xref: /netbsd-src/sys/ufs/lfs/lfs_bio.c (revision 81e0d2b0af8485d94ed5da487d4253841a2e6e45)
1 /*	$NetBSD: lfs_bio.c,v 1.77 2004/01/28 10:54:23 yamt Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003 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) 1991, 1993
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_bio.c	8.10 (Berkeley) 6/10/95
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: lfs_bio.c,v 1.77 2004/01/28 10:54:23 yamt Exp $");
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/proc.h>
75 #include <sys/buf.h>
76 #include <sys/vnode.h>
77 #include <sys/resourcevar.h>
78 #include <sys/mount.h>
79 #include <sys/kernel.h>
80 
81 #include <ufs/ufs/inode.h>
82 #include <ufs/ufs/ufsmount.h>
83 #include <ufs/ufs/ufs_extern.h>
84 
85 #include <ufs/lfs/lfs.h>
86 #include <ufs/lfs/lfs_extern.h>
87 
88 #include <uvm/uvm.h>
89 
90 /* Macros to clear/set/test flags. */
91 # define	SET(t, f)	(t) |= (f)
92 # define	CLR(t, f)	(t) &= ~(f)
93 # define	ISSET(t, f)	((t) & (f))
94 
95 /*
96  * LFS block write function.
97  *
98  * XXX
99  * No write cost accounting is done.
100  * This is almost certainly wrong for synchronous operations and NFS.
101  *
102  * protected by lfs_subsys_lock.
103  */
104 int	locked_queue_count   = 0;	/* Count of locked-down buffers. */
105 long	locked_queue_bytes   = 0L;	/* Total size of locked buffers. */
106 int	lfs_subsys_pages     = 0L;	/* Total number LFS-written pages */
107 int	lfs_writing	     = 0;	/* Set if already kicked off a writer
108 					   because of buffer space */
109 /* Lock for aboves */
110 struct simplelock lfs_subsys_lock = SIMPLELOCK_INITIALIZER;
111 
112 extern int lfs_dostats;
113 
114 /*
115  * reserved number/bytes of locked buffers
116  */
117 int locked_queue_rcount = 0;
118 long locked_queue_rbytes = 0L;
119 
120 int lfs_fits_buf(struct lfs *, int, int);
121 int lfs_reservebuf(struct lfs *, struct vnode *vp, struct vnode *vp2,
122     int, int);
123 int lfs_reserveavail(struct lfs *, struct vnode *vp, struct vnode *vp2, int);
124 
125 int
126 lfs_fits_buf(struct lfs *fs, int n, int bytes)
127 {
128 	int count_fit, bytes_fit;
129 
130 	LOCK_ASSERT(simple_lock_held(&lfs_subsys_lock));
131 
132 	count_fit =
133 	    (locked_queue_count + locked_queue_rcount + n < LFS_WAIT_BUFS);
134 	bytes_fit =
135 	    (locked_queue_bytes + locked_queue_rbytes + bytes < LFS_WAIT_BYTES);
136 
137 #ifdef DEBUG_LFS
138 	if (!count_fit) {
139 		printf("lfs_fits_buf: no fit count: %d + %d + %d >= %d\n",
140 			locked_queue_count, locked_queue_rcount,
141 			n, LFS_WAIT_BUFS);
142 	}
143 	if (!bytes_fit) {
144 		printf("lfs_fits_buf: no fit bytes: %ld + %ld + %d >= %ld\n",
145 			locked_queue_bytes, locked_queue_rbytes,
146 			bytes, LFS_WAIT_BYTES);
147 	}
148 #endif /* DEBUG_LFS */
149 
150 	return (count_fit && bytes_fit);
151 }
152 
153 /* ARGSUSED */
154 int
155 lfs_reservebuf(struct lfs *fs, struct vnode *vp, struct vnode *vp2,
156     int n, int bytes)
157 {
158 	KASSERT(locked_queue_rcount >= 0);
159 	KASSERT(locked_queue_rbytes >= 0);
160 
161 	simple_lock(&lfs_subsys_lock);
162 	while (n > 0 && !lfs_fits_buf(fs, n, bytes)) {
163 		int error;
164 
165 		lfs_flush(fs, 0);
166 
167 		error = ltsleep(&locked_queue_count, PCATCH | PUSER,
168 		    "lfsresbuf", hz * LFS_BUFWAIT, &lfs_subsys_lock);
169 		if (error && error != EWOULDBLOCK) {
170 			simple_unlock(&lfs_subsys_lock);
171 			return error;
172 		}
173 	}
174 
175 	locked_queue_rcount += n;
176 	locked_queue_rbytes += bytes;
177 
178 	simple_unlock(&lfs_subsys_lock);
179 
180 	KASSERT(locked_queue_rcount >= 0);
181 	KASSERT(locked_queue_rbytes >= 0);
182 
183 	return 0;
184 }
185 
186 /*
187  * Try to reserve some blocks, prior to performing a sensitive operation that
188  * requires the vnode lock to be honored.  If there is not enough space, give
189  * up the vnode lock temporarily and wait for the space to become available.
