Home | History | Annotate | Line # | Download | only in lfs
lfs_subr.c revision 1.27
      1 /*	$NetBSD: lfs_subr.c,v 1.27 2002/07/06 01:30:13 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 (at) 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 1.27 2002/07/06 01:30:13 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 
    135 	/*
    136 	 * SEGM_PROT specifies a shared lock, which can be held by multiple
    137 	 * processes simultaneously.  It is not possible to upgrade from a
    138 	 * SEGM_PROT lock to a normal seglock.
    139 	 */
    140 	if (fs->lfs_seglock) {
    141 		if (fs->lfs_lockpid == curproc->p_pid) {
    142 			++fs->lfs_seglock;
    143 			fs->lfs_sp->seg_flags |= flags;
    144 			return;
    145 		} else while (fs->lfs_seglock)
    146 			(void)tsleep(&fs->lfs_seglock, PRIBIO + 1,
    147 				     "lfs seglock", 0);
    148 	}
    149 
    150 	fs->lfs_seglock = 1;
    151 	fs->lfs_lockpid = curproc->p_pid;
    152 
    153 	/* Drain fragment size changes out */
    154 	lockmgr(&fs->lfs_fraglock, LK_EXCLUSIVE, 0);
    155 
    156 	sp = fs->lfs_sp = malloc(sizeof(struct segment), M_SEGMENT, M_WAITOK);
    157 	sp->bpp = malloc(((fs->lfs_sumsize - SEGSUM_SIZE(fs)) /
    158 			  sizeof(ufs_daddr_t) + 1) * sizeof(struct buf *),
    159 			 M_SEGMENT, M_WAITOK);
    160 	sp->seg_flags = flags;
    161 	sp->vp = NULL;
    162 	sp->seg_iocount = 0;
    163 	(void) lfs_initseg(fs);
    164 
    165 	/*
    166 	 * Keep a cumulative count of the outstanding I/O operations.  If the
    167 	 * disk drive catches up with us it could go to zero before we finish,
    168 	 * so we artificially increment it by one until we've scheduled all of
    169 	 * the writes we intend to do.
    170 	 */
    171 	++fs->lfs_iocount;
    172 }
    173 
    174 /*
    175  * lfs_segunlock --
    176  *	Single thread the segment writer.
    177  */
    178 void
    179 lfs_segunlock(struct lfs *fs)
    180 {
    181 	struct segment *sp;
    182 	unsigned long sync, ckp;
    183 	struct buf *bp;
    184 	struct vnode *vp, *nvp;
    185 	struct mount *mp;
    186 	extern int lfs_dirvcount;
    187 #ifdef LFS_MALLOC_SUMMARY
    188 	extern int locked_queue_count;
    189 	extern long locked_queue_bytes;
    190 #endif
    191 
    192 	sp = fs->lfs_sp;
    193 
    194 	if (fs->lfs_seglock == 1 && !(sp->seg_flags & SEGM_PROT)) {
    195 
    196 		mp = fs->lfs_ivnode->v_mount;
    197 		/*
    198 		 * Go through and unmark all DIROP vnodes, possibly
    199 		 * calling VOP_INACTIVE (through vrele).  This is
    200 		 * delayed until now in order not to accidentally
    201 		 * write a DIROP node through lfs_flush.
    202 		 */
    203 #ifndef LFS_NO_BACKVP_HACK
    204 	/* BEGIN HACK */
    205 #define	VN_OFFSET	(((caddr_t)&LIST_NEXT(vp, v_mntvnodes)) - (caddr_t)vp)
    206 #define	BACK_VP(VP)	((struct vnode *)(((caddr_t)(VP)->v_mntvnodes.le_prev) - VN_OFFSET))
    207 #define	BEG_OF_VLIST	((struct vnode *)(((caddr_t)&LIST_FIRST(&mp->mnt_vnodelist)) - VN_OFFSET))
    208 
    209 		/* Find last vnode. */
    210 	loop:	for (vp = LIST_FIRST(&mp->mnt_vnodelist);
    211 		     vp && LIST_NEXT(vp, v_mntvnodes) != NULL;
    212 		     vp = LIST_NEXT(vp, v_mntvnodes));
    213 		for (; vp && vp != BEG_OF_VLIST; vp = nvp) {
    214 			nvp = BACK_VP(vp);
    215 #else
    216 	loop:
    217 		 for (vp = LIST_FIRST(&mp->mnt_vnodelist);
    218 		     vp != NULL;
    219 		     vp = nvp) {
    220 			nvp = LIST_NEXT(vp, v_mntvnodes);
    221 #endif
    222 			if (vp->v_mount != mp) {
    223 				printf("lfs_segunlock: starting over\n");
    224 				goto loop;
    225 			}
    226 			if (vp->v_type == VNON)
    227 				continue;
    228 			if (lfs_vref(vp))
    229 				continue;
