xref: /netbsd-src/sys/ufs/lfs/lfs_syscalls.c (revision b83ebeba7f767758d2778bb0f9d7a76534253621)
1 /*	$NetBSD: lfs_syscalls.c,v 1.172 2015/10/15 06:15:48 dholland 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 1.172 2015/10/15 06:15:48 dholland 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/lfs/ulfs_inode.h>
81 #include <ufs/lfs/ulfsmount.h>
82 #include <ufs/lfs/ulfs_extern.h>
83 
84 #include <ufs/lfs/lfs.h>
85 #include <ufs/lfs/lfs_accessors.h>
86 #include <ufs/lfs/lfs_kernel.h>
87 #include <ufs/lfs/lfs_extern.h>
88 
89 static int lfs_fastvget(struct mount *, ino_t, BLOCK_INFO *, int,
90     struct vnode **);
91 static struct buf *lfs_fakebuf(struct lfs *, struct vnode *, daddr_t,
92     size_t, void *);
93 
94 /*
95  * sys_lfs_markv:
96  *
97  * This will mark inodes and blocks dirty, so they are written into the log.
98  * It will block until all the blocks have been written.  The segment create
99  * time passed in the block_info and inode_info structures is used to decide
100  * if the data is valid for each block (in case some process dirtied a block
101  * or inode that is being cleaned between the determination that a block is
102  * live and the lfs_markv call).
103  *
104  *  0 on success
105  * -1/errno is return on error.
106  */
107 #ifdef USE_64BIT_SYSCALLS
108 int
109 sys_lfs_markv(struct lwp *l, const struct sys_lfs_markv_args *uap, register_t *retval)
110 {
111 	/* {
112 		syscallarg(fsid_t *) fsidp;
113 		syscallarg(struct block_info *) blkiov;
114 		syscallarg(int) blkcnt;
115 	} */
116 	BLOCK_INFO *blkiov;
117 	int blkcnt, error;
118 	fsid_t fsid;
119 	struct lfs *fs;
120 	struct mount *mntp;
121 
122 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
123 		return (error);
124 
125 	if ((mntp = vfs_getvfs(&fsid)) == NULL)
126 		return (ENOENT);
127 	fs = VFSTOULFS(mntp)->um_lfs;
128 
129 	blkcnt = SCARG(uap, blkcnt);
130 	if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT)
131 		return (EINVAL);
132 
133 	KERNEL_LOCK(1, NULL);
134 	blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV);
135 	if ((error = copyin(SCARG(uap, blkiov), blkiov,
136 			    blkcnt * sizeof(BLOCK_INFO))) != 0)
137 		goto out;
138 
139 	if ((error = lfs_markv(l, &fsid, blkiov, blkcnt)) == 0)
140 		copyout(blkiov, SCARG(uap, blkiov),
141 			blkcnt * sizeof(BLOCK_INFO));
142     out:
143 	lfs_free(fs, blkiov, LFS_NB_BLKIOV);
144 	KERNEL_UNLOCK_ONE(NULL);
145 	return error;
146 }
147 #else
148 int
149 sys_lfs_markv(struct lwp *l, const struct sys_lfs_markv_args *uap, register_t *retval)
150 {
151 	/* {
152 		syscallarg(fsid_t *) fsidp;
153 		syscallarg(struct block_info *) blkiov;
154 		syscallarg(int) blkcnt;
155 	} */
156 	BLOCK_INFO *blkiov;
157 	BLOCK_INFO_15 *blkiov15;
158 	int i, blkcnt, error;
159 	fsid_t fsid;
160 	struct lfs *fs;
161 	struct mount *mntp;
162 
163 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
164 		return (error);
165 
166 	if ((mntp = vfs_getvfs(&fsid)) == NULL)
167 		return (ENOENT);
168 	fs = VFSTOULFS(mntp)->um_lfs;
169 
170 	blkcnt = SCARG(uap, blkcnt);
171 	if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT)
172 		return (EINVAL);
173 
174 	KERNEL_LOCK(1, NULL);
175 	blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV);
176 	blkiov15 = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO_15), LFS_NB_BLKIOV);
177 	if ((error = copyin(SCARG(uap, blkiov), blkiov15,
178 			    blkcnt * sizeof(BLOCK_INFO_15))) != 0)