190  *
191  * Called with vp locked.  (Note nowever that if fsb < 0, vp is ignored.)
192  *
193  * XXX YAMT - it isn't safe to unlock vp here
194  * because the node might be modified while we sleep.
195  * (eg. cached states like i_offset might be stale,
196  *  the vnode might be truncated, etc..)
197  * maybe we should have a way to restart the vnodeop (EVOPRESTART?)
198  * or rearrange vnodeop interface to leave vnode locking to file system
199  * specific code so that each file systems can have their own vnode locking and
200  * vnode re-using strategies.
201  */
202 int
203 lfs_reserveavail(struct lfs *fs, struct vnode *vp, struct vnode *vp2, int fsb)
204 {
205 	CLEANERINFO *cip;
206 	struct buf *bp;
207 	int error, slept;
208 
209 	slept = 0;
210 	while (fsb > 0 && !lfs_fits(fs, fsb + fs->lfs_ravail)) {
211 #if 0
212 		/*
213 		 * XXX ideally, we should unlock vnodes here
214 		 * because we might sleep very long time.
215 		 */
216 		VOP_UNLOCK(vp, 0);
217 		if (vp2 != NULL) {
218 			VOP_UNLOCK(vp2, 0);
219 		}
220 #else
221 		/*
222 		 * XXX since we'll sleep for cleaner with vnode lock holding,
223 		 * deadlock will occur if cleaner tries to lock the vnode.
224 		 * (eg. lfs_markv -> lfs_fastvget -> getnewvnode -> vclean)
225 		 */
226 #endif
227 
228 		if (!slept) {
229 #ifdef DEBUG
230 			printf("lfs_reserve: waiting for %ld (bfree = %d,"
231 			       " est_bfree = %d)\n",
232 			       fsb + fs->lfs_ravail, fs->lfs_bfree,
233 			       LFS_EST_BFREE(fs));
234 #endif
235 		}
236 		++slept;
237 
238 		/* Wake up the cleaner */
239 		LFS_CLEANERINFO(cip, fs, bp);
240 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 0);
241 		wakeup(&lfs_allclean_wakeup);
242 		wakeup(&fs->lfs_nextseg);
243 
244 		error = tsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_reserve",
245 			       0);
246 #if 0
247 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX use lockstatus */
248 		vn_lock(vp2, LK_EXCLUSIVE | LK_RETRY); /* XXX use lockstatus */
249 #endif
250 		if (error)
251 			return error;
252 	}
253 #ifdef DEBUG
254 	if (slept)
255 		printf("lfs_reserve: woke up\n");
256 #endif
257 	fs->lfs_ravail += fsb;
258 
259 	return 0;
260 }
261 
262 #ifdef DIAGNOSTIC
263 int lfs_rescount;
264 int lfs_rescountdirop;
265 #endif
266 
267 int
268 lfs_reserve(struct lfs *fs, struct vnode *vp, struct vnode *vp2, int fsb)
269 {
270 	int error;
271 	int cantwait;
272 
273 	KASSERT(fsb < 0 || VOP_ISLOCKED(vp));
274 	KASSERT(vp2 == NULL || fsb < 0 || VOP_ISLOCKED(vp2));
275 	KASSERT(vp2 == NULL || !(VTOI(vp2)->i_flag & IN_ADIROP));
276 	KASSERT(vp2 == NULL || vp2 != fs->lfs_unlockvp);
277 
278 	cantwait = (VTOI(vp)->i_flag & IN_ADIROP) || fs->lfs_unlockvp == vp;
279 #ifdef DIAGNOSTIC
280 	if (cantwait) {
