xref: /netbsd-src/sys/ufs/lfs/lfs_vfsops.c (revision 7fa608457b817eca6e0977b37f758ae064f3c99c)
1 /*	$NetBSD: lfs_vfsops.c,v 1.247 2007/11/10 18:53:57 rmind Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003, 2007 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Konrad E. Schroder <perseant@hhhh.org>.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 /*-
39  * Copyright (c) 1989, 1991, 1993, 1994
40  *	The Regents of the University of California.  All rights reserved.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)lfs_vfsops.c	8.20 (Berkeley) 6/10/95
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.247 2007/11/10 18:53:57 rmind Exp $");
71 
72 #if defined(_KERNEL_OPT)
73 #include "opt_quota.h"
74 #endif
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/namei.h>
79 #include <sys/proc.h>
80 #include <sys/kernel.h>
81 #include <sys/vnode.h>
82 #include <sys/mount.h>
83 #include <sys/kthread.h>
84 #include <sys/buf.h>
85 #include <sys/device.h>
86 #include <sys/mbuf.h>
87 #include <sys/file.h>
88 #include <sys/disklabel.h>
89 #include <sys/ioctl.h>
90 #include <sys/errno.h>
91 #include <sys/malloc.h>
92 #include <sys/pool.h>
93 #include <sys/socket.h>
94 #include <sys/syslog.h>
95 #include <uvm/uvm_extern.h>
96 #include <sys/sysctl.h>
97 #include <sys/conf.h>
98 #include <sys/kauth.h>
99 
100 #include <miscfs/specfs/specdev.h>
101 
102 #include <ufs/ufs/quota.h>
103 #include <ufs/ufs/inode.h>
104 #include <ufs/ufs/ufsmount.h>
105 #include <ufs/ufs/ufs_extern.h>
106 
107 #include <uvm/uvm.h>
108 #include <uvm/uvm_stat.h>
109 #include <uvm/uvm_pager.h>
110 #include <uvm/uvm_pdaemon.h>
111 
112 #include <ufs/lfs/lfs.h>
113 #include <ufs/lfs/lfs_extern.h>
114 
115 #include <miscfs/genfs/genfs.h>
116 #include <miscfs/genfs/genfs_node.h>
117 
118 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
119 static bool lfs_issequential_hole(const struct ufsmount *,
120     daddr_t, daddr_t);
121 
122 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *);
123 
124 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc;
125 extern const struct vnodeopv_desc lfs_specop_opv_desc;
126 extern const struct vnodeopv_desc lfs_fifoop_opv_desc;
127 
128 pid_t lfs_writer_daemon = 0;
129 int lfs_do_flush = 0;
130 #ifdef LFS_KERNEL_RFW
131 int lfs_do_rfw = 0;
132 #endif
133 
134 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
135 	&lfs_vnodeop_opv_desc,
136 	&lfs_specop_opv_desc,
137 	&lfs_fifoop_opv_desc,
138 	NULL,
139 };
140 
141 struct vfsops lfs_vfsops = {
142 	MOUNT_LFS,
143 	sizeof (struct ufs_args),
144 	lfs_mount,
145 	ufs_start,
146 	lfs_unmount,
147 	ufs_root,
148 	ufs_quotactl,
149 	lfs_statvfs,
150 	lfs_sync,
151 	lfs_vget,
152 	lfs_fhtovp,
153 	lfs_vptofh,
154 	lfs_init,
155 	lfs_reinit,
156 	lfs_done,
157 	lfs_mountroot,
158 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
159 	vfs_stdextattrctl,
160 	(void *)eopnotsupp,	/* vfs_suspendctl */
161 	lfs_vnodeopv_descs,
162 	0,
163 	{ NULL, NULL },
164 };
165 VFS_ATTACH(lfs_vfsops);
166 
167 const struct genfs_ops lfs_genfsops = {
168 	.gop_size = lfs_gop_size,
169 	.gop_alloc = ufs_gop_alloc,
170 	.gop_write = lfs_gop_write,
171 	.gop_markupdate = ufs_gop_markupdate,
172 };
173 
174 static const struct ufs_ops lfs_ufsops = {
175 	.uo_itimes = NULL,
176 	.uo_update = lfs_update,
177 	.uo_truncate = lfs_truncate,
178 	.uo_valloc = lfs_valloc,
179 	.uo_vfree = lfs_vfree,
180 	.uo_balloc = lfs_balloc,
181 };
182 
183 /*
184  * XXX Same structure as FFS inodes?  Should we share a common pool?
185  */
186 struct pool lfs_inode_pool;
187 struct pool lfs_dinode_pool;
188 struct pool lfs_inoext_pool;
189 struct pool lfs_lbnentry_pool;
190 
191 /*
192  * The writer daemon.  UVM keeps track of how many dirty pages we are holding
193  * in lfs_subsys_pages; the daemon flushes the filesystem when this value
194  * crosses the (user-defined) threshhold LFS_MAX_PAGES.
195  */
196 static void
197 lfs_writerd(void *arg)
198 {
199 	struct mount *mp, *nmp;
200 	struct lfs *fs;
201 	int fsflags;
202 	int loopcount;
203 
204 	lfs_writer_daemon = curproc->p_pid;
205 
206 	simple_lock(&lfs_subsys_lock);
207 	for (;;) {
208 		ltsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", hz/10,
209 		    &lfs_subsys_lock);
210 
211 		/*
212 		 * Look through the list of LFSs to see if any of them
213 		 * have requested pageouts.
214 		 */
215 		mutex_enter(&mountlist_lock);
216 		for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
217 		     mp = nmp) {
218 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_lock)) {
219 				nmp = CIRCLEQ_NEXT(mp, mnt_list);
220 				continue;
221 			}
222 			if (strncmp(mp->mnt_stat.f_fstypename, MOUNT_LFS,
223 			    sizeof(mp->mnt_stat.f_fstypename)) == 0) {
224 				fs = VFSTOUFS(mp)->um_lfs;
225 				simple_lock(&fs->lfs_interlock);
226 				fsflags = 0;
227 				if ((fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) ||
228 				     lfs_dirvcount > LFS_MAX_DIROP) &&
229 				    fs->lfs_dirops == 0)
230 					fsflags |= SEGM_CKP;
231 				if (fs->lfs_pdflush) {
232 					DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n"));
233 					fs->lfs_pdflush = 0;
234 					lfs_flush_fs(fs, fsflags);
235 					simple_unlock(&fs->lfs_interlock);
236 				} else if (!TAILQ_EMPTY(&fs->lfs_pchainhd)) {
237 					DLOG((DLOG_FLUSH, "lfs_writerd: pchain non-empty\n"));
238 					simple_unlock(&fs->lfs_interlock);
239 					lfs_writer_enter(fs, "wrdirop");
240 					lfs_flush_pchain(fs);
241 					lfs_writer_leave(fs);
242 				} else
243 					simple_unlock(&fs->lfs_interlock);
244 			}
245 
246 			mutex_enter(&mountlist_lock);
247 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
248 			vfs_unbusy(mp);
249 		}
250 		mutex_exit(&mountlist_lock);
251 
252 		/*
253 		 * If global state wants a flush, flush everything.
254 		 */
255 		simple_lock(&lfs_subsys_lock);
256 		loopcount = 0;
257 		if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
258 			locked_queue_bytes > LFS_MAX_BYTES ||
259 			lfs_subsys_pages > LFS_MAX_PAGES) {
260 
261 			if (lfs_do_flush) {
262 				DLOG((DLOG_FLUSH, "daemon: lfs_do_flush\n"));
263 			}
264 			if (locked_queue_count > LFS_MAX_BUFS) {
265 				DLOG((DLOG_FLUSH, "daemon: lqc = %d, max %d\n",
266 				      locked_queue_count, LFS_MAX_BUFS));
267 			}
268 			if (locked_queue_bytes > LFS_MAX_BYTES) {
269 				DLOG((DLOG_FLUSH, "daemon: lqb = %ld, max %ld\n",
270 				      locked_queue_bytes, LFS_MAX_BYTES));
271 			}
272 			if (lfs_subsys_pages > LFS_MAX_PAGES) {
273 				DLOG((DLOG_FLUSH, "daemon: lssp = %d, max %d\n",
274 				      lfs_subsys_pages, LFS_MAX_PAGES));
275 			}
276 
277 			lfs_flush(NULL, SEGM_WRITERD, 0);
278 			lfs_do_flush = 0;
279 		}
280 	}
281 	/* NOTREACHED */
282 }
283 
284 /*
285  * Initialize the filesystem, most work done by ufs_init.
286  */
287 void
288 lfs_init()
289 {
290 
291 	malloc_type_attach(M_SEGMENT);
292 	pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
293 	    "lfsinopl", &pool_allocator_nointr, IPL_NONE);
294 	pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0,
295 	    "lfsdinopl", &pool_allocator_nointr, IPL_NONE);
296 	pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
297 	    "lfsinoextpl", &pool_allocator_nointr, IPL_NONE);
298 	pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0,
299 	    "lfslbnpool", &pool_allocator_nointr, IPL_NONE);
300 	ufs_init();
301 
302 #ifdef DEBUG
303 	memset(lfs_log, 0, sizeof(lfs_log));
304 #endif
305 	simple_lock_init(&lfs_subsys_lock);
306 }
307 
308 void
309 lfs_reinit()
310 {
311 	ufs_reinit();
312 }
313 
314 void
315 lfs_done()
316 {
317 
318 	ufs_done();
319 	pool_destroy(&lfs_inode_pool);
320 	pool_destroy(&lfs_dinode_pool);
321 	pool_destroy(&lfs_inoext_pool);
322 	pool_destroy(&lfs_lbnentry_pool);
323 	malloc_type_detach(M_SEGMENT);
324 }
325 
326 /*
327  * Called by main() when ufs is going to be mounted as root.
