	 */
798 
799 	if (dfs->dlfs_sboffs[1] &&
800 	    dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
801 	{
802 		error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize),
803 			LFS_SBPAD, cred, 0, &abp);
804 		if (error)
805 			goto out;
806 		adfs = (struct dlfs *)abp->b_data;
807 
808 		if (dfs->dlfs_version == 1) {
809 			/* 1s resolution comparison */
810 			if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
811 				tdfs = adfs;
812 			else
813 				tdfs = dfs;
814 		} else {
815 			/* monotonic infinite-resolution comparison */
816 			if (adfs->dlfs_serial < dfs->dlfs_serial)
817 				tdfs = adfs;
818 			else
819 				tdfs = dfs;
820 		}
821 
822 		/* Check the basics. */
823 		if (tdfs->dlfs_magic != LFS_MAGIC ||
824 		    tdfs->dlfs_bsize > MAXBSIZE ||
825 		    tdfs->dlfs_version > LFS_VERSION ||
826 		    tdfs->dlfs_bsize < sizeof(struct dlfs)) {
827 			DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock"
828 			      " sanity failed\n"));
829 			error = EINVAL;		/* XXX needs translation */
830 			goto out;
831 		}
832 	} else {
833 		DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock"
834 		      " daddr=0x%x\n", dfs->dlfs_sboffs[1]));
835 		error = EINVAL;
836 		goto out;
837 	}
838 
839 	/* Allocate the mount structure, copy the superblock into it. */
840 	fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO);
841 	memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs));
842 
843 	/* Compatibility */
844 	if (fs->lfs_version < 2) {
845 		fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
846 		fs->lfs_ibsize = fs->lfs_bsize;
847 		fs->lfs_start = fs->lfs_sboffs[0];
848 		fs->lfs_tstamp = fs->lfs_otstamp;
849 		fs->lfs_fsbtodb = 0;
850 	}
851 	if (fs->lfs_resvseg == 0)
852 		fs->lfs_resvseg = MIN(fs->lfs_minfreeseg - 1, \
853 			MAX(MIN_RESV_SEGS, fs->lfs_minfreeseg / 2 + 1));
854 
855 	/*
856 	 * If we aren't going to be able to write meaningfully to this
857 	 * filesystem, and were not mounted readonly, bomb out now.
858 	 */
859 	if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) {
860 		DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write,"
861 		      " we need BUFPAGES >= %lld\n",
862 		      (long long)((bufmem_hiwater / bufmem_lowater) *
863 				  LFS_INVERSE_MAX_BYTES(
864 					  fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT)));
865 		free(fs, M_UFSMNT);
866 		error = EFBIG; /* XXX needs translation */
867 		goto out;
868 	}
869 
870 	/* Before rolling forward, lock so vget will sleep for other procs */
871 	if (l != NULL) {
872 		fs->lfs_flags = LFS_NOTYET;
873 		fs->lfs_rfpid = l->l_proc->p_pid;
874 	}
875 
876 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
877 	ump->um_lfs = fs;
878 	ump->um_ops = &lfs_ufsops;
879 	ump->um_fstype = UFS1;
880 	if (sizeof(struct lfs) < LFS_SBPAD) {			/* XXX why? */
881 		brelse(bp, BC_INVAL);
882 		brelse(abp, BC_INVAL);
883 	} else {
884 		brelse(bp, 0);
885 		brelse(abp, 0);
886 	}
887 	bp = NULL;
888 	abp = NULL;
889 
890 
891 	/* Set up the I/O information */
892 	fs->lfs_devbsize = secsize;
893 	fs->lfs_iocount = 0;
894 	fs->lfs_diropwait = 0;
895 	fs->lfs_activesb = 0;
896 	fs->lfs_uinodes = 0;
897 	fs->lfs_ravail = 0;
898 	fs->lfs_favail = 0;
899 	fs->lfs_sbactive = 0;
900 
901 	/* Set up the ifile and lock aflags */
902 	fs->lfs_doifile = 0;
903 	fs->lfs_writer = 0;
904 	fs->lfs_dirops = 0;
905 	fs->lfs_nadirop = 0;
906 	fs->lfs_seglock = 0;
907 	fs->lfs_pdflush = 0;
908 	fs->lfs_sleepers = 0;
909 	fs->lfs_pages = 0;
910 	rw_init(&fs->lfs_fraglock);
911 	rw_init(&fs->lfs_iflock);
912 	cv_init(&fs->lfs_stopcv, "lfsstop");
913 
914 	/* Set the file system readonly/modify bits. */
915 	fs->lfs_ronly = ronly;
916 	if (ronly == 0)
917 		fs->lfs_fmod = 1;
918 
919 	/* Initialize the mount structure. */
920 	dev = devvp->v_rdev;
921 	mp->mnt_data = ump;
922 	mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