281 		if (fsb > 0)
282 			lfs_rescountdirop++;
283 		else if (fsb < 0)
284 			lfs_rescountdirop--;
285 		if (lfs_rescountdirop < 0)
286 			panic("lfs_rescountdirop");
287 	}
288 	else {
289 		if (fsb > 0)
290 			lfs_rescount++;
291 		else if (fsb < 0)
292 			lfs_rescount--;
293 		if (lfs_rescount < 0)
294 			panic("lfs_rescount");
295 	}
296 #endif
297 	if (cantwait)
298 		return 0;
299 
300 	/*
301 	 * XXX
302 	 * vref vnodes here so that cleaner doesn't try to reuse them.
303 	 * (see XXX comment in lfs_reserveavail)
304 	 */
305 	lfs_vref(vp);
306 	if (vp2 != NULL) {
307 		lfs_vref(vp2);
308 	}
309 
310 	error = lfs_reserveavail(fs, vp, vp2, fsb);
311 	if (error)
312 		goto done;
313 
314 	/*
315 	 * XXX just a guess. should be more precise.
316 	 */
317 	error = lfs_reservebuf(fs, vp, vp2,
318 	    fragstoblks(fs, fsb), fsbtob(fs, fsb));
319 	if (error)
320 		lfs_reserveavail(fs, vp, vp2, -fsb);
321 
322 done:
323 	lfs_vunref(vp);
324 	if (vp2 != NULL) {
325 		lfs_vunref(vp2);
326 	}
327 
328 	return error;
329 }
330 
331 int
332 lfs_bwrite(void *v)
333 {
334 	struct vop_bwrite_args /* {
335 		struct buf *a_bp;
336 	} */ *ap = v;
337 	struct buf *bp = ap->a_bp;
338 
339 #ifdef DIAGNOSTIC
340 	if (VTOI(bp->b_vp)->i_lfs->lfs_ronly == 0 && (bp->b_flags & B_ASYNC)) {
341 		panic("bawrite LFS buffer");
342 	}
343 #endif /* DIAGNOSTIC */
344 	return lfs_bwrite_ext(bp,0);
345 }
346 
347 /*
348  * Determine if there is enough room currently available to write fsb
349  * blocks.  We need enough blocks for the new blocks, the current
350  * inode blocks (including potentially the ifile inode), a summary block,
351  * and the segment usage table, plus an ifile block.
352  */
353 int
354 lfs_fits(struct lfs *fs, int fsb)
355 {
356 	int needed;
357 
358 	needed = fsb + btofsb(fs, fs->lfs_sumsize) +
359 		 ((howmany(fs->lfs_uinodes + 1, INOPB(fs)) + fs->lfs_segtabsz +
360 		   1) << (fs->lfs_blktodb - fs->lfs_fsbtodb));
361 
362 	if (needed >= fs->lfs_avail) {
363 #ifdef DEBUG
364 		printf("lfs_fits: no fit: fsb = %d, uinodes = %d, "
365 		       "needed = %d, avail = %d\n",
366 		       fsb, fs->lfs_uinodes, needed, fs->lfs_avail);
367 #endif
368 		return 0;
369 	}
370 	return 1;
371 }
372 
373 int
374 lfs_availwait(struct lfs *fs, int fsb)
375 {
376 	int error;
377 	CLEANERINFO *cip;
378 	struct buf *cbp;
379 
380 	/* Push cleaner blocks through regardless */
381 	simple_lock(&fs->lfs_interlock);
382 	if (fs->lfs_seglock &&
383 	    fs->lfs_lockpid == curproc->p_pid &&
384 	    fs->lfs_sp->seg_flags & (SEGM_CLEAN | SEGM_FORCE_CKP)) {
385 		simple_unlock(&fs->lfs_interlock);
386 		return 0;
387 	}
388 	simple_unlock(&fs->lfs_interlock);
389 
390 	while (!lfs_fits(fs, fsb)) {
391 		/*
392 		 * Out of space, need cleaner to run.
