xref: /netbsd-src/sys/ufs/lfs/lfs_syscalls.c (revision b757af438b42b93f8c6571f026d8b8ef3eaf5fc9)
1 /*	$NetBSD: lfs_syscalls.c,v 1.142 2012/03/13 18:41:13 elad Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003, 2007, 2007, 2008
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) 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_syscalls.c	8.10 (Berkeley) 5/14/95
61  */
62 
63 #include <sys/cdefs.h>
64 __KERNEL_RCSID(0, "$NetBSD: lfs_syscalls.c,v 1.142 2012/03/13 18:41:13 elad Exp $");
65 
66 #ifndef LFS
67 # define LFS		/* for prototypes in syscallargs.h */
68 #endif
69 
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/proc.h>
73 #include <sys/buf.h>
74 #include <sys/mount.h>
75 #include <sys/vnode.h>
76 #include <sys/kernel.h>
77 #include <sys/kauth.h>
78 #include <sys/syscallargs.h>
79 
80 #include <ufs/ufs/inode.h>
81 #include <ufs/ufs/ufsmount.h>
82 #include <ufs/ufs/ufs_extern.h>
83 
84 #include <ufs/lfs/lfs.h>
85 #include <ufs/lfs/lfs_extern.h>
86 
87 struct buf *lfs_fakebuf(struct lfs *, struct vnode *, int, size_t, void *);
88 int lfs_fasthashget(dev_t, ino_t, struct vnode **);
89 
90 pid_t lfs_cleaner_pid = 0;
91 
92 /*
93  * sys_lfs_markv:
94  *
95  * This will mark inodes and blocks dirty, so they are written into the log.
96  * It will block until all the blocks have been written.  The segment create
97  * time passed in the block_info and inode_info structures is used to decide
98  * if the data is valid for each block (in case some process dirtied a block
99  * or inode that is being cleaned between the determination that a block is
100  * live and the lfs_markv call).
101  *
102  *  0 on success
103  * -1/errno is return on error.
104  */
105 #ifdef USE_64BIT_SYSCALLS
106 int
107 sys_lfs_markv(struct lwp *l, const struct sys_lfs_markv_args *uap, register_t *retval)
108 {
109 	/* {
110 		syscallarg(fsid_t *) fsidp;
111 		syscallarg(struct block_info *) blkiov;
112 		syscallarg(int) blkcnt;
113 	} */
114 	BLOCK_INFO *blkiov;
115 	int blkcnt, error;
116 	fsid_t fsid;
117 	struct lfs *fs;
118 	struct mount *mntp;
119 
120 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
121 	    KAUTH_REQ_SYSTEM_LFS_MARKV, NULL, NULL, NULL);
122 	if (error)
123 		return (error);
124 
125 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
126 		return (error);
127 
128 	if ((mntp = vfs_getvfs(fsidp)) == NULL)
129 		return (ENOENT);
130 	fs = VFSTOUFS(mntp)->um_lfs;
131 
132 	blkcnt = SCARG(uap, blkcnt);
133 	if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT)
134 		return (EINVAL);
135 
136 	KERNEL_LOCK(1, NULL);
137 	blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV);
138 	if ((error = copyin(SCARG(uap, blkiov), blkiov,
139 			    blkcnt * sizeof(BLOCK_INFO))) != 0)
140 		goto out;
141 
142 	if ((error = lfs_markv(p, &fsid, blkiov, blkcnt)) == 0)
143 		copyout(blkiov, SCARG(uap, blkiov),
144 			blkcnt * sizeof(BLOCK_INFO));
145     out:
146 	lfs_free(fs, blkiov, LFS_NB_BLKIOV);
147 	KERNEL_UNLOCK_ONE(NULL);
148 	return error;
149 }
150 #else
151 int
152 sys_lfs_markv(struct lwp *l, const struct sys_lfs_markv_args *uap, register_t *retval)
153 {
154 	/* {
155 		syscallarg(fsid_t *) fsidp;
156 		syscallarg(struct block_info *) blkiov;
157 		syscallarg(int) blkcnt;
158 	} */
159 	BLOCK_INFO *blkiov;
160 	BLOCK_INFO_15 *blkiov15;
161 	int i, blkcnt, error;
162 	fsid_t fsid;
163 	struct lfs *fs;
164 	struct mount *mntp;
165 
166 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
167 	    KAUTH_REQ_SYSTEM_LFS_MARKV, NULL, NULL, NULL);
168 	if (error)
169 		return (error);
170 
171 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
172 		return (error);
173 
174 	if ((mntp = vfs_getvfs(&fsid)) == NULL)
175 		return (ENOENT);
176 	fs = VFSTOUFS(mntp)->um_lfs;
177 
178 	blkcnt = SCARG(uap, blkcnt);
179 	if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT)
180 		return (EINVAL);
181 
182 	KERNEL_LOCK(1, NULL);
183 	blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV);
184 	blkiov15 = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO_15), LFS_NB_BLKIOV);
185 	if ((error = copyin(SCARG(uap, blkiov), blkiov15,
186 			    blkcnt * sizeof(BLOCK_INFO_15))) != 0)
187 		goto out;
188 
189 	for (i = 0; i < blkcnt; i++) {
190 		blkiov[i].bi_inode     = blkiov15[i].bi_inode;
191 		blkiov[i].bi_lbn       = blkiov15[i].bi_lbn;
192 		blkiov[i].bi_daddr     = blkiov15[i].bi_daddr;
193 		blkiov[i].bi_segcreate = blkiov15[i].bi_segcreate;