179 		goto out;
180 
181 	for (i = 0; i < blkcnt; i++) {
182 		blkiov[i].bi_inode     = blkiov15[i].bi_inode;
183 		blkiov[i].bi_lbn       = blkiov15[i].bi_lbn;
184 		blkiov[i].bi_daddr     = blkiov15[i].bi_daddr;
185 		blkiov[i].bi_segcreate = blkiov15[i].bi_segcreate;
186 		blkiov[i].bi_version   = blkiov15[i].bi_version;
187 		blkiov[i].bi_bp	       = blkiov15[i].bi_bp;
188 		blkiov[i].bi_size      = blkiov15[i].bi_size;
189 	}
190 
191 	if ((error = lfs_markv(l, &fsid, blkiov, blkcnt)) == 0) {
192 		for (i = 0; i < blkcnt; i++) {
193 			blkiov15[i].bi_inode	 = blkiov[i].bi_inode;
194 			blkiov15[i].bi_lbn	 = blkiov[i].bi_lbn;
195 			blkiov15[i].bi_daddr	 = blkiov[i].bi_daddr;
196 			blkiov15[i].bi_segcreate = blkiov[i].bi_segcreate;
197 			blkiov15[i].bi_version	 = blkiov[i].bi_version;
198 			blkiov15[i].bi_bp	 = blkiov[i].bi_bp;
199 			blkiov15[i].bi_size	 = blkiov[i].bi_size;
200 		}
201 		copyout(blkiov15, SCARG(uap, blkiov),
202 			blkcnt * sizeof(BLOCK_INFO_15));
203 	}
204     out:
205 	lfs_free(fs, blkiov, LFS_NB_BLKIOV);
206 	lfs_free(fs, blkiov15, LFS_NB_BLKIOV);
207 	KERNEL_UNLOCK_ONE(NULL);
208 	return error;
209 }
210 #endif
211 
212 #define	LFS_MARKV_MAX_BLOCKS	(LFS_MAX_BUFS)
213 
214 int
215 lfs_markv(struct lwp *l, fsid_t *fsidp, BLOCK_INFO *blkiov,
216     int blkcnt)
217 {
218 	BLOCK_INFO *blkp;
219 	IFILE *ifp;
220 	struct buf *bp;
221 	struct inode *ip = NULL;
222 	struct lfs *fs;
223 	struct mount *mntp;
224 	struct ulfsmount *ump;
225 	struct vnode *vp;
226 	ino_t lastino;
227 	daddr_t b_daddr;
228 	int cnt, error;
229 	int do_again = 0;
230 	int numrefed = 0;
231 	ino_t maxino;
232 	size_t obsize;
233 
234 	/* number of blocks/inodes that we have already bwrite'ed */
235 	int nblkwritten, ninowritten;
236 
237 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
238 	    KAUTH_REQ_SYSTEM_LFS_MARKV, NULL, NULL, NULL);
239 	if (error)
240 		return (error);
241 
242 	if ((mntp = vfs_getvfs(fsidp)) == NULL)
243 		return (ENOENT);
244 
245 	ump = VFSTOULFS(mntp);
246 	fs = ump->um_lfs;
247 
248 	if (fs->lfs_ronly)
249 		return EROFS;
250 
251 	maxino = (lfs_fragstoblks(fs, lfs_dino_getblocks(fs, VTOI(fs->lfs_ivnode)->i_din)) -
252 		      lfs_sb_getcleansz(fs) - lfs_sb_getsegtabsz(fs)) * lfs_sb_getifpb(fs);
253 
254 	cnt = blkcnt;
255 
256 	if ((error = vfs_busy(mntp, NULL)) != 0)
257 		return (error);
258 
259 	/*
260 	 * This seglock is just to prevent the fact that we might have to sleep
261 	 * from allowing the possibility that our blocks might become
262 	 * invalid.
263 	 *
264 	 * It is also important to note here that unless we specify SEGM_CKP,
265 	 * any Ifile blocks that we might be asked to clean will never get
266 	 * to the disk.
267 	 */
268 	lfs_seglock(fs, SEGM_CLEAN | SEGM_CKP | SEGM_SYNC);
269 
270 	/* Mark blocks/inodes dirty.  */
271 	error = 0;
272 
273 	/* these were inside the initialization for the for loop */
274 	vp = NULL;
275 	lastino = LFS_UNUSED_INUM;
276 	nblkwritten = ninowritten = 0;
277 	for (blkp = blkiov; cnt--; ++blkp)
278 	{
279 		/* Bounds-check incoming data, avoid panic for failed VGET */
280 		if (blkp->bi_inode <= 0 || blkp->bi_inode >= maxino) {
281 			error = EINVAL;
282 			goto err3;
283 		}
284 		/*
285 		 * Get the IFILE entry (only once) and see if the file still
286 		 * exists.
287 		 */
288 		if (lastino != blkp->bi_inode) {
289 			/*
290 			 * Finish the old file, if there was one.