393 		 * Update the cleaner info, then wake it up.
394 		 * Note the cleanerinfo block is on the ifile
395 		 * so it CANT_WAIT.
396 		 */
397 		LFS_CLEANERINFO(cip, fs, cbp);
398 		LFS_SYNC_CLEANERINFO(cip, fs, cbp, 0);
399 
400 		printf("lfs_availwait: out of available space, "
401 		       "waiting on cleaner\n");
402 
403 		wakeup(&lfs_allclean_wakeup);
404 		wakeup(&fs->lfs_nextseg);
405 #ifdef DIAGNOSTIC
406 		if (fs->lfs_seglock && fs->lfs_lockpid == curproc->p_pid)
407 			panic("lfs_availwait: deadlock");
408 #endif
409 		error = tsleep(&fs->lfs_avail, PCATCH | PUSER, "cleaner", 0);
410 		if (error)
411 			return (error);
412 	}
413 	return 0;
414 }
415 
416 int
417 lfs_bwrite_ext(struct buf *bp, int flags)
418 {
419 	struct lfs *fs;
420 	struct inode *ip;
421 	int fsb, s;
422 
423 	KASSERT(bp->b_flags & B_BUSY);
424 	KASSERT(flags & BW_CLEAN || !LFS_IS_MALLOC_BUF(bp));
425 	KASSERT((bp->b_flags & (B_DELWRI|B_LOCKED)) != B_DELWRI);
426 	KASSERT((bp->b_flags & (B_DELWRI|B_LOCKED)) != B_LOCKED);
427 
428 	/*
429 	 * Don't write *any* blocks if we're mounted read-only.
430 	 * In particular the cleaner can't write blocks either.
431 	 */
432 	if (VTOI(bp->b_vp)->i_lfs->lfs_ronly) {
433 		bp->b_flags &= ~(B_DELWRI | B_READ | B_ERROR);
434 		LFS_UNLOCK_BUF(bp);
435 		if (LFS_IS_MALLOC_BUF(bp))
436 			bp->b_flags &= ~B_BUSY;
437 		else
438 			brelse(bp);
439 		return EROFS;
440 	}
441 
442 	/*
443 	 * Set the delayed write flag and use reassignbuf to move the buffer
444 	 * from the clean list to the dirty one.
445 	 *
446 	 * Set the B_LOCKED flag and unlock the buffer, causing brelse to move
447 	 * the buffer onto the LOCKED free list.  This is necessary, otherwise
448 	 * getnewbuf() would try to reclaim the buffers using bawrite, which
449 	 * isn't going to work.
450 	 *
451 	 * XXX we don't let meta-data writes run out of space because they can
452 	 * come from the segment writer.  We need to make sure that there is
453 	 * enough space reserved so that there's room to write meta-data
454 	 * blocks.