194 		blkiov[i].bi_version   = blkiov15[i].bi_version;
195 		blkiov[i].bi_bp	       = blkiov15[i].bi_bp;
196 		blkiov[i].bi_size      = blkiov15[i].bi_size;
197 	}
198 
199 	if ((error = lfs_markv(l->l_proc, &fsid, blkiov, blkcnt)) == 0) {
200 		for (i = 0; i < blkcnt; i++) {
201 			blkiov15[i].bi_inode	 = blkiov[i].bi_inode;
202 			blkiov15[i].bi_lbn	 = blkiov[i].bi_lbn;
203 			blkiov15[i].bi_daddr	 = blkiov[i].bi_daddr;
204 			blkiov15[i].bi_segcreate = blkiov[i].bi_segcreate;
205 			blkiov15[i].bi_version	 = blkiov[i].bi_version;
206 			blkiov15[i].bi_bp	 = blkiov[i].bi_bp;
207 			blkiov15[i].bi_size	 = blkiov[i].bi_size;
208 		}
209 		copyout(blkiov15, SCARG(uap, blkiov),
210 			blkcnt * sizeof(BLOCK_INFO_15));
211 	}
212     out:
213 	lfs_free(fs, blkiov, LFS_NB_BLKIOV);
214 	lfs_free(fs, blkiov15, LFS_NB_BLKIOV);
215 	KERNEL_UNLOCK_ONE(NULL);
216 	return error;
217 }
218 #endif
219 
220 #define	LFS_MARKV_MAX_BLOCKS	(LFS_MAX_BUFS)
221 
222 int
223 lfs_markv(struct proc *p, fsid_t *fsidp, BLOCK_INFO *blkiov,
224     int blkcnt)
225 {
226 	BLOCK_INFO *blkp;
227 	IFILE *ifp;
228 	struct buf *bp;
229 	struct inode *ip = NULL;
230 	struct lfs *fs;
231 	struct mount *mntp;
232 	struct vnode *vp = NULL;
233 	ino_t lastino;
234 	daddr_t b_daddr, v_daddr;
235 	int cnt, error;
236 	int do_again = 0;
237 	int numrefed = 0;
238 	ino_t maxino;
239 	size_t obsize;
240 
241 	/* number of blocks/inodes that we have already bwrite'ed */
242 	int nblkwritten, ninowritten;
243 
244 	if ((mntp = vfs_getvfs(fsidp)) == NULL)
245 		return (ENOENT);
246 
247 	fs = VFSTOUFS(mntp)->um_lfs;
248 
249 	if (fs->lfs_ronly)
250 		return EROFS;
251 
252 	maxino = (fragstoblks(fs, VTOI(fs->lfs_ivnode)->i_ffs1_blocks) -
253 		      fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb;
254 
255 	cnt = blkcnt;
256 
257 	if ((error = vfs_busy(mntp, NULL)) != 0)
258 		return (error);
259 
260 	/*
261 	 * This seglock is just to prevent the fact that we might have to sleep
262 	 * from allowing the possibility that our blocks might become
263 	 * invalid.
264 	 *
265 	 * It is also important to note here that unless we specify SEGM_CKP,
266 	 * any Ifile blocks that we might be asked to clean will never get
267 	 * to the disk.
268 	 */
269 	lfs_seglock(fs, SEGM_CLEAN | SEGM_CKP | SEGM_SYNC);
270 
271 	/* Mark blocks/inodes dirty.  */
272 	error = 0;
273 
274 	/* these were inside the initialization for the for loop */
275 	v_daddr = LFS_UNUSED_DADDR;
276 	lastino = LFS_UNUSED_INUM;
277 	nblkwritten = ninowritten = 0;
278 	for (blkp = blkiov; cnt--; ++blkp)
279 	{
280 		/* Bounds-check incoming data, avoid panic for failed VGET */
281 		if (blkp->bi_inode <= 0 || blkp->bi_inode >= maxino) {
282 			error = EINVAL;
283 			goto err3;
284 		}
285 		/*
286 		 * Get the IFILE entry (only once) and see if the file still
287 		 * exists.
288 		 */
289 		if (lastino != blkp->bi_inode) {
290 			/*
291 			 * Finish the old file, if there was one.  The presence
292 			 * of a usable vnode in vp is signaled by a valid v_daddr.
293 			 */
294 			if (v_daddr != LFS_UNUSED_DADDR) {
295 				lfs_vunref(vp);
296 				numrefed--;
297 			}
298 
299 			/*
300 			 * Start a new file
301 			 */
302 			lastino = blkp->bi_inode;
303 			if (blkp->bi_inode == LFS_IFILE_INUM)
304 				v_daddr = fs->lfs_idaddr;
305 			else {
306 				LFS_IENTRY(ifp, fs, blkp->bi_inode, bp);
307 				/* XXX fix for force write */
308 				v_daddr = ifp->if_daddr;
309 				brelse(bp, 0);
310 			}
311 			if (v_daddr == LFS_UNUSED_DADDR)
312 				continue;
313 
314 			/* Get the vnode/inode. */
315 			error = lfs_fastvget(mntp, blkp->bi_inode, v_daddr,
316 					   &vp,
317 					   (blkp->bi_lbn == LFS_UNUSED_LBN
318 					    ? blkp->bi_bp
319 					    : NULL));
320 
321 			if (!error) {
322 				numrefed++;
323 			}
324 			if (error) {
325 				DLOG((DLOG_CLEAN, "lfs_markv: lfs_fastvget"
326 				      " failed with %d (ino %d, segment %d)\n",
327 				      error, blkp->bi_inode,
328 				      dtosn(fs, blkp->bi_daddr)));
329 				/*
330 				 * If we got EAGAIN, that means that the
331 				 * Inode was locked.  This is
332 				 * recoverable: just clean the rest of
333 				 * this segment, and let the cleaner try
334 				 * again with another.	(When the
335 				 * cleaner runs again, this segment will
336 				 * sort high on the list, since it is
337 				 * now almost entirely empty.) But, we
338 				 * still set v_daddr = LFS_UNUSED_ADDR
339 				 * so as not to test this over and over
340 				 * again.