923 	mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS);
924 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
925 	mp->mnt_stat.f_namemax = LFS_MAXNAMLEN;
926 	mp->mnt_stat.f_iosize = fs->lfs_bsize;
927 	mp->mnt_flag |= MNT_LOCAL;
928 	mp->mnt_fs_bshift = fs->lfs_bshift;
929 	ump->um_flags = 0;
930 	ump->um_mountp = mp;
931 	ump->um_dev = dev;
932 	ump->um_devvp = devvp;
933 	ump->um_bptrtodb = fs->lfs_fsbtodb;
934 	ump->um_seqinc = fragstofsb(fs, fs->lfs_frag);
935 	ump->um_nindir = fs->lfs_nindir;
936 	ump->um_lognindir = ffs(fs->lfs_nindir) - 1;
937 	for (i = 0; i < MAXQUOTAS; i++)
938 		ump->um_quotas[i] = NULLVP;
939 	ump->um_maxsymlinklen = fs->lfs_maxsymlinklen;
940 	ump->um_dirblksiz = DIRBLKSIZ;
941 	ump->um_maxfilesize = fs->lfs_maxfilesize;
942 	if (ump->um_maxsymlinklen > 0)
943 		mp->mnt_iflag |= IMNT_DTYPE;
944 	devvp->v_specmountpoint = mp;
945 
946 	/* Set up reserved memory for pageout */
947 	lfs_setup_resblks(fs);
948 	/* Set up vdirop tailq */
949 	TAILQ_INIT(&fs->lfs_dchainhd);
950 	/* and paging tailq */
951 	TAILQ_INIT(&fs->lfs_pchainhd);
952 	/* and delayed segment accounting for truncation list */
953 	LIST_INIT(&fs->lfs_segdhd);
954 
955 	/*
956 	 * We use the ifile vnode for almost every operation.  Instead of
957 	 * retrieving it from the hash table each time we retrieve it here,
958 	 * artificially increment the reference count and keep a pointer
959 	 * to it in the incore copy of the superblock.
960 	 */
961 	if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
962 		DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error));
963 		goto out;
964 	}
965 	fs->lfs_ivnode = vp;
966 	VREF(vp);
967 
968 	/* Set up inode bitmap and order free list */
969 	lfs_order_freelist(fs);
970 
971 	/* Set up segment usage flags for the autocleaner. */
972 	fs->lfs_nactive = 0;
973 	fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *),
974 						M_SEGMENT, M_WAITOK);
975 	fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
976 						 M_SEGMENT, M_WAITOK);
977 	fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
978 						 M_SEGMENT, M_WAITOK);
979 	memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t));
980 	for (i = 0; i < fs->lfs_nseg; i++) {
981 		int changed;
982 
983 		LFS_SEGENTRY(sup, fs, i, bp);
984 		changed = 0;
985 		if (!ronly) {
986 			if (sup->su_nbytes == 0 &&
987 			    !(sup->su_flags & SEGUSE_EMPTY)) {
988 				sup->su_flags |= SEGUSE_EMPTY;
989 				++changed;
990 			} else if (!(sup->su_nbytes == 0) &&
991 				   (sup->su_flags & SEGUSE_EMPTY)) {
992 				sup->su_flags &= ~SEGUSE_EMPTY;
993 				++changed;
994 			}
995 			if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) {
996 				sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL);
997 				++changed;
998 			}
999 		}
1000 		fs->lfs_suflags[0][i] = sup->su_flags;
1001 		if (changed)
1002 			LFS_WRITESEGENTRY(sup, fs, i, bp);
1003 		else
1004 			brelse(bp, 0);
1005 	}
1006 
1007 #ifdef LFS_KERNEL_RFW
1008 	lfs_roll_forward(fs, mp, l);
1009 #endif
1010 
1011 	/* If writing, sb is not clean; record in case of immediate crash */
1012 	if (!fs->lfs_ronly) {
1013 		fs->lfs_pflags &= ~LFS_PF_CLEAN;
1014 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
1015 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
1016 	}
1017 
1018 	/* Allow vget now that roll-forward is complete */
1019 	fs->lfs_flags &= ~(LFS_NOTYET);
1020 	wakeup(&fs->lfs_flags);
1021 
1022 	/*
1023 	 * Initialize the ifile cleaner info with information from
1024 	 * the superblock.
1025 	 */
1026 	LFS_CLEANERINFO(cip, fs, bp);
1027 	cip->clean = fs->lfs_nclean;
1028 	cip->dirty = fs->lfs_nseg - fs->lfs_nclean;
1029 	cip->avail = fs->lfs_avail;
1030 	cip->bfree = fs->lfs_bfree;
1031 	(void) LFS_BWRITE_LOG(bp); /* Ifile */
1032 
1033 	/*
1034 	 * Mark the current segment as ACTIVE, since we're going to
1035 	 * be writing to it.