291 			 */
292 			if (vp != NULL) {
293 				vput(vp);
294 				vp = NULL;
295 				numrefed--;
296 			}
297 
298 			/*
299 			 * Start a new file
300 			 */
301 			lastino = blkp->bi_inode;
302 
303 			/* Get the vnode/inode. */
304 			error = lfs_fastvget(mntp, blkp->bi_inode, blkp,
305 			    LK_EXCLUSIVE | LK_NOWAIT, &vp);
306 			if (error) {
307 				DLOG((DLOG_CLEAN, "lfs_markv: lfs_fastvget"
308 				      " failed with %d (ino %d, segment %d)\n",
309 				      error, blkp->bi_inode,
310 				      lfs_dtosn(fs, blkp->bi_daddr)));
311 				/*
312 				 * If we got EAGAIN, that means that the
313 				 * Inode was locked.  This is
314 				 * recoverable: just clean the rest of
315 				 * this segment, and let the cleaner try
316 				 * again with another.	(When the
317 				 * cleaner runs again, this segment will
318 				 * sort high on the list, since it is
319 				 * now almost entirely empty.)
320 				 */
321 				if (error == EAGAIN) {
322 					error = 0;
323 					do_again++;
324 				} else
325 					KASSERT(error == ENOENT);
326 				KASSERT(vp == NULL);
327 				ip = NULL;
328 				continue;
329 			}
330 
331 			ip = VTOI(vp);
332 			numrefed++;
333 			ninowritten++;
334 		} else if (vp == NULL) {
335 			/*
336 			 * This can only happen if the vnode is dead (or
337 			 * in any case we can't get it...e.g., it is
338 			 * inlocked).  Keep going.
339 			 */
340 			continue;
341 		}
342 
343 		/* Past this point we are guaranteed that vp, ip are valid. */
344 
345 		/* Can't clean VU_DIROP directories in case of truncation */
346 		/* XXX - maybe we should mark removed dirs specially? */
347 		if (vp->v_type == VDIR && (vp->v_uflag & VU_DIROP)) {
348 			do_again++;
349 			continue;
350 		}
351 
352 		/* If this BLOCK_INFO didn't contain a block, keep going. */
353 		if (blkp->bi_lbn == LFS_UNUSED_LBN) {
354 			/* XXX need to make sure that the inode gets written in this case */
355 			/* XXX but only write the inode if it's the right one */
356 			if (blkp->bi_inode != LFS_IFILE_INUM) {
357 				LFS_IENTRY(ifp, fs, blkp->bi_inode, bp);
358 				if (lfs_if_getdaddr(fs, ifp) == blkp->bi_daddr) {
359 					mutex_enter(&lfs_lock);
360 					LFS_SET_UINO(ip, IN_CLEANING);
361 					mutex_exit(&lfs_lock);
362 				}
363 				brelse(bp, 0);
364 			}
365 			continue;
366 		}
367 
368 		b_daddr = 0;
369 		if (VOP_BMAP(vp, blkp->bi_lbn, NULL, &b_daddr, NULL) ||
370 		    LFS_DBTOFSB(fs, b_daddr) != blkp->bi_daddr)
371 		{
372 			if (lfs_dtosn(fs, LFS_DBTOFSB(fs, b_daddr)) ==
373 			    lfs_dtosn(fs, blkp->bi_daddr))
374 			{
375 				DLOG((DLOG_CLEAN, "lfs_markv: wrong da same seg: %jx vs %jx\n",
376 				      (intmax_t)blkp->bi_daddr, (intmax_t)LFS_DBTOFSB(fs, b_daddr)));
377 			}
378 			do_again++;
379 			continue;
380 		}
381 
382 		/*
383 		 * Check block sizes.  The blocks being cleaned come from
384 		 * disk, so they should have the same size as their on-disk
385 		 * counterparts.
386 		 */
387 		if (blkp->bi_lbn >= 0)
388 			obsize = lfs_blksize(fs, ip, blkp->bi_lbn);
389 		else
390 			obsize = lfs_sb_getbsize(fs);
391 		/* Check for fragment size change */
392 		if (blkp->bi_lbn >= 0 && blkp->bi_lbn < ULFS_NDADDR) {
393 			obsize = ip->i_lfs_fragsize[blkp->bi_lbn];
394 		}
395 		if (obsize != blkp->bi_size) {
396 			DLOG((DLOG_CLEAN, "lfs_markv: ino %d lbn %jd wrong"
397 			      " size (%ld != %d), try again\n",
398 			      blkp->bi_inode, (intmax_t)blkp->bi_lbn,
399 			      (long) obsize, blkp->bi_size));
400 			do_again++;
401 			continue;
402 		}
403 
404 		/*
405 		 * If we get to here, then we are keeping the block.  If
406 		 * it is an indirect block, we want to actually put it
407 		 * in the buffer cache so that it can be updated in the
408 		 * finish_meta section.	 If it's not, we need to
409 		 * allocate a fake buffer so that writeseg can perform
410 		 * the copyin and write the buffer.