341 				 */
342 				if (error == EAGAIN) {
343 					error = 0;
344 					do_again++;
345 				}
346 #ifdef DIAGNOSTIC
347 				else if (error != ENOENT)
348 					panic("lfs_markv VFS_VGET FAILED");
349 #endif
350 				/* lastino = LFS_UNUSED_INUM; */
351 				v_daddr = LFS_UNUSED_DADDR;
352 				vp = NULL;
353 				ip = NULL;
354 				continue;
355 			}
356 			ip = VTOI(vp);
357 			ninowritten++;
358 		} else if (v_daddr == LFS_UNUSED_DADDR) {
359 			/*
360 			 * This can only happen if the vnode is dead (or
361 			 * in any case we can't get it...e.g., it is
362 			 * inlocked).  Keep going.
363 			 */
364 			continue;
365 		}
366 
367 		/* Past this point we are guaranteed that vp, ip are valid. */
368 
369 		/* Can't clean VU_DIROP directories in case of truncation */
370 		/* XXX - maybe we should mark removed dirs specially? */
371 		if (vp->v_type == VDIR && (vp->v_uflag & VU_DIROP)) {
372 			do_again++;
373 			continue;
374 		}
375 
376 		/* If this BLOCK_INFO didn't contain a block, keep going. */
377 		if (blkp->bi_lbn == LFS_UNUSED_LBN) {
378 			/* XXX need to make sure that the inode gets written in this case */
379 			/* XXX but only write the inode if it's the right one */
380 			if (blkp->bi_inode != LFS_IFILE_INUM) {
381 				LFS_IENTRY(ifp, fs, blkp->bi_inode, bp);
382 				if (ifp->if_daddr == blkp->bi_daddr) {
383 					mutex_enter(&lfs_lock);
384 					LFS_SET_UINO(ip, IN_CLEANING);
385 					mutex_exit(&lfs_lock);
386 				}
387 				brelse(bp, 0);
388 			}
389 			continue;
390 		}
391 
392 		b_daddr = 0;
393 		if (VOP_BMAP(vp, blkp->bi_lbn, NULL, &b_daddr, NULL) ||
394 		    dbtofsb(fs, b_daddr) != blkp->bi_daddr)
395 		{
396 			if (dtosn(fs, dbtofsb(fs, b_daddr)) ==
397 			    dtosn(fs, blkp->bi_daddr))
398 			{
399 				DLOG((DLOG_CLEAN, "lfs_markv: wrong da same seg: %llx vs %llx\n",
400 				      (long long)blkp->bi_daddr, (long long)dbtofsb(fs, b_daddr)));
401 			}
402 			do_again++;
403 			continue;
404 		}
405 
406 		/*
407 		 * Check block sizes.  The blocks being cleaned come from
408 		 * disk, so they should have the same size as their on-disk
409 		 * counterparts.
410 		 */
411 		if (blkp->bi_lbn >= 0)
412 			obsize = blksize(fs, ip, blkp->bi_lbn);
413 		else
414 			obsize = fs->lfs_bsize;
415 		/* Check for fragment size change */
416 		if (blkp->bi_lbn >= 0 && blkp->bi_lbn < NDADDR) {
417 			obsize = ip->i_lfs_fragsize[blkp->bi_lbn];
418 		}
419 		if (obsize != blkp->bi_size) {
420 			DLOG((DLOG_CLEAN, "lfs_markv: ino %d lbn %lld wrong"
421 			      " size (%ld != %d), try again\n",
422 			      blkp->bi_inode, (long long)blkp->bi_lbn,
423 			      (long) obsize, blkp->bi_size));
424 			do_again++;
425 			continue;
426 		}
427 
428 		/*
429 		 * If we get to here, then we are keeping the block.  If
430 		 * it is an indirect block, we want to actually put it
431 		 * in the buffer cache so that it can be updated in the
432 		 * finish_meta section.	 If it's not, we need to
433 		 * allocate a fake buffer so that writeseg can perform
434 		 * the copyin and write the buffer.
435 		 */
436 		if (ip->i_number != LFS_IFILE_INUM && blkp->bi_lbn >= 0) {
437 			/* Data Block */
438 			bp = lfs_fakebuf(fs, vp, blkp->bi_lbn,
439 					 blkp->bi_size, blkp->bi_bp);
440 			/* Pretend we used bread() to get it */
441 			bp->b_blkno = fsbtodb(fs, blkp->bi_daddr);
442 		} else {
443 			/* Indirect block or ifile */
444 			if (blkp->bi_size != fs->lfs_bsize &&
445 			    ip->i_number != LFS_IFILE_INUM)
446 				panic("lfs_markv: partial indirect block?"
447 				    " size=%d\n", blkp->bi_size);
448 			bp = getblk(vp, blkp->bi_lbn, blkp->bi_size, 0, 0);
449 			if (!(bp->b_oflags & (BO_DONE|BO_DELWRI))) {
450 				/*
451 				 * The block in question was not found
452 				 * in the cache; i.e., the block that
453 				 * getblk() returned is empty.	So, we
454 				 * can (and should) copy in the
455 				 * contents, because we've already
456 				 * determined that this was the right
457 				 * version of this block on disk.
458 				 *
459 				 * And, it can't have changed underneath
460 				 * us, because we have the segment lock.
461 				 */
462 				error = copyin(blkp->bi_bp, bp->b_data, blkp->bi_size);
463 				if (error)
464 					goto err2;
465 			}
466 		}
467 		if ((error = lfs_bwrite_ext(bp, BW_CLEAN)) != 0)
468 			goto err2;
469 
470 		nblkwritten++;
471 		/*
472 		 * XXX should account indirect blocks and ifile pages as well
473 		 */
474 		if (nblkwritten + lblkno(fs, ninowritten * sizeof (struct ufs1_dinode))
475 		    > LFS_MARKV_MAX_BLOCKS) {
476 			DLOG((DLOG_CLEAN, "lfs_markv: writing %d blks %d inos\n",
477 			      nblkwritten, ninowritten));
478 			lfs_segwrite(mntp, SEGM_CLEAN);
479 			nblkwritten = ninowritten = 0;
480 		}
481 	}
482 
483 	/*
484 	 * Finish the old file, if there was one
485 	 */
486 	if (v_daddr != LFS_UNUSED_DADDR) {
487 		lfs_vunref(vp);
488 		numrefed--;
489 	}
490 
491 #ifdef DIAGNOSTIC
492 	if (numrefed != 0)
493 		panic("lfs_markv: numrefed=%d", numrefed);
494 #endif
495 	DLOG((DLOG_CLEAN, "lfs_markv: writing %d blks %d inos (check point)\n",
496 	      nblkwritten, ninowritten));
497 
498 	/*
499 	 * The last write has to be SEGM_SYNC, because of calling semantics.
