xref: /netbsd/sys/ufs/lfs/lfs_subr.c (revision bf9ec67e)
1 /*	$NetBSD: lfs_subr.c,v 1.25 2002/05/24 22:13:57 perseant Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000 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. All advertising materials mentioning features or use of this software
51  *    must display the following acknowledgement:
52  *	This product includes software developed by the University of
53  *	California, Berkeley and its contributors.
54  * 4. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)lfs_subr.c	8.4 (Berkeley) 5/8/95
71  */
72 
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: lfs_subr.c,v 1.25 2002/05/24 22:13:57 perseant Exp $");
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/namei.h>
79 #include <sys/vnode.h>
80 #include <sys/buf.h>
81 #include <sys/mount.h>
82 #include <sys/malloc.h>
83 #include <sys/proc.h>
84 
85 #include <ufs/ufs/inode.h>
86 #include <ufs/lfs/lfs.h>
87 #include <ufs/lfs/lfs_extern.h>
88 
89 /*
90  * Return buffer with the contents of block "offset" from the beginning of
91  * directory "ip".  If "res" is non-zero, fill it in with a pointer to the
92  * remaining space in the directory.
93  */
94 int
95 lfs_blkatoff(void *v)
96 {
97 	struct vop_blkatoff_args /* {
98 		struct vnode *a_vp;
99 		off_t a_offset;
100 		char **a_res;
101 		struct buf **a_bpp;
102 		} */ *ap = v;
103 	struct lfs *fs;
104 	struct inode *ip;
105 	struct buf *bp;
106 	ufs_daddr_t lbn;
107 	int bsize, error;
108 
109 	ip = VTOI(ap->a_vp);
110 	fs = ip->i_lfs;
111 	lbn = lblkno(fs, ap->a_offset);
112 	bsize = blksize(fs, ip, lbn);
113 
114 	*ap->a_bpp = NULL;
115 	if ((error = bread(ap->a_vp, lbn, bsize, NOCRED, &bp)) != 0) {
116 		brelse(bp);
117 		return (error);
118 	}
119 	if (ap->a_res)
120 		*ap->a_res = (char *)bp->b_data + blkoff(fs, ap->a_offset);
121 	*ap->a_bpp = bp;
122 	return (0);
123 }
124 
125 
126 /*
127  * lfs_seglock --
128  *	Single thread the segment writer.
129  */
130 void
131 lfs_seglock(struct lfs *fs, unsigned long flags)
132 {
133 	struct segment *sp;
134 	int s;
135 
136 	if (fs->lfs_seglock) {
137 		if (fs->lfs_lockpid == curproc->p_pid) {
138 			++fs->lfs_seglock;
139 			fs->lfs_sp->seg_flags |= flags;
140 			return;
141 		} else while (fs->lfs_seglock)
142 			(void)tsleep(&fs->lfs_seglock, PRIBIO + 1,
143 				     "lfs seglock", 0);
144 	}
145 
146 	fs->lfs_seglock = 1;
147 	fs->lfs_lockpid = curproc->p_pid;
148 
149 	sp = fs->lfs_sp = malloc(sizeof(struct segment), M_SEGMENT, M_WAITOK);
150 	sp->bpp = malloc(((fs->lfs_sumsize - SEGSUM_SIZE(fs)) /
151 			  sizeof(ufs_daddr_t) + 1) * sizeof(struct buf *),
152 			 M_SEGMENT, M_WAITOK);
153 	sp->seg_flags = flags;
154 	sp->vp = NULL;
155 	(void) lfs_initseg(fs);
156 
157 	/*
158 	 * Keep a cumulative count of the outstanding I/O operations.  If the
159 	 * disk drive catches up with us it could go to zero before we finish,
160 	 * so we artificially increment it by one until we've scheduled all of
161 	 * the writes we intend to do.
162 	 */
163 	s = splbio();
164 	++fs->lfs_iocount;
165 	splx(s);
166 }
167 
168 /*
169  * lfs_segunlock --
170  *	Single thread the segment writer.
171  */
172 void
173 lfs_segunlock(struct lfs *fs)
174 {
175 	struct segment *sp;
176 	unsigned long sync, ckp;
177 	int s;
178 	struct buf *bp;
179 	struct vnode *vp, *nvp;
180 	struct mount *mp;
181 	extern int lfs_dirvcount;
182 #ifdef LFS_MALLOC_SUMMARY
183 	extern int locked_queue_count;
184 	extern long locked_queue_bytes;
185 #endif
186 
187 	sp = fs->lfs_sp;
188 
189 	if (fs->lfs_seglock == 1 && !(sp->seg_flags & SEGM_PROT)) {
190 
191 		mp = fs->lfs_ivnode->v_mount;
192 		/*
193 		 * Go through and unmark all DIROP vnodes, possibly
194 		 * calling VOP_INACTIVE (through vrele).  This is
195 		 * delayed until now in order not to accidentally
196 		 * write a DIROP node through lfs_flush.