411 		 */
412 		if (ip->i_number != LFS_IFILE_INUM && blkp->bi_lbn >= 0) {
413 			/* Data Block */
414 			bp = lfs_fakebuf(fs, vp, blkp->bi_lbn,
415 					 blkp->bi_size, blkp->bi_bp);
416 			/* Pretend we used bread() to get it */
417 			bp->b_blkno = LFS_FSBTODB(fs, blkp->bi_daddr);
418 		} else {
419 			/* Indirect block or ifile */
420 			if (blkp->bi_size != lfs_sb_getbsize(fs) &&
421 			    ip->i_number != LFS_IFILE_INUM)
422 				panic("lfs_markv: partial indirect block?"
423 				    " size=%d\n", blkp->bi_size);
424 			bp = getblk(vp, blkp->bi_lbn, blkp->bi_size, 0, 0);
425 			if (!(bp->b_oflags & (BO_DONE|BO_DELWRI))) {
426 				/*
427 				 * The block in question was not found
428 				 * in the cache; i.e., the block that
429 				 * getblk() returned is empty.	So, we
430 				 * can (and should) copy in the
431 				 * contents, because we've already
432 				 * determined that this was the right
433 				 * version of this block on disk.
434 				 *
435 				 * And, it can't have changed underneath
436 				 * us, because we have the segment lock.
437 				 */
438 				error = copyin(blkp->bi_bp, bp->b_data, blkp->bi_size);
439 				if (error)
440 					goto err2;
441 			}
442 		}
443 		if ((error = lfs_bwrite_ext(bp, BW_CLEAN)) != 0)
444 			goto err2;
445 
446 		nblkwritten++;
447 		/*
448 		 * XXX should account indirect blocks and ifile pages as well
449 		 */
450 		if (nblkwritten + lfs_lblkno(fs, ninowritten * DINOSIZE(fs))
451 		    > LFS_MARKV_MAX_BLOCKS) {
452 			DLOG((DLOG_CLEAN, "lfs_markv: writing %d blks %d inos\n",
453 			      nblkwritten, ninowritten));
454 			lfs_segwrite(mntp, SEGM_CLEAN);
455 			nblkwritten = ninowritten = 0;
456 		}
457 	}
458 
459 	/*
460 	 * Finish the old file, if there was one
461 	 */
462 	if (vp != NULL) {
463 		vput(vp);
464 		vp = NULL;
465 		numrefed--;
466 	}
467 
468 #ifdef DIAGNOSTIC
469 	if (numrefed != 0)
470 		panic("lfs_markv: numrefed=%d", numrefed);
471 #endif
472 	DLOG((DLOG_CLEAN, "lfs_markv: writing %d blks %d inos (check point)\n",
473 	      nblkwritten, ninowritten));
474 
475 	/*
476 	 * The last write has to be SEGM_SYNC, because of calling semantics.
477 	 * It also has to be SEGM_CKP, because otherwise we could write
478 	 * over the newly cleaned data contained in a checkpoint, and then
479 	 * we'd be unhappy at recovery time.
480 	 */
481 	lfs_segwrite(mntp, SEGM_CLEAN | SEGM_CKP | SEGM_SYNC);
482 
483 	lfs_segunlock(fs);
484 
485 	vfs_unbusy(mntp, false, NULL);
486 	if (error)
487 		return (error);
488 	else if (do_again)
489 		return EAGAIN;
490 
491 	return 0;
492 
493 err2:
494 	DLOG((DLOG_CLEAN, "lfs_markv err2\n"));
495 
496 	/*
497 	 * XXX we're here because copyin() failed.
498 	 * XXX it means that we can't trust the cleanerd.  too bad.
499 	 * XXX how can we recover from this?
500 	 */
501 
502 err3:
503 	/*
504 	 * XXX should do segwrite here anyway?
505 	 */
506 
507 	if (vp != NULL) {
508 		vput(vp);
509 		vp = NULL;
510 		--numrefed;
511 	}
512 
513 	lfs_segunlock(fs);
514 	vfs_unbusy(mntp, false, NULL);
515 #ifdef DIAGNOSTIC
516 	if (numrefed != 0)
517 		panic("lfs_markv: numrefed=%d", numrefed);
518 #endif
519 
520 	return (error);
521 }
522 
523 /*
524  * sys_lfs_bmapv:
525  *
526  * This will fill in the current disk address for arrays of blocks.
527  *
528  *  0 on success
529  * -1/errno is return on error.