500 	 * It also has to be SEGM_CKP, because otherwise we could write
501 	 * over the newly cleaned data contained in a checkpoint, and then
502 	 * we'd be unhappy at recovery time.
503 	 */
504 	lfs_segwrite(mntp, SEGM_CLEAN | SEGM_CKP | SEGM_SYNC);
505 
506 	lfs_segunlock(fs);
507 
508 	vfs_unbusy(mntp, false, NULL);
509 	if (error)
510 		return (error);
511 	else if (do_again)
512 		return EAGAIN;
513 
514 	return 0;
515 
516 err2:
517 	DLOG((DLOG_CLEAN, "lfs_markv err2\n"));
518 
519 	/*
520 	 * XXX we're here because copyin() failed.
521 	 * XXX it means that we can't trust the cleanerd.  too bad.
522 	 * XXX how can we recover from this?
523 	 */
524 
525 err3:
526 	KERNEL_UNLOCK_ONE(NULL);
527 	/*
528 	 * XXX should do segwrite here anyway?
529 	 */
530 
531 	if (v_daddr != LFS_UNUSED_DADDR) {
532 		lfs_vunref(vp);
533 		--numrefed;
534 	}
535 
536 	lfs_segunlock(fs);
537 	vfs_unbusy(mntp, false, NULL);
538 #ifdef DIAGNOSTIC
539 	if (numrefed != 0)
540 		panic("lfs_markv: numrefed=%d", numrefed);
541 #endif
542 
543 	return (error);
544 }
545 
546 /*
547  * sys_lfs_bmapv:
548  *
549  * This will fill in the current disk address for arrays of blocks.
550  *
551  *  0 on success
552  * -1/errno is return on error.
553  */
554 #ifdef USE_64BIT_SYSCALLS
555 int
556 sys_lfs_bmapv(struct lwp *l, const struct sys_lfs_bmapv_args *uap, register_t *retval)
557 {
558 	/* {
559 		syscallarg(fsid_t *) fsidp;
560 		syscallarg(struct block_info *) blkiov;
561 		syscallarg(int) blkcnt;
562 	} */
563 	BLOCK_INFO *blkiov;
564 	int blkcnt, error;
565 	fsid_t fsid;
566 	struct lfs *fs;
567 	struct mount *mntp;
568 
569 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
570 	    KAUTH_REQ_SYSTEM_LFS_BMAPV, NULL, NULL, NULL);
571 	if (error)
572 		return (error);
573 
574 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
575 		return (error);
576 
577 	if ((mntp = vfs_getvfs(&fsid)) == NULL)
578 		return (ENOENT);
579 	fs = VFSTOUFS(mntp)->um_lfs;
580 
581 	blkcnt = SCARG(uap, blkcnt);
582 	if ((u_int) blkcnt > SIZE_T_MAX / sizeof(BLOCK_INFO))
583 		return (EINVAL);
584 	KERNEL_LOCK(1, NULL);
585 	blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV);
586 	if ((error = copyin(SCARG(uap, blkiov), blkiov,
587 			    blkcnt * sizeof(BLOCK_INFO))) != 0)
588 		goto out;
589 
590 	if ((error = lfs_bmapv(p, &fsid, blkiov, blkcnt)) == 0)
591 		copyout(blkiov, SCARG(uap, blkiov),
592 			blkcnt * sizeof(BLOCK_INFO));
593     out:
594 	lfs_free(fs, blkiov, LFS_NB_BLKIOV);
595 	KERNEL_UNLOCK_ONE(NULL);
596 	return error;
597 }
598 #else
599 int
600 sys_lfs_bmapv(struct lwp *l, const struct sys_lfs_bmapv_args *uap, register_t *retval)
601 {
602 	/* {
603 		syscallarg(fsid_t *) fsidp;
604 		syscallarg(struct block_info *) blkiov;
605 		syscallarg(int) blkcnt;
606 	} */
607 	BLOCK_INFO *blkiov;
608 	BLOCK_INFO_15 *blkiov15;
609 	int i, blkcnt, error;
610 	fsid_t fsid;
611 	struct lfs *fs;
612 	struct mount *mntp;
613 
614 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
615 	    KAUTH_REQ_SYSTEM_LFS_BMAPV, NULL, NULL, NULL);
616 	if (error)
617 		return (error);
618 
619 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
620 		return (error);
621 
622 	if ((mntp = vfs_getvfs(&fsid)) == NULL)
623 		return (ENOENT);
624 	fs = VFSTOUFS(mntp)->um_lfs;
625 
626 	blkcnt = SCARG(uap, blkcnt);
627 	if ((size_t) blkcnt > SIZE_T_MAX / sizeof(BLOCK_INFO))
628 		return (EINVAL);
629 	KERNEL_LOCK(1, NULL);
630 	blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV);
631 	blkiov15 = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO_15), LFS_NB_BLKIOV);
632 	if ((error = copyin(SCARG(uap, blkiov), blkiov15,
633 			    blkcnt * sizeof(BLOCK_INFO_15))) != 0)
634 		goto out;
635 
636 	for (i = 0; i < blkcnt; i++) {
637 		blkiov[i].bi_inode     = blkiov15[i].bi_inode;
638 		blkiov[i].bi_lbn       = blkiov15[i].bi_lbn;
639 		blkiov[i].bi_daddr     = blkiov15[i].bi_daddr;
640 		blkiov[i].bi_segcreate = blkiov15[i].bi_segcreate;
641 		blkiov[i].bi_version   = blkiov15[i].bi_version;
642 		blkiov[i].bi_bp	       = blkiov15[i].bi_bp;