1036 	 */
1037 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);
1038 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1039 	fs->lfs_nactive++;
1040 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);  /* Ifile */
1041 
1042 	/* Now that roll-forward is done, unlock the Ifile */
1043 	vput(vp);
1044 
1045 	/* Start the pagedaemon-anticipating daemon */
1046 	if (lfs_writer_daemon == 0 && kthread_create(PRI_BIO, 0, NULL,
1047 	    lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1048 		panic("fork lfs_writer");
1049 
1050 	printf("WARNING: the log-structured file system is experimental\n"
1051 	    "WARNING: it may cause system crashes and/or corrupt data\n");
1052 
1053 	return (0);
1054 
1055 out:
1056 	if (bp)
1057 		brelse(bp, 0);
1058 	if (abp)
1059 		brelse(abp, 0);
1060 	if (ump) {
1061 		free(ump->um_lfs, M_UFSMNT);
1062 		free(ump, M_UFSMNT);
1063 		mp->mnt_data = NULL;
1064 	}
1065 
1066 	return (error);
1067 }
1068 
1069 /*
1070  * unmount system call
1071  */
1072 int
1073 lfs_unmount(struct mount *mp, int mntflags)
1074 {
1075 	struct lwp *l = curlwp;
1076 	struct ufsmount *ump;
1077 	struct lfs *fs;
1078 	int error, flags, ronly;
1079 	vnode_t *vp;
1080 
1081 	flags = 0;
1082 	if (mntflags & MNT_FORCE)
1083 		flags |= FORCECLOSE;
1084 
1085 	ump = VFSTOUFS(mp);
1086 	fs = ump->um_lfs;
1087 
1088 	/* Two checkpoints */
1089 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1090 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1091 
1092 	/* wake up the cleaner so it can die */
1093 	lfs_wakeup_cleaner(fs);
1094 	mutex_enter(&lfs_lock);
1095 	while (fs->lfs_sleepers)
1096 		mtsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1097 			&lfs_lock);
1098 	mutex_exit(&lfs_lock);
1099 
1100 #ifdef QUOTA
1101 	if (mp->mnt_flag & MNT_QUOTA) {
1102 		int i;
1103 		error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags);
1104 		if (error)
1105 			return (error);
1106 		for (i = 0; i < MAXQUOTAS; i++) {
1107 			if (ump->um_quotas[i] == NULLVP)
1108 				continue;
1109 			quotaoff(l, mp, i);
1110 		}
1111 		/*
1112 		 * Here we fall through to vflush again to ensure
1113 		 * that we have gotten rid of all the system vnodes.
1114 		 */
1115 	}
1116 #endif
1117 	if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1118 		return (error);
1119 	if ((error = VFS_SYNC(mp, 1, l->l_cred)) != 0)
1120 		return (error);
1121 	vp = fs->lfs_ivnode;
1122 	mutex_enter(&vp->v_interlock);
1123 	if (LIST_FIRST(&vp->v_dirtyblkhd))
1124 		panic("lfs_unmount: still dirty blocks on ifile vnode");
1125 	mutex_exit(&vp->v_interlock);
1126 
1127 	/* Explicitly write the superblock, to update serial and pflags */
1128 	fs->lfs_pflags |= LFS_PF_CLEAN;
1129 	lfs_writesuper(fs, fs->lfs_sboffs[0]);
1130 	lfs_writesuper(fs, fs->lfs_sboffs[1]);
1131 	mutex_enter(&lfs_lock);
1132 	while (fs->lfs_iocount)
1133 		mtsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0,
1134 			&lfs_lock);
1135 	mutex_exit(&lfs_lock);
1136 
1137 	/* Finish with the Ifile, now that we're done with it */
1138 	vgone(fs->lfs_ivnode);
1139 
1140 	ronly = !fs->lfs_ronly;
1141 	if (ump->um_devvp->v_type != VBAD)
1142 		ump->um_devvp->v_specmountpoint = NULL;
1143 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1144 	error = VOP_CLOSE(ump->um_devvp,
1145 	    ronly ? FREAD : FREAD|FWRITE, NOCRED);
1146 	vput(ump->um_devvp);
1147 
1148 	/* Complain about page leakage */
1149 	if (fs->lfs_pages > 0)
1150 		printf("lfs_unmount: still claim %d pages (%d in subsystem)\n",
1151 			fs->lfs_pages, lfs_subsys_pages);
1152 
1153 	/* Free per-mount data structures */
1154 	free(fs->lfs_ino_bitmap, M_SEGMENT);
1155 	free(fs->lfs_suflags[0], M_SEGMENT);
1156 	free(fs->lfs_suflags[1], M_SEGMENT);
1157 	free(fs->lfs_suflags, M_SEGMENT);
1158 	lfs_free_resblks(fs);
1159 	cv_destroy(&fs->lfs_stopcv);
1160 	rw_destroy(&fs->lfs_fraglock);
1161 	rw_destroy(&fs->lfs_iflock);
1162 	free(fs, M_UFSMNT);
1163 	free(ump, M_UFSMNT);
1164 
1165 	mp->mnt_data = NULL;
1166 	mp->mnt_flag &= ~MNT_LOCAL;
1167 	return (error);
1168 }
1169 
1170 /*
1171  * Get file system statistics.