530  */
531 #ifdef USE_64BIT_SYSCALLS
532 int
533 sys_lfs_bmapv(struct lwp *l, const struct sys_lfs_bmapv_args *uap, register_t *retval)
534 {
535 	/* {
536 		syscallarg(fsid_t *) fsidp;
537 		syscallarg(struct block_info *) blkiov;
538 		syscallarg(int) blkcnt;
539 	} */
540 	BLOCK_INFO *blkiov;
541 	int blkcnt, error;
542 	fsid_t fsid;
543 	struct lfs *fs;
544 	struct mount *mntp;
545 
546 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
547 		return (error);
548 
549 	if ((mntp = vfs_getvfs(&fsid)) == NULL)
550 		return (ENOENT);
551 	fs = VFSTOULFS(mntp)->um_lfs;
552 
553 	blkcnt = SCARG(uap, blkcnt);
554 #if SIZE_T_MAX <= UINT_MAX
555 	if ((u_int) blkcnt > SIZE_T_MAX / sizeof(BLOCK_INFO))
556 		return (EINVAL);
557 #endif
558 	KERNEL_LOCK(1, NULL);
559 	blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV);
560 	if ((error = copyin(SCARG(uap, blkiov), blkiov,
561 			    blkcnt * sizeof(BLOCK_INFO))) != 0)
562 		goto out;
563 
564 	if ((error = lfs_bmapv(l, &fsid, blkiov, blkcnt)) == 0)
565 		copyout(blkiov, SCARG(uap, blkiov),
566 			blkcnt * sizeof(BLOCK_INFO));
567     out:
568 	lfs_free(fs, blkiov, LFS_NB_BLKIOV);
569 	KERNEL_UNLOCK_ONE(NULL);
570 	return error;
571 }
572 #else
573 int
574 sys_lfs_bmapv(struct lwp *l, const struct sys_lfs_bmapv_args *uap, register_t *retval)
575 {
576 	/* {
577 		syscallarg(fsid_t *) fsidp;
578 		syscallarg(struct block_info *) blkiov;
579 		syscallarg(int) blkcnt;
580 	} */
581 	BLOCK_INFO *blkiov;
582 	BLOCK_INFO_15 *blkiov15;
583 	int i, blkcnt, error;
584 	fsid_t fsid;
585 	struct lfs *fs;
586 	struct mount *mntp;
587 
588 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
589 		return (error);
590 
591 	if ((mntp = vfs_getvfs(&fsid)) == NULL)
592 		return (ENOENT);
593 	fs = VFSTOULFS(mntp)->um_lfs;
594 
595 	blkcnt = SCARG(uap, blkcnt);
596 	if ((size_t) blkcnt > SIZE_T_MAX / sizeof(BLOCK_INFO))
597 		return (EINVAL);
598 	KERNEL_LOCK(1, NULL);
599 	blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV);
600 	blkiov15 = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO_15), LFS_NB_BLKIOV);
601 	if ((error = copyin(SCARG(uap, blkiov), blkiov15,
602 			    blkcnt * sizeof(BLOCK_INFO_15))) != 0)
603 		goto out;
604 
605 	for (i = 0; i < blkcnt; i++) {
606 		blkiov[i].bi_inode     = blkiov15[i].bi_inode;
607 		blkiov[i].bi_lbn       = blkiov15[i].bi_lbn;
608 		blkiov[i].bi_daddr     = blkiov15[i].bi_daddr;
609 		blkiov[i].bi_segcreate = blkiov15[i].bi_segcreate;
610 		blkiov[i].bi_version   = blkiov15[i].bi_version;
611 		blkiov[i].bi_bp	       = blkiov15[i].bi_bp;
612 		blkiov[i].bi_size      = blkiov15[i].bi_size;
613 	}
614 
615 	if ((error = lfs_bmapv(l, &fsid, blkiov, blkcnt)) == 0) {
616 		for (i = 0; i < blkcnt; i++) {
617 			blkiov15[i].bi_inode	 = blkiov[i].bi_inode;
618 			blkiov15[i].bi_lbn	 = blkiov[i].bi_lbn;
619 			blkiov15[i].bi_daddr	 = blkiov[i].bi_daddr;
620 			blkiov15[i].bi_segcreate = blkiov[i].bi_segcreate;
621 			blkiov15[i].bi_version	 = blkiov[i].bi_version;
622 			blkiov15[i].bi_bp	 = blkiov[i].bi_bp;
623 			blkiov15[i].bi_size	 = blkiov[i].bi_size;
624 		}
625 		copyout(blkiov15, SCARG(uap, blkiov),
626 			blkcnt * sizeof(BLOCK_INFO_15));
627 	}
628     out:
629 	lfs_free(fs, blkiov, LFS_NB_BLKIOV);
630 	lfs_free(fs, blkiov15, LFS_NB_BLKIOV);
631 	KERNEL_UNLOCK_ONE(NULL);
632 	return error;
633 }
634 #endif
635 
636 int
637 lfs_bmapv(struct lwp *l, fsid_t *fsidp, BLOCK_INFO *blkiov, int blkcnt)
638 {
639 	BLOCK_INFO *blkp;
640 	IFILE *ifp;
641 	struct buf *bp;
642 	struct inode *ip = NULL;
643 	struct lfs *fs;
644 	struct mount *mntp;
645 	struct ulfsmount *ump;
646 	struct vnode *vp;
647 	ino_t lastino;
648 	daddr_t v_daddr;
649 	int cnt, error;
650 	int numrefed = 0;
651 
652 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
653 	    KAUTH_REQ_SYSTEM_LFS_BMAPV, NULL, NULL, NULL);
654 	if (error)
655 		return (error);
656 
657 	if ((mntp = vfs_getvfs(fsidp)) == NULL)
658 		return (ENOENT);
659 
660 	if ((error = vfs_busy(mntp, NULL)) != 0)
661 		return (error);
662 
663 	ump = VFSTOULFS(mntp);
664 	fs = ump->um_lfs;
665 
666 	if (fs->lfs_cleaner_thread == NULL)
667 		fs->lfs_cleaner_thread = curlwp;
668 	KASSERT(fs->lfs_cleaner_thread == curlwp);
669 
670 	cnt = blkcnt;
671 
672 	error = 0;
673 
674 	/* these were inside the initialization for the for loop */
675 	vp = NULL;
676 	v_daddr = LFS_UNUSED_DADDR;
677 	lastino = LFS_UNUSED_INUM;
678 	for (blkp = blkiov; cnt--; ++blkp)
679 	{
680 		/*
681 		 * Get the IFILE entry (only once) and see if the file still
682 		 * exists.