643 		blkiov[i].bi_size      = blkiov15[i].bi_size;
644 	}
645 
646 	if ((error = lfs_bmapv(l->l_proc, &fsid, blkiov, blkcnt)) == 0) {
647 		for (i = 0; i < blkcnt; i++) {
648 			blkiov15[i].bi_inode	 = blkiov[i].bi_inode;
649 			blkiov15[i].bi_lbn	 = blkiov[i].bi_lbn;
650 			blkiov15[i].bi_daddr	 = blkiov[i].bi_daddr;
651 			blkiov15[i].bi_segcreate = blkiov[i].bi_segcreate;
652 			blkiov15[i].bi_version	 = blkiov[i].bi_version;
653 			blkiov15[i].bi_bp	 = blkiov[i].bi_bp;
654 			blkiov15[i].bi_size	 = blkiov[i].bi_size;
655 		}
656 		copyout(blkiov15, SCARG(uap, blkiov),
657 			blkcnt * sizeof(BLOCK_INFO_15));
658 	}
659     out:
660 	lfs_free(fs, blkiov, LFS_NB_BLKIOV);
661 	lfs_free(fs, blkiov15, LFS_NB_BLKIOV);
662 	KERNEL_UNLOCK_ONE(NULL);
663 	return error;
664 }
665 #endif
666 
667 int
668 lfs_bmapv(struct proc *p, fsid_t *fsidp, BLOCK_INFO *blkiov, int blkcnt)
669 {
670 	BLOCK_INFO *blkp;
671 	IFILE *ifp;
672 	struct buf *bp;
673 	struct inode *ip = NULL;
674 	struct lfs *fs;
675 	struct mount *mntp;
676 	struct ufsmount *ump;
677 	struct vnode *vp;
678 	ino_t lastino;
679 	daddr_t v_daddr;
680 	int cnt, error;
681 	int numrefed = 0;
682 
683 	lfs_cleaner_pid = p->p_pid;
684 
685 	if ((mntp = vfs_getvfs(fsidp)) == NULL)
686 		return (ENOENT);
687 
688 	ump = VFSTOUFS(mntp);
689 	if ((error = vfs_busy(mntp, NULL)) != 0)
690 		return (error);
691 
692 	cnt = blkcnt;
693 
694 	fs = VFSTOUFS(mntp)->um_lfs;
695 
696 	error = 0;
697 
698 	/* these were inside the initialization for the for loop */
699 	v_daddr = LFS_UNUSED_DADDR;
700 	lastino = LFS_UNUSED_INUM;
701 	for (blkp = blkiov; cnt--; ++blkp)
702 	{
703 		/*
704 		 * Get the IFILE entry (only once) and see if the file still
705 		 * exists.
706 		 */
707 		if (lastino != blkp->bi_inode) {
708 			/*
709 			 * Finish the old file, if there was one.  The presence
710 			 * of a usable vnode in vp is signaled by a valid
711 			 * v_daddr.
712 			 */
713 			if (v_daddr != LFS_UNUSED_DADDR) {
714 				lfs_vunref(vp);
715 				if (VTOI(vp)->i_lfs_iflags & LFSI_BMAP)
716 					vrecycle(vp, NULL, NULL);
717 				numrefed--;
718 			}
719 
720 			/*
721 			 * Start a new file
722 			 */
723 			lastino = blkp->bi_inode;
724 			if (blkp->bi_inode == LFS_IFILE_INUM)
725 				v_daddr = fs->lfs_idaddr;
726 			else {
727 				LFS_IENTRY(ifp, fs, blkp->bi_inode, bp);
728 				v_daddr = ifp->if_daddr;
729 				brelse(bp, 0);
730 			}
731 			if (v_daddr == LFS_UNUSED_DADDR) {
732 				blkp->bi_daddr = LFS_UNUSED_DADDR;
733 				continue;
734 			}
735 			/*
736 			 * A regular call to VFS_VGET could deadlock
737 			 * here.  Instead, we try an unlocked access.
738 			 */
739 			mutex_enter(&ufs_ihash_lock);
740 			vp = ufs_ihashlookup(ump->um_dev, blkp->bi_inode);
741 			if (vp != NULL && !(vp->v_iflag & VI_XLOCK)) {
742 				ip = VTOI(vp);
743 				mutex_enter(vp->v_interlock);
744 				mutex_exit(&ufs_ihash_lock);
745 				if (lfs_vref(vp)) {
746 					v_daddr = LFS_UNUSED_DADDR;
747 					continue;
748 				}
749 				numrefed++;
750 			} else {
751 				mutex_exit(&ufs_ihash_lock);
752 				/*
753 				 * Don't VFS_VGET if we're being unmounted,
754 				 * since we hold vfs_busy().
755 				 */
756 				if (mntp->mnt_iflag & IMNT_UNMOUNT) {
757 					v_daddr = LFS_UNUSED_DADDR;
758 					continue;
759 				}
760 				error = VFS_VGET(mntp, blkp->bi_inode, &vp);
761 				if (error) {
762 					DLOG((DLOG_CLEAN, "lfs_bmapv: vget ino"
763 					      "%d failed with %d",
764 					      blkp->bi_inode,error));
765 					v_daddr = LFS_UNUSED_DADDR;
766 					continue;
767 				} else {
768 					KASSERT(VOP_ISLOCKED(vp));
769 					VTOI(vp)->i_lfs_iflags |= LFSI_BMAP;
770 					VOP_UNLOCK(vp);
771 					numrefed++;
772 				}
773 			}
774 			ip = VTOI(vp);
775 		} else if (v_daddr == LFS_UNUSED_DADDR) {
776 			/*
777 			 * This can only happen if the vnode is dead.
778 			 * Keep going.	Note that we DO NOT set the
779 			 * bi_addr to anything -- if we failed to get
780 			 * the vnode, for example, we want to assume
781 			 * conservatively that all of its blocks *are*
782 			 * located in the segment in question.