1172  *
1173  * NB: We don't lock to access the superblock here, because it's not
1174  * really that important if we get it wrong.
1175  */
1176 int
1177 lfs_statvfs(struct mount *mp, struct statvfs *sbp)
1178 {
1179 	struct lfs *fs;
1180 	struct ufsmount *ump;
1181 
1182 	ump = VFSTOUFS(mp);
1183 	fs = ump->um_lfs;
1184 	if (fs->lfs_magic != LFS_MAGIC)
1185 		panic("lfs_statvfs: magic");
1186 
1187 	sbp->f_bsize = fs->lfs_bsize;
1188 	sbp->f_frsize = fs->lfs_fsize;
1189 	sbp->f_iosize = fs->lfs_bsize;
1190 	sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks);
1191 
1192 	sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs));
1193 	KASSERT(sbp->f_bfree <= fs->lfs_dsize);
1194 #if 0
1195 	if (sbp->f_bfree < 0)
1196 		sbp->f_bfree = 0;
1197 #endif
1198 
1199 	sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs));
1200 	if (sbp->f_bfree > sbp->f_bresvd)
1201 		sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
1202 	else
1203 		sbp->f_bavail = 0;
1204 
1205 	sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs);
1206 	sbp->f_ffree = sbp->f_files - fs->lfs_nfiles;
1207 	sbp->f_favail = sbp->f_ffree;
1208 	sbp->f_fresvd = 0;
1209 	copy_statvfs_info(sbp, mp);
1210 	return (0);
1211 }
1212 
1213 /*
1214  * Go through the disk queues to initiate sandbagged IO;
1215  * go through the inodes to write those that have been modified;
1216  * initiate the writing of the super block if it has been modified.
1217  *
1218  * Note: we are always called with the filesystem marked `MPBUSY'.
1219  */
1220 int
1221 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
1222 {
1223 	int error;
1224 	struct lfs *fs;
1225 
1226 	fs = VFSTOUFS(mp)->um_lfs;
1227 	if (fs->lfs_ronly)
1228 		return 0;
1229 
1230 	/* Snapshots should not hose the syncer */
1231 	/*
1232 	 * XXX Sync can block here anyway, since we don't have a very
1233 	 * XXX good idea of how much data is pending.  If it's more
1234 	 * XXX than a segment and lfs_nextseg is close to the end of
1235 	 * XXX the log, we'll likely block.
1236 	 */
1237 	mutex_enter(&lfs_lock);
1238 	if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) {
1239 		mutex_exit(&lfs_lock);
1240 		return 0;
1241 	}
1242 	mutex_exit(&lfs_lock);
1243 
1244 	lfs_writer_enter(fs, "lfs_dirops");
1245 
1246 	/* All syncs must be checkpoints until roll-forward is implemented. */
1247 	DLOG((DLOG_FLUSH, "lfs_sync at 0x%x\n", fs->lfs_offset));
1248 	error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1249 	lfs_writer_leave(fs);
1250 #ifdef QUOTA
1251 	qsync(mp);
1252 #endif
1253 	return (error);
1254 }
1255 
1256 extern kmutex_t ufs_hashlock;
1257 
1258 /*
1259  * Look up an LFS dinode number to find its incore vnode.  If not already
1260  * in core, read it in from the specified device.  Return the inode locked.
1261  * Detection and handling of mount points must be done by the calling routine.
1262  */
1263 int
1264 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1265 {
1266 	struct lfs *fs;
1267 	struct ufs1_dinode *dip;
1268 	struct inode *ip;
1269 	struct buf *bp;
1270 	struct ifile *ifp;
1271 	struct vnode *vp;
1272 	struct ufsmount *ump;
1273 	daddr_t daddr;
1274 	dev_t dev;
1275 	int error, retries;
1276 	struct timespec ts;
1277 
1278 	memset(&ts, 0, sizeof ts);	/* XXX gcc */
1279 
1280 	ump = VFSTOUFS(mp);
1281 	dev = ump->um_dev;
1282 	fs = ump->um_lfs;
1283 
1284 	/*
1285 	 * If the filesystem is not completely mounted yet, suspend
1286 	 * any access requests (wait for roll-forward to complete).