328  */
329 int
330 lfs_mountroot()
331 {
332 	extern struct vnode *rootvp;
333 	struct mount *mp;
334 	struct lwp *l = curlwp;	/* XXX */
335 	int error;
336 
337 	if (device_class(root_device) != DV_DISK)
338 		return (ENODEV);
339 
340 	if (rootdev == NODEV)
341 		return (ENODEV);
342 	if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
343 		vrele(rootvp);
344 		return (error);
345 	}
346 	if ((error = lfs_mountfs(rootvp, mp, l))) {
347 		mp->mnt_op->vfs_refcount--;
348 		vfs_unbusy(mp);
349 		free(mp, M_MOUNT);
350 		return (error);
351 	}
352 	mutex_enter(&mountlist_lock);
353 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
354 	mutex_exit(&mountlist_lock);
355 	(void)lfs_statvfs(mp, &mp->mnt_stat, l);
356 	vfs_unbusy(mp);
357 	setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp));
358 	return (0);
359 }
360 
361 /*
362  * VFS Operations.
363  *
364  * mount system call
365  */
366 int
367 lfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len,
368     struct lwp *l)
369 {
370 	struct nameidata nd;
371 	struct vnode *devvp;
372 	struct ufs_args *args = data;
373 	struct ufsmount *ump = NULL;
374 	struct lfs *fs = NULL;				/* LFS */
375 	int error = 0, update;
376 	mode_t accessmode;
377 
378 	if (*data_len < sizeof *args)
379 		return EINVAL;
380 
381 	if (mp->mnt_flag & MNT_GETARGS) {
382 		ump = VFSTOUFS(mp);
383 		if (ump == NULL)
384 			return EIO;
385 		args->fspec = NULL;
386 		*data_len = sizeof *args;
387 		return 0;
388 	}
389 
390 	update = mp->mnt_flag & MNT_UPDATE;
391 
392 	/* Check arguments */
393 	if (args->fspec != NULL) {
394 		/*
395 		 * Look up the name and verify that it's sane.
396 		 */
397 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, args->fspec, l);
398 		if ((error = namei(&nd)) != 0)
399 			return (error);
400 		devvp = nd.ni_vp;
401 
402 		if (!update) {
403 			/*
404 			 * Be sure this is a valid block device
405 			 */
406 			if (devvp->v_type != VBLK)
407 				error = ENOTBLK;
408 			else if (bdevsw_lookup(devvp->v_rdev) == NULL)
409 				error = ENXIO;
410 		} else {
411 			/*
412 			 * Be sure we're still naming the same device
413 			 * used for our initial mount
414 			 */
415 			ump = VFSTOUFS(mp);
416 			if (devvp != ump->um_devvp)
417 				error = EINVAL;
418 		}
419 	} else {
420 		if (!update) {
421 			/* New mounts must have a filename for the device */
422 			return (EINVAL);
423 		} else {
424 			/* Use the extant mount */
425 			ump = VFSTOUFS(mp);
426 			devvp = ump->um_devvp;
427 			vref(devvp);
428 		}
429 	}
430 
431 
432 	/*
433 	 * If mount by non-root, then verify that user has necessary
434 	 * permissions on the device.
435 	 */
436 	if (error == 0 && kauth_authorize_generic(l->l_cred,
437 	    KAUTH_GENERIC_ISSUSER, NULL) != 0) {
438 		accessmode = VREAD;
439 		if (update ?
440 		    (mp->mnt_iflag & IMNT_WANTRDWR) != 0 :
441 		    (mp->mnt_flag & MNT_RDONLY) == 0)
442 			accessmode |= VWRITE;
443 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
444 		error = VOP_ACCESS(devvp, accessmode, l->l_cred, l);
445 		VOP_UNLOCK(devvp, 0);
446 	}
447 
448 	if (error) {
449 		vrele(devvp);
450 		return (error);
451 	}
452 
453 	if (!update) {
454 		int flags;
455 
456 		/*
457 		 * Disallow multiple mounts of the same device.
458 		 * Disallow mounting of a device that is currently in use
459 		 * (except for root, which might share swap device for
460 		 * miniroot).
461 		 */
462 		error = vfs_mountedon(devvp);
463 		if (error)
464 			goto fail;
465 		if (vcount(devvp) > 1 && devvp != rootvp) {
466 			error = EBUSY;
467 			goto fail;
468 		}
469 		if (mp->mnt_flag & MNT_RDONLY)
470 			flags = FREAD;
471 		else
472 			flags = FREAD|FWRITE;
473 		error = VOP_OPEN(devvp, flags, FSCRED, l);
474 		if (error)
475 			goto fail;
476 		error = lfs_mountfs(devvp, mp, l);		/* LFS */
477 		if (error) {
478 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
479 			(void)VOP_CLOSE(devvp, flags, NOCRED, l);
480 			VOP_UNLOCK(devvp, 0);
481 			goto fail;
482 		}
483 
484 		ump = VFSTOUFS(mp);
485 		fs = ump->um_lfs;
486 	} else {
487 		/*
488 		 * Update the mount.
489 		 */
490 
491 		/*
492 		 * The initial mount got a reference on this
493 		 * device, so drop the one obtained via
494 		 * namei(), above.
495 		 */
496 		vrele(devvp);
497 
498 		ump = VFSTOUFS(mp);
499 		fs = ump->um_lfs;
500 		if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
501 			/*
502 			 * Changing from read-only to read/write.
503 			 * Note in the superblocks that we're writing.
504 			 */
505 			fs->lfs_ronly = 0;
506 			if (fs->lfs_pflags & LFS_PF_CLEAN) {
507 				fs->lfs_pflags &= ~LFS_PF_CLEAN;
508 				lfs_writesuper(fs, fs->lfs_sboffs[0]);
509 				lfs_writesuper(fs, fs->lfs_sboffs[1]);
510 			}
511 		}
512 		if (args->fspec == NULL)
513 			return EINVAL;
514 	}
515 
516 	error = set_statvfs_info(path, UIO_USERSPACE, args->fspec,
517 	    UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l);
518 	if (error == 0)
519 		(void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname,
520 			      sizeof(fs->lfs_fsmnt));
521 	return error;
522 
523 fail:
524 	vrele(devvp);
525 	return (error);
526 }
527 
528 
529 /*
530  * Common code for mount and mountroot
531  * LFS specific
532  */
533 int
534 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l)
535 {
536 	struct dlfs *tdfs, *dfs, *adfs;
537 	struct lfs *fs;
538 	struct ufsmount *ump;
539 	struct vnode *vp;
540 	struct buf *bp, *abp;
541 	struct partinfo dpart;
542 	dev_t dev;
543 	int error, i, ronly, secsize, fsbsize;
544 	kauth_cred_t cred;
545 	CLEANERINFO *cip;
546 	SEGUSE *sup;
547 	daddr_t sb_addr;
548 
549 	cred = l ? l->l_cred : NOCRED;
550 
551 	/*
552 	 * Flush out any old buffers remaining from a previous use.
553 	 */
554 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
555 	error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0);
556 	VOP_UNLOCK(devvp, 0);
557 	if (error)
558 		return (error);
559 
560 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
561 	if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, l) != 0)
562 		secsize = DEV_BSIZE;
563 	else
564 		secsize = dpart.disklab->d_secsize;
565 
566 	/* Don't free random space on error. */
567 	bp = NULL;
568 	abp = NULL;
569 	ump = NULL;
570 
571 	sb_addr = LFS_LABELPAD / secsize;
572 	while (1) {
573 		/* Read in the superblock. */
574 		error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp);
575 		if (error)
576 			goto out;
577 		dfs = (struct dlfs *)bp->b_data;
578 
579 		/* Check the basics. */
580 		if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE ||
581 		    dfs->dlfs_version > LFS_VERSION ||
582 		    dfs->dlfs_bsize < sizeof(struct dlfs)) {
583 			DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n"));
584 			error = EINVAL;		/* XXX needs translation */
585 			goto out;
586 		}
587 		if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) {
588 			DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n",
589 			       dfs->dlfs_inodefmt));
590 			error = EINVAL;
591 			goto out;
592 		}
593 
594 		if (dfs->dlfs_version == 1)
595 			fsbsize = secsize;
596 		else {
597 			fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb +
598 				dfs->dlfs_fsbtodb);
599 			/*
600 			 * Could be, if the frag size is large enough, that we
601 			 * don't have the "real" primary superblock.  If that's
602 			 * the case, get the real one, and try again.
603 			 */
604 			if (sb_addr != dfs->dlfs_sboffs[0] <<
605 				       dfs->dlfs_fsbtodb) {
606 				DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr"
607 				      " 0x%llx is not right, trying 0x%llx\n",
608 				      (long long)sb_addr,
609 				      (long long)(dfs->dlfs_sboffs[0] <<
610 						  dfs->dlfs_fsbtodb)));
611 				sb_addr = dfs->dlfs_sboffs[0] <<
612 					  dfs->dlfs_fsbtodb;
613 				brelse(bp, 0);
614 				continue;
615 			}
616 		}
617 		break;
618 	}
619 
620 	/*
621 	 * Check the second superblock to see which is newer; then mount
622 	 * using the older of the two.	This is necessary to ensure that
623 	 * the filesystem is valid if it was not unmounted cleanly.