783 			 * lfs_markv will throw them out if we are
784 			 * wrong.
785 			 */
786 			/* blkp->bi_daddr = LFS_UNUSED_DADDR; */
787 			continue;
788 		}
789 
790 		/* Past this point we are guaranteed that vp, ip are valid. */
791 
792 		if (blkp->bi_lbn == LFS_UNUSED_LBN) {
793 			/*
794 			 * We just want the inode address, which is
795 			 * conveniently in v_daddr.
796 			 */
797 			blkp->bi_daddr = v_daddr;
798 		} else {
799 			daddr_t bi_daddr;
800 
801 			/* XXX ondisk32 */
802 			error = VOP_BMAP(vp, blkp->bi_lbn, NULL,
803 					 &bi_daddr, NULL);
804 			if (error)
805 			{
806 				blkp->bi_daddr = LFS_UNUSED_DADDR;
807 				continue;
808 			}
809 			blkp->bi_daddr = dbtofsb(fs, bi_daddr);
810 			/* Fill in the block size, too */
811 			if (blkp->bi_lbn >= 0)
812 				blkp->bi_size = blksize(fs, ip, blkp->bi_lbn);
813 			else
814 				blkp->bi_size = fs->lfs_bsize;
815 		}
816 	}
817 
818 	/*
819 	 * Finish the old file, if there was one.  The presence
820 	 * of a usable vnode in vp is signaled by a valid v_daddr.
821 	 */
822 	if (v_daddr != LFS_UNUSED_DADDR) {
823 		lfs_vunref(vp);
824 		/* Recycle as above. */
825 		if (ip->i_lfs_iflags & LFSI_BMAP)
826 			vrecycle(vp, NULL, NULL);
827 		numrefed--;
828 	}
829 
830 #ifdef DIAGNOSTIC
831 	if (numrefed != 0)
832 		panic("lfs_bmapv: numrefed=%d", numrefed);
833 #endif
834 
835 	vfs_unbusy(mntp, false, NULL);
836 
837 	return 0;
838 }
839 
840 /*
841  * sys_lfs_segclean:
842  *
843  * Mark the segment clean.
844  *
845  *  0 on success
846  * -1/errno is return on error.
847  */
848 int
849 sys_lfs_segclean(struct lwp *l, const struct sys_lfs_segclean_args *uap, register_t *retval)
850 {
851 	/* {
852 		syscallarg(fsid_t *) fsidp;
853 		syscallarg(u_long) segment;
854 	} */
855 	struct lfs *fs;
856 	struct mount *mntp;
857 	fsid_t fsid;
858 	int error;
859 	unsigned long segnum;
860 
861 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
862 	    KAUTH_REQ_SYSTEM_LFS_SEGCLEAN, NULL, NULL, NULL);
863 	if (error)
864 		return (error);
865 
866 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
867 		return (error);
868 	if ((mntp = vfs_getvfs(&fsid)) == NULL)
869 		return (ENOENT);
870 
871 	fs = VFSTOUFS(mntp)->um_lfs;
872 	segnum = SCARG(uap, segment);
873 
874 	if ((error = vfs_busy(mntp, NULL)) != 0)
875 		return (error);
876 
877 	KERNEL_LOCK(1, NULL);
878 	lfs_seglock(fs, SEGM_PROT);
879 	error = lfs_do_segclean(fs, segnum);
880 	lfs_segunlock(fs);
881 	KERNEL_UNLOCK_ONE(NULL);
882 	vfs_unbusy(mntp, false, NULL);
883 	return error;
884 }
885 
886 /*
887  * Actually mark the segment clean.
888  * Must be called with the segment lock held.
889  */
890 int
891 lfs_do_segclean(struct lfs *fs, unsigned long segnum)
892 {
893 	extern int lfs_dostats;
894 	struct buf *bp;
895 	CLEANERINFO *cip;
896 	SEGUSE *sup;
897 
898 	if (dtosn(fs, fs->lfs_curseg) == segnum) {
899 		return (EBUSY);
900 	}
901 
902 	LFS_SEGENTRY(sup, fs, segnum, bp);
903 	if (sup->su_nbytes) {
904 		DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
905 		      " %d live bytes\n", segnum, sup->su_nbytes));
906 		brelse(bp, 0);
907 		return (EBUSY);
908 	}
909 	if (sup->su_flags & SEGUSE_ACTIVE) {
910 		DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
911 		      " segment is active\n", segnum));
912 		brelse(bp, 0);
913 		return (EBUSY);
914 	}
915 	if (!(sup->su_flags & SEGUSE_DIRTY)) {
916 		DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
917 		      " segment is already clean\n", segnum));
918 		brelse(bp, 0);
919 		return (EALREADY);
920 	}
921 
922 	fs->lfs_avail += segtod(fs, 1);
923 	if (sup->su_flags & SEGUSE_SUPERBLOCK)
924 		fs->lfs_avail -= btofsb(fs, LFS_SBPAD);
925 	if (fs->lfs_version > 1 && segnum == 0 &&
926 	    fs->lfs_start < btofsb(fs, LFS_LABELPAD))
927 		fs->lfs_avail -= btofsb(fs, LFS_LABELPAD) - fs->lfs_start;
928 	mutex_enter(&lfs_lock);
929 	fs->lfs_bfree += sup->su_nsums * btofsb(fs, fs->lfs_sumsize) +
930 		btofsb(fs, sup->su_ninos * fs->lfs_ibsize);
931 	fs->lfs_dmeta -= sup->su_nsums * btofsb(fs, fs->lfs_sumsize) +
932 		btofsb(fs, sup->su_ninos * fs->lfs_ibsize);
933 	if (fs->lfs_dmeta < 0)
934 		fs->lfs_dmeta = 0;
935 	mutex_exit(&lfs_lock);
936 	sup->su_flags &= ~SEGUSE_DIRTY;
937 	LFS_WRITESEGENTRY(sup, fs, segnum, bp);
938 
939 	LFS_CLEANERINFO(cip, fs, bp);
940 	++cip->clean;
941 	--cip->dirty;
942 	fs->lfs_nclean = cip->clean;
943 	cip->bfree = fs->lfs_bfree;
944 	mutex_enter(&lfs_lock);
945 	cip->avail = fs->lfs_avail - fs->lfs_ravail - fs->lfs_favail;
946 	wakeup(&fs->lfs_avail);
947 	mutex_exit(&lfs_lock);
948 	(void) LFS_BWRITE_LOG(bp);
949 
950 	if (lfs_dostats)
951 		++lfs_stats.segs_reclaimed;
952 
953 	return (0);
954 }
955 
956 /*
957  * This will block until a segment in file system fsid is written.  A timeout
958  * in milliseconds may be specified which will awake the cleaner automatically.