1287 	 */
1288 	mutex_enter(&lfs_lock);
1289 	while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1290 		mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0,
1291 			&lfs_lock);
1292 	mutex_exit(&lfs_lock);
1293 
1294 retry:
1295 	if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
1296 		return (0);
1297 
1298 	if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) {
1299 		*vpp = NULL;
1300 		 return (error);
1301 	}
1302 
1303 	mutex_enter(&ufs_hashlock);
1304 	if (ufs_ihashget(dev, ino, 0) != NULL) {
1305 		mutex_exit(&ufs_hashlock);
1306 		ungetnewvnode(vp);
1307 		goto retry;
1308 	}
1309 
1310 	/* Translate the inode number to a disk address. */
1311 	if (ino == LFS_IFILE_INUM)
1312 		daddr = fs->lfs_idaddr;
1313 	else {
1314 		/* XXX bounds-check this too */
1315 		LFS_IENTRY(ifp, fs, ino, bp);
1316 		daddr = ifp->if_daddr;
1317 		if (fs->lfs_version > 1) {
1318 			ts.tv_sec = ifp->if_atime_sec;
1319 			ts.tv_nsec = ifp->if_atime_nsec;
1320 		}
1321 
1322 		brelse(bp, 0);
1323 		if (daddr == LFS_UNUSED_DADDR) {
1324 			*vpp = NULLVP;
1325 			mutex_exit(&ufs_hashlock);
1326 			ungetnewvnode(vp);
1327 			return (ENOENT);
1328 		}
1329 	}
1330 
1331 	/* Allocate/init new vnode/inode. */
1332 	lfs_vcreate(mp, ino, vp);
1333 
1334 	/*
1335 	 * Put it onto its hash chain and lock it so that other requests for
1336 	 * this inode will block if they arrive while we are sleeping waiting
1337 	 * for old data structures to be purged or for the contents of the
1338 	 * disk portion of this inode to be read.
1339 	 */
1340 	ip = VTOI(vp);
1341 	ufs_ihashins(ip);
1342 	mutex_exit(&ufs_hashlock);
1343 
1344 	/*
1345 	 * XXX
1346 	 * This may not need to be here, logically it should go down with
1347 	 * the i_devvp initialization.
1348 	 * Ask Kirk.
1349 	 */
1350 	ip->i_lfs = ump->um_lfs;
1351 
1352 	/* Read in the disk contents for the inode, copy into the inode. */
1353 	retries = 0;
1354     again:
1355 	error = bread(ump->um_devvp, fsbtodb(fs, daddr),
1356 		(fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize),
1357 		NOCRED, 0, &bp);
1358 	if (error) {
1359 		/*
1360 		 * The inode does not contain anything useful, so it would
1361 		 * be misleading to leave it on its hash chain. With mode
1362 		 * still zero, it will be unlinked and returned to the free
1363 		 * list by vput().
1364 		 */
1365 		vput(vp);
1366 		brelse(bp, 0);
1367 		*vpp = NULL;
1368 		return (error);
1369 	}
1370 
1371 	dip = lfs_ifind(fs, ino, bp);
1372 	if (dip == NULL) {
1373 		/* Assume write has not completed yet; try again */
1374 		brelse(bp, BC_INVAL);
1375 		++retries;
1376 		if (retries > LFS_IFIND_RETRIES) {
1377 #ifdef DEBUG
1378 			/* If the seglock is held look at the bpp to see
1379 			   what is there anyway */
1380 			mutex_enter(&lfs_lock);
1381 			if (fs->lfs_seglock > 0) {
1382 				struct buf **bpp;
1383 				struct ufs1_dinode *dp;
1384 				int i;
1385 
1386 				for (bpp = fs->lfs_sp->bpp;
1387 				     bpp != fs->lfs_sp->cbpp; ++bpp) {
1388 					if ((*bpp)->b_vp == fs->lfs_ivnode &&
1389 					    bpp != fs->lfs_sp->bpp) {
1390 						/* Inode block */
1391 						printf("lfs_vget: block 0x%" PRIx64 ": ",
1392 						       (*bpp)->b_blkno);
1393 						dp = (struct ufs1_dinode *)(*bpp)->b_data;
1394 						for (i = 0; i < INOPB(fs); i++)
1395 							if (dp[i].di_u.inumber)
1396 								printf("%d ", dp[i].di_u.inumber);
1397 						printf("\n");
1398 					}
1399 				}
1400 			}
1401 			mutex_exit(&lfs_lock);
1402 #endif /* DEBUG */
1403 			panic("lfs_vget: dinode not found");
1404 		}
1405 		mutex_enter(&lfs_lock);
1406 		if (fs->lfs_iocount) {
1407 			DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino));
1408 			(void)mtsleep(&fs->lfs_iocount, PRIBIO + 1,
1409 				      "lfs ifind", 1, &lfs_lock);
1410 		} else
1411 			retries = LFS_IFIND_RETRIES;
1412 		mutex_exit(&lfs_lock);
1413 		goto again;
1414 	}