624 	 */
625 
626 	if (dfs->dlfs_sboffs[1] &&
627 	    dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
628 	{
629 		error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize),
630 			LFS_SBPAD, cred, &abp);
631 		if (error)
632 			goto out;
633 		adfs = (struct dlfs *)abp->b_data;
634 
635 		if (dfs->dlfs_version == 1) {
636 			/* 1s resolution comparison */
637 			if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
638 				tdfs = adfs;
639 			else
640 				tdfs = dfs;
641 		} else {
642 			/* monotonic infinite-resolution comparison */
643 			if (adfs->dlfs_serial < dfs->dlfs_serial)
644 				tdfs = adfs;
645 			else
646 				tdfs = dfs;
647 		}
648 
649 		/* Check the basics. */
650 		if (tdfs->dlfs_magic != LFS_MAGIC ||
651 		    tdfs->dlfs_bsize > MAXBSIZE ||
652 		    tdfs->dlfs_version > LFS_VERSION ||
653 		    tdfs->dlfs_bsize < sizeof(struct dlfs)) {
654 			DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock"
655 			      " sanity failed\n"));
656 			error = EINVAL;		/* XXX needs translation */
657 			goto out;
658 		}
659 	} else {
660 		DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock"
661 		      " daddr=0x%x\n", dfs->dlfs_sboffs[1]));
662 		error = EINVAL;
663 		goto out;
664 	}
665 
666 	/* Allocate the mount structure, copy the superblock into it. */
667 	fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO);
668 	memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs));
669 
670 	/* Compatibility */
671 	if (fs->lfs_version < 2) {
672 		fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
673 		fs->lfs_ibsize = fs->lfs_bsize;
674 		fs->lfs_start = fs->lfs_sboffs[0];
675 		fs->lfs_tstamp = fs->lfs_otstamp;
676 		fs->lfs_fsbtodb = 0;
677 	}
678 	if (fs->lfs_resvseg == 0)
679 		fs->lfs_resvseg = MIN(fs->lfs_minfreeseg - 1, \
680 			MAX(MIN_RESV_SEGS, fs->lfs_minfreeseg / 2 + 1));
681 
682 	/*
683 	 * If we aren't going to be able to write meaningfully to this
684 	 * filesystem, and were not mounted readonly, bomb out now.
685 	 */
686 	if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) {
687 		DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write,"
688 		      " we need BUFPAGES >= %lld\n",
689 		      (long long)((bufmem_hiwater / bufmem_lowater) *
690 				  LFS_INVERSE_MAX_BYTES(
691 					  fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT)));
692 		free(fs, M_UFSMNT);
693 		error = EFBIG; /* XXX needs translation */
694 		goto out;
695 	}
696 
697 	/* Before rolling forward, lock so vget will sleep for other procs */
698 	if (l != NULL) {
699 		fs->lfs_flags = LFS_NOTYET;
700 		fs->lfs_rfpid = l->l_proc->p_pid;
701 	}
702 
703 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
704 	ump->um_lfs = fs;
705 	ump->um_ops = &lfs_ufsops;
706 	ump->um_fstype = UFS1;
707 	if (sizeof(struct lfs) < LFS_SBPAD) {			/* XXX why? */
708 		brelse(bp, BC_INVAL);
709 		brelse(abp, BC_INVAL);
710 	} else {
711 		brelse(bp, 0);
712 		brelse(abp, 0);
713 	}
714 	bp = NULL;
715 	abp = NULL;
716 
717 
718 	/* Set up the I/O information */
719 	fs->lfs_devbsize = secsize;
720 	fs->lfs_iocount = 0;
721 	fs->lfs_diropwait = 0;
722 	fs->lfs_activesb = 0;
723 	fs->lfs_uinodes = 0;
724 	fs->lfs_ravail = 0;
725 	fs->lfs_favail = 0;
726 	fs->lfs_sbactive = 0;
727 
728 	/* Set up the ifile and lock aflags */
729 	fs->lfs_doifile = 0;
730 	fs->lfs_writer = 0;
731 	fs->lfs_dirops = 0;
732 	fs->lfs_nadirop = 0;
733 	fs->lfs_seglock = 0;
734 	fs->lfs_pdflush = 0;
735 	fs->lfs_sleepers = 0;
736 	fs->lfs_pages = 0;
737 	simple_lock_init(&fs->lfs_interlock);
738 	rw_init(&fs->lfs_fraglock);
739 	lockinit(&fs->lfs_iflock, PINOD, "lfs_iflock", 0, 0);
740 	lockinit(&fs->lfs_stoplock, PINOD, "lfs_stoplock", 0, 0);
741 
742 	/* Set the file system readonly/modify bits. */
743 	fs->lfs_ronly = ronly;
744 	if (ronly == 0)
745 		fs->lfs_fmod = 1;
746 
747 	/* Initialize the mount structure. */
748 	dev = devvp->v_rdev;
749 	mp->mnt_data = ump;
750 	mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
751 	mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS);
752 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
753 	mp->mnt_stat.f_namemax = LFS_MAXNAMLEN;
754 	mp->mnt_stat.f_iosize = fs->lfs_bsize;
755 	mp->mnt_flag |= MNT_LOCAL;
756 	mp->mnt_fs_bshift = fs->lfs_bshift;
757 	ump->um_flags = 0;
758 	ump->um_mountp = mp;
759 	ump->um_dev = dev;
760 	ump->um_devvp = devvp;
761 	ump->um_bptrtodb = fs->lfs_fsbtodb;
762 	ump->um_seqinc = fragstofsb(fs, fs->lfs_frag);
763 	ump->um_nindir = fs->lfs_nindir;
764 	ump->um_lognindir = ffs(fs->lfs_nindir) - 1;
765 	for (i = 0; i < MAXQUOTAS; i++)
766 		ump->um_quotas[i] = NULLVP;
767 	ump->um_maxsymlinklen = fs->lfs_maxsymlinklen;
768 	ump->um_dirblksiz = DIRBLKSIZ;
769 	ump->um_maxfilesize = fs->lfs_maxfilesize;
770 	if (ump->um_maxsymlinklen > 0)
771 		mp->mnt_iflag |= IMNT_DTYPE;
772 	devvp->v_specmountpoint = mp;
773 
774 	/* Set up reserved memory for pageout */
775 	lfs_setup_resblks(fs);
776 	/* Set up vdirop tailq */
777 	TAILQ_INIT(&fs->lfs_dchainhd);
778 	/* and paging tailq */
779 	TAILQ_INIT(&fs->lfs_pchainhd);
780 	/* and delayed segment accounting for truncation list */
781 	LIST_INIT(&fs->lfs_segdhd);
782 
783 	/*
784 	 * We use the ifile vnode for almost every operation.  Instead of
785 	 * retrieving it from the hash table each time we retrieve it here,
786 	 * artificially increment the reference count and keep a pointer
787 	 * to it in the incore copy of the superblock.
788 	 */
789 	if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
790 		DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error));
791 		goto out;
792 	}
793 	fs->lfs_ivnode = vp;
794 	VREF(vp);
795 
796 	/* Set up inode bitmap and order free list */
797 	lfs_order_freelist(fs);
798 
799 	/* Set up segment usage flags for the autocleaner. */
800 	fs->lfs_nactive = 0;
801 	fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *),
802 						M_SEGMENT, M_WAITOK);
803 	fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
804 						 M_SEGMENT, M_WAITOK);
805 	fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
806 						 M_SEGMENT, M_WAITOK);
807 	memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t));
808 	for (i = 0; i < fs->lfs_nseg; i++) {
809 		int changed;
810 
811 		LFS_SEGENTRY(sup, fs, i, bp);
812 		changed = 0;
813 		if (!ronly) {
814 			if (sup->su_nbytes == 0 &&
815 			    !(sup->su_flags & SEGUSE_EMPTY)) {
816 				sup->su_flags |= SEGUSE_EMPTY;
817 				++changed;
818 			} else if (!(sup->su_nbytes == 0) &&
819 				   (sup->su_flags & SEGUSE_EMPTY)) {
820 				sup->su_flags &= ~SEGUSE_EMPTY;
821 				++changed;
822 			}
823 			if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) {
824 				sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL);
825 				++changed;
826 			}
827 		}
828 		fs->lfs_suflags[0][i] = sup->su_flags;
829 		if (changed)
830 			LFS_WRITESEGENTRY(sup, fs, i, bp);
831 		else
832 			brelse(bp, 0);
833 	}
834 
835 #ifdef LFS_KERNEL_RFW
836 	lfs_roll_forward(fs, mp, l);
837 #endif
838 
839 	/* If writing, sb is not clean; record in case of immediate crash */
840 	if (!fs->lfs_ronly) {
841 		fs->lfs_pflags &= ~LFS_PF_CLEAN;
842 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
843 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
844 	}
845 
846 	/* Allow vget now that roll-forward is complete */
847 	fs->lfs_flags &= ~(LFS_NOTYET);
848 	wakeup(&fs->lfs_flags);
849 
850 	/*
851 	 * Initialize the ifile cleaner info with information from
852 	 * the superblock.
853 	 */
854 	LFS_CLEANERINFO(cip, fs, bp);
855 	cip->clean = fs->lfs_nclean;
856 	cip->dirty = fs->lfs_nseg - fs->lfs_nclean;
857 	cip->avail = fs->lfs_avail;
858 	cip->bfree = fs->lfs_bfree;
859 	(void) LFS_BWRITE_LOG(bp); /* Ifile */
860 
861 	/*
862 	 * Mark the current segment as ACTIVE, since we're going to
863 	 * be writing to it.