683 		 */
684 		if (lastino != blkp->bi_inode) {
685 			/*
686 			 * Finish the old file, if there was one.
687 			 */
688 			if (vp != NULL) {
689 				vput(vp);
690 				vp = NULL;
691 				numrefed--;
692 			}
693 
694 			/*
695 			 * Start a new file
696 			 */
697 			lastino = blkp->bi_inode;
698 			if (blkp->bi_inode == LFS_IFILE_INUM)
699 				v_daddr = lfs_sb_getidaddr(fs);
700 			else {
701 				LFS_IENTRY(ifp, fs, blkp->bi_inode, bp);
702 				v_daddr = lfs_if_getdaddr(fs, ifp);
703 				brelse(bp, 0);
704 			}
705 			if (v_daddr == LFS_UNUSED_DADDR) {
706 				blkp->bi_daddr = LFS_UNUSED_DADDR;
707 				continue;
708 			}
709 			error = lfs_fastvget(mntp, blkp->bi_inode, NULL,
710 			    LK_SHARED, &vp);
711 			if (error) {
712 				DLOG((DLOG_CLEAN, "lfs_bmapv: lfs_fastvget ino"
713 				      "%d failed with %d",
714 				      blkp->bi_inode,error));
715 				KASSERT(vp == NULL);
716 				continue;
717 			} else {
718 				KASSERT(VOP_ISLOCKED(vp));
719 				numrefed++;
720 			}
721 			ip = VTOI(vp);
722 		} else if (vp == NULL) {
723 			/*
724 			 * This can only happen if the vnode is dead.
725 			 * Keep going.	Note that we DO NOT set the
726 			 * bi_addr to anything -- if we failed to get
727 			 * the vnode, for example, we want to assume
728 			 * conservatively that all of its blocks *are*
729 			 * located in the segment in question.
730 			 * lfs_markv will throw them out if we are
731 			 * wrong.
732 			 */
733 			continue;
734 		}
735 
736 		/* Past this point we are guaranteed that vp, ip are valid. */
737 
738 		if (blkp->bi_lbn == LFS_UNUSED_LBN) {
739 			/*
740 			 * We just want the inode address, which is
741 			 * conveniently in v_daddr.
742 			 */
743 			blkp->bi_daddr = v_daddr;
744 		} else {
745 			daddr_t bi_daddr;
746 
747 			error = VOP_BMAP(vp, blkp->bi_lbn, NULL,
748 					 &bi_daddr, NULL);
749 			if (error)
750 			{
751 				blkp->bi_daddr = LFS_UNUSED_DADDR;
752 				continue;
753 			}
754 			blkp->bi_daddr = LFS_DBTOFSB(fs, bi_daddr);
755 			/* Fill in the block size, too */
756 			if (blkp->bi_lbn >= 0)
757 				blkp->bi_size = lfs_blksize(fs, ip, blkp->bi_lbn);
758 			else
759 				blkp->bi_size = lfs_sb_getbsize(fs);
760 		}
761 	}
762 
763 	/*
764 	 * Finish the old file, if there was one.
765 	 */
766 	if (vp != NULL) {
767 		vput(vp);
768 		vp = NULL;
769 		numrefed--;
770 	}
771 
772 #ifdef DIAGNOSTIC
773 	if (numrefed != 0)
774 		panic("lfs_bmapv: numrefed=%d", numrefed);
775 #endif
776 
777 	vfs_unbusy(mntp, false, NULL);
778 
779 	return 0;
780 }
781 
782 /*
783  * sys_lfs_segclean:
784  *
785  * Mark the segment clean.
786  *
787  *  0 on success
788  * -1/errno is return on error.