1415 	*ip->i_din.ffs1_din = *dip;
1416 	brelse(bp, 0);
1417 
1418 	if (fs->lfs_version > 1) {
1419 		ip->i_ffs1_atime = ts.tv_sec;
1420 		ip->i_ffs1_atimensec = ts.tv_nsec;
1421 	}
1422 
1423 	lfs_vinit(mp, &vp);
1424 
1425 	*vpp = vp;
1426 
1427 	KASSERT(VOP_ISLOCKED(vp));
1428 
1429 	return (0);
1430 }
1431 
1432 /*
1433  * File handle to vnode
1434  */
1435 int
1436 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1437 {
1438 	struct lfid lfh;
1439 	struct buf *bp;
1440 	IFILE *ifp;
1441 	int32_t daddr;
1442 	struct lfs *fs;
1443 	vnode_t *vp;
1444 
1445 	if (fhp->fid_len != sizeof(struct lfid))
1446 		return EINVAL;
1447 
1448 	memcpy(&lfh, fhp, sizeof(lfh));
1449 	if (lfh.lfid_ino < LFS_IFILE_INUM)
1450 		return ESTALE;
1451 
1452 	fs = VFSTOUFS(mp)->um_lfs;
1453 	if (lfh.lfid_ident != fs->lfs_ident)
1454 		return ESTALE;
1455 
1456 	if (lfh.lfid_ino >
1457 	    ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) -
1458 	     fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb)
1459 		return ESTALE;
1460 
1461 	mutex_enter(&ufs_ihash_lock);
1462 	vp = ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfh.lfid_ino);
1463 	mutex_exit(&ufs_ihash_lock);
1464 	if (vp == NULL) {
1465 		LFS_IENTRY(ifp, fs, lfh.lfid_ino, bp);
1466 		daddr = ifp->if_daddr;
1467 		brelse(bp, 0);
1468 		if (daddr == LFS_UNUSED_DADDR)
1469 			return ESTALE;
1470 	}
1471 
1472 	return (ufs_fhtovp(mp, &lfh.lfid_ufid, vpp));
1473 }
1474 
1475 /*
1476  * Vnode pointer to File handle
1477  */
1478 /* ARGSUSED */
1479 int
1480 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size)
1481 {
1482 	struct inode *ip;
1483 	struct lfid lfh;
1484 
1485 	if (*fh_size < sizeof(struct lfid)) {
1486 		*fh_size = sizeof(struct lfid);
1487 		return E2BIG;
1488 	}
1489 	*fh_size = sizeof(struct lfid);
1490 	ip = VTOI(vp);
1491 	memset(&lfh, 0, sizeof(lfh));
1492 	lfh.lfid_len = sizeof(struct lfid);
1493 	lfh.lfid_ino = ip->i_number;
1494 	lfh.lfid_gen = ip->i_gen;
1495 	lfh.lfid_ident = ip->i_lfs->lfs_ident;
1496 	memcpy(fhp, &lfh, sizeof(lfh));
1497 	return (0);
1498 }
1499 
1500 /*
1501  * ufs_bmaparray callback function for writing.
1502  *
1503  * Since blocks will be written to the new segment anyway,
1504  * we don't care about current daddr of them.
1505  */
1506 static bool
1507 lfs_issequential_hole(const struct ufsmount *ump,
1508     daddr_t daddr0, daddr_t daddr1)
1509 {
1510 	daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */
1511 	daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */
1512 
1513 	KASSERT(daddr0 == UNWRITTEN ||
1514 	    (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR));
1515 	KASSERT(daddr1 == UNWRITTEN ||
1516 	    (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR));
1517 
1518 	/* NOTE: all we want to know here is 'hole or not'. */
1519 	/* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */
1520 
1521 	/*
1522 	 * treat UNWRITTENs and all resident blocks as 'contiguous'
1523 	 */
1524 	if (daddr0 != 0 && daddr1 != 0)
1525 		return true;
1526 
1527 	/*
1528 	 * both are in hole?
1529 	 */
1530 	if (daddr0 == 0 && daddr1 == 0)
1531 		return true; /* all holes are 'contiguous' for us. */
1532 
1533 	return false;
1534 }
1535 
1536 /*
1537  * lfs_gop_write functions exactly like genfs_gop_write, except that
1538  * (1) it requires the seglock to be held by its caller, and sp->fip
1539  *     to be properly initialized (it will return without re-initializing
1540  *     sp->fip, and without calling lfs_writeseg).
1541  * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1542  *     to determine how large a block it can write at once (though it does
1543  *     still use VOP_BMAP to find holes in the file);
1544  * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1545  *     (leaving lfs_writeseg to deal with the cluster blocks, so we might
1546  *     now have clusters of clusters, ick.)