197 		 */
198 #ifndef LFS_NO_BACKVP_HACK
199 	/* BEGIN HACK */
200 #define	VN_OFFSET	(((caddr_t)&LIST_NEXT(vp, v_mntvnodes)) - (caddr_t)vp)
201 #define	BACK_VP(VP)	((struct vnode *)(((caddr_t)(VP)->v_mntvnodes.le_prev) - VN_OFFSET))
202 #define	BEG_OF_VLIST	((struct vnode *)(((caddr_t)&LIST_FIRST(&mp->mnt_vnodelist)) - VN_OFFSET))
203 
204 		/* Find last vnode. */
205 	loop:	for (vp = LIST_FIRST(&mp->mnt_vnodelist);
206 		     vp && LIST_NEXT(vp, v_mntvnodes) != NULL;
207 		     vp = LIST_NEXT(vp, v_mntvnodes));
208 		for (; vp && vp != BEG_OF_VLIST; vp = nvp) {
209 			nvp = BACK_VP(vp);
210 #else
211 	loop:
212 		 for (vp = LIST_FIRST(&mp->mnt_vnodelist);
213 		     vp != NULL;
214 		     vp = nvp) {
215 			nvp = LIST_NEXT(vp, v_mntvnodes);
216 #endif
217 			if (vp->v_mount != mp) {
218 				printf("lfs_segunlock: starting over\n");
219 				goto loop;
220 			}
221 			if (vp->v_type == VNON)
222 				continue;
223 			if (lfs_vref(vp))
224 				continue;
225 			if (VOP_ISLOCKED(vp) &&
226                             vp->v_lock.lk_lockholder != curproc->p_pid) {
227 				lfs_vunref(vp);
228 				continue;
229 			}
230 			if ((vp->v_flag & VDIROP) &&
231 			    !(VTOI(vp)->i_flag & IN_ADIROP)) {
232 				--lfs_dirvcount;
233 				vp->v_flag &= ~VDIROP;
234 				wakeup(&lfs_dirvcount);
235 				fs->lfs_unlockvp = vp;
236 				lfs_vunref(vp);
237 				vrele(vp);
238 				fs->lfs_unlockvp = NULL;
239 			} else {
240 				lfs_vunref(vp);
241 			}
242 		}
243 	}
244 
245 	if (fs->lfs_seglock == 1) {
246 		sync = sp->seg_flags & SEGM_SYNC;
247 		ckp = sp->seg_flags & SEGM_CKP;
248 		if (sp->bpp != sp->cbpp) {
249 			/* Free allocated segment summary */
250 			fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize);
251 			bp = *sp->bpp;
252 #ifdef LFS_MALLOC_SUMMARY
253 			lfs_freebuf(bp);
254 #else
255 			s = splbio();
256 			bremfree(bp);
257 			bp->b_flags |= B_DONE|B_INVAL;
258 			bp->b_flags &= ~B_DELWRI;
259 			reassignbuf(bp,bp->b_vp);
260 			splx(s);
261 			brelse(bp);
262 #endif
263 		} else
264 			printf ("unlock to 0 with no summary");
265 
266 		free(sp->bpp, M_SEGMENT);
267 		sp->bpp = NULL;
268 		free(sp, M_SEGMENT);
269 		fs->lfs_sp = NULL;
270 
271 		/*
272 		 * If the I/O count is non-zero, sleep until it reaches zero.
273 		 * At the moment, the user's process hangs around so we can
274 		 * sleep.
275 		 */
276 		s = splbio();
277 		if (--fs->lfs_iocount < LFS_THROTTLE)
278 			wakeup(&fs->lfs_iocount);
279 		if(fs->lfs_iocount == 0) {
280 			lfs_countlocked(&locked_queue_count,
281 					&locked_queue_bytes, "lfs_segunlock");
282 			wakeup(&locked_queue_count);
283 			wakeup(&fs->lfs_iocount);
284 		}
285 		/*
286 		 * We let checkpoints happen asynchronously.  That means
287 		 * that during recovery, we have to roll forward between
288 		 * the two segments described by the first and second
289 		 * superblocks to make sure that the checkpoint described
290 		 * by a superblock completed.
291 		 */
292 		while (sync && fs->lfs_iocount)
293 			(void)tsleep(&fs->lfs_iocount, PRIBIO + 1,
294 				     "lfs vflush", 0);
295 		splx(s);
296 		if (ckp) {
297 			fs->lfs_nactive = 0;
298 			/* If we *know* everything's on disk, write both sbs */
299 			if (sync)
300 				lfs_writesuper(fs,fs->lfs_sboffs[fs->lfs_activesb]);
301 			fs->lfs_activesb = 1 - fs->lfs_activesb;
302 			lfs_writesuper(fs,fs->lfs_sboffs[fs->lfs_activesb]);
303 		}
304 		--fs->lfs_seglock;
305 		fs->lfs_lockpid = 0;
306 		wakeup(&fs->lfs_seglock);
307 	} else if (fs->lfs_seglock == 0) {
308 		panic ("Seglock not held");
309 	} else {
310 		--fs->lfs_seglock;
311 	}
312 }
313