455 	 */
456 	if (!(bp->b_flags & B_LOCKED)) {
457 		fs = VFSTOUFS(bp->b_vp->v_mount)->um_lfs;
458 		fsb = fragstofsb(fs, numfrags(fs, bp->b_bcount));
459 
460 		ip = VTOI(bp->b_vp);
461 		if (flags & BW_CLEAN) {
462 			LFS_SET_UINO(ip, IN_CLEANING);
463 		} else {
464 			LFS_SET_UINO(ip, IN_MODIFIED);
465 		}
466 		fs->lfs_avail -= fsb;
467 		bp->b_flags |= B_DELWRI;
468 
469 		LFS_LOCK_BUF(bp);
470 		bp->b_flags &= ~(B_READ | B_DONE | B_ERROR);
471 		s = splbio();
472 		reassignbuf(bp, bp->b_vp);
473 		splx(s);
474 	}
475 
476 	if (bp->b_flags & B_CALL)
477 		bp->b_flags &= ~B_BUSY;
478 	else
479 		brelse(bp);
480 
481 	return (0);
482 }
483 
484 void
485 lfs_flush_fs(struct lfs *fs, int flags)
486 {
487 	if (fs->lfs_ronly)
488 		return;
489 
490 	lfs_writer_enter(fs, "fldirop");
491 
492 	if (lfs_dostats)
493 		++lfs_stats.flush_invoked;
494 	lfs_segwrite(fs->lfs_ivnode->v_mount, flags);
495 
496 	lfs_writer_leave(fs);
497 }
498 
499 /*
500  * XXX
501  * This routine flushes buffers out of the B_LOCKED queue when LFS has too
502  * many locked down.  Eventually the pageout daemon will simply call LFS
503  * when pages need to be reclaimed.  Note, we have one static count of locked
504  * buffers, so we can't have more than a single file system.  To make this
505  * work for multiple file systems, put the count into the mount structure.
506  *
507  * called and return with lfs_subsys_lock held.
508  */
509 void
510 lfs_flush(struct lfs *fs, int flags)
511 {
512 	struct mount *mp, *nmp;
513 
514 	LOCK_ASSERT(simple_lock_held(&lfs_subsys_lock));
515 	KDASSERT(fs == NULL || !LFS_SEGLOCK_HELD(fs));
516 
517 	if (lfs_dostats)
518 		++lfs_stats.write_exceeded;
519 	if (lfs_writing && flags == 0) {/* XXX flags */
520 #ifdef DEBUG_LFS
521 		printf("lfs_flush: not flushing because another flush is active\n");
522 #endif
523 		return;
524 	}
525 	while (lfs_writing && (flags & SEGM_WRITERD))
526 		ltsleep(&lfs_writing, PRIBIO + 1, "lfsflush", 0,
527 		    &lfs_subsys_lock);
528 	lfs_writing = 1;
529 
530 	lfs_subsys_pages = 0; /* XXXUBC need a better way to count this */
531 	simple_unlock(&lfs_subsys_lock);
532 	wakeup(&lfs_subsys_pages);
533 
534 	simple_lock(&mountlist_slock);
535 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
536 	    mp = nmp) {
537 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
538 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
539 			continue;
540 		}
541 		if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS,
542 		    MFSNAMELEN) == 0)
543 			lfs_flush_fs(VFSTOUFS(mp)->um_lfs, flags);
544 		simple_lock(&mountlist_slock);
545 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
546 		vfs_unbusy(mp);
547 	}
548 	simple_unlock(&mountlist_slock);
549 	LFS_DEBUG_COUNTLOCKED("flush");
550 
551 	simple_lock(&lfs_subsys_lock);
552 	KASSERT(lfs_writing);
553 	lfs_writing = 0;
554 	wakeup(&lfs_writing);
555 }
556 
557 #define INOCOUNT(fs) howmany((fs)->lfs_uinodes, INOPB(fs))
558 #define INOBYTES(fs) ((fs)->lfs_uinodes * sizeof (struct ufs1_dinode))
559 
560 /*
561  * make sure that we don't have too many locked buffers.
562  * flush buffers if needed.