1547  */
1548 static int
1549 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages,
1550     int flags)
1551 {
1552 	int i, error, run, haveeof = 0;
1553 	int fs_bshift;
1554 	vaddr_t kva;
1555 	off_t eof, offset, startoffset = 0;
1556 	size_t bytes, iobytes, skipbytes;
1557 	daddr_t lbn, blkno;
1558 	struct vm_page *pg;
1559 	struct buf *mbp, *bp;
1560 	struct vnode *devvp = VTOI(vp)->i_devvp;
1561 	struct inode *ip = VTOI(vp);
1562 	struct lfs *fs = ip->i_lfs;
1563 	struct segment *sp = fs->lfs_sp;
1564 	UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1565 
1566 	ASSERT_SEGLOCK(fs);
1567 
1568 	/* The Ifile lives in the buffer cache */
1569 	KASSERT(vp != fs->lfs_ivnode);
1570 
1571 	/*
1572 	 * We don't want to fill the disk before the cleaner has a chance
1573 	 * to make room for us.  If we're in danger of doing that, fail
1574 	 * with EAGAIN.  The caller will have to notice this, unlock
1575 	 * so the cleaner can run, relock and try again.
1576 	 *
1577 	 * We must write everything, however, if our vnode is being
1578 	 * reclaimed.
1579 	 */
1580 	if (LFS_STARVED_FOR_SEGS(fs) && vp != fs->lfs_flushvp)
1581 		goto tryagain;
1582 
1583 	/*
1584 	 * Sometimes things slip past the filters in lfs_putpages,
1585 	 * and the pagedaemon tries to write pages---problem is
1586 	 * that the pagedaemon never acquires the segment lock.
1587 	 *
1588 	 * Alternatively, pages that were clean when we called
1589 	 * genfs_putpages may have become dirty in the meantime.  In this
1590 	 * case the segment header is not properly set up for blocks
1591 	 * to be added to it.
1592 	 *
1593 	 * Unbusy and unclean the pages, and put them on the ACTIVE
1594 	 * queue under the hypothesis that they couldn't have got here
1595 	 * unless they were modified *quite* recently.
1596 	 *
1597 	 * XXXUBC that last statement is an oversimplification of course.
1598 	 */
1599 	if (!LFS_SEGLOCK_HELD(fs) ||
1600 	    (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) ||
1601 	    (pgs[0]->offset & fs->lfs_bmask) != 0) {
1602 		goto tryagain;
1603 	}
1604 
1605 	UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1606 	    vp, pgs, npages, flags);
1607 
1608 	GOP_SIZE(vp, vp->v_size, &eof, 0);
1609 	haveeof = 1;
1610 
1611 	if (vp->v_type == VREG)
1612 		fs_bshift = vp->v_mount->mnt_fs_bshift;
1613 	else
1614 		fs_bshift = DEV_BSHIFT;
1615 	error = 0;
1616 	pg = pgs[0];
1617 	startoffset = pg->offset;
1618 	KASSERT(eof >= 0);
1619 
1620 	if (startoffset >= eof) {
1621 		goto tryagain;
1622 	} else
1623 		bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1624 	skipbytes = 0;
1625 
1626 	KASSERT(bytes != 0);
1627 
1628 	/* Swap PG_DELWRI for PG_PAGEOUT */
1629 	for (i = 0; i < npages; i++) {
1630 		if (pgs[i]->flags & PG_DELWRI) {
1631 			KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
1632 			pgs[i]->flags &= ~PG_DELWRI;
1633 			pgs[i]->flags |= PG_PAGEOUT;
1634 			uvm_pageout_start(1);
1635 			mutex_enter(&uvm_pageqlock);
1636 			uvm_pageunwire(pgs[i]);
1637 			mutex_exit(&uvm_pageqlock);
1638 		}
1639 	}
1640 
1641 	/*
1642 	 * Check to make sure we're starting on a block boundary.
1643 	 * We'll check later to make sure we always write entire
1644 	 * blocks (or fragments).
1645 	 */
1646 	if (startoffset & fs->lfs_bmask)
1647 		printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n",
1648 		       startoffset, fs->lfs_bmask,
1649 		       startoffset & fs->lfs_bmask);
1650 	KASSERT((startoffset & fs->lfs_bmask) == 0);
1651 	if (bytes & fs->lfs_ffmask) {
1652 		printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
1653 		panic("lfs_gop_write: non-integer blocks");
1654 	}
1655 
1656 	/*
1657 	 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK.
1658 	 * If we would, write what we have and try again.  If we don't
1659 	 * have anything to write, we'll have to sleep.
1660 	 */
1661 	if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
1662 				      (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ?