789  */
790 int
791 sys_lfs_segclean(struct lwp *l, const struct sys_lfs_segclean_args *uap, register_t *retval)
792 {
793 	/* {
794 		syscallarg(fsid_t *) fsidp;
795 		syscallarg(u_long) segment;
796 	} */
797 	struct lfs *fs;
798 	struct mount *mntp;
799 	fsid_t fsid;
800 	int error;
801 	unsigned long segnum;
802 
803 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
804 	    KAUTH_REQ_SYSTEM_LFS_SEGCLEAN, NULL, NULL, NULL);
805 	if (error)
806 		return (error);
807 
808 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
809 		return (error);
810 	if ((mntp = vfs_getvfs(&fsid)) == NULL)
811 		return (ENOENT);
812 
813 	fs = VFSTOULFS(mntp)->um_lfs;
814 	segnum = SCARG(uap, segment);
815 
816 	if ((error = vfs_busy(mntp, NULL)) != 0)
817 		return (error);
818 
819 	KERNEL_LOCK(1, NULL);
820 	lfs_seglock(fs, SEGM_PROT);
821 	error = lfs_do_segclean(fs, segnum);
822 	lfs_segunlock(fs);
823 	KERNEL_UNLOCK_ONE(NULL);
824 	vfs_unbusy(mntp, false, NULL);
825 	return error;
826 }
827 
828 /*
829  * Actually mark the segment clean.
830  * Must be called with the segment lock held.
831  */
832 int
833 lfs_do_segclean(struct lfs *fs, unsigned long segnum)
834 {
835 	extern int lfs_dostats;
836 	struct buf *bp;
837 	CLEANERINFO *cip;
838 	SEGUSE *sup;
839 
840 	if (lfs_dtosn(fs, lfs_sb_getcurseg(fs)) == segnum) {
841 		return (EBUSY);
842 	}
843 
844 	LFS_SEGENTRY(sup, fs, segnum, bp);
845 	if (sup->su_nbytes) {
846 		DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
847 		      " %d live bytes\n", segnum, sup->su_nbytes));
848 		brelse(bp, 0);
849 		return (EBUSY);
850 	}
851 	if (sup->su_flags & SEGUSE_ACTIVE) {
852 		DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
853 		      " segment is active\n", segnum));
854 		brelse(bp, 0);
855 		return (EBUSY);
856 	}
857 	if (!(sup->su_flags & SEGUSE_DIRTY)) {
858 		DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
859 		      " segment is already clean\n", segnum));
860 		brelse(bp, 0);
861 		return (EALREADY);
862 	}
863 
864 	lfs_sb_addavail(fs, lfs_segtod(fs, 1));
865 	if (sup->su_flags & SEGUSE_SUPERBLOCK)
866 		lfs_sb_subavail(fs, lfs_btofsb(fs, LFS_SBPAD));
867 	if (lfs_sb_getversion(fs) > 1 && segnum == 0 &&
868 	    lfs_sb_gets0addr(fs) < lfs_btofsb(fs, LFS_LABELPAD))
869 		lfs_sb_subavail(fs, lfs_btofsb(fs, LFS_LABELPAD) - lfs_sb_gets0addr(fs));
870 	mutex_enter(&lfs_lock);
871 	lfs_sb_addbfree(fs, sup->su_nsums * lfs_btofsb(fs, lfs_sb_getsumsize(fs)) +
872 		lfs_btofsb(fs, sup->su_ninos * lfs_sb_getibsize(fs)));
873 	lfs_sb_subdmeta(fs, sup->su_nsums * lfs_btofsb(fs, lfs_sb_getsumsize(fs)) +
874 		lfs_btofsb(fs, sup->su_ninos * lfs_sb_getibsize(fs)));
875 	if (lfs_sb_getdmeta(fs) < 0)
876 		lfs_sb_setdmeta(fs, 0);
877 	mutex_exit(&lfs_lock);
878 	sup->su_flags &= ~SEGUSE_DIRTY;
879 	LFS_WRITESEGENTRY(sup, fs, segnum, bp);
880 
881 	LFS_CLEANERINFO(cip, fs, bp);
882 	lfs_ci_shiftdirtytoclean(fs, cip, 1);
883 	lfs_sb_setnclean(fs, lfs_ci_getclean(fs, cip));
884 	mutex_enter(&lfs_lock);
885 	lfs_ci_setbfree(fs, cip, lfs_sb_getbfree(fs));
886 	lfs_ci_setavail(fs, cip, lfs_sb_getavail(fs)
887 			- fs->lfs_ravail - fs->lfs_favail);
888 	wakeup(&fs->lfs_availsleep);
889 	mutex_exit(&lfs_lock);
890 	(void) LFS_BWRITE_LOG(bp);
891 
892 	if (lfs_dostats)
893 		++lfs_stats.segs_reclaimed;
894 
895 	return (0);
896 }
897 
898 /*
899  * This will block until a segment in file system fsid is written.  A timeout
900  * in milliseconds may be specified which will awake the cleaner automatically.