959  * An fsid of -1 means any file system, and a timeout of 0 means forever.
960  */
961 int
962 lfs_segwait(fsid_t *fsidp, struct timeval *tv)
963 {
964 	struct mount *mntp;
965 	void *addr;
966 	u_long timeout;
967 	int error;
968 
969 	KERNEL_LOCK(1, NULL);
970 	if (fsidp == NULL || (mntp = vfs_getvfs(fsidp)) == NULL)
971 		addr = &lfs_allclean_wakeup;
972 	else
973 		addr = &VFSTOUFS(mntp)->um_lfs->lfs_nextseg;
974 	/*
975 	 * XXX THIS COULD SLEEP FOREVER IF TIMEOUT IS {0,0}!
976 	 * XXX IS THAT WHAT IS INTENDED?
977 	 */
978 	timeout = tvtohz(tv);
979 	error = tsleep(addr, PCATCH | PVFS, "segment", timeout);
980 	KERNEL_UNLOCK_ONE(NULL);
981 	return (error == ERESTART ? EINTR : 0);
982 }
983 
984 /*
985  * sys_lfs_segwait:
986  *
987  * System call wrapper around lfs_segwait().
988  *
989  *  0 on success
990  *  1 on timeout
991  * -1/errno is return on error.
992  */
993 int
994 sys___lfs_segwait50(struct lwp *l, const struct sys___lfs_segwait50_args *uap,
995     register_t *retval)
996 {
997 	/* {
998 		syscallarg(fsid_t *) fsidp;
999 		syscallarg(struct timeval *) tv;
1000 	} */
1001 	struct timeval atv;
1002 	fsid_t fsid;
1003 	int error;
1004 
1005 	/* XXX need we be su to segwait? */
1006 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
1007 	    KAUTH_REQ_SYSTEM_LFS_SEGWAIT, NULL, NULL, NULL);
1008 	if (error)
1009 		return (error);
1010 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
1011 		return (error);
1012 
1013 	if (SCARG(uap, tv)) {
1014 		error = copyin(SCARG(uap, tv), &atv, sizeof(struct timeval));
1015 		if (error)
1016 			return (error);
1017 		if (itimerfix(&atv))
1018 			return (EINVAL);
1019 	} else /* NULL or invalid */
1020 		atv.tv_sec = atv.tv_usec = 0;
1021 	return lfs_segwait(&fsid, &atv);
1022 }
1023 
1024 /*
1025  * VFS_VGET call specialized for the cleaner.  The cleaner already knows the
1026  * daddr from the ifile, so don't look it up again.  If the cleaner is
1027  * processing IINFO structures, it may have the ondisk inode already, so
1028  * don't go retrieving it again.
1029  *
1030  * we lfs_vref, and it is the caller's responsibility to lfs_vunref
1031  * when finished.
1032  */
1033 
1034 int
1035 lfs_fasthashget(dev_t dev, ino_t ino, struct vnode **vpp)
1036 {
1037 	struct vnode *vp;
1038 
1039 	mutex_enter(&ufs_ihash_lock);
1040 	if ((vp = ufs_ihashlookup(dev, ino)) != NULL) {
1041 		mutex_enter(vp->v_interlock);
1042 		mutex_exit(&ufs_ihash_lock);
1043 		if (vp->v_iflag & VI_XLOCK) {
1044 			DLOG((DLOG_CLEAN, "lfs_fastvget: ino %d VI_XLOCK\n",
1045 			      ino));
1046 			lfs_stats.clean_vnlocked++;
1047 			mutex_exit(vp->v_interlock);
1048 			return EAGAIN;
1049 		}
1050 		if (lfs_vref(vp)) {
1051 			DLOG((DLOG_CLEAN, "lfs_fastvget: lfs_vref failed"
1052 			      " for ino %d\n", ino));
1053 			lfs_stats.clean_inlocked++;
1054 			return EAGAIN;
1055 		}
1056 	} else {
1057 		mutex_exit(&ufs_ihash_lock);
1058 	}
1059 	*vpp = vp;
1060 
1061 	return (0);
1062 }
1063 
1064 int
1065 lfs_fastvget(struct mount *mp, ino_t ino, daddr_t daddr, struct vnode **vpp,
1066 	     struct ufs1_dinode *dinp)
1067 {
1068 	struct inode *ip;
1069 	struct ufs1_dinode *dip;
1070 	struct vnode *vp;
1071 	struct ufsmount *ump;
1072 	dev_t dev;
1073 	int error, retries;
1074 	struct buf *bp;
1075 	struct lfs *fs;
1076 
1077 	ump = VFSTOUFS(mp);
1078 	dev = ump->um_dev;
1079 	fs = ump->um_lfs;
1080 
1081 	/*
1082 	 * Wait until the filesystem is fully mounted before allowing vget
1083 	 * to complete.	 This prevents possible problems with roll-forward.
1084 	 */
1085 	mutex_enter(&lfs_lock);
1086 	while (fs->lfs_flags & LFS_NOTYET) {
1087 		mtsleep(&fs->lfs_flags, PRIBIO+1, "lfs_fnotyet", 0,
1088 			&lfs_lock);
1089 	}
1090 	mutex_exit(&lfs_lock);
1091 
1092 	/*
1093 	 * This is playing fast and loose.  Someone may have the inode
1094 	 * locked, in which case they are going to be distinctly unhappy
1095 	 * if we trash something.