1663 				       UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) {
1664 		DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n"));
1665 #if 0
1666 		      " with nfinfo=%d at offset 0x%x\n",
1667 		      (int)((SEGSUM *)(sp->segsum))->ss_nfinfo,
1668 		      (unsigned)fs->lfs_offset));
1669 #endif
1670 		lfs_updatemeta(sp);
1671 		lfs_release_finfo(fs);
1672 		(void) lfs_writeseg(fs, sp);
1673 
1674 		lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
1675 
1676 		/*
1677 		 * Having given up all of the pager_map we were holding,
1678 		 * we can now wait for aiodoned to reclaim it for us
1679 		 * without fear of deadlock.
1680 		 */
1681 		kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
1682 				     UVMPAGER_MAPIN_WAITOK);
1683 	}
1684 
1685 	mutex_enter(&vp->v_interlock);
1686 	vp->v_numoutput += 2; /* one for biodone, one for aiodone */
1687 	mutex_exit(&vp->v_interlock);
1688 
1689 	mbp = getiobuf(NULL, true);
1690 	UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
1691 	    vp, mbp, vp->v_numoutput, bytes);
1692 	mbp->b_bufsize = npages << PAGE_SHIFT;
1693 	mbp->b_data = (void *)kva;
1694 	mbp->b_resid = mbp->b_bcount = bytes;
1695 	mbp->b_cflags = BC_BUSY|BC_AGE;
1696 	mbp->b_iodone = uvm_aio_biodone;
1697 
1698 	bp = NULL;
1699 	for (offset = startoffset;
1700 	    bytes > 0;
1701 	    offset += iobytes, bytes -= iobytes) {
1702 		lbn = offset >> fs_bshift;
1703 		error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
1704 		    lfs_issequential_hole);
1705 		if (error) {
1706 			UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d",
1707 			    error,0,0,0);
1708 			skipbytes += bytes;
1709 			bytes = 0;
1710 			break;
1711 		}
1712 
1713 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
1714 		    bytes);
1715 		if (blkno == (daddr_t)-1) {
1716 			skipbytes += iobytes;
1717 			continue;
1718 		}
1719 
1720 		/*
1721 		 * Discover how much we can really pack into this buffer.
1722 		 */
1723 		/* If no room in the current segment, finish it up */
1724 		if (sp->sum_bytes_left < sizeof(int32_t) ||
1725 		    sp->seg_bytes_left < (1 << fs->lfs_bshift)) {
1726 			int vers;
1727 
1728 			lfs_updatemeta(sp);
1729 			vers = sp->fip->fi_version;
1730 			lfs_release_finfo(fs);
1731 			(void) lfs_writeseg(fs, sp);
1732 
1733 			lfs_acquire_finfo(fs, ip->i_number, vers);
1734 		}
1735 		/* Check both for space in segment and space in segsum */
1736 		iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
1737 					<< fs_bshift);
1738 		iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
1739 				       << fs_bshift);
1740 		KASSERT(iobytes > 0);
1741 
1742 		/* if it's really one i/o, don't make a second buf */
1743 		if (offset == startoffset && iobytes == bytes) {
1744 			bp = mbp;
1745 			/* correct overcount if there is no second buffer */
1746 			mutex_enter(&vp->v_interlock);
1747 			--vp->v_numoutput;
1748 			mutex_exit(&vp->v_interlock);
1749 		} else {
1750 			bp = getiobuf(NULL, true);
1751 			UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
1752 			    vp, bp, vp->v_numoutput, 0);
1753 			bp->b_data = (char *)kva +
1754 			    (vaddr_t)(offset - pg->offset);
1755 			bp->b_resid = bp->b_bcount = iobytes;
1756 			bp->b_cflags = BC_BUSY;
1757 			bp->b_iodone = uvm_aio_biodone1;
1758 		}
1759 
1760 		/* XXX This is silly ... is this necessary? */
1761 		mutex_enter(&bufcache_lock);
1762 		mutex_enter(&vp->v_interlock);
1763 		bgetvp(vp, bp);
1764 		mutex_exit(&vp->v_interlock);
1765 		mutex_exit(&bufcache_lock);
1766 
1767 		bp->b_lblkno = lblkno(fs, offset);
1768 		bp->b_private = mbp;
1769 		if (devvp->v_type == VBLK) {
1770 			bp->b_dev = devvp->v_rdev;
1771 		}
1772 		VOP_BWRITE(bp);
1773 		while (lfs_gatherblock(sp, bp, NULL))
1774 			continue;
1775 	}
1776 
1777 	if (skipbytes) {
1778 		UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
1779 		mutex_enter(mbp->b_objlock);
1780 		if (error) {
1781 			mbp->b_error = error;
1782 		}
1783 		mbp->b_resid -= skipbytes;
1784 		mutex_exit(mbp->b_objlock);
1785 		if (mbp->b_resid == 0) {
1786 			biodone(mbp);
1787 		}
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