Home | History | Annotate | Line # | Download | only in lfs
lfs_segment.c revision 1.180.2.1
      1  1.180.2.1      chap /*	$NetBSD: lfs_segment.c,v 1.180.2.1 2006/06/19 04:11:44 chap Exp $	*/
      2        1.2       cgd 
      3       1.15  perseant /*-
      4      1.101  perseant  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
      5       1.15  perseant  * All rights reserved.
      6       1.15  perseant  *
      7       1.15  perseant  * This code is derived from software contributed to The NetBSD Foundation
      8       1.15  perseant  * by Konrad E. Schroder <perseant (at) hhhh.org>.
      9       1.15  perseant  *
     10       1.15  perseant  * Redistribution and use in source and binary forms, with or without
     11       1.15  perseant  * modification, are permitted provided that the following conditions
     12       1.15  perseant  * are met:
     13       1.15  perseant  * 1. Redistributions of source code must retain the above copyright
     14       1.15  perseant  *    notice, this list of conditions and the following disclaimer.
     15       1.15  perseant  * 2. Redistributions in binary form must reproduce the above copyright
     16       1.15  perseant  *    notice, this list of conditions and the following disclaimer in the
     17       1.15  perseant  *    documentation and/or other materials provided with the distribution.
     18       1.15  perseant  * 3. All advertising materials mentioning features or use of this software
     19       1.15  perseant  *    must display the following acknowledgement:
     20      1.103  perseant  *	This product includes software developed by the NetBSD
     21      1.103  perseant  *	Foundation, Inc. and its contributors.
     22       1.15  perseant  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23       1.15  perseant  *    contributors may be used to endorse or promote products derived
     24       1.15  perseant  *    from this software without specific prior written permission.
     25       1.15  perseant  *
     26       1.15  perseant  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27       1.15  perseant  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28       1.15  perseant  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29       1.15  perseant  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30       1.15  perseant  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31       1.15  perseant  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32       1.15  perseant  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33       1.15  perseant  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34       1.15  perseant  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35       1.15  perseant  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36       1.15  perseant  * POSSIBILITY OF SUCH DAMAGE.
     37       1.15  perseant  */
     38        1.1   mycroft /*
     39        1.1   mycroft  * Copyright (c) 1991, 1993
     40        1.1   mycroft  *	The Regents of the University of California.  All rights reserved.
     41        1.1   mycroft  *
     42        1.1   mycroft  * Redistribution and use in source and binary forms, with or without
     43        1.1   mycroft  * modification, are permitted provided that the following conditions
     44        1.1   mycroft  * are met:
     45        1.1   mycroft  * 1. Redistributions of source code must retain the above copyright
     46        1.1   mycroft  *    notice, this list of conditions and the following disclaimer.
     47        1.1   mycroft  * 2. Redistributions in binary form must reproduce the above copyright
     48        1.1   mycroft  *    notice, this list of conditions and the following disclaimer in the
     49        1.1   mycroft  *    documentation and/or other materials provided with the distribution.
     50      1.131       agc  * 3. Neither the name of the University nor the names of its contributors
     51        1.1   mycroft  *    may be used to endorse or promote products derived from this software
     52        1.1   mycroft  *    without specific prior written permission.
     53        1.1   mycroft  *
     54        1.1   mycroft  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     55        1.1   mycroft  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     56        1.1   mycroft  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     57        1.1   mycroft  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     58        1.1   mycroft  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     59        1.1   mycroft  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     60        1.1   mycroft  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     61        1.1   mycroft  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     62        1.1   mycroft  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     63        1.1   mycroft  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     64        1.1   mycroft  * SUCH DAMAGE.
     65        1.1   mycroft  *
     66       1.10      fvdl  *	@(#)lfs_segment.c	8.10 (Berkeley) 6/10/95
     67        1.1   mycroft  */
     68       1.72     lukem 
     69       1.72     lukem #include <sys/cdefs.h>
     70  1.180.2.1      chap __KERNEL_RCSID(0, "$NetBSD: lfs_segment.c,v 1.180.2.1 2006/06/19 04:11:44 chap Exp $");
     71        1.1   mycroft 
     72      1.158  perseant #ifdef DEBUG
     73      1.158  perseant # define vndebug(vp, str) do {						\
     74      1.158  perseant 	if (VTOI(vp)->i_flag & IN_CLEANING)				\
     75      1.158  perseant 		DLOG((DLOG_WVNODE, "not writing ino %d because %s (op %d)\n", \
     76      1.158  perseant 		     VTOI(vp)->i_number, (str), op));			\
     77      1.158  perseant } while(0)
     78      1.158  perseant #else
     79      1.158  perseant # define vndebug(vp, str)
     80      1.158  perseant #endif
     81      1.158  perseant #define ivndebug(vp, str) \
     82      1.158  perseant 	DLOG((DLOG_WVNODE, "ino %d: %s\n", VTOI(vp)->i_number, (str)))
     83       1.16  perseant 
     84       1.68       mrg #if defined(_KERNEL_OPT)
     85       1.30  perseant #include "opt_ddb.h"
     86       1.65  jdolecek #endif
     87       1.65  jdolecek 
     88        1.1   mycroft #include <sys/param.h>
     89        1.1   mycroft #include <sys/systm.h>
     90        1.1   mycroft #include <sys/namei.h>
     91        1.1   mycroft #include <sys/kernel.h>
     92        1.1   mycroft #include <sys/resourcevar.h>
     93        1.1   mycroft #include <sys/file.h>
     94        1.1   mycroft #include <sys/stat.h>
     95        1.1   mycroft #include <sys/buf.h>
     96        1.1   mycroft #include <sys/proc.h>
     97        1.1   mycroft #include <sys/vnode.h>
     98        1.1   mycroft #include <sys/mount.h>
     99      1.179      elad #include <sys/kauth.h>
    100        1.1   mycroft 
    101        1.1   mycroft #include <miscfs/specfs/specdev.h>
    102        1.1   mycroft #include <miscfs/fifofs/fifo.h>
    103        1.1   mycroft 
    104        1.1   mycroft #include <ufs/ufs/inode.h>
    105        1.1   mycroft #include <ufs/ufs/dir.h>
    106        1.1   mycroft #include <ufs/ufs/ufsmount.h>
    107        1.1   mycroft #include <ufs/ufs/ufs_extern.h>
    108        1.1   mycroft 
    109        1.1   mycroft #include <ufs/lfs/lfs.h>
    110        1.1   mycroft #include <ufs/lfs/lfs_extern.h>
    111        1.1   mycroft 
    112       1.79  perseant #include <uvm/uvm.h>
    113       1.74  perseant #include <uvm/uvm_extern.h>
    114       1.99   thorpej 
    115       1.99   thorpej MALLOC_DEFINE(M_SEGMENT, "LFS segment", "Segment for LFS");
    116       1.74  perseant 
    117       1.69  perseant extern int count_lock_queue(void);
    118       1.10      fvdl extern struct simplelock vnode_free_list_slock;		/* XXX */
    119      1.132      yamt extern struct simplelock bqueue_slock;			/* XXX */
    120        1.1   mycroft 
    121       1.79  perseant static void lfs_generic_callback(struct buf *, void (*)(struct buf *));
    122       1.79  perseant static void lfs_super_aiodone(struct buf *);
    123       1.79  perseant static void lfs_cluster_aiodone(struct buf *);
    124       1.74  perseant static void lfs_cluster_callback(struct buf *);
    125       1.74  perseant 
    126        1.1   mycroft /*
    127        1.1   mycroft  * Determine if it's OK to start a partial in this segment, or if we need
    128        1.1   mycroft  * to go on to a new segment.
    129        1.1   mycroft  */
    130        1.1   mycroft #define	LFS_PARTIAL_FITS(fs) \
    131       1.69  perseant 	((fs)->lfs_fsbpseg - ((fs)->lfs_offset - (fs)->lfs_curseg) > \
    132       1.69  perseant 	fragstofsb((fs), (fs)->lfs_frag))
    133        1.1   mycroft 
    134      1.171  perseant /*
    135      1.171  perseant  * Figure out whether we should do a checkpoint write or go ahead with
    136      1.171  perseant  * an ordinary write.
    137      1.171  perseant  */
    138      1.171  perseant #define LFS_SHOULD_CHECKPOINT(fs, flags) \
    139      1.171  perseant 	(fs->lfs_nactive > LFS_MAX_ACTIVE ||				\
    140      1.171  perseant 	 (flags & SEGM_CKP) ||						\
    141      1.171  perseant 	 fs->lfs_nclean < LFS_MAX_ACTIVE)
    142      1.171  perseant 
    143       1.69  perseant int	 lfs_match_fake(struct lfs *, struct buf *);
    144       1.69  perseant void	 lfs_newseg(struct lfs *);
    145       1.91      fvdl /* XXX ondisk32 */
    146      1.104  perseant void	 lfs_shellsort(struct buf **, int32_t *, int, int);
    147       1.69  perseant void	 lfs_supercallback(struct buf *);
    148       1.69  perseant void	 lfs_updatemeta(struct segment *);
    149       1.69  perseant void	 lfs_writesuper(struct lfs *, daddr_t);
    150       1.69  perseant int	 lfs_writevnodes(struct lfs *fs, struct mount *mp,
    151       1.69  perseant 	    struct segment *sp, int dirops);
    152        1.1   mycroft 
    153        1.1   mycroft int	lfs_allclean_wakeup;		/* Cleaner wakeup address. */
    154      1.103  perseant int	lfs_writeindir = 1;		/* whether to flush indir on non-ckp */
    155       1.25  perseant int	lfs_clean_vnhead = 0;		/* Allow freeing to head of vn list */
    156       1.32  perseant int	lfs_dirvcount = 0;		/* # active dirops */
    157        1.1   mycroft 
    158        1.1   mycroft /* Statistics Counters */
    159       1.15  perseant int lfs_dostats = 1;
    160        1.1   mycroft struct lfs_stats lfs_stats;
    161        1.1   mycroft 
    162        1.1   mycroft /* op values to lfs_writevnodes */
    163      1.103  perseant #define	VN_REG		0
    164        1.1   mycroft #define	VN_DIROP	1
    165        1.1   mycroft #define	VN_EMPTY	2
    166      1.103  perseant #define VN_CLEAN	3
    167       1.15  perseant 
    168       1.15  perseant /*
    169       1.15  perseant  * XXX KS - Set modification time on the Ifile, so the cleaner can
    170       1.15  perseant  * read the fs mod time off of it.  We don't set IN_UPDATE here,
    171       1.15  perseant  * since we don't really need this to be flushed to disk (and in any
    172       1.15  perseant  * case that wouldn't happen to the Ifile until we checkpoint).