864 	 */
865 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);
866 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
867 	fs->lfs_nactive++;
868 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);  /* Ifile */
869 
870 	/* Now that roll-forward is done, unlock the Ifile */
871 	vput(vp);
872 
873 	/* Start the pagedaemon-anticipating daemon */
874 	if (lfs_writer_daemon == 0 && kthread_create(PRI_BIO, 0, NULL,
875 	    lfs_writerd, NULL, NULL, "lfs_writer") != 0)
876 		panic("fork lfs_writer");
877 
878 	return (0);
879 
880 out:
881 	if (bp)
882 		brelse(bp, 0);
883 	if (abp)
884 		brelse(abp, 0);
885 	if (ump) {
886 		free(ump->um_lfs, M_UFSMNT);
887 		free(ump, M_UFSMNT);
888 		mp->mnt_data = NULL;
889 	}
890 
891 	return (error);
892 }
893 
894 /*
895  * unmount system call
896  */
897 int
898 lfs_unmount(struct mount *mp, int mntflags, struct lwp *l)
899 {
900 	struct ufsmount *ump;
901 	struct lfs *fs;
902 	int error, flags, ronly;
903 	int s;
904 
905 	flags = 0;
906 	if (mntflags & MNT_FORCE)
907 		flags |= FORCECLOSE;
908 
909 	ump = VFSTOUFS(mp);
910 	fs = ump->um_lfs;
911 
912 	/* Two checkpoints */
913 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
914 	lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
915 
916 	/* wake up the cleaner so it can die */
917 	lfs_wakeup_cleaner(fs);
918 	simple_lock(&fs->lfs_interlock);
919 	while (fs->lfs_sleepers)
920 		ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
921 			&fs->lfs_interlock);
922 	simple_unlock(&fs->lfs_interlock);
923 
924 #ifdef QUOTA
925 	if (mp->mnt_flag & MNT_QUOTA) {
926 		int i;
927 		error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags);
928 		if (error)
929 			return (error);
930 		for (i = 0; i < MAXQUOTAS; i++) {
931 			if (ump->um_quotas[i] == NULLVP)
932 				continue;
933 			quotaoff(l, mp, i);
934 		}
935 		/*
936 		 * Here we fall through to vflush again to ensure
937 		 * that we have gotten rid of all the system vnodes.
938 		 */
939 	}
940 #endif
941 	if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
942 		return (error);
943 	if ((error = VFS_SYNC(mp, 1, l->l_cred, l)) != 0)
944 		return (error);
945 	s = splbio();
946 	if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd))
947 		panic("lfs_unmount: still dirty blocks on ifile vnode");
948 	splx(s);
949 
950 	/* Explicitly write the superblock, to update serial and pflags */
951 	fs->lfs_pflags |= LFS_PF_CLEAN;
952 	lfs_writesuper(fs, fs->lfs_sboffs[0]);
953 	lfs_writesuper(fs, fs->lfs_sboffs[1]);
954 	simple_lock(&fs->lfs_interlock);
955 	while (fs->lfs_iocount)
956 		ltsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0,
957 			&fs->lfs_interlock);
958 	simple_unlock(&fs->lfs_interlock);
959 
960 	/* Finish with the Ifile, now that we're done with it */
961 	vrele(fs->lfs_ivnode);
962 	vgone(fs->lfs_ivnode);
963 
964 	ronly = !fs->lfs_ronly;
965 	if (ump->um_devvp->v_type != VBAD)
966 		ump->um_devvp->v_specmountpoint = NULL;
967 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
968 	error = VOP_CLOSE(ump->um_devvp,
969 	    ronly ? FREAD : FREAD|FWRITE, NOCRED, l);
970 	vput(ump->um_devvp);
971 
972 	/* Complain about page leakage */
973 	if (fs->lfs_pages > 0)
974 		printf("lfs_unmount: still claim %d pages (%d in subsystem)\n",
975 			fs->lfs_pages, lfs_subsys_pages);
976 
977 	/* Free per-mount data structures */
978 	free(fs->lfs_ino_bitmap, M_SEGMENT);
979 	free(fs->lfs_suflags[0], M_SEGMENT);
980 	free(fs->lfs_suflags[1], M_SEGMENT);
981 	free(fs->lfs_suflags, M_SEGMENT);
982 	lfs_free_resblks(fs);
983 	rw_destroy(&fs->lfs_fraglock);
984 	free(fs, M_UFSMNT);
985 	free(ump, M_UFSMNT);
986 
987 	mp->mnt_data = NULL;
988 	mp->mnt_flag &= ~MNT_LOCAL;
989 	return (error);
990 }
991 
992 /*
993  * Get file system statistics.
994  *
995  * NB: We don't lock to access the superblock here, because it's not
996  * really that important if we get it wrong.
997  */
998 int
999 lfs_statvfs(struct mount *mp, struct statvfs *sbp, struct lwp *l)
1000 {
1001 	struct lfs *fs;
1002 	struct ufsmount *ump;
1003 
1004 	ump = VFSTOUFS(mp);
1005 	fs = ump->um_lfs;
1006 	if (fs->lfs_magic != LFS_MAGIC)
1007 		panic("lfs_statvfs: magic");
1008 
1009 	sbp->f_bsize = fs->lfs_bsize;
1010 	sbp->f_frsize = fs->lfs_fsize;
1011 	sbp->f_iosize = fs->lfs_bsize;
1012 	sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks);
1013 
1014 	sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs));
1015 	KASSERT(sbp->f_bfree <= fs->lfs_dsize);
1016 #if 0
1017 	if (sbp->f_bfree < 0)
1018 		sbp->f_bfree = 0;
1019 #endif
1020 
1021 	sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs));
1022 	if (sbp->f_bfree > sbp->f_bresvd)
1023 		sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
1024 	else
1025 		sbp->f_bavail = 0;
1026 
1027 	sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs);
1028 	sbp->f_ffree = sbp->f_files - fs->lfs_nfiles;
1029 	sbp->f_favail = sbp->f_ffree;
1030 	sbp->f_fresvd = 0;
1031 	copy_statvfs_info(sbp, mp);
1032 	return (0);
1033 }
1034 
1035 /*
1036  * Go through the disk queues to initiate sandbagged IO;
1037  * go through the inodes to write those that have been modified;
1038  * initiate the writing of the super block if it has been modified.
1039  *
1040  * Note: we are always called with the filesystem marked `MPBUSY'.
1041  */
1042 int
1043 lfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred,
1044     struct lwp *l)
1045 {
1046 	int error;
1047 	struct lfs *fs;
1048 
1049 	fs = VFSTOUFS(mp)->um_lfs;
1050 	if (fs->lfs_ronly)
1051 		return 0;
1052 
1053 	/* Snapshots should not hose the syncer */
1054 	/*
1055 	 * XXX Sync can block here anyway, since we don't have a very
1056 	 * XXX good idea of how much data is pending.  If it's more
1057 	 * XXX than a segment and lfs_nextseg is close to the end of
1058 	 * XXX the log, we'll likely block.
1059 	 */
1060 	simple_lock(&fs->lfs_interlock);
1061 	if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) {
1062 		simple_unlock(&fs->lfs_interlock);
1063 		return 0;
1064 	}
1065 	simple_unlock(&fs->lfs_interlock);
1066 
1067 	lfs_writer_enter(fs, "lfs_dirops");
1068 
1069 	/* All syncs must be checkpoints until roll-forward is implemented. */
1070 	DLOG((DLOG_FLUSH, "lfs_sync at 0x%x\n", fs->lfs_offset));
1071 	error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1072 	lfs_writer_leave(fs);
1073 #ifdef QUOTA
1074 	qsync(mp);
1075 #endif
1076 	return (error);
1077 }
1078 
1079 extern kmutex_t ufs_hashlock;
1080 
1081 /*
1082  * Look up an LFS dinode number to find its incore vnode.  If not already
1083  * in core, read it in from the specified device.  Return the inode locked.
1084  * Detection and handling of mount points must be done by the calling routine.
1085  */
1086 int
1087 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1088 {
1089 	struct lfs *fs;
1090 	struct ufs1_dinode *dip;
1091 	struct inode *ip;
1092 	struct buf *bp;
1093 	struct ifile *ifp;
1094 	struct vnode *vp;
1095 	struct ufsmount *ump;
1096 	daddr_t daddr;
1097 	dev_t dev;
1098 	int error, retries;
1099 	struct timespec ts;
1100 
1101 	memset(&ts, 0, sizeof ts);	/* XXX gcc */
1102 
1103 	ump = VFSTOUFS(mp);
1104 	dev = ump->um_dev;
1105 	fs = ump->um_lfs;
1106 
1107 	/*
1108 	 * If the filesystem is not completely mounted yet, suspend
1109 	 * any access requests (wait for roll-forward to complete).
1110 	 */
1111 	simple_lock(&fs->lfs_interlock);
1112 	while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1113 		ltsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0,
1114 			&fs->lfs_interlock);
1115 	simple_unlock(&fs->lfs_interlock);
1116 
1117 retry:
1118 	if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
1119 		return (0);
1120 
1121 	if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) {
1122 		*vpp = NULL;
1123 		 return (error);
1124 	}
1125 
1126 	mutex_enter(&ufs_hashlock);
1127 	if (ufs_ihashget(dev, ino, 0) != NULL) {
1128 		mutex_exit(&ufs_hashlock);
1129 		ungetnewvnode(vp);
1130 		goto retry;
1131 	}
1132 
1133 	/* Translate the inode number to a disk address. */
1134 	if (ino == LFS_IFILE_INUM)
1135 		daddr = fs->lfs_idaddr;
1136 	else {
1137 		/* XXX bounds-check this too */
1138 		LFS_IENTRY(ifp, fs, ino, bp);
1139 		daddr = ifp->if_daddr;
1140 		if (fs->lfs_version > 1) {
1141 			ts.tv_sec = ifp->if_atime_sec;
1142 			ts.tv_nsec = ifp->if_atime_nsec;
1143 		}
1144 
1145 		brelse(bp, 0);
1146 		if (daddr == LFS_UNUSED_DADDR) {
1147 			*vpp = NULLVP;
1148 			mutex_exit(&ufs_hashlock);
1149 			ungetnewvnode(vp);
1150 			return (ENOENT);
1151 		}
1152 	}
1153 
1154 	/* Allocate/init new vnode/inode. */
1155 	lfs_vcreate(mp, ino, vp);
1156 
1157 	/*
1158 	 * Put it onto its hash chain and lock it so that other requests for
1159 	 * this inode will block if they arrive while we are sleeping waiting
1160 	 * for old data structures to be purged or for the contents of the
1161 	 * disk portion of this inode to be read.