563  */
564 int
565 lfs_check(struct vnode *vp, daddr_t blkno, int flags)
566 {
567 	int error;
568 	struct lfs *fs;
569 	struct inode *ip;
570 
571 	error = 0;
572 	ip = VTOI(vp);
573 
574 	/* If out of buffers, wait on writer */
575 	/* XXX KS - if it's the Ifile, we're probably the cleaner! */
576 	if (ip->i_number == LFS_IFILE_INUM)
577 		return 0;
578 	/* If we're being called from inside a dirop, don't sleep */
579 	if (ip->i_flag & IN_ADIROP)
580 		return 0;
581 
582 	fs = ip->i_lfs;
583 
584 	/*
585 	 * If we would flush below, but dirops are active, sleep.
586 	 * Note that a dirop cannot ever reach this code!
587 	 */
588 	simple_lock(&lfs_subsys_lock);
589 	while (fs->lfs_dirops > 0 &&
590 	       (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS ||
591 		locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES ||
592 		lfs_subsys_pages > LFS_MAX_PAGES ||
593 		lfs_dirvcount > LFS_MAX_DIROP || fs->lfs_diropwait > 0))
594 	{
595 		++fs->lfs_diropwait;
596 		ltsleep(&fs->lfs_writer, PRIBIO+1, "bufdirop", 0,
597 			&lfs_subsys_lock);
598 		--fs->lfs_diropwait;
599 	}
600 
601 #ifdef DEBUG_LFS_FLUSH
602 	if (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS)
603 		printf("lqc = %d, max %d\n",
604 		    locked_queue_count + INOCOUNT(fs), LFS_MAX_BUFS);
605 	if (locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES)
606 		printf("lqb = %ld, max %ld\n",
607 		    locked_queue_bytes + INOBYTES(fs), LFS_MAX_BYTES);
608 	if (lfs_subsys_pages > LFS_MAX_PAGES)
609 		printf("lssp = %d, max %d\n", lfs_subsys_pages, LFS_MAX_PAGES);
610 	if (lfs_dirvcount > LFS_MAX_DIROP)
611 		printf("ldvc = %d, max %d\n", lfs_dirvcount, LFS_MAX_DIROP);
612 	if (fs->lfs_diropwait > 0)
613 		printf("ldvw = %d\n", fs->lfs_diropwait);
614 #endif
615 
616 	if (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS ||
617 	    locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES ||
618 	    lfs_subsys_pages > LFS_MAX_PAGES ||
619 	    lfs_dirvcount > LFS_MAX_DIROP || fs->lfs_diropwait > 0) {
620 		lfs_flush(fs, flags);
621 	}
622 
623 	while (locked_queue_count + INOCOUNT(fs) > LFS_WAIT_BUFS ||
624 		locked_queue_bytes + INOBYTES(fs) > LFS_WAIT_BYTES ||
625 		lfs_subsys_pages > LFS_WAIT_PAGES ||
626 		lfs_dirvcount > LFS_MAX_DIROP) {
627 		simple_unlock(&lfs_subsys_lock);
628 		if (lfs_dostats)
629 			++lfs_stats.wait_exceeded;
630 #ifdef DEBUG_LFS
631 		printf("lfs_check: waiting: count=%d, bytes=%ld\n",
632 			locked_queue_count, locked_queue_bytes);
633 #endif
634 		error = tsleep(&locked_queue_count, PCATCH | PUSER,
635 			       "buffers", hz * LFS_BUFWAIT);
636 		if (error != EWOULDBLOCK) {
637 			simple_lock(&lfs_subsys_lock);
638 			break;
639 		}
640 		/*
641 		 * lfs_flush might not flush all the buffers, if some of the
642 		 * inodes were locked or if most of them were Ifile blocks
643 		 * and we weren't asked to checkpoint.	Try flushing again
644 		 * to keep us from blocking indefinitely.
645 		 */
646 		simple_lock(&lfs_subsys_lock);
647 		if (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS ||
648 		    locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES) {
649 			lfs_flush(fs, flags | SEGM_CKP);
650 		}
651 	}
652 	simple_unlock(&lfs_subsys_lock);
653 	return (error);
654 }
655 
656 /*
657  * Allocate a new buffer header.