1096 	 */
1097 
1098 	error = lfs_fasthashget(dev, ino, vpp);
1099 	if (error != 0 || *vpp != NULL)
1100 		return (error);
1101 
1102 	/*
1103 	 * getnewvnode(9) will call vfs_busy, which will block if the
1104 	 * filesystem is being unmounted; but umount(9) is waiting for
1105 	 * us because we're already holding the fs busy.
1106 	 * XXXMP
1107 	 */
1108 	if (mp->mnt_iflag & IMNT_UNMOUNT) {
1109 		*vpp = NULL;
1110 		return EDEADLK;
1111 	}
1112 	error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, NULL, &vp);
1113 	if (error) {
1114 		*vpp = NULL;
1115 		return (error);
1116 	}
1117 
1118 	mutex_enter(&ufs_hashlock);
1119 	error = lfs_fasthashget(dev, ino, vpp);
1120 	if (error != 0 || *vpp != NULL) {
1121 		mutex_exit(&ufs_hashlock);
1122 		ungetnewvnode(vp);
1123 		return (error);
1124 	}
1125 
1126 	/* Allocate new vnode/inode. */
1127 	lfs_vcreate(mp, ino, vp);
1128 
1129 	/*
1130 	 * Put it onto its hash chain and lock it so that other requests for
1131 	 * this inode will block if they arrive while we are sleeping waiting
1132 	 * for old data structures to be purged or for the contents of the
1133 	 * disk portion of this inode to be read.
1134 	 */
1135 	ip = VTOI(vp);
1136 	ufs_ihashins(ip);
1137 	mutex_exit(&ufs_hashlock);
1138 
1139 #ifdef notyet
1140 	/* Not found in the cache => this vnode was loaded only for cleaning. */
1141 	ip->i_lfs_iflags |= LFSI_BMAP;
1142 #endif
1143 
1144 	/*
1145 	 * XXX
1146 	 * This may not need to be here, logically it should go down with
1147 	 * the i_devvp initialization.
1148 	 * Ask Kirk.
1149 	 */
1150 	ip->i_lfs = fs;
1151 
1152 	/* Read in the disk contents for the inode, copy into the inode. */
1153 	if (dinp) {
1154 		error = copyin(dinp, ip->i_din.ffs1_din, sizeof (struct ufs1_dinode));
1155 		if (error) {
1156 			DLOG((DLOG_CLEAN, "lfs_fastvget: dinode copyin failed"
1157 			      " for ino %d\n", ino));
1158 			ufs_ihashrem(ip);
1159 
1160 			/* Unlock and discard unneeded inode. */
1161 			VOP_UNLOCK(vp);
1162 			lfs_vunref(vp);
1163 			*vpp = NULL;
1164 			return (error);
1165 		}
1166 		if (ip->i_number != ino)
1167 			panic("lfs_fastvget: I was fed the wrong inode!");
1168 	} else {
1169 		retries = 0;
1170 	    again:
1171 		error = bread(ump->um_devvp, fsbtodb(fs, daddr), fs->lfs_ibsize,
1172 			      NOCRED, 0, &bp);
1173 		if (error) {
1174 			DLOG((DLOG_CLEAN, "lfs_fastvget: bread failed (%d)\n",
1175 			      error));
1176 			/*
1177 			 * The inode does not contain anything useful, so it
1178 			 * would be misleading to leave it on its hash chain.
1179 			 * Iput() will return it to the free list.
1180 			 */
1181 			ufs_ihashrem(ip);
1182 
1183 			/* Unlock and discard unneeded inode. */
1184 			VOP_UNLOCK(vp);
1185 			lfs_vunref(vp);
1186 			brelse(bp, 0);
1187 			*vpp = NULL;
1188 			return (error);
1189 		}
1190 		dip = lfs_ifind(ump->um_lfs, ino, bp);
1191 		if (dip == NULL) {
1192 			/* Assume write has not completed yet; try again */
1193 			brelse(bp, BC_INVAL);
1194 			++retries;
1195 			if (retries > LFS_IFIND_RETRIES)
1196 				panic("lfs_fastvget: dinode not found");
1197 			DLOG((DLOG_CLEAN, "lfs_fastvget: dinode not found,"
1198 			      " retrying...\n"));
1199 			goto again;
1200 		}
1201 		*ip->i_din.ffs1_din = *dip;
1202 		brelse(bp, 0);
1203 	}
1204 	lfs_vinit(mp, &vp);
1205 
1206 	*vpp = vp;
1207 
1208 	KASSERT(VOP_ISLOCKED(vp));
1209 	VOP_UNLOCK(vp);
1210 
1211 	return (0);
1212 }
1213 
1214 /*
1215  * Make up a "fake" cleaner buffer, copy the data from userland into it.
1216  */
1217 struct buf *
1218 lfs_fakebuf(struct lfs *fs, struct vnode *vp, int lbn, size_t size, void *uaddr)
1219 {
1220 	struct buf *bp;
1221 	int error;
1222 
1223 	KASSERT(VTOI(vp)->i_number != LFS_IFILE_INUM);
1224 
1225 	bp = lfs_newbuf(VTOI(vp)->i_lfs, vp, lbn, size, LFS_NB_CLEAN);
1226 	error = copyin(uaddr, bp->b_data, size);
1227 	if (error) {
1228 		lfs_freebuf(fs, bp);
1229 		return NULL;
1230 	}
1231 	KDASSERT(bp->b_iodone == lfs_callback);
1232 
1233 #if 0
1234 	mutex_enter(&lfs_lock);
1235 	++fs->lfs_iocount;
1236 	mutex_exit(&lfs_lock);
1237 #endif
1238 	bp->b_bufsize = size;
1239 	bp->b_bcount = size;
1240 	return (bp);
1241 }
1242