1162 	 */
1163 	ip = VTOI(vp);
1164 	ufs_ihashins(ip);
1165 	mutex_exit(&ufs_hashlock);
1166 
1167 	/*
1168 	 * XXX
1169 	 * This may not need to be here, logically it should go down with
1170 	 * the i_devvp initialization.
1171 	 * Ask Kirk.
1172 	 */
1173 	ip->i_lfs = ump->um_lfs;
1174 
1175 	/* Read in the disk contents for the inode, copy into the inode. */
1176 	retries = 0;
1177     again:
1178 	error = bread(ump->um_devvp, fsbtodb(fs, daddr),
1179 		(fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize),
1180 		NOCRED, &bp);
1181 	if (error) {
1182 		/*
1183 		 * The inode does not contain anything useful, so it would
1184 		 * be misleading to leave it on its hash chain. With mode
1185 		 * still zero, it will be unlinked and returned to the free
1186 		 * list by vput().
1187 		 */
1188 		vput(vp);
1189 		brelse(bp, 0);
1190 		*vpp = NULL;
1191 		return (error);
1192 	}
1193 
1194 	dip = lfs_ifind(fs, ino, bp);
1195 	if (dip == NULL) {
1196 		/* Assume write has not completed yet; try again */
1197 		brelse(bp, BC_INVAL);
1198 		++retries;
1199 		if (retries > LFS_IFIND_RETRIES) {
1200 #ifdef DEBUG
1201 			/* If the seglock is held look at the bpp to see
1202 			   what is there anyway */
1203 			simple_lock(&fs->lfs_interlock);
1204 			if (fs->lfs_seglock > 0) {
1205 				struct buf **bpp;
1206 				struct ufs1_dinode *dp;
1207 				int i;
1208 
1209 				for (bpp = fs->lfs_sp->bpp;
1210 				     bpp != fs->lfs_sp->cbpp; ++bpp) {
1211 					if ((*bpp)->b_vp == fs->lfs_ivnode &&
1212 					    bpp != fs->lfs_sp->bpp) {
1213 						/* Inode block */
1214 						printf("lfs_vget: block 0x%" PRIx64 ": ",
1215 						       (*bpp)->b_blkno);
1216 						dp = (struct ufs1_dinode *)(*bpp)->b_data;
1217 						for (i = 0; i < INOPB(fs); i++)
1218 							if (dp[i].di_u.inumber)
1219 								printf("%d ", dp[i].di_u.inumber);
1220 						printf("\n");
1221 					}
1222 				}
1223 			}
1224 			simple_unlock(&fs->lfs_interlock);
1225 #endif /* DEBUG */
1226 			panic("lfs_vget: dinode not found");
1227 		}
1228 		simple_lock(&fs->lfs_interlock);
1229 		if (fs->lfs_iocount) {
1230 			DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino));
1231 			(void)ltsleep(&fs->lfs_iocount, PRIBIO + 1,
1232 				      "lfs ifind", 1, &fs->lfs_interlock);
1233 		} else
1234 			retries = LFS_IFIND_RETRIES;
1235 		simple_unlock(&fs->lfs_interlock);
1236 		goto again;
1237 	}
1238 	*ip->i_din.ffs1_din = *dip;
1239 	brelse(bp, 0);
1240 
1241 	if (fs->lfs_version > 1) {
1242 		ip->i_ffs1_atime = ts.tv_sec;
1243 		ip->i_ffs1_atimensec = ts.tv_nsec;
1244 	}
1245 
1246 	lfs_vinit(mp, &vp);
1247 
1248 	*vpp = vp;
1249 
1250 	KASSERT(VOP_ISLOCKED(vp));
1251 
1252 	return (0);
1253 }
1254 
1255 /*
1256  * File handle to vnode
1257  */
1258 int
1259 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1260 {
1261 	struct lfid lfh;
1262 	struct buf *bp;
1263 	IFILE *ifp;
1264 	int32_t daddr;
1265 	struct lfs *fs;
1266 
1267 	if (fhp->fid_len != sizeof(struct lfid))
1268 		return EINVAL;
1269 
1270 	memcpy(&lfh, fhp, sizeof(lfh));
1271 	if (lfh.lfid_ino < LFS_IFILE_INUM)
1272 		return ESTALE;
1273 
1274 	fs = VFSTOUFS(mp)->um_lfs;
1275 	if (lfh.lfid_ident != fs->lfs_ident)
1276 		return ESTALE;
1277 
1278 	if (lfh.lfid_ino >
1279 	    ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) -
1280 	     fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb)
1281 		return ESTALE;
1282 
1283 	if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfh.lfid_ino) == NULLVP) {
1284 		LFS_IENTRY(ifp, fs, lfh.lfid_ino, bp);
1285 		daddr = ifp->if_daddr;
1286 		brelse(bp, 0);
1287 		if (daddr == LFS_UNUSED_DADDR)
1288 			return ESTALE;
1289 	}
1290 
1291 	return (ufs_fhtovp(mp, &lfh.lfid_ufid, vpp));
1292 }
1293 
1294 /*
1295  * Vnode pointer to File handle
1296  */
1297 /* ARGSUSED */
1298 int
1299 lfs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size)
1300 {
1301 	struct inode *ip;
1302 	struct lfid lfh;
1303 
1304 	if (*fh_size < sizeof(struct lfid)) {
1305 		*fh_size = sizeof(struct lfid);
1306 		return E2BIG;
1307 	}
1308 	*fh_size = sizeof(struct lfid);
1309 	ip = VTOI(vp);
1310 	memset(&lfh, 0, sizeof(lfh));
1311 	lfh.lfid_len = sizeof(struct lfid);
1312 	lfh.lfid_ino = ip->i_number;
1313 	lfh.lfid_gen = ip->i_gen;
1314 	lfh.lfid_ident = ip->i_lfs->lfs_ident;
1315 	memcpy(fhp, &lfh, sizeof(lfh));
1316 	return (0);
1317 }
1318 
1319 static int
1320 sysctl_lfs_dostats(SYSCTLFN_ARGS)
1321 {
1322 	extern struct lfs_stats lfs_stats;
1323 	extern int lfs_dostats;
1324 	int error;
1325 
1326 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
1327 	if (error || newp == NULL)
1328 		return (error);
1329 
1330 	if (lfs_dostats == 0)
1331 		memset(&lfs_stats, 0, sizeof(lfs_stats));
1332 
1333 	return (0);
1334 }
1335 
1336 struct shortlong {
1337 	const char *sname;
1338 	const char *lname;
1339 };
1340 
1341 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs subtree setup")
1342 {
1343 	int i;
1344 	extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead,
1345 		   lfs_fs_pagetrip, lfs_ignore_lazy_sync;
1346 #ifdef DEBUG
1347 	extern int lfs_debug_log_subsys[DLOG_MAX];
1348 	struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */
1349 		{ "rollforward", "Debug roll-forward code" },
1350 		{ "alloc",	"Debug inode allocation and free list" },
1351 		{ "avail",	"Debug space-available-now accounting" },
1352 		{ "flush",	"Debug flush triggers" },
1353 		{ "lockedlist",	"Debug locked list accounting" },
1354 		{ "vnode_verbose", "Verbose per-vnode-written debugging" },
1355 		{ "vnode",	"Debug vnode use during segment write" },
1356 		{ "segment",	"Debug segment writing" },
1357 		{ "seguse",	"Debug segment used-bytes accounting" },
1358 		{ "cleaner",	"Debug cleaning routines" },
1359 		{ "mount",	"Debug mount/unmount routines" },
1360 		{ "pagecache",	"Debug UBC interactions" },
1361 		{ "dirop",	"Debug directory-operation accounting" },
1362 		{ "malloc",	"Debug private malloc accounting" },
1363 	};
1364 #endif /* DEBUG */
1365 	struct shortlong stat_names[] = { /* Must match lfs.h! */
1366 		{ "segsused",	    "Number of new segments allocated" },
1367 		{ "psegwrites",	    "Number of partial-segment writes" },
1368 		{ "psyncwrites",    "Number of synchronous partial-segment"
1369 				    " writes" },
1370 		{ "pcleanwrites",   "Number of partial-segment writes by the"
1371 				    " cleaner" },
1372 		{ "blocktot",       "Number of blocks written" },
1373 		{ "cleanblocks",    "Number of blocks written by the cleaner" },
1374 		{ "ncheckpoints",   "Number of checkpoints made" },
1375 		{ "nwrites",        "Number of whole writes" },
1376 		{ "nsync_writes",   "Number of synchronous writes" },
1377 		{ "wait_exceeded",  "Number of times writer waited for"
1378 				    " cleaner" },
1379 		{ "write_exceeded", "Number of times writer invoked flush" },
1380 		{ "flush_invoked",  "Number of times flush was invoked" },
1381 		{ "vflush_invoked", "Number of time vflush was called" },
1382 		{ "clean_inlocked", "Number of vnodes skipped for VI_XLOCK" },
1383 		{ "clean_vnlocked", "Number of vnodes skipped for vget failure" },
1384 		{ "segs_reclaimed", "Number of segments reclaimed" },
1385 	};
1386 
1387 	sysctl_createv(clog, 0, NULL, NULL,
1388 		       CTLFLAG_PERMANENT,
1389 		       CTLTYPE_NODE, "vfs", NULL,
1390 		       NULL, 0, NULL, 0,
1391 		       CTL_VFS, CTL_EOL);
1392 	sysctl_createv(clog, 0, NULL, NULL,
1393 		       CTLFLAG_PERMANENT,
1394 		       CTLTYPE_NODE, "lfs",
1395 		       SYSCTL_DESCR("Log-structured file system"),
1396 		       NULL, 0, NULL, 0,
1397 		       CTL_VFS, 5, CTL_EOL);
1398 	/*
1399 	 * XXX the "5" above could be dynamic, thereby eliminating one
1400 	 * more instance of the "number to vfs" mapping problem, but
1401 	 * "5" is the order as taken from sys/mount.h
1402 	 */
1403 
1404 	sysctl_createv(clog, 0, NULL, NULL,
1405 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1406 		       CTLTYPE_INT, "flushindir", NULL,