    173       1.15  perseant  */
    174       1.15  perseant void
    175       1.69  perseant lfs_imtime(struct lfs *fs)
    176       1.15  perseant {
    177       1.15  perseant 	struct timespec ts;
    178       1.15  perseant 	struct inode *ip;
    179      1.157     perry 
    180      1.159  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    181  1.180.2.1      chap 	getnanotime(&ts);
    182       1.15  perseant 	ip = VTOI(fs->lfs_ivnode);
    183      1.119      fvdl 	ip->i_ffs1_mtime = ts.tv_sec;
    184      1.119      fvdl 	ip->i_ffs1_mtimensec = ts.tv_nsec;
    185       1.15  perseant }
    186        1.1   mycroft 
    187        1.1   mycroft /*
    188        1.1   mycroft  * Ifile and meta data blocks are not marked busy, so segment writes MUST be
    189        1.1   mycroft  * single threaded.  Currently, there are two paths into lfs_segwrite, sync()
    190        1.1   mycroft  * and getnewbuf().  They both mark the file system busy.  Lfs_vflush()
    191        1.1   mycroft  * explicitly marks the file system busy.  So lfs_segwrite is safe.  I think.
    192        1.1   mycroft  */
    193        1.1   mycroft 
    194       1.15  perseant #define IS_FLUSHING(fs,vp)  ((fs)->lfs_flushvp == (vp))
    195       1.15  perseant 
    196        1.1   mycroft int
    197       1.69  perseant lfs_vflush(struct vnode *vp)
    198        1.1   mycroft {
    199        1.1   mycroft 	struct inode *ip;
    200        1.1   mycroft 	struct lfs *fs;
    201        1.1   mycroft 	struct segment *sp;
    202       1.38  perseant 	struct buf *bp, *nbp, *tbp, *tnbp;
    203       1.30  perseant 	int error, s;
    204      1.101  perseant 	int flushed;
    205      1.171  perseant 	int relock;
    206       1.19  perseant 
    207       1.22  perseant 	ip = VTOI(vp);
    208       1.22  perseant 	fs = VFSTOUFS(vp->v_mount)->um_lfs;
    209      1.171  perseant 	relock = 0;
    210       1.22  perseant 
    211      1.171  perseant     top:
    212      1.159  perseant 	ASSERT_NO_SEGLOCK(fs);
    213       1.73       chs 	if (ip->i_flag & IN_CLEANING) {
    214       1.19  perseant 		ivndebug(vp,"vflush/in_cleaning");
    215       1.56  perseant 		LFS_CLR_UINO(ip, IN_CLEANING);
    216       1.56  perseant 		LFS_SET_UINO(ip, IN_MODIFIED);
    217       1.56  perseant 
    218       1.38  perseant 		/*
    219       1.38  perseant 		 * Toss any cleaning buffers that have real counterparts
    220      1.101  perseant 		 * to avoid losing new data.
    221       1.38  perseant 		 */
    222       1.38  perseant 		s = splbio();
    223       1.75  perseant 		for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
    224       1.75  perseant 			nbp = LIST_NEXT(bp, b_vnbufs);
    225      1.101  perseant 			if (!LFS_IS_MALLOC_BUF(bp))
    226      1.101  perseant 				continue;
    227      1.101  perseant 			/*
    228      1.106  perseant 			 * Look for pages matching the range covered
    229      1.106  perseant 			 * by cleaning blocks.  It's okay if more dirty
    230      1.106  perseant 			 * pages appear, so long as none disappear out
    231      1.106  perseant 			 * from under us.
    232      1.101  perseant 			 */
    233      1.101  perseant 			if (bp->b_lblkno > 0 && vp->v_type == VREG &&
    234      1.101  perseant 			    vp != fs->lfs_ivnode) {
    235      1.101  perseant 				struct vm_page *pg;
    236      1.101  perseant 				voff_t off;
    237      1.101  perseant 
    238      1.106  perseant 				simple_lock(&vp->v_interlock);
    239      1.101  perseant 				for (off = lblktosize(fs, bp->b_lblkno);
    240      1.101  perseant 				     off < lblktosize(fs, bp->b_lblkno + 1);
    241      1.101  perseant 				     off += PAGE_SIZE) {
    242      1.101  perseant 					pg = uvm_pagelookup(&vp->v_uobj, off);
    243      1.106  perseant 					if (pg == NULL)
    244      1.106  perseant 						continue;
    245      1.106  perseant 					if ((pg->flags & PG_CLEAN) == 0 ||
    246      1.106  perseant 					    pmap_is_modified(pg)) {
    247      1.101  perseant 						fs->lfs_avail += btofsb(fs,
    248      1.101  perseant 							bp->b_bcount);
    249       1.62  perseant 						wakeup(&fs->lfs_avail);
    250      1.101  perseant 						lfs_freebuf(fs, bp);
    251       1.69  perseant 						bp = NULL;
    252      1.173  perseant 						simple_unlock(&vp->v_interlock);
    253      1.101  perseant 						goto nextbp;
    254       1.38  perseant 					}
    255       1.38  perseant 				}
    256      1.106  perseant 				simple_unlock(&vp->v_interlock);
    257       1.38  perseant 			}
    258      1.101  perseant 			for (tbp = LIST_FIRST(&vp->v_dirtyblkhd); tbp;
    259      1.101  perseant 			    tbp = tnbp)
    260      1.101  perseant 			{
    261      1.101  perseant 				tnbp = LIST_NEXT(tbp, b_vnbufs);
    262      1.101  perseant 				if (tbp->b_vp == bp->b_vp
    263      1.101  perseant 				   && tbp->b_lblkno == bp->b_lblkno
    264      1.101  perseant 				   && tbp != bp)
    265      1.101  perseant 				{
    266      1.101  perseant 					fs->lfs_avail += btofsb(fs,
    267      1.101  perseant 						bp->b_bcount);
    268      1.101  perseant 					wakeup(&fs->lfs_avail);
    269      1.101  perseant 					lfs_freebuf(fs, bp);
    270      1.101  perseant 					bp = NULL;
    271      1.101  perseant 					break;
    272      1.101  perseant 				}
    273      1.101  perseant 			}
    274      1.101  perseant 		    nextbp:
    275      1.110  kristerw 			;
    276       1.38  perseant 		}
    277       1.38  perseant 		splx(s);
    278       1.19  perseant 	}
    279       1.19  perseant 
    280       1.19  perseant 	/* If the node is being written, wait until that is done */
    281      1.159  perseant 	simple_lock(&vp->v_interlock);
    282       1.74  perseant 	s = splbio();
    283       1.73       chs 	if (WRITEINPROG(vp)) {
    284       1.19  perseant 		ivndebug(vp,"vflush/writeinprog");
    285      1.159  perseant 		ltsleep(vp, (PRIBIO+1), "lfs_vw", 0, &vp->v_interlock);
    286       1.19  perseant 	}
    287       1.74  perseant 	splx(s);
    288      1.159  perseant 	simple_unlock(&vp->v_interlock);
    289        1.1   mycroft 
    290       1.15  perseant 	/* Protect against VXLOCK deadlock in vinvalbuf() */
    291        1.1   mycroft 	lfs_seglock(fs, SEGM_SYNC);
    292       1.30  perseant 
    293       1.30  perseant 	/* If we're supposed to flush a freed inode, just toss it */
    294      1.178  perseant 	if (ip->i_lfs_iflags & LFSI_DELETED) {
    295      1.158  perseant 		DLOG((DLOG_VNODE, "lfs_vflush: ino %d freed, not flushing\n",
    296      1.158  perseant 		      ip->i_number));
    297       1.30  perseant 		s = splbio();
    298      1.178  perseant 		/* Drain v_numoutput */
    299      1.178  perseant 		simple_lock(&global_v_numoutput_slock);
    300      1.178  perseant 		while (vp->v_numoutput > 0) {
    301      1.178  perseant 			vp->v_flag |= VBWAIT;
    302      1.178  perseant 			ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vf4", 0,
    303      1.178  perseant 				&global_v_numoutput_slock);
    304      1.178  perseant 		}
    305      1.178  perseant 		simple_unlock(&global_v_numoutput_slock);
    306      1.178  perseant 		KASSERT(vp->v_numoutput == 0);
    307      1.178  perseant 
    308       1.75  perseant 		for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
    309       1.75  perseant 			nbp = LIST_NEXT(bp, b_vnbufs);
    310      1.178  perseant 
    311      1.178  perseant 			KASSERT((bp->b_flags & B_GATHERED) == 0);
    312       1.62  perseant 			if (bp->b_flags & B_DELWRI) { /* XXX always true? */
    313       1.69  perseant 				fs->lfs_avail += btofsb(fs, bp->b_bcount);