658  */
659 struct buf *
660 lfs_newbuf(struct lfs *fs, struct vnode *vp, daddr_t daddr, size_t size, int type)
661 {
662 	struct buf *bp;
663 	size_t nbytes;
664 	int s;
665 
666 	nbytes = roundup(size, fsbtob(fs, 1));
667 
668 	s = splbio();
669 	bp = pool_get(&bufpool, PR_WAITOK);
670 	splx(s);
671 	memset(bp, 0, sizeof(struct buf));
672 	BUF_INIT(bp);
673 	if (nbytes) {
674 		bp->b_data = lfs_malloc(fs, nbytes, type);
675 		/* memset(bp->b_data, 0, nbytes); */
676 	}
677 #ifdef DIAGNOSTIC
678 	if (vp == NULL)
679 		panic("vp is NULL in lfs_newbuf");
680 	if (bp == NULL)
681 		panic("bp is NULL after malloc in lfs_newbuf");
682 #endif
683 	s = splbio();
684 	bgetvp(vp, bp);
685 	splx(s);
686 
687 	bp->b_bufsize = size;
688 	bp->b_bcount = size;
689 	bp->b_lblkno = daddr;
690 	bp->b_blkno = daddr;
691 	bp->b_error = 0;
692 	bp->b_resid = 0;
693 	bp->b_iodone = lfs_callback;
694 	bp->b_flags |= B_BUSY | B_CALL | B_NOCACHE;
695 	bp->b_private = fs;
696 
697 	return (bp);
698 }
699 
700 void
701 lfs_freebuf(struct lfs *fs, struct buf *bp)
702 {
703 	int s;
704 
705 	s = splbio();
706 	if (bp->b_vp)
707 		brelvp(bp);
708 	if (!(bp->b_flags & B_INVAL)) { /* B_INVAL indicates a "fake" buffer */
709 		lfs_free(fs, bp->b_data, LFS_NB_UNKNOWN);
710 		bp->b_data = NULL;
711 	}
712 	pool_put(&bufpool, bp);
713 	splx(s);
714 }
715 
716 /*
717  * Definitions for the buffer free lists.
718  */
719 #define BQUEUES		4		/* number of free buffer queues */
720 
721 #define BQ_LOCKED	0		/* super-blocks &c */
722 #define BQ_LRU		1		/* lru, useful buffers */
723 #define BQ_AGE		2		/* rubbish */
724 #define BQ_EMPTY	3		/* buffer headers with no memory */
725 
726 extern TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES];
727 extern struct simplelock bqueue_slock;
728 
729 /*
730  * Return a count of buffers on the "locked" queue.
731  * Don't count malloced buffers, since they don't detract from the total.
732  */
733 void
734 lfs_countlocked(int *count, long *bytes, char *msg)
735 {
736 	struct buf *bp;
737 	int n = 0;
738 	long int size = 0L;
739 	int s;
740 
741 	s = splbio();
742 	simple_lock(&bqueue_slock);
743 	TAILQ_FOREACH(bp, &bufqueues[BQ_LOCKED], b_freelist) {
744 		KASSERT(!(bp->b_flags & B_CALL));
745 		n++;
746 		size += bp->b_bufsize;
747 #ifdef DEBUG_LOCKED_LIST
748 		if (n > nbuf)
749 			panic("lfs_countlocked: this can't happen: more"
750 			      " buffers locked than exist");
751 #endif
752 	}
753 #ifdef DEBUG_LOCKED_LIST
754 	/* Theoretically this function never really does anything */
755 	if (n != *count)
756 		printf("lfs_countlocked: %s: adjusted buf count from %d to %d\n",
757 		       msg, *count, n);
758 	if (size != *bytes)
759 		printf("lfs_countlocked: %s: adjusted byte count from %ld to %ld\n",
760 		       msg, *bytes, size);
761 #endif
762 	*count = n;
763 	*bytes = size;
764 	simple_unlock(&bqueue_slock);
765 	splx(s);
766 	return;
767 }
768