1407 		       NULL, 0, &lfs_writeindir, 0,
1408 		       CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
1409 	sysctl_createv(clog, 0, NULL, NULL,
1410 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1411 		       CTLTYPE_INT, "clean_vnhead", NULL,
1412 		       NULL, 0, &lfs_clean_vnhead, 0,
1413 		       CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
1414 	sysctl_createv(clog, 0, NULL, NULL,
1415 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1416 		       CTLTYPE_INT, "dostats",
1417 		       SYSCTL_DESCR("Maintain statistics on LFS operations"),
1418 		       sysctl_lfs_dostats, 0, &lfs_dostats, 0,
1419 		       CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
1420 	sysctl_createv(clog, 0, NULL, NULL,
1421 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1422 		       CTLTYPE_INT, "pagetrip",
1423 		       SYSCTL_DESCR("How many dirty pages in fs triggers"
1424 				    " a flush"),
1425 		       NULL, 0, &lfs_fs_pagetrip, 0,
1426 		       CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL);
1427 	sysctl_createv(clog, 0, NULL, NULL,
1428 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1429 		       CTLTYPE_INT, "ignore_lazy_sync",
1430 		       SYSCTL_DESCR("Lazy Sync is ignored entirely"),
1431 		       NULL, 0, &lfs_ignore_lazy_sync, 0,
1432 		       CTL_VFS, 5, LFS_IGNORE_LAZY_SYNC, CTL_EOL);
1433 #ifdef LFS_KERNEL_RFW
1434 	sysctl_createv(clog, 0, NULL, NULL,
1435 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1436 		       CTLTYPE_INT, "rfw",
1437 		       SYSCTL_DESCR("Use in-kernel roll-forward on mount"),
1438 		       NULL, 0, &lfs_do_rfw, 0,
1439 		       CTL_VFS, 5, LFS_DO_RFW, CTL_EOL);
1440 #endif
1441 
1442 	sysctl_createv(clog, 0, NULL, NULL,
1443 		       CTLFLAG_PERMANENT,
1444 		       CTLTYPE_NODE, "stats",
1445 		       SYSCTL_DESCR("Debugging options"),
1446 		       NULL, 0, NULL, 0,
1447 		       CTL_VFS, 5, LFS_STATS, CTL_EOL);
1448 	for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) {
1449 		sysctl_createv(clog, 0, NULL, NULL,
1450 			       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
1451 			       CTLTYPE_INT, stat_names[i].sname,
1452 			       SYSCTL_DESCR(stat_names[i].lname),
1453 			       NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]),
1454 			       0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL);
1455 	}
1456 
1457 #ifdef DEBUG
1458 	sysctl_createv(clog, 0, NULL, NULL,
1459 		       CTLFLAG_PERMANENT,
1460 		       CTLTYPE_NODE, "debug",
1461 		       SYSCTL_DESCR("Debugging options"),
1462 		       NULL, 0, NULL, 0,
1463 		       CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL);
1464 	for (i = 0; i < DLOG_MAX; i++) {
1465 		sysctl_createv(clog, 0, NULL, NULL,
1466 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1467 			       CTLTYPE_INT, dlog_names[i].sname,
1468 			       SYSCTL_DESCR(dlog_names[i].lname),
1469 			       NULL, 0, &(lfs_debug_log_subsys[i]), 0,
1470 			       CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL);
1471 	}
1472 #endif
1473 }
1474 
1475 /*
1476  * ufs_bmaparray callback function for writing.
1477  *
1478  * Since blocks will be written to the new segment anyway,
1479  * we don't care about current daddr of them.
1480  */
1481 static bool
1482 lfs_issequential_hole(const struct ufsmount *ump,
1483     daddr_t daddr0, daddr_t daddr1)
1484 {
1485 	daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */
1486 	daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */
1487 
1488 	KASSERT(daddr0 == UNWRITTEN ||
1489 	    (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR));
1490 	KASSERT(daddr1 == UNWRITTEN ||
1491 	    (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR));
1492 
1493 	/* NOTE: all we want to know here is 'hole or not'. */
1494 	/* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */
1495 
1496 	/*
1497 	 * treat UNWRITTENs and all resident blocks as 'contiguous'
1498 	 */
1499 	if (daddr0 != 0 && daddr1 != 0)
1500 		return true;
1501 
1502 	/*
1503 	 * both are in hole?
1504 	 */
1505 	if (daddr0 == 0 && daddr1 == 0)
1506 		return true; /* all holes are 'contiguous' for us. */
1507 
1508 	return false;
1509 }
1510 
1511 /*
1512  * lfs_gop_write functions exactly like genfs_gop_write, except that
1513  * (1) it requires the seglock to be held by its caller, and sp->fip
1514  *     to be properly initialized (it will return without re-initializing
1515  *     sp->fip, and without calling lfs_writeseg).
1516  * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1517  *     to determine how large a block it can write at once (though it does
1518  *     still use VOP_BMAP to find holes in the file);
1519  * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1520  *     (leaving lfs_writeseg to deal with the cluster blocks, so we might
1521  *     now have clusters of clusters, ick.)
1522  */
1523 static int
1524 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages,
1525     int flags)
1526 {
1527 	int i, s, error, run, haveeof = 0;
1528 	int fs_bshift;
1529 	vaddr_t kva;
1530 	off_t eof, offset, startoffset = 0;
1531 	size_t bytes, iobytes, skipbytes;
1532 	daddr_t lbn, blkno;
1533 	struct vm_page *pg;
1534 	struct buf *mbp, *bp;
1535 	struct vnode *devvp = VTOI(vp)->i_devvp;
1536 	struct inode *ip = VTOI(vp);
1537 	struct lfs *fs = ip->i_lfs;
1538 	struct segment *sp = fs->lfs_sp;
1539 	UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1540 
1541 	ASSERT_SEGLOCK(fs);
1542 
1543 	/* The Ifile lives in the buffer cache */
1544 	KASSERT(vp != fs->lfs_ivnode);
1545 
1546 	/*
1547 	 * We don't want to fill the disk before the cleaner has a chance
1548 	 * to make room for us.  If we're in danger of doing that, fail
1549 	 * with EAGAIN.  The caller will have to notice this, unlock
1550 	 * so the cleaner can run, relock and try again.
1551 	 *
1552 	 * We must write everything, however, if our vnode is being
1553 	 * reclaimed.
1554 	 */
1555 	if (LFS_STARVED_FOR_SEGS(fs) && vp != fs->lfs_flushvp)
1556 		goto tryagain;
1557 
1558 	/*
1559 	 * Sometimes things slip past the filters in lfs_putpages,
1560 	 * and the pagedaemon tries to write pages---problem is
1561 	 * that the pagedaemon never acquires the segment lock.
1562 	 *
1563 	 * Alternatively, pages that were clean when we called
1564 	 * genfs_putpages may have become dirty in the meantime.  In this
1565 	 * case the segment header is not properly set up for blocks
1566 	 * to be added to it.
1567 	 *
1568 	 * Unbusy and unclean the pages, and put them on the ACTIVE
1569 	 * queue under the hypothesis that they couldn't have got here
1570 	 * unless they were modified *quite* recently.
1571 	 *
1572 	 * XXXUBC that last statement is an oversimplification of course.
1573 	 */
1574 	if (!LFS_SEGLOCK_HELD(fs) ||
1575 	    (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) ||
1576 	    (pgs[0]->offset & fs->lfs_bmask) != 0) {
1577 		goto tryagain;
1578 	}
1579 
1580 	UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1581 	    vp, pgs, npages, flags);
1582 
1583 	GOP_SIZE(vp, vp->v_size, &eof, 0);
1584 	haveeof = 1;
1585 
1586 	if (vp->v_type == VREG)
1587 		fs_bshift = vp->v_mount->mnt_fs_bshift;
1588 	else
1589 		fs_bshift = DEV_BSHIFT;
1590 	error = 0;
1591 	pg = pgs[0];
1592 	startoffset = pg->offset;
1593 	KASSERT(eof >= 0);
1594 
1595 	if (startoffset >= eof) {
1596 		goto tryagain;
1597 	} else
1598 		bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1599 	skipbytes = 0;
1600 
1601 	KASSERT(bytes != 0);
1602 
1603 	/* Swap PG_DELWRI for PG_PAGEOUT */
1604 	for (i = 0; i < npages; i++) {
1605 		if (pgs[i]->flags & PG_DELWRI) {
1606 			KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
1607 			pgs[i]->flags &= ~PG_DELWRI;
1608 			pgs[i]->flags |= PG_PAGEOUT;
1609 			uvmexp.paging++;
1610 			uvm_lock_pageq();
1611 			uvm_pageunwire(pgs[i]);
1612 			uvm_unlock_pageq();
1613 		}
1614 	}
1615 
1616 	/*
1617 	 * Check to make sure we're starting on a block boundary.