    314       1.62  perseant 				wakeup(&fs->lfs_avail);
    315       1.62  perseant 			}
    316       1.30  perseant 			/* Copied from lfs_writeseg */
    317       1.30  perseant 			if (bp->b_flags & B_CALL) {
    318      1.101  perseant 				biodone(bp);
    319       1.30  perseant 			} else {
    320       1.30  perseant 				bremfree(bp);
    321       1.62  perseant 				LFS_UNLOCK_BUF(bp);
    322       1.30  perseant 				bp->b_flags &= ~(B_ERROR | B_READ | B_DELWRI |
    323      1.103  perseant 					 B_GATHERED);
    324       1.30  perseant 				bp->b_flags |= B_DONE;
    325       1.30  perseant 				reassignbuf(bp, vp);
    326      1.157     perry 				brelse(bp);
    327       1.30  perseant 			}
    328       1.30  perseant 		}
    329       1.30  perseant 		splx(s);
    330       1.56  perseant 		LFS_CLR_UINO(ip, IN_CLEANING);
    331       1.56  perseant 		LFS_CLR_UINO(ip, IN_MODIFIED | IN_ACCESSED);
    332       1.47  perseant 		ip->i_flag &= ~IN_ALLMOD;
    333      1.158  perseant 		DLOG((DLOG_VNODE, "lfs_vflush: done not flushing ino %d\n",
    334      1.158  perseant 		      ip->i_number));
    335       1.30  perseant 		lfs_segunlock(fs);
    336      1.178  perseant 
    337      1.178  perseant 		KASSERT(LIST_FIRST(&vp->v_dirtyblkhd) == NULL);
    338      1.178  perseant 
    339       1.30  perseant 		return 0;
    340       1.30  perseant 	}
    341       1.30  perseant 
    342      1.173  perseant 	fs->lfs_flushvp = vp;
    343      1.171  perseant 	if (LFS_SHOULD_CHECKPOINT(fs, fs->lfs_sp->seg_flags)) {
    344       1.79  perseant 		error = lfs_segwrite(vp->v_mount, SEGM_CKP | SEGM_SYNC);
    345      1.173  perseant 		fs->lfs_flushvp = NULL;
    346      1.173  perseant 		KASSERT(fs->lfs_flushvp_fakevref == 0);
    347       1.15  perseant 		lfs_segunlock(fs);
    348      1.178  perseant 
    349      1.178  perseant 		/* Make sure that any pending buffers get written */
    350      1.178  perseant 		s = splbio();
    351      1.178  perseant 		simple_lock(&global_v_numoutput_slock);
    352      1.178  perseant 		while (vp->v_numoutput > 0) {
    353      1.178  perseant 			vp->v_flag |= VBWAIT;
    354      1.178  perseant 			ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vf3", 0,
    355      1.178  perseant 				&global_v_numoutput_slock);
    356      1.178  perseant 		}
    357      1.178  perseant 		simple_unlock(&global_v_numoutput_slock);
    358      1.178  perseant 		splx(s);
    359      1.178  perseant 
    360      1.178  perseant 		KASSERT(LIST_FIRST(&vp->v_dirtyblkhd) == NULL);
    361      1.178  perseant 		KASSERT(vp->v_numoutput == 0);
    362      1.178  perseant 
    363       1.15  perseant 		return error;
    364       1.15  perseant 	}
    365        1.1   mycroft 	sp = fs->lfs_sp;
    366        1.1   mycroft 
    367      1.101  perseant 	flushed = 0;
    368      1.101  perseant 	if (VPISEMPTY(vp)) {
    369        1.1   mycroft 		lfs_writevnodes(fs, vp->v_mount, sp, VN_EMPTY);
    370      1.101  perseant 		++flushed;
    371       1.73       chs 	} else if ((ip->i_flag & IN_CLEANING) &&
    372       1.58  perseant 		  (fs->lfs_sp->seg_flags & SEGM_CLEAN)) {
    373       1.19  perseant 		ivndebug(vp,"vflush/clean");
    374       1.19  perseant 		lfs_writevnodes(fs, vp->v_mount, sp, VN_CLEAN);
    375      1.101  perseant 		++flushed;
    376       1.74  perseant 	} else if (lfs_dostats) {
    377      1.101  perseant 		if (!VPISEMPTY(vp) || (VTOI(vp)->i_flag & IN_ALLMOD))
    378       1.15  perseant 			++lfs_stats.vflush_invoked;
    379       1.19  perseant 		ivndebug(vp,"vflush");
    380       1.15  perseant 	}
    381       1.15  perseant 
    382       1.19  perseant #ifdef DIAGNOSTIC
    383       1.73       chs 	if (vp->v_flag & VDIROP) {
    384      1.158  perseant 		DLOG((DLOG_VNODE, "lfs_vflush: flushing VDIROP\n"));
    385      1.158  perseant 		/* panic("lfs_vflush: VDIROP being flushed...this can\'t happen"); */
    386       1.19  perseant 	}
    387       1.73       chs 	if (vp->v_usecount < 0) {
    388       1.69  perseant 		printf("usecount=%ld\n", (long)vp->v_usecount);
    389       1.19  perseant 		panic("lfs_vflush: usecount<0");
    390       1.19  perseant 	}
    391       1.15  perseant #endif
    392        1.1   mycroft 
    393        1.1   mycroft 	do {
    394        1.1   mycroft 		do {
    395      1.171  perseant 			if (LIST_FIRST(&vp->v_dirtyblkhd) != NULL) {
    396      1.171  perseant 				relock = lfs_writefile(fs, sp, vp);
    397      1.171  perseant 				if (relock) {
    398      1.171  perseant 					/*
    399      1.171  perseant 					 * Might have to wait for the
    400      1.171  perseant 					 * cleaner to run; but we're
    401      1.171  perseant 					 * still not done with this vnode.
    402      1.171  perseant 					 */
    403      1.176  perseant 					lfs_writeinode(fs, sp, ip);
    404      1.176  perseant 					LFS_SET_UINO(ip, IN_MODIFIED);
    405      1.171  perseant 					lfs_writeseg(fs, sp);
    406      1.171  perseant 					lfs_segunlock(fs);
    407      1.171  perseant 					lfs_segunlock_relock(fs);
    408      1.171  perseant 					goto top;
    409      1.171  perseant 				}
    410      1.171  perseant 			}
    411      1.174  perseant 			/*
    412      1.174  perseant 			 * If we begin a new segment in the middle of writing
    413      1.174  perseant 			 * the Ifile, it creates an inconsistent checkpoint,
    414      1.174  perseant 			 * since the Ifile information for the new segment
    415      1.174  perseant 			 * is not up-to-date.  Take care of this here by
    416      1.174  perseant 			 * sending the Ifile through again in case there
    417      1.174  perseant 			 * are newly dirtied blocks.  But wait, there's more!
    418      1.174  perseant 			 * This second Ifile write could *also* cross a segment
    419      1.174  perseant 			 * boundary, if the first one was large.  The second
    420      1.174  perseant 			 * one is guaranteed to be no more than 8 blocks,
    421      1.174  perseant 			 * though (two segment blocks and supporting indirects)
    422      1.174  perseant 			 * so the third write *will not* cross the boundary.
    423      1.174  perseant 			 */
    424      1.174  perseant 			if (vp == fs->lfs_ivnode) {
    425      1.174  perseant 				lfs_writefile(fs, sp, vp);
    426      1.174  perseant 				lfs_writefile(fs, sp, vp);
    427      1.174  perseant 			}
    428        1.1   mycroft 		} while (lfs_writeinode(fs, sp, ip));
    429        1.1   mycroft 	} while (lfs_writeseg(fs, sp) && ip->i_number == LFS_IFILE_INUM);
    430      1.171  perseant 
    431       1.73       chs 	if (lfs_dostats) {
    432       1.15  perseant 		++lfs_stats.nwrites;
    433       1.15  perseant 		if (sp->seg_flags & SEGM_SYNC)
    434       1.15  perseant 			++lfs_stats.nsync_writes;
    435       1.15  perseant 		if (sp->seg_flags & SEGM_CKP)
    436       1.15  perseant 			++lfs_stats.ncheckpoints;
    437       1.15  perseant 	}
    438       1.74  perseant 	/*
    439       1.74  perseant 	 * If we were called from somewhere that has already held the seglock
    440       1.74  perseant 	 * (e.g., lfs_markv()), the lfs_segunlock will not wait for
    441       1.74  perseant 	 * the write to complete because we are still locked.
    442       1.74  perseant 	 * Since lfs_vflush() must return the vnode with no dirty buffers,
    443       1.74  perseant 	 * we must explicitly wait, if that is the case.
    444       1.74  perseant 	 *
    445       1.74  perseant 	 * We compare the iocount against 1, not 0, because it is
    446       1.74  perseant 	 * artificially incremented by lfs_seglock().