901  * An fsid of -1 means any file system, and a timeout of 0 means forever.
902  */
903 int
904 lfs_segwait(fsid_t *fsidp, struct timeval *tv)
905 {
906 	struct mount *mntp;
907 	void *addr;
908 	u_long timeout;
909 	int error;
910 
911 	KERNEL_LOCK(1, NULL);
912 	if (fsidp == NULL || (mntp = vfs_getvfs(fsidp)) == NULL)
913 		addr = &lfs_allclean_wakeup;
914 	else
915 		addr = &VFSTOULFS(mntp)->um_lfs->lfs_nextsegsleep;
916 	/*
917 	 * XXX THIS COULD SLEEP FOREVER IF TIMEOUT IS {0,0}!
918 	 * XXX IS THAT WHAT IS INTENDED?
919 	 */
920 	timeout = tvtohz(tv);
921 	error = tsleep(addr, PCATCH | PVFS, "segment", timeout);
922 	KERNEL_UNLOCK_ONE(NULL);
923 	return (error == ERESTART ? EINTR : 0);
924 }
925 
926 /*
927  * sys_lfs_segwait:
928  *
929  * System call wrapper around lfs_segwait().
930  *
931  *  0 on success
932  *  1 on timeout
933  * -1/errno is return on error.
934  */
935 int
936 sys___lfs_segwait50(struct lwp *l, const struct sys___lfs_segwait50_args *uap,
937     register_t *retval)
938 {
939 	/* {
940 		syscallarg(fsid_t *) fsidp;
941 		syscallarg(struct timeval *) tv;
942 	} */
943 	struct timeval atv;
944 	fsid_t fsid;
945 	int error;
946 
947 	/* XXX need we be su to segwait? */
948 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_LFS,
949 	    KAUTH_REQ_SYSTEM_LFS_SEGWAIT, NULL, NULL, NULL);
950 	if (error)
951 		return (error);
952 	if ((error = copyin(SCARG(uap, fsidp), &fsid, sizeof(fsid_t))) != 0)
953 		return (error);
954 
955 	if (SCARG(uap, tv)) {
956 		error = copyin(SCARG(uap, tv), &atv, sizeof(struct timeval));
957 		if (error)
958 			return (error);
959 		if (itimerfix(&atv))
960 			return (EINVAL);
961 	} else /* NULL or invalid */
962 		atv.tv_sec = atv.tv_usec = 0;
963 	return lfs_segwait(&fsid, &atv);
964 }
965 
966 /*
967  * VFS_VGET call specialized for the cleaner.  If the cleaner is
968  * processing IINFO structures, it may have the ondisk inode already, so
969  * don't go retrieving it again.
970  *
971  * Return the vnode referenced and locked.
972  */
973 
974 static int
975 lfs_fastvget(struct mount *mp, ino_t ino, BLOCK_INFO *blkp, int lk_flags,
976     struct vnode **vpp)
977 {
978 	struct ulfsmount *ump;
979 	struct lfs *fs;
980 	int error;
981 
982 	ump = VFSTOULFS(mp);
983 	fs = ump->um_lfs;
984 	fs->lfs_cleaner_hint = blkp;
985 	error = vcache_get(mp, &ino, sizeof(ino), vpp);
986 	fs->lfs_cleaner_hint = NULL;
987 	if (error)
988 		return error;
989 	error = vn_lock(*vpp, lk_flags);
990 	if (error) {
991 		if (error == EBUSY)
992 			error = EAGAIN;
993 		vrele(*vpp);
994 		*vpp = NULL;
995 		return error;
996 	}
997 
998 	return 0;
999 }
1000 
1001 /*
1002  * Make up a "fake" cleaner buffer, copy the data from userland into it.
1003  */
1004 static struct buf *
1005 lfs_fakebuf(struct lfs *fs, struct vnode *vp, daddr_t lbn, size_t size, void *uaddr)
1006 {
1007 	struct buf *bp;
1008 	int error;
1009 
1010 	KASSERT(VTOI(vp)->i_number != LFS_IFILE_INUM);
1011 
1012 	bp = lfs_newbuf(VTOI(vp)->i_lfs, vp, lbn, size, LFS_NB_CLEAN);
1013 	error = copyin(uaddr, bp->b_data, size);
1014 	if (error) {
1015 		lfs_freebuf(fs, bp);
1016 		return NULL;
1017 	}
1018 	KDASSERT(bp->b_iodone == lfs_callback);
1019 
1020 #if 0
1021 	mutex_enter(&lfs_lock);
1022 	++fs->lfs_iocount;
1023 	mutex_exit(&lfs_lock);
1024 #endif
1025 	bp->b_bufsize = size;
1026 	bp->b_bcount = size;
1027 	return (bp);
1028 }
1029