1618 	 * We'll check later to make sure we always write entire
1619 	 * blocks (or fragments).
1620 	 */
1621 	if (startoffset & fs->lfs_bmask)
1622 		printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n",
1623 		       startoffset, fs->lfs_bmask,
1624 		       startoffset & fs->lfs_bmask);
1625 	KASSERT((startoffset & fs->lfs_bmask) == 0);
1626 	if (bytes & fs->lfs_ffmask) {
1627 		printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
1628 		panic("lfs_gop_write: non-integer blocks");
1629 	}
1630 
1631 	/*
1632 	 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK.
1633 	 * If we would, write what we have and try again.  If we don't
1634 	 * have anything to write, we'll have to sleep.
1635 	 */
1636 	if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
1637 				      (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ?
1638 				       UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) {
1639 		DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n"));
1640 #if 0
1641 		      " with nfinfo=%d at offset 0x%x\n",
1642 		      (int)((SEGSUM *)(sp->segsum))->ss_nfinfo,
1643 		      (unsigned)fs->lfs_offset));
1644 #endif
1645 		lfs_updatemeta(sp);
1646 		lfs_release_finfo(fs);
1647 		(void) lfs_writeseg(fs, sp);
1648 
1649 		lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
1650 
1651 		/*
1652 		 * Having given up all of the pager_map we were holding,
1653 		 * we can now wait for aiodoned to reclaim it for us
1654 		 * without fear of deadlock.
1655 		 */
1656 		kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE |
1657 				     UVMPAGER_MAPIN_WAITOK);
1658 	}
1659 
1660 	s = splbio();
1661 	simple_lock(&global_v_numoutput_slock);
1662 	vp->v_numoutput += 2; /* one for biodone, one for aiodone */
1663 	simple_unlock(&global_v_numoutput_slock);
1664 	splx(s);
1665 
1666 	mbp = getiobuf();
1667 	UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
1668 	    vp, mbp, vp->v_numoutput, bytes);
1669 	mbp->b_bufsize = npages << PAGE_SHIFT;
1670 	mbp->b_data = (void *)kva;
1671 	mbp->b_resid = mbp->b_bcount = bytes;
1672 	mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL;
1673 	mbp->b_iodone = uvm_aio_biodone;
1674 	mbp->b_vp = vp;
1675 
1676 	bp = NULL;
1677 	for (offset = startoffset;
1678 	    bytes > 0;
1679 	    offset += iobytes, bytes -= iobytes) {
1680 		lbn = offset >> fs_bshift;
1681 		error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
1682 		    lfs_issequential_hole);
1683 		if (error) {
1684 			UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d",
1685 			    error,0,0,0);
1686 			skipbytes += bytes;
1687 			bytes = 0;
1688 			break;
1689 		}
1690 
1691 		iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
1692 		    bytes);
1693 		if (blkno == (daddr_t)-1) {
1694 			skipbytes += iobytes;
1695 			continue;
1696 		}
1697 
1698 		/*
1699 		 * Discover how much we can really pack into this buffer.
1700 		 */
1701 		/* If no room in the current segment, finish it up */
1702 		if (sp->sum_bytes_left < sizeof(int32_t) ||
1703 		    sp->seg_bytes_left < (1 << fs->lfs_bshift)) {
1704 			int vers;
1705 
1706 			lfs_updatemeta(sp);
1707 			vers = sp->fip->fi_version;
1708 			lfs_release_finfo(fs);
1709 			(void) lfs_writeseg(fs, sp);
1710 
1711 			lfs_acquire_finfo(fs, ip->i_number, vers);
1712 		}
1713 		/* Check both for space in segment and space in segsum */
1714 		iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
1715 					<< fs_bshift);
1716 		iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
1717 				       << fs_bshift);
1718 		KASSERT(iobytes > 0);
1719 
1720 		/* if it's really one i/o, don't make a second buf */
1721 		if (offset == startoffset && iobytes == bytes) {
1722 			bp = mbp;
1723 			/* correct overcount if there is no second buffer */
1724 			s = splbio();
1725 			simple_lock(&global_v_numoutput_slock);
1726 			--vp->v_numoutput;
1727 			simple_unlock(&global_v_numoutput_slock);
1728 			splx(s);
1729 		} else {
1730 			bp = getiobuf();
1731 			UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
1732 			    vp, bp, vp->v_numoutput, 0);
1733 			bp->b_data = (char *)kva +
1734 			    (vaddr_t)(offset - pg->offset);
1735 			bp->b_resid = bp->b_bcount = iobytes;
1736 			bp->b_flags = B_BUSY|B_WRITE|B_CALL;
1737 			bp->b_iodone = uvm_aio_biodone1;
1738 		}
1739 
1740 		/* XXX This is silly ... is this necessary? */
1741 		bp->b_vp = NULL;
1742 		s = splbio();
1743 		bgetvp(vp, bp);
1744 		splx(s);
1745 
1746 		bp->b_lblkno = lblkno(fs, offset);
1747 		bp->b_private = mbp;
1748 		if (devvp->v_type == VBLK) {
1749 			bp->b_dev = devvp->v_rdev;
1750 		}
1751 		VOP_BWRITE(bp);
1752 		while (lfs_gatherblock(sp, bp, NULL))
1753 			continue;
1754 	}
1755 
1756 	if (skipbytes) {
1757 		UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
1758 		s = splbio();
1759 		if (error) {
1760 			mbp->b_error = error;
1761 		}
1762 		mbp->b_resid -= skipbytes;
1763 		if (mbp->b_resid == 0) {
1764 			biodone(mbp);
1765 		}
1766 		splx(s);
1767 	}
1768 	UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
1769 	return (0);
1770 
1771     tryagain:
1772 	/*
1773 	 * We can't write the pages, for whatever reason.
1774 	 * Clean up after ourselves, and make the caller try again.
1775 	 */
1776 	simple_lock(&vp->v_interlock);
1777 
1778 	/* Tell why we're here, if we know */
1779 	if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) {
1780 		DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n"));
1781 	} else if ((pgs[0]->offset & fs->lfs_bmask) != 0) {
1782 		DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n"));
1783 	} else if (haveeof && startoffset >= eof) {
1784 		DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64
1785 		      " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number,
1786 		      pgs[0]->offset, eof, npages));
1787 	} else if (LFS_STARVED_FOR_SEGS(fs)) {
1788 		DLOG((DLOG_PAGE, "lfs_gop_write: avail too low\n"));
1789 	} else {
1790 		DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n"));
1791 	}
1792 
1793 	uvm_lock_pageq();
1794 	for (i = 0; i < npages; i++) {
1795 		pg = pgs[i];
1796 
1797 		if (pg->flags & PG_PAGEOUT)
1798 			uvmexp.paging--;
1799 		if (pg->flags & PG_DELWRI) {
1800 			uvm_pageunwire(pg);
1801 		}
1802 		uvm_pageactivate(pg);
1803 		pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
1804 		DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg,
1805 			vp, pg->offset));
1806 		DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags));
1807 		DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags));
1808 		DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon));
1809 		DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject));
1810 		DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i,
1811 		      pg->wire_count));
1812 		DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i,
1813 		      pg->loan_count));
1814 	}
1815 	/* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
1816 	uvm_page_unbusy(pgs, npages);
1817 	uvm_unlock_pageq();
1818 	simple_unlock(&vp->v_interlock);
1819 	return EAGAIN;
1820 }
1821 
1822 /*
1823  * finish vnode/inode initialization.
1824  * used by lfs_vget and lfs_fastvget.
1825  */
1826 void
1827 lfs_vinit(struct mount *mp, struct vnode **vpp)
1828 {
1829 	struct vnode *vp = *vpp;
1830 	struct inode *ip = VTOI(vp);
1831 	struct ufsmount *ump = VFSTOUFS(mp);
1832 	struct lfs *fs = ump->um_lfs;
1833 	int i;
1834 
1835 	ip->i_mode = ip->i_ffs1_mode;
1836 	ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink;
1837 	ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size;
1838 	ip->i_flags = ip->i_ffs1_flags;
1839 	ip->i_gen = ip->i_ffs1_gen;
1840 	ip->i_uid = ip->i_ffs1_uid;
1841 	ip->i_gid = ip->i_ffs1_gid;
1842 
1843 	ip->i_lfs_effnblks = ip->i_ffs1_blocks;
1844 	ip->i_lfs_odnlink = ip->i_ffs1_nlink;
1845 
1846 	/*
1847 	 * Initialize the vnode from the inode, check for aliases.  In all
1848 	 * cases re-init ip, the underlying vnode/inode may have changed.