    447       1.74  perseant 	 */
    448      1.111  perseant 	simple_lock(&fs->lfs_interlock);
    449       1.74  perseant 	if (fs->lfs_seglock > 1) {
    450       1.74  perseant 		while (fs->lfs_iocount > 1)
    451      1.159  perseant 			(void)ltsleep(&fs->lfs_iocount, PRIBIO + 1,
    452      1.159  perseant 				     "lfs_vflush", 0, &fs->lfs_interlock);
    453      1.159  perseant 	}
    454      1.159  perseant 	simple_unlock(&fs->lfs_interlock);
    455      1.111  perseant 
    456       1.15  perseant 	lfs_segunlock(fs);
    457        1.1   mycroft 
    458      1.120  perseant 	/* Wait for these buffers to be recovered by aiodoned */
    459      1.120  perseant 	s = splbio();
    460      1.120  perseant 	simple_lock(&global_v_numoutput_slock);
    461      1.120  perseant 	while (vp->v_numoutput > 0) {
    462      1.146      yamt 		vp->v_flag |= VBWAIT;
    463      1.120  perseant 		ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vf2", 0,
    464      1.120  perseant 			&global_v_numoutput_slock);
    465      1.120  perseant 	}
    466      1.120  perseant 	simple_unlock(&global_v_numoutput_slock);
    467      1.120  perseant 	splx(s);
    468      1.120  perseant 
    469      1.178  perseant 	KASSERT(LIST_FIRST(&vp->v_dirtyblkhd) == NULL);
    470      1.178  perseant 	KASSERT(vp->v_numoutput == 0);
    471      1.178  perseant 
    472      1.173  perseant 	fs->lfs_flushvp = NULL;
    473      1.173  perseant 	KASSERT(fs->lfs_flushvp_fakevref == 0);
    474      1.173  perseant 
    475        1.1   mycroft 	return (0);
    476        1.1   mycroft }
    477        1.1   mycroft 
    478       1.15  perseant int
    479       1.69  perseant lfs_writevnodes(struct lfs *fs, struct mount *mp, struct segment *sp, int op)
    480        1.1   mycroft {
    481        1.1   mycroft 	struct inode *ip;
    482       1.77  perseant 	struct vnode *vp, *nvp;
    483       1.73       chs 	int inodes_written = 0, only_cleaning;
    484      1.171  perseant 	int error = 0;
    485        1.1   mycroft 
    486      1.159  perseant 	ASSERT_SEGLOCK(fs);
    487       1.15  perseant #ifndef LFS_NO_BACKVP_HACK
    488       1.15  perseant 	/* BEGIN HACK */
    489      1.128      yamt #define	VN_OFFSET	\
    490      1.128      yamt 	(((caddr_t)&LIST_NEXT(vp, v_mntvnodes)) - (caddr_t)vp)
    491      1.128      yamt #define	BACK_VP(VP)	\
    492      1.128      yamt 	((struct vnode *)(((caddr_t)(VP)->v_mntvnodes.le_prev) - VN_OFFSET))
    493      1.128      yamt #define	BEG_OF_VLIST	\
    494      1.128      yamt 	((struct vnode *)(((caddr_t)&LIST_FIRST(&mp->mnt_vnodelist)) \
    495      1.128      yamt 	- VN_OFFSET))
    496      1.157     perry 
    497       1.15  perseant 	/* Find last vnode. */
    498       1.75  perseant  loop:	for (vp = LIST_FIRST(&mp->mnt_vnodelist);
    499       1.75  perseant 	     vp && LIST_NEXT(vp, v_mntvnodes) != NULL;
    500       1.75  perseant 	     vp = LIST_NEXT(vp, v_mntvnodes));
    501       1.77  perseant 	for (; vp && vp != BEG_OF_VLIST; vp = nvp) {
    502       1.77  perseant 		nvp = BACK_VP(vp);
    503       1.15  perseant #else
    504       1.15  perseant 	loop:
    505       1.77  perseant 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
    506       1.77  perseant 		nvp = LIST_NEXT(vp, v_mntvnodes);
    507       1.15  perseant #endif
    508        1.1   mycroft 		/*
    509        1.1   mycroft 		 * If the vnode that we are about to sync is no longer
    510        1.1   mycroft 		 * associated with this mount point, start over.
    511        1.1   mycroft 		 */
    512       1.58  perseant 		if (vp->v_mount != mp) {
    513      1.158  perseant 			DLOG((DLOG_VNODE, "lfs_writevnodes: starting over\n"));
    514      1.113      yamt 			/*
    515      1.113      yamt 			 * After this, pages might be busy
    516      1.113      yamt 			 * due to our own previous putpages.
    517      1.113      yamt 			 * Start actual segment write here to avoid deadlock.
    518      1.113      yamt 			 */
    519      1.113      yamt 			(void)lfs_writeseg(fs, sp);
    520        1.1   mycroft 			goto loop;
    521       1.58  perseant 		}
    522      1.157     perry 
    523       1.85      yamt 		if (vp->v_type == VNON) {
    524       1.85      yamt 			continue;
    525       1.85      yamt 		}
    526        1.1   mycroft 
    527       1.15  perseant 		ip = VTOI(vp);
    528       1.15  perseant 		if ((op == VN_DIROP && !(vp->v_flag & VDIROP)) ||
    529      1.128      yamt 		    (op != VN_DIROP && op != VN_CLEAN &&
    530      1.128      yamt 		    (vp->v_flag & VDIROP))) {
    531       1.15  perseant 			vndebug(vp,"dirop");
    532       1.15  perseant 			continue;
    533       1.15  perseant 		}
    534      1.157     perry 
    535      1.101  perseant 		if (op == VN_EMPTY && !VPISEMPTY(vp)) {
    536       1.15  perseant 			vndebug(vp,"empty");
    537       1.15  perseant 			continue;
    538       1.15  perseant 		}
    539        1.1   mycroft 
    540       1.73       chs 		if (op == VN_CLEAN && ip->i_number != LFS_IFILE_INUM
    541       1.38  perseant 		   && vp != fs->lfs_flushvp
    542       1.15  perseant 		   && !(ip->i_flag & IN_CLEANING)) {
    543       1.15  perseant 			vndebug(vp,"cleaning");
    544        1.1   mycroft 			continue;
    545       1.15  perseant 		}
    546        1.1   mycroft 
    547       1.15  perseant 		if (lfs_vref(vp)) {
    548       1.15  perseant 			vndebug(vp,"vref");
    549        1.1   mycroft 			continue;
    550       1.15  perseant 		}
    551        1.1   mycroft 
    552       1.23  perseant 		only_cleaning = 0;
    553        1.1   mycroft 		/*
    554       1.55  perseant 		 * Write the inode/file if dirty and it's not the IFILE.
    555        1.1   mycroft 		 */
    556      1.101  perseant 		if ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp)) {
    557      1.128      yamt 			only_cleaning =
    558      1.128      yamt 			    ((ip->i_flag & IN_ALLMOD) == IN_CLEANING);
    559       1.20  perseant 
    560      1.159  perseant 			if (ip->i_number != LFS_IFILE_INUM) {
    561      1.171  perseant 				error = lfs_writefile(fs, sp, vp);
    562      1.171  perseant 				if (error) {
    563      1.171  perseant 					lfs_vunref(vp);
    564      1.171  perseant 					if (error == EAGAIN) {
    565      1.171  perseant 						/*
    566      1.171  perseant 						 * This error from lfs_putpages
    567      1.171  perseant 						 * indicates we need to drop
    568      1.171  perseant 						 * the segment lock and start
    569      1.171  perseant 						 * over after the cleaner has
    570      1.171  perseant 						 * had a chance to run.
    571      1.171  perseant 						 */
    572      1.176  perseant 						lfs_writeinode(fs, sp, ip);
    573      1.171  perseant 						lfs_writeseg(fs, sp);
    574      1.172  perseant 						if (!VPISEMPTY(vp) &&
    575      1.172  perseant 						    !WRITEINPROG(vp) &&
    576      1.172  perseant 						    !(ip->i_flag & IN_ALLMOD))
    577      1.172  perseant 							LFS_SET_UINO(ip, IN_MODIFIED);
    578      1.171  perseant 						break;
    579      1.171  perseant 					}
    580      1.171  perseant 					error = 0; /* XXX not quite right */
    581      1.171  perseant 					continue;
    582      1.171  perseant 				}
    583      1.171  perseant 
    584      1.159  perseant 				if (!VPISEMPTY(vp)) {
    585      1.159  perseant 					if (WRITEINPROG(vp)) {
    586      1.159  perseant 						ivndebug(vp,"writevnodes/write2");
    587      1.159  perseant 					} else if (!(ip->i_flag & IN_ALLMOD)) {
    588      1.159  perseant 						LFS_SET_UINO(ip, IN_MODIFIED);
    589      1.159  perseant 					}
    590       1.15  perseant 				}
    591      1.159  perseant 				(void) lfs_writeinode(fs, sp, ip);
    592      1.159  perseant 				inodes_written++;
    593       1.15  perseant 			}
    594       1.15  perseant 		}
    595       1.43  perseant 
    596       1.52  perseant 		if (lfs_clean_vnhead && only_cleaning)
    597       1.20  perseant 			lfs_vunref_head(vp);
    598       1.20  perseant 		else
    599       1.20  perseant 			lfs_vunref(vp);
    600        1.1   mycroft 	}
    601      1.171  perseant 	return error;
    602        1.1   mycroft }
    603        1.1   mycroft 
    604       1.69  perseant /*
    605       1.69  perseant  * Do a checkpoint.