    230 			if (VOP_ISLOCKED(vp) &&
    231                             vp->v_lock.lk_lockholder != curproc->p_pid) {
    232 				lfs_vunref(vp);
    233 				continue;
    234 			}
    235 			if ((vp->v_flag & VDIROP) &&
    236 			    !(VTOI(vp)->i_flag & IN_ADIROP)) {
    237 				--lfs_dirvcount;
    238 				vp->v_flag &= ~VDIROP;
    239 				wakeup(&lfs_dirvcount);
    240 				fs->lfs_unlockvp = vp;
    241 				lfs_vunref(vp);
    242 				vrele(vp);
    243 				fs->lfs_unlockvp = NULL;
    244 			} else {
    245 				lfs_vunref(vp);
    246 			}
    247 		}
    248 	}
    249 
    250 	if (fs->lfs_seglock == 1) {
    251 		sync = sp->seg_flags & SEGM_SYNC;
    252 		ckp = sp->seg_flags & SEGM_CKP;
    253 		if (sp->bpp != sp->cbpp) {
    254 			/* Free allocated segment summary */
    255 			fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize);
    256 			bp = *sp->bpp;
    257 #ifdef LFS_MALLOC_SUMMARY
    258 			lfs_freebuf(bp);
    259 #else
    260 			s = splbio();
    261 			bremfree(bp);
    262 			bp->b_flags |= B_DONE|B_INVAL;
    263 			bp->b_flags &= ~B_DELWRI;
    264 			reassignbuf(bp,bp->b_vp);
    265 			splx(s);
    266 			brelse(bp);
    267 #endif
    268 		} else
    269 			printf ("unlock to 0 with no summary");
    270 
    271 		free(sp->bpp, M_SEGMENT);
    272 		sp->bpp = NULL;
    273 		/* The sync case holds a reference in `sp' to be freed below */
    274 		if (!sync)
    275 			free(sp, M_SEGMENT);
    276 		fs->lfs_sp = NULL;
    277 
    278 		/*
    279 		 * If the I/O count is non-zero, sleep until it reaches zero.
    280 		 * At the moment, the user's process hangs around so we can
    281 		 * sleep.
    282 		 */
    283 		if (--fs->lfs_iocount < LFS_THROTTLE)
    284 			wakeup(&fs->lfs_iocount);
    285 		if(fs->lfs_iocount == 0) {
    286 			lfs_countlocked(&locked_queue_count,
    287 					&locked_queue_bytes, "lfs_segunlock");
    288 			wakeup(&locked_queue_count);
    289 			wakeup(&fs->lfs_iocount);
    290 		}
    291 		/*
    292 		 * If we're not checkpointing, we don't have to block
    293 		 * other processes to wait for a synchronous write
    294 		 * to complete.
    295 		 */
    296 		if (!ckp) {
    297 			--fs->lfs_seglock;
    298 			fs->lfs_lockpid = 0;
    299 			wakeup(&fs->lfs_seglock);
    300 		}
    301 		/*
    302 		 * We let checkpoints happen asynchronously.  That means
    303 		 * that during recovery, we have to roll forward between
    304 		 * the two segments described by the first and second
    305 		 * superblocks to make sure that the checkpoint described
    306 		 * by a superblock completed.
    307 		 */
    308 		while (ckp && sync && fs->lfs_iocount)
    309 			(void)tsleep(&fs->lfs_iocount, PRIBIO + 1,
    310 				     "lfs_iocount", 0);
    311 		while (sync && sp->seg_iocount) {
    312 			(void)tsleep(&sp->seg_iocount, PRIBIO + 1,
    313 				     "seg_iocount", 0);
    314 			/* printf("sleeping on iocount %x == %d\n", sp, sp->seg_iocount); */
    315 		}
    316 		if (sync)
    317 			free(sp, M_SEGMENT);
    318 		if (ckp) {
    319 			fs->lfs_nactive = 0;
    320 			/* If we *know* everything's on disk, write both sbs */
    321 			if (sync)
    322 				lfs_writesuper(fs,fs->lfs_sboffs[fs->lfs_activesb]);
    323 			fs->lfs_activesb = 1 - fs->lfs_activesb;
    324 			lfs_writesuper(fs,fs->lfs_sboffs[fs->lfs_activesb]);
    325 
    326 			--fs->lfs_seglock;
    327 			fs->lfs_lockpid = 0;
    328 			wakeup(&fs->lfs_seglock);
    329 		}
    330 		/* Reenable fragment size changes */
    331 		lockmgr(&fs->lfs_fraglock, LK_RELEASE, 0);
    332 	} else if (fs->lfs_seglock == 0) {
    333 		panic ("Seglock not held");
    334 	} else {
    335 		--fs->lfs_seglock;
    336 	}
    337 }
    338