1849 	 */
1850 	ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
1851 	ip = VTOI(vp);
1852 
1853 	memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize));
1854 	if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) {
1855 #ifdef DEBUG
1856 		for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift;
1857 		    i < NDADDR; i++) {
1858 			if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
1859 			    i == 0)
1860 				continue;
1861 			if (ip->i_ffs1_db[i] != 0) {
1862 inconsistent:
1863 				lfs_dump_dinode(ip->i_din.ffs1_din);
1864 				panic("inconsistent inode");
1865 			}
1866 		}
1867 		for ( ; i < NDADDR + NIADDR; i++) {
1868 			if (ip->i_ffs1_ib[i - NDADDR] != 0) {
1869 				goto inconsistent;
1870 			}
1871 		}
1872 #endif /* DEBUG */
1873 		for (i = 0; i < NDADDR; i++)
1874 			if (ip->i_ffs1_db[i] != 0)
1875 				ip->i_lfs_fragsize[i] = blksize(fs, ip, i);
1876 	}
1877 
1878 #ifdef DIAGNOSTIC
1879 	if (vp->v_type == VNON) {
1880 # ifdef DEBUG
1881 		lfs_dump_dinode(ip->i_din.ffs1_din);
1882 # endif
1883 		panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n",
1884 		      (unsigned long long)ip->i_number,
1885 		      (ip->i_mode & IFMT) >> 12);
1886 	}
1887 #endif /* DIAGNOSTIC */
1888 
1889 	/*
1890 	 * Finish inode initialization now that aliasing has been resolved.
1891 	 */
1892 
1893 	ip->i_devvp = ump->um_devvp;
1894 	VREF(ip->i_devvp);
1895 	genfs_node_init(vp, &lfs_genfsops);
1896 	uvm_vnp_setsize(vp, ip->i_size);
1897 
1898 	/* Initialize hiblk from file size */
1899 	ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1;
1900 
1901 	*vpp = vp;
1902 }
1903 
1904 /*
1905  * Resize the filesystem to contain the specified number of segments.
1906  */
1907 int
1908 lfs_resize_fs(struct lfs *fs, int newnsegs)
1909 {
1910 	SEGUSE *sup;
1911 	struct buf *bp, *obp;
1912 	daddr_t olast, nlast, ilast, noff, start, end;
1913 	struct vnode *ivp;
1914 	struct inode *ip;
1915 	int error, badnews, inc, oldnsegs;
1916 	int sbbytes, csbbytes, gain, cgain;
1917 	int i;
1918 
1919 	/* Only support v2 and up */
1920 	if (fs->lfs_version < 2)
1921 		return EOPNOTSUPP;
1922 
1923 	/* If we're doing nothing, do it fast */
1924 	oldnsegs = fs->lfs_nseg;
1925 	if (newnsegs == oldnsegs)
1926 		return 0;
1927 
1928 	/* We always have to have two superblocks */
1929 	if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1]))
1930 		return EFBIG;
1931 
1932 	ivp = fs->lfs_ivnode;
1933 	ip = VTOI(ivp);
1934 	error = 0;
1935 
1936 	/* Take the segment lock so no one else calls lfs_newseg() */
1937 	lfs_seglock(fs, SEGM_PROT);
1938 
1939 	/*
1940 	 * Make sure the segments we're going to be losing, if any,
1941 	 * are in fact empty.  We hold the seglock, so their status
1942 	 * cannot change underneath us.  Count the superblocks we lose,
1943 	 * while we're at it.
1944 	 */
1945 	sbbytes = csbbytes = 0;
1946 	cgain = 0;
1947 	for (i = newnsegs; i < oldnsegs; i++) {
1948 		LFS_SEGENTRY(sup, fs, i, bp);
1949 		badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL);
1950 		if (sup->su_flags & SEGUSE_SUPERBLOCK)
1951 			sbbytes += LFS_SBPAD;
1952 		if (!(sup->su_flags & SEGUSE_DIRTY)) {
1953 			++cgain;
1954 			if (sup->su_flags & SEGUSE_SUPERBLOCK)
1955 				csbbytes += LFS_SBPAD;
1956 		}
1957 		brelse(bp, 0);
1958 		if (badnews) {
1959 			error = EBUSY;
1960 			goto out;
1961 		}
1962 	}
1963 
1964 	/* Note old and new segment table endpoints, and old ifile size */
1965 	olast = fs->lfs_cleansz + fs->lfs_segtabsz;
1966 	nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz;
1967 	ilast = ivp->v_size >> fs->lfs_bshift;
1968 	noff = nlast - olast;
1969 
1970 	/*
1971 	 * Make sure no one can use the Ifile while we change it around.
1972 	 * Even after taking the iflock we need to make sure no one still
1973 	 * is holding Ifile buffers, so we get each one, to drain them.
1974 	 * (XXX this could be done better.)
1975 	 */
1976 	simple_lock(&fs->lfs_interlock);
1977 	lockmgr(&fs->lfs_iflock, LK_EXCLUSIVE, &fs->lfs_interlock);
1978 	simple_unlock(&fs->lfs_interlock);
1979 	vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY);
1980 	for (i = 0; i < ilast; i++) {
1981 		bread(ivp, i, fs->lfs_bsize, NOCRED, &bp);
1982 		brelse(bp, 0);
1983 	}
1984 
1985 	/* Allocate new Ifile blocks */
1986 	for (i = ilast; i < ilast + noff; i++) {
1987 		if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0,
1988 			       &bp) != 0)
1989 			panic("balloc extending ifile");
1990 		memset(bp->b_data, 0, fs->lfs_bsize);
1991 		VOP_BWRITE(bp);
1992 	}
1993 
1994 	/* Register new ifile size */
1995 	ip->i_size += noff * fs->lfs_bsize;
1996 	ip->i_ffs1_size = ip->i_size;
1997 	uvm_vnp_setsize(ivp, ip->i_size);
1998 
1999 	/* Copy the inode table to its new position */
2000 	if (noff != 0) {
2001 		if (noff < 0) {
2002 			start = nlast;
2003 			end = ilast + noff;
2004 			inc = 1;
2005 		} else {
2006 			start = ilast + noff - 1;
2007 			end = nlast - 1;
2008 			inc = -1;
2009 		}
2010 		for (i = start; i != end; i += inc) {
2011 			if (bread(ivp, i, fs->lfs_bsize, NOCRED, &bp) != 0)
2012 				panic("resize: bread dst blk failed");
2013 			if (bread(ivp, i - noff, fs->lfs_bsize, NOCRED, &obp))
2014 				panic("resize: bread src blk failed");
2015 			memcpy(bp->b_data, obp->b_data, fs->lfs_bsize);
2016 			VOP_BWRITE(bp);
2017 			brelse(obp, 0);
2018 		}
2019 	}
2020 
2021 	/* If we are expanding, write the new empty SEGUSE entries */
2022 	if (newnsegs > oldnsegs) {
2023 		for (i = oldnsegs; i < newnsegs; i++) {
2024 			if ((error = bread(ivp, i / fs->lfs_sepb +
2025 					   fs->lfs_cleansz,
2026 					   fs->lfs_bsize, NOCRED, &bp)) != 0)
2027 				panic("lfs: ifile read: %d", error);
2028 			while ((i + 1) % fs->lfs_sepb && i < newnsegs) {
2029 				sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb];
2030 				memset(sup, 0, sizeof(*sup));
2031 				i++;
2032 			}
2033 			VOP_BWRITE(bp);
2034 		}
2035 	}
2036 
2037 	/* Zero out unused superblock offsets */
2038 	for (i = 2; i < LFS_MAXNUMSB; i++)
2039 		if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs)
2040 			fs->lfs_sboffs[i] = 0x0;
2041 
2042 	/*
2043 	 * Correct superblock entries that depend on fs size.
2044 	 * The computations of these are as follows:
2045 	 *
2046 	 * size  = segtod(fs, nseg)
2047 	 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD)
2048 	 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used
2049 	 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD)
2050 	 *         + (segtod(fs, 1) - (offset - curseg))
2051 	 *	   - segtod(fs, minfreeseg - (minfreeseg / 2))
2052 	 *
2053 	 * XXX - we should probably adjust minfreeseg as well.
2054 	 */
2055 	gain = (newnsegs - oldnsegs);
2056 	fs->lfs_nseg = newnsegs;
2057 	fs->lfs_segtabsz = nlast - fs->lfs_cleansz;
2058 	fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize);
2059 	fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes);
2060 	fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes)
2061 		       - gain * btofsb(fs, fs->lfs_bsize / 2);
2062 	if (gain > 0) {
2063 		fs->lfs_nclean += gain;
2064 		fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize);
2065 	} else {
2066 		fs->lfs_nclean -= cgain;
2067 		fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) -
2068 				 btofsb(fs, csbbytes);
2069 	}
2070 
2071 	/* Resize segment flag cache */
2072 	fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0],
2073 						  fs->lfs_nseg * sizeof(u_int32_t),
2074 						  M_SEGMENT, M_WAITOK);
2075 	fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[1],
2076 						  fs->lfs_nseg * sizeof(u_int32_t),
2077 						  M_SEGMENT, M_WAITOK);
2078 	for (i = oldnsegs; i < newnsegs; i++)
2079 		fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0;
2080 
2081 	/* Truncate Ifile if necessary */
2082 	if (noff < 0)
2083 		lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0,
2084 			     NOCRED, curlwp);
2085 
2086 	/* Update cleaner info so the cleaner can die */
2087 	bread(ivp, 0, fs->lfs_bsize, NOCRED, &bp);
2088 	((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean;
2089 	((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean;
2090 	VOP_BWRITE(bp);
2091 
2092 	/* Let Ifile accesses proceed */
2093 	VOP_UNLOCK(ivp, 0);
2094 	simple_lock(&fs->lfs_interlock);
2095 	lockmgr(&fs->lfs_iflock, LK_RELEASE, &fs->lfs_interlock);
2096 	simple_unlock(&fs->lfs_interlock);
2097 
2098     out:
2099 	lfs_segunlock(fs);
2100 	return error;
2101 }
2102