    606       1.69  perseant  */
    607        1.1   mycroft int
    608       1.69  perseant lfs_segwrite(struct mount *mp, int flags)
    609        1.1   mycroft {
    610        1.1   mycroft 	struct buf *bp;
    611        1.1   mycroft 	struct inode *ip;
    612        1.1   mycroft 	struct lfs *fs;
    613        1.1   mycroft 	struct segment *sp;
    614        1.1   mycroft 	struct vnode *vp;
    615        1.1   mycroft 	SEGUSE *segusep;
    616      1.117      fvdl 	int do_ckp, did_ckp, error, s;
    617      1.117      fvdl 	unsigned n, segleft, maxseg, sn, i, curseg;
    618       1.15  perseant 	int writer_set = 0;
    619       1.61  perseant 	int dirty;
    620       1.74  perseant 	int redo;
    621      1.171  perseant 	int um_error;
    622      1.157     perry 
    623        1.1   mycroft 	fs = VFSTOUFS(mp)->um_lfs;
    624      1.159  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    625        1.1   mycroft 
    626       1.53  perseant 	if (fs->lfs_ronly)
    627       1.53  perseant 		return EROFS;
    628       1.53  perseant 
    629       1.15  perseant 	lfs_imtime(fs);
    630       1.58  perseant 
    631        1.1   mycroft 	/*
    632        1.1   mycroft 	 * Allocate a segment structure and enough space to hold pointers to
    633        1.1   mycroft 	 * the maximum possible number of buffers which can be described in a
    634        1.1   mycroft 	 * single summary block.
    635        1.1   mycroft 	 */
    636      1.171  perseant 	do_ckp = LFS_SHOULD_CHECKPOINT(fs, flags);
    637      1.171  perseant 
    638        1.1   mycroft 	lfs_seglock(fs, flags | (do_ckp ? SEGM_CKP : 0));
    639        1.1   mycroft 	sp = fs->lfs_sp;
    640        1.1   mycroft 
    641       1.15  perseant 	/*
    642       1.16  perseant 	 * If lfs_flushvp is non-NULL, we are called from lfs_vflush,
    643       1.16  perseant 	 * in which case we have to flush *all* buffers off of this vnode.
    644       1.37  perseant 	 * We don't care about other nodes, but write any non-dirop nodes
    645       1.37  perseant 	 * anyway in anticipation of another getnewvnode().
    646       1.37  perseant 	 *
    647       1.37  perseant 	 * If we're cleaning we only write cleaning and ifile blocks, and
    648       1.37  perseant 	 * no dirops, since otherwise we'd risk corruption in a crash.
    649       1.15  perseant 	 */
    650       1.73       chs 	if (sp->seg_flags & SEGM_CLEAN)
    651       1.15  perseant 		lfs_writevnodes(fs, mp, sp, VN_CLEAN);
    652      1.105  perseant 	else if (!(sp->seg_flags & SEGM_FORCE_CKP)) {
    653      1.171  perseant 		do {
    654      1.171  perseant 			um_error = lfs_writevnodes(fs, mp, sp, VN_REG);
    655      1.171  perseant 			if (!fs->lfs_dirops || !fs->lfs_flushvp) {
    656      1.171  perseant 				if (!writer_set) {
    657      1.171  perseant 					lfs_writer_enter(fs, "lfs writer");
    658      1.171  perseant 					writer_set = 1;
    659      1.171  perseant 				}
    660      1.171  perseant 				error = lfs_writevnodes(fs, mp, sp, VN_DIROP);
    661      1.171  perseant 				if (um_error == 0)
    662      1.171  perseant 					um_error = error;
    663      1.177  perseant 				/* In case writevnodes errored out */
    664      1.177  perseant 				lfs_flush_dirops(fs);
    665      1.171  perseant 				((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT);
    666      1.176  perseant 				lfs_finalize_fs_seguse(fs);
    667      1.171  perseant 			}
    668      1.171  perseant 			if (do_ckp && um_error) {
    669      1.171  perseant 				lfs_segunlock_relock(fs);
    670      1.171  perseant 				sp = fs->lfs_sp;
    671      1.171  perseant 			}
    672      1.171  perseant 		} while (do_ckp && um_error != 0);
    673      1.157     perry 	}
    674        1.1   mycroft 
    675        1.1   mycroft 	/*
    676        1.1   mycroft 	 * If we are doing a checkpoint, mark everything since the
    677        1.1   mycroft 	 * last checkpoint as no longer ACTIVE.
    678        1.1   mycroft 	 */
    679       1.15  perseant 	if (do_ckp) {
    680      1.117      fvdl 		segleft = fs->lfs_nseg;
    681      1.117      fvdl 		curseg = 0;
    682      1.117      fvdl 		for (n = 0; n < fs->lfs_segtabsz; n++) {
    683       1.61  perseant 			dirty = 0;
    684      1.157     perry 			if (bread(fs->lfs_ivnode,
    685      1.117      fvdl 			    fs->lfs_cleansz + n, fs->lfs_bsize, NOCRED, &bp))
    686       1.15  perseant 				panic("lfs_segwrite: ifile read");
    687        1.1   mycroft 			segusep = (SEGUSE *)bp->b_data;
    688      1.117      fvdl 			maxseg = min(segleft, fs->lfs_sepb);
    689      1.117      fvdl 			for (i = 0; i < maxseg; i++) {
    690      1.117      fvdl 				sn = curseg + i;
    691      1.134      yamt 				if (sn != dtosn(fs, fs->lfs_curseg) &&
    692      1.117      fvdl 				    segusep->su_flags & SEGUSE_ACTIVE) {
    693       1.61  perseant 					segusep->su_flags &= ~SEGUSE_ACTIVE;
    694      1.105  perseant 					--fs->lfs_nactive;
    695       1.61  perseant 					++dirty;
    696       1.61  perseant 				}
    697      1.105  perseant 				fs->lfs_suflags[fs->lfs_activesb][sn] =
    698      1.105  perseant 					segusep->su_flags;
    699       1.69  perseant 				if (fs->lfs_version > 1)
    700       1.69  perseant 					++segusep;
    701       1.69  perseant 				else
    702       1.69  perseant 					segusep = (SEGUSE *)
    703       1.69  perseant 						((SEGUSE_V1 *)segusep + 1);
    704       1.61  perseant 			}
    705      1.157     perry 
    706       1.61  perseant 			if (dirty)
    707       1.74  perseant 				error = LFS_BWRITE_LOG(bp); /* Ifile */
    708       1.61  perseant 			else
    709       1.61  perseant 				brelse(bp);
    710      1.117      fvdl 			segleft -= fs->lfs_sepb;
    711      1.117      fvdl 			curseg += fs->lfs_sepb;
    712        1.1   mycroft 		}
    713       1.15  perseant 	}
    714       1.61  perseant 
    715      1.159  perseant 	LOCK_ASSERT(LFS_SEGLOCK_HELD(fs));
    716      1.159  perseant 
    717       1.61  perseant 	did_ckp = 0;
    718        1.1   mycroft 	if (do_ckp || fs->lfs_doifile) {
    719      1.159  perseant 		vp = fs->lfs_ivnode;
    720      1.159  perseant 		vn_lock(vp, LK_EXCLUSIVE);
    721       1.63  perseant 		do {
    722       1.74  perseant #ifdef DEBUG
    723      1.159  perseant 			LFS_ENTER_LOG("pretend", __FILE__, __LINE__, 0, 0, curproc->p_pid);
    724       1.74  perseant #endif
    725      1.159  perseant 			simple_lock(&fs->lfs_interlock);
    726       1.74  perseant 			fs->lfs_flags &= ~LFS_IFDIRTY;
    727      1.159  perseant 			simple_unlock(&fs->lfs_interlock);
    728       1.55  perseant 
    729       1.63  perseant 			ip = VTOI(vp);
    730      1.101  perseant 
    731      1.171  perseant 			if (LIST_FIRST(&vp->v_dirtyblkhd) != NULL) {
    732      1.171  perseant 				/*
    733      1.171  perseant 				 * Ifile has no pages, so we don't need
    734      1.171  perseant 				 * to check error return here.
    735      1.171  perseant 				 */
    736       1.63  perseant 				lfs_writefile(fs, sp, vp);
    737      1.174  perseant 				/*
    738      1.174  perseant 				 * Ensure the Ifile takes the current segment
    739      1.174  perseant 				 * into account.  See comment in lfs_vflush.
    740      1.174  perseant 				 */
    741      1.174  perseant 				lfs_writefile(fs, sp, vp);
    742      1.174  perseant 				lfs_writefile(fs, sp, vp);
    743      1.171  perseant 			}
    744      1.101  perseant 
    745       1.63  perseant 			if (ip->i_flag & IN_ALLMOD)
    746       1.63  perseant 				++did_ckp;
    747       1.74  perseant 			redo = lfs_writeinode(fs, sp, ip);
    748      1.101  perseant 			redo += lfs_writeseg(fs, sp);
    749      1.159  perseant 			simple_lock(&fs->lfs_interlock);
    750       1.74  perseant 			redo += (fs->lfs_flags & LFS_IFDIRTY);
    751      1.159  perseant 			simple_unlock(&fs->lfs_interlock);
    752      1.112  perseant 		} while (redo && do_ckp);
    753      1.112  perseant 
    754      1.112  perseant 		/*
    755      1.112  perseant 		 * Unless we are unmounting, the Ifile may continue to have
    756      1.112  perseant 		 * dirty blocks even after a checkpoint, due to changes to
    757      1.112  perseant 		 * inodes' atime.  If we're checkpointing, it's "impossible"
    758      1.112  perseant 		 * for other parts of the Ifile to be dirty after the loop
    759      1.112  perseant 		 * above, since we hold the segment lock.
    760      1.112  perseant 		 */
    761      1.114  perseant 		s = splbio();
    762      1.112  perseant 		if (LIST_EMPTY(&vp->v_dirtyblkhd)) {
    763      1.112  perseant 			LFS_CLR_UINO(ip, IN_ALLMOD);
    764      1.112  perseant 		}
    765      1.112  perseant #ifdef DIAGNOSTIC
    766      1.112  perseant 		else if (do_ckp) {
    767      1.159  perseant 			int do_panic = 0;
    768      1.112  perseant 			LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
    769      1.112  perseant 				if (bp->b_lblkno < fs->lfs_cleansz +
    770      1.112  perseant 				    fs->lfs_segtabsz &&
    771      1.112  perseant 				    !(bp->b_flags & B_GATHERED)) {
    772      1.159  perseant 					printf("ifile lbn %ld still dirty (flags %lx)\n",
    773      1.159  perseant 						(long)bp->b_lblkno,
    774      1.159  perseant 						(long)bp->b_flags);
    775      1.159  perseant 					++do_panic;
    776       1.74  perseant 				}
    777       1.74  perseant 			}
    778      1.159  perseant 			if (do_panic)
    779      1.159  perseant 				panic("dirty blocks");
    780       1.74  perseant 		}
    781      1.112  perseant #endif
    782      1.114  perseant 		splx(s);
    783      1.159  perseant 		VOP_UNLOCK(vp, 0);
    784       1.15  perseant 	} else {
    785        1.1   mycroft 		(void) lfs_writeseg(fs, sp);
    786       1.15  perseant 	}
    787      1.157     perry 
    788      1.112  perseant 	/* Note Ifile no longer needs to be written */
    789      1.112  perseant 	fs->lfs_doifile = 0;
    790      1.125      yamt 	if (writer_set)
    791      1.125      yamt 		lfs_writer_leave(fs);
    792       1.61  perseant 
    793       1.61  perseant 	/*
    794       1.61  perseant 	 * If we didn't write the Ifile, we didn't really do anything.
    795       1.61  perseant 	 * That means that (1) there is a checkpoint on disk and (2)
    796       1.61  perseant 	 * nothing has changed since it was written.
    797       1.61  perseant 	 *
    798       1.61  perseant 	 * Take the flags off of the segment so that lfs_segunlock
    799       1.61  perseant 	 * doesn't have to write the superblock either.
    800       1.61  perseant 	 */
    801       1.79  perseant 	if (do_ckp && !did_ckp) {
    802       1.79  perseant 		sp->seg_flags &= ~SEGM_CKP;
    803       1.61  perseant 	}
    804       1.61  perseant 
    805       1.73       chs 	if (lfs_dostats) {
    806       1.15  perseant 		++lfs_stats.nwrites;
    807       1.15  perseant 		if (sp->seg_flags & SEGM_SYNC)
    808       1.15  perseant 			++lfs_stats.nsync_writes;
    809       1.15  perseant 		if (sp->seg_flags & SEGM_CKP)
    810       1.15  perseant 			++lfs_stats.ncheckpoints;
    811       1.15  perseant 	}
    812        1.1   mycroft 	lfs_segunlock(fs);
    813        1.1   mycroft 	return (0);
    814        1.1   mycroft }
    815        1.1   mycroft 
    816        1.1   mycroft /*
    817        1.1   mycroft  * Write the dirty blocks associated with a vnode.
    818        1.1   mycroft  */
    819      1.171  perseant int
    820       1.69  perseant lfs_writefile(struct lfs *fs, struct segment *sp, struct vnode *vp)
    821        1.1   mycroft {
    822        1.1   mycroft 	struct finfo *fip;
    823       1.80  perseant 	struct inode *ip;
    824  1.180.2.1      chap 	int i, frag;
    825      1.171  perseant 	int error;
    826      1.157     perry 
    827      1.159  perseant 	ASSERT_SEGLOCK(fs);
    828      1.171  perseant 	error = 0;
    829       1.80  perseant 	ip = VTOI(vp);
    830       1.80  perseant 
    831      1.180  perseant 	fip = sp->fip;
    832  1.180.2.1      chap 	lfs_acquire_finfo(fs, ip->i_number, ip->i_gen);
    833        1.1   mycroft 
    834       1.73       chs 	if (vp->v_flag & VDIROP)
    835       1.15  perseant 		((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT);
    836      1.157     perry 
    837       1.74  perseant 	if (sp->seg_flags & SEGM_CLEAN) {
    838       1.38  perseant 		lfs_gather(fs, sp, vp, lfs_match_fake);
    839       1.38  perseant 		/*
    840       1.38  perseant 		 * For a file being flushed, we need to write *all* blocks.
    841       1.38  perseant 		 * This means writing the cleaning blocks first, and then
    842       1.38  perseant 		 * immediately following with any non-cleaning blocks.
    843       1.38  perseant 		 * The same is true of the Ifile since checkpoints assume
    844       1.38  perseant 		 * that all valid Ifile blocks are written.
    845       1.38  perseant 		 */
    846      1.178  perseant 		if (IS_FLUSHING(fs, vp) || vp == fs->lfs_ivnode) {
    847       1.38  perseant 			lfs_gather(fs, sp, vp, lfs_match_data);
    848      1.101  perseant 			/*
    849      1.101  perseant 			 * Don't call VOP_PUTPAGES: if we're flushing,
    850      1.101  perseant 			 * we've already done it, and the Ifile doesn't
    851      1.101  perseant 			 * use the page cache.
    852      1.101  perseant 			 */
    853      1.101  perseant 		}
    854      1.101  perseant 	} else {
    855       1.38  perseant 		lfs_gather(fs, sp, vp, lfs_match_data);
    856      1.101  perseant 		/*
    857      1.101  perseant 		 * If we're flushing, we've already called VOP_PUTPAGES
    858      1.101  perseant 		 * so don't do it again.  Otherwise, we want to write
    859      1.101  perseant 		 * everything we've got.
    860      1.101  perseant 		 */
    861      1.101  perseant 		if (!IS_FLUSHING(fs, vp)) {
    862      1.102      yamt 			simple_lock(&vp->v_interlock);
    863      1.171  perseant 			error = VOP_PUTPAGES(vp, 0, 0,
    864      1.171  perseant 				PGO_CLEANIT | PGO_ALLPAGES | PGO_LOCKED);
    865      1.101  perseant 		}
    866      1.101  perseant 	}
    867       1.38  perseant 
    868        1.1   mycroft 	/*
    869        1.1   mycroft 	 * It may not be necessary to write the meta-data blocks at this point,
    870        1.1   mycroft 	 * as the roll-forward recovery code should be able to reconstruct the
    871        1.1   mycroft 	 * list.
    872       1.15  perseant 	 *
    873       1.15  perseant 	 * We have to write them anyway, though, under two conditions: (1) the
    874       1.15  perseant 	 * vnode is being flushed (for reuse by vinvalbuf); or (2) we are
    875       1.15  perseant 	 * checkpointing.
    876       1.80  perseant 	 *
    877       1.80  perseant 	 * BUT if we are cleaning, we might have indirect blocks that refer to
    878       1.80  perseant 	 * new blocks not being written yet, in addition to fragments being
    879       1.80  perseant 	 * moved out of a cleaned segment.  If that is the case, don't
    880       1.80  perseant 	 * write the indirect blocks, or the finfo will have a small block
    881       1.80  perseant 	 * in the middle of it!
    882       1.80  perseant 	 * XXX in this case isn't the inode size wrong too?
    883        1.1   mycroft 	 */
    884       1.80  perseant 	frag = 0;
    885       1.80  perseant 	if (sp->seg_flags & SEGM_CLEAN) {
    886       1.80  perseant 		for (i = 0; i < NDADDR; i++)
    887       1.80  perseant 			if (ip->i_lfs_fragsize[i] > 0 &&
    888       1.80  perseant 			    ip->i_lfs_fragsize[i] < fs->lfs_bsize)
    889       1.80  perseant 				++frag;
    890       1.80  perseant 	}
    891       1.80  perseant #ifdef DIAGNOSTIC
    892       1.80  perseant 	if (frag > 1)
    893       1.80  perseant 		panic("lfs_writefile: more than one fragment!");
    894       1.80  perseant #endif
    895       1.80  perseant 	if (IS_FLUSHING(fs, vp) ||
    896       1.80  perseant 	    (frag == 0 && (lfs_writeindir || (sp->seg_flags & SEGM_CKP)))) {
    897       1.15  perseant 		lfs_gather(fs, sp, vp, lfs_match_indir);
    898       1.15  perseant 		lfs_gather(fs, sp, vp, lfs_match_dindir);
    899       1.15  perseant 		lfs_gather(fs, sp, vp, lfs_match_tindir);
    900       1.15  perseant 	}
    901        1.1   mycroft 	fip = sp->fip;
    902      1.180  perseant 	lfs_release_finfo(fs);
    903      1.171  perseant 
    904      1.171  perseant 	return error;
    905        1.1   mycroft }
    906        1.1   mycroft 
    907        1.1   mycroft int
    908       1.69  perseant lfs_writeinode(struct lfs *fs, struct segment *sp, struct inode *ip)
    909        1.1   mycroft {
    910        1.1   mycroft 	struct buf *bp, *ibp;
    911      1.119      fvdl 	struct ufs1_dinode *cdp;
    912        1.1   mycroft 	IFILE *ifp;
    913        1.1   mycroft 	SEGUSE *sup;
    914       1.91      fvdl 	daddr_t daddr;
    915       1.91      fvdl 	int32_t *daddrp;	/* XXX ondisk32 */
    916        1.1   mycroft 	ino_t ino;
    917       1.69  perseant 	int error, i, ndx, fsb = 0;
    918        1.1   mycroft 	int redo_ifile = 0;
    919       1.69  perseant 	int gotblk = 0;
    920      1.157     perry 
    921      1.159  perseant 	ASSERT_SEGLOCK(fs);
    922       1.47  perseant 	if (!(ip->i_flag & IN_ALLMOD))
    923       1.73       chs 		return (0);
    924      1.157     perry 
    925        1.1   mycroft 	/* Allocate a new inode block if necessary. */
    926      1.128      yamt 	if ((ip->i_number != LFS_IFILE_INUM || sp->idp == NULL) &&
    927      1.128      yamt 	    sp->ibp == NULL) {
    928        1.1   mycroft 		/* Allocate a new segment if necessary. */
    929       1.69  perseant 		if (sp->seg_bytes_left < fs->lfs_ibsize ||
    930       1.98      yamt 		    sp->sum_bytes_left < sizeof(int32_t))
    931        1.1   mycroft 			(void) lfs_writeseg(fs, sp);
    932        1.1   mycroft 
    933        1.1   mycroft 		/* Get next inode block. */
    934        1.1   mycroft 		daddr = fs->lfs_offset;
    935       1.69  perseant 		fs->lfs_offset += btofsb(fs, fs->lfs_ibsize);
    936        1.1   mycroft 		sp->ibp = *sp->cbpp++ =
    937      1.128      yamt 			getblk(VTOI(fs->lfs_ivnode)->i_devvp,
    938      1.128      yamt 			    fsbtodb(fs, daddr), fs->lfs_ibsize, 0, 0);
    939       1.24  perseant 		gotblk++;
    940       1.24  perseant 
    941        1.1   mycroft 		/* Zero out inode numbers */
    942        1.1   mycroft 		for (i = 0; i < INOPB(fs); ++i)
    943      1.128      yamt 			((struct ufs1_dinode *)sp->ibp->b_data)[i].di_inumber =
    944      1.128      yamt 			    0;
    945       1.15  perseant 
    946        1.1   mycroft 		++sp->start_bpp;
    947       1.69  perseant 		fs->lfs_avail -= btofsb(fs, fs->lfs_ibsize);
    948        1.1   mycroft 		/* Set remaining space counters. */
    949       1.69  perseant 		sp->seg_bytes_left -= fs->lfs_ibsize;
    950       1.98      yamt 		sp->sum_bytes_left -= sizeof(int32_t);
    951       1.98      yamt 		ndx = fs->lfs_sumsize / sizeof(int32_t) -
    952       1.15  perseant 			sp->ninodes / INOPB(fs) - 1;
    953       1.98      yamt 		((int32_t *)(sp->segsum))[ndx] = daddr;
    954        1.1   mycroft 	}
    955       1.27  perseant 
    956        1.1   mycroft 	/* Update the inode times and copy the inode onto the inode page. */
    957       1.74  perseant 	/* XXX kludge --- don't redirty the ifile just to put times on it */
    958       1.74  perseant 	if (ip->i_number != LFS_IFILE_INUM)
    959      1.166  christos 		LFS_ITIMES(ip, NULL, NULL, NULL);
    960       1.16  perseant 
    961       1.27  perseant 	/*
    962       1.27  perseant 	 * If this is the Ifile, and we've already written the Ifile in this
    963       1.27  perseant 	 * partial segment, just overwrite it (it's not on disk yet) and
    964       1.27  perseant 	 * continue.
    965       1.27  perseant 	 *
    966       1.27  perseant 	 * XXX we know that the bp that we get the second time around has
    967       1.27  perseant 	 * already been gathered.
    968       1.27  perseant 	 */
    969       1.73       chs 	if (ip->i_number == LFS_IFILE_INUM && sp->idp) {
    970      1.119      fvdl 		*(sp->idp) = *ip->i_din.ffs1_din;
    971      1.119      fvdl 		ip->i_lfs_osize = ip->i_size;
    972       1.27  perseant 		return 0;
    973       1.27  perseant 	}
    974       1.27  perseant 
    975        1.1   mycroft 	bp = sp->ibp;
    976      1.119      fvdl 	cdp = ((struct ufs1_dinode *)bp->b_data) + (sp->ninodes % INOPB(fs));
    977      1.119      fvdl 	*cdp = *ip->i_din.ffs1_din;
    978       1.53  perseant 
    979      1.176  perseant 	/* We can finish the segment accounting for truncations now */
    980      1.176  perseant 	lfs_finalize_ino_seguse(fs, ip);
    981      1.176  perseant 
    982       1.53  perseant 	/*
    983       1.53  perseant 	 * If we are cleaning, ensure that we don't write UNWRITTEN disk
    984      1.160  perseant 	 * addresses to disk; possibly change the on-disk record of
    985      1.160  perseant 	 * the inode size, either by reverting to the previous size
    986      1.160  perseant 	 * (in the case of cleaning) or by verifying the inode's block
    987      1.160  perseant 	 * holdings (in the case of files being allocated as they are being
    988      1.160  perseant 	 * written).
    989      1.160  perseant 	 * XXX By not writing UNWRITTEN blocks, we are making the lfs_avail
    990      1.103  perseant 	 * XXX count on disk wrong by the same amount.	We should be
    991      1.101  perseant 	 * XXX able to "borrow" from lfs_avail and return it after the
    992      1.101  perseant 	 * XXX Ifile is written.  See also in lfs_writeseg.
    993       1.53  perseant 	 */
    994      1.160  perseant 
    995      1.160  perseant 	/* Check file size based on highest allocated block */
    996      1.160  perseant 	if (((ip->i_ffs1_mode & IFMT) == IFREG ||
    997      1.161  perseant 	     (ip->i_ffs1_mode & IFMT) == IFDIR) &&
    998      1.160  perseant 	    ip->i_size > ((ip->i_lfs_hiblk + 1) << fs->lfs_bshift)) {
    999      1.160  perseant 		cdp->di_size = (ip->i_lfs_hiblk + 1) << fs->lfs_bshift;
   1000      1.160  perseant 		DLOG((DLOG_SEG, "lfs_writeinode: ino %d size %" PRId64 " -> %"
   1001      1.160  perseant 		      PRId64 "\n", (int)ip->i_number, ip->i_size, cdp->di_size));
   1002      1.160  perseant 	}
   1003      1.119      fvdl 	if (ip->i_lfs_effnblks != ip->i_ffs1_blocks) {
   1004      1.160  perseant 		if (ip->i_flags & IN_CLEANING)
   1005      1.160  perseant 			cdp->di_size = ip->i_lfs_osize;
   1006      1.160  perseant 		DLOG((DLOG_SEG, "lfs_writeinode: cleansing ino %d eff %d != nblk %d)"
   1007      1.160  perseant 		      " at %x\n", ip->i_number, ip->i_lfs_effnblks,
   1008      1.160  perseant 		      ip->i_ffs1_blocks, fs->lfs_offset));
   1009       1.53  perseant 		for (daddrp = cdp->di_db; daddrp < cdp->di_ib + NIADDR;
   1010       1.53  perseant 		     daddrp++) {
   1011       1.53  perseant 			if (*daddrp == UNWRITTEN) {
   1012      1.158  perseant 				DLOG((DLOG_SEG, "lfs_writeinode: wiping UNWRITTEN\n"));
   1013       1.53  perseant 				*daddrp = 0;
   1014       1.53  perseant 			}
   1015       1.53  perseant 		}
   1016       1.80  perseant 	} else {
   1017      1.160  perseant 		/* If all blocks are going to disk, update "size on disk" */
   1018      1.119      fvdl 		ip->i_lfs_osize = ip->i_size;
   1019       1.53  perseant 	}
   1020      1.157     perry 
   1021      1.160  perseant #ifdef DIAGNOSTIC
   1022      1.160  perseant 	/*
   1023      1.160  perseant 	 * Check dinode held blocks against dinode size.
   1024      1.160  perseant 	 * This should be identical to the check in lfs_vget().
   1025      1.160  perseant 	 */
   1026      1.160  perseant 	for (i = (cdp->di_size + fs->lfs_bsize - 1) >> fs->lfs_bshift;
   1027      1.160  perseant 	     i < NDADDR; i++) {
   1028      1.160  perseant 		KASSERT(i >= 0);
   1029      1.160  perseant 		if ((cdp->di_mode & IFMT) == IFLNK)
   1030      1.160  perseant 			continue;
   1031      1.160  perseant 		if (((cdp->di_mode & IFMT) == IFBLK ||
   1032      1.160  perseant 		     (cdp->di_mode & IFMT) == IFCHR) && i == 0)
   1033      1.160  perseant 			continue;
   1034      1.160  perseant 		if (cdp->di_db[i] != 0) {
   1035      1.160  perseant # ifdef DEBUG
   1036      1.160  perseant 			lfs_dump_dinode(cdp);
   1037      1.160  perseant # endif
   1038      1.160  perseant 			panic("writing inconsistent inode");
   1039      1.160  perseant 		}
   1040      1.160  perseant 	}
   1041      1.160  perseant #endif /* DIAGNOSTIC */
   1042      1.160  perseant 
   1043       1.73       chs 	if (ip->i_flag & IN_CLEANING)
   1044       1.56  perseant 		LFS_CLR_UINO(ip, IN_CLEANING);
   1045       1.55  perseant 	else {
   1046       1.56  perseant 		/* XXX IN_ALLMOD */
   1047       1.56  perseant 		LFS_CLR_UINO(ip, IN_ACCESSED | IN_ACCESS | IN_CHANGE |
   1048      1.154   mycroft 			     IN_UPDATE | IN_MODIFY);
   1049      1.119      fvdl 		if (ip->i_lfs_effnblks == ip->i_ffs1_blocks)
   1050       1.56  perseant 			LFS_CLR_UINO(ip, IN_MODIFIED);
   1051       1.63  perseant 		else
   1052      1.158  perseant 			DLOG((DLOG_VNODE, "lfs_writeinode: ino %d: real blks=%d, "
   1053      1.158  perseant 			      "eff=%d\n", ip->i_number, ip->i_ffs1_blocks,
   1054      1.158  perseant 			      ip->i_lfs_effnblks));
   1055       1.55  perseant 	}
   1056       1.55  perseant 
   1057       1.73       chs 	if (ip->i_number == LFS_IFILE_INUM) /* We know sp->idp == NULL */
   1058      1.157     perry 		sp->idp = ((struct ufs1_dinode *)bp->b_data) +
   1059       1.53  perseant 			(sp->ninodes % INOPB(fs));
   1060       1.73       chs 	if (gotblk) {
   1061       1.62  perseant 		LFS_LOCK_BUF(bp);
   1062       1.24  perseant 		brelse(bp);
   1063       1.24  perseant 	}
   1064      1.157     perry 
   1065        1.1   mycroft 	/* Increment inode count in segment summary block. */
   1066        1.1   mycroft 	++((SEGSUM *)(sp->segsum))->ss_ninos;
   1067      1.157     perry 
   1068        1.1   mycroft 	/* If this page is full, set flag to allocate a new page. */
   1069        1.1   mycroft 	if (++sp->ninodes % INOPB(fs) == 0)
   1070        1.1   mycroft 		sp->ibp = NULL;
   1071      1.157     perry 
   1072        1.1   mycroft 	/*
   1073        1.1   mycroft 	 * If updating the ifile, update the super-block.  Update the disk
   1074        1.1   mycroft 	 * address and access times for this inode in the ifile.
   1075        1.1   mycroft 	 */
   1076        1.1   mycroft 	ino = ip->i_number;
   1077        1.1   mycroft 	if (ino == LFS_IFILE_INUM) {
   1078        1.1   mycroft 		daddr = fs->lfs_idaddr;
   1079       1.69  perseant 		fs->lfs_idaddr = dbtofsb(fs, bp->b_blkno);
   1080        1.1   mycroft 	} else {
   1081        1.1   mycroft 		LFS_IENTRY(ifp, fs, ino, ibp);
   1082        1.1   mycroft 		daddr = ifp->if_daddr;
   1083       1.69  perseant 		ifp->if_daddr = dbtofsb(fs, bp->b_blkno) + fsb;
   1084       1.74  perseant 		error = LFS_BWRITE_LOG(ibp); /* Ifile */
   1085        1.1   mycroft 	}
   1086      1.157     perry 
   1087        1.1   mycroft 	/*
   1088       1.60    toshii 	 * The inode's last address should not be in the current partial
   1089       1.60    toshii 	 * segment, except under exceptional circumstances (lfs_writevnodes
   1090       1.60    toshii 	 * had to start over, and in the meantime more blocks were written
   1091      1.103  perseant 	 * to a vnode).	 Both inodes will be accounted to this segment
   1092       1.80  perseant 	 * in lfs_writeseg so we need to subtract the earlier version
   1093      1.103  perseant 	 * here anyway.	 The segment count can temporarily dip below
   1094      1.157     perry 	 * zero here; keep track of how many duplicates we have in
   1095       1.80  perseant 	 * "dupino" so we don't panic below.
   1096       1.60    toshii 	 */
   1097       1.80  perseant 	if (daddr >= fs->lfs_lastpseg && daddr <= dbtofsb(fs, bp->b_blkno)) {
   1098       1.80  perseant 		++sp->ndupino;
   1099      1.158  perseant 		DLOG((DLOG_SEG, "lfs_writeinode: last inode addr in current pseg "
   1100      1.158  perseant 		      "(ino %d daddr 0x%llx) ndupino=%d\n", ino,
   1101      1.158  perseant 		      (long long)daddr, sp->ndupino));
   1102       1.80  perseant 	}
   1103       1.80  perseant 	/*
   1104       1.80  perseant 	 * Account the inode: it no longer belongs to its former segment,
   1105       1.80  perseant 	 * though it will not belong to the new segment until that segment
   1106       1.80  perseant 	 * is actually written.
   1107       1.80  perseant 	 */
   1108       1.49  perseant 	if (daddr != LFS_UNUSED_DADDR) {
   1109       1.83      yamt 		u_int32_t oldsn = dtosn(fs, daddr);
   1110        1.1   mycroft #ifdef DIAGNOSTIC
   1111       1.83      yamt 		int ndupino = (sp->seg_number == oldsn) ? sp->ndupino : 0;
   1112       1.83      yamt #endif
   1113       1.83      yamt 		LFS_SEGENTRY(sup, fs, oldsn, bp);
   1114       1.83      yamt #ifdef DIAGNOSTIC
   1115      1.119      fvdl 		if (sup->su_nbytes +
   1116      1.119      fvdl 		    sizeof (struct ufs1_dinode) * ndupino
   1117      1.119      fvdl 		      < sizeof (struct ufs1_dinode)) {
   1118       1.53  perseant 			printf("lfs_writeinode: negative bytes "
   1119       1.96      yamt 			       "(segment %" PRIu32 " short by %d, "
   1120       1.96      yamt 			       "oldsn=%" PRIu32 ", cursn=%" PRIu32
   1121       1.96      yamt 			       ", daddr=%" PRId64 ", su_nbytes=%u, "
   1122       1.83      yamt 			       "ndupino=%d)\n",
   1123       1.69  perseant 			       dtosn(fs, daddr),
   1124      1.128      yamt 			       (int)sizeof (struct ufs1_dinode) *
   1125      1.128      yamt 				   (1 - sp->ndupino) - sup->su_nbytes,
   1126       1.96      yamt 			       oldsn, sp->seg_number, daddr,
   1127       1.83      yamt 			       (unsigned int)sup->su_nbytes,
   1128       1.83      yamt 			       sp->ndupino);
   1129       1.27  perseant 			panic("lfs_writeinode: negative bytes");
   1130      1.119      fvdl 			sup->su_nbytes = sizeof (struct ufs1_dinode);
   1131        1.1   mycroft 		}
   1132        1.1   mycroft #endif
   1133      1.158  perseant 		DLOG((DLOG_SU, "seg %d -= %d for ino %d inode\n",
   1134      1.158  perseant 		      dtosn(fs, daddr), sizeof (struct ufs1_dinode), ino));
   1135      1.119      fvdl 		sup->su_nbytes -= sizeof (struct ufs1_dinode);
   1136        1.1   mycroft 		redo_ifile =
   1137       1.15  perseant 			(ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED));
   1138      1.159  perseant 		if (redo_ifile) {
   1139      1.159  perseant 			simple_lock(&fs->lfs_interlock);
   1140       1.74  perseant 			fs->lfs_flags |= LFS_IFDIRTY;
   1141      1.159  perseant 			simple_unlock(&fs->lfs_interlock);
   1142      1.159  perseant 		}
   1143      1.101  perseant 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp); /* Ifile */
   1144        1.1   mycroft 	}
   1145        1.1   mycroft 	return (redo_ifile);
   1146        1.1   mycroft }
   1147        1.1   mycroft 
   1148        1.1   mycroft int
   1149       1.69  perseant lfs_gatherblock(struct segment *sp, struct buf *bp, int *sptr)
   1150        1.1   mycroft {
   1151        1.1   mycroft 	struct lfs *fs;
   1152      1.164  christos 	int vers;
   1153      1.101  perseant 	int j, blksinblk;
   1154      1.101  perseant 
   1155      1.159  perseant 	ASSERT_SEGLOCK(sp->fs);
   1156        1.1   mycroft 	/*
   1157        1.1   mycroft 	 * If full, finish this segment.  We may be doing I/O, so
   1158        1.1   mycroft 	 * release and reacquire the splbio().
   1159        1.1   mycroft 	 */
   1160        1.1   mycroft #ifdef DIAGNOSTIC
   1161        1.1   mycroft 	if (sp->vp == NULL)
   1162        1.1   mycroft 		panic ("lfs_gatherblock: Null vp in segment");
   1163        1.1   mycroft #endif
   1164        1.1   mycroft 	fs = sp->fs;
   1165      1.101  perseant 	blksinblk = howmany(bp->b_bcount, fs->lfs_bsize);
   1166      1.101  perseant 	if (sp->sum_bytes_left < sizeof(int32_t) * blksinblk ||
   1167       1.10      fvdl 	    sp->seg_bytes_left < bp->b_bcount) {
   1168        1.1   mycroft 		if (sptr)
   1169        1.1   mycroft 			splx(*sptr);
   1170        1.1   mycroft 		lfs_updatemeta(sp);
   1171      1.157     perry 
   1172      1.164  christos 		vers = sp->fip->fi_version;
   1173        1.1   mycroft 		(void) lfs_writeseg(fs, sp);
   1174      1.157     perry 
   1175        1.1   mycroft 		/* Add the current file to the segment summary. */
   1176      1.180  perseant 		lfs_acquire_finfo(fs, VTOI(sp->vp)->i_number, vers);
   1177      1.157     perry 
   1178        1.1   mycroft 		if (sptr)
   1179        1.1   mycroft 			*sptr = splbio();
   1180       1.73       chs 		return (1);
   1181        1.1   mycroft 	}
   1182      1.157     perry 
   1183       1.73       chs 	if (bp->b_flags & B_GATHERED) {
   1184      1.158  perseant 		DLOG((DLOG_SEG, "lfs_gatherblock: already gathered! Ino %d,"
   1185      1.158  perseant 		      " lbn %" PRId64 "\n",
   1186      1.158  perseant 		      sp->fip->fi_ino, bp->b_lblkno));
   1187       1.73       chs 		return (0);
   1188       1.15  perseant 	}
   1189      1.158  perseant 
   1190        1.1   mycroft 	/* Insert into the buffer list, update the FINFO block. */
   1191        1.1   mycroft 	bp->b_flags |= B_GATHERED;
   1192       1.74  perseant 
   1193        1.1   mycroft 	*sp->cbpp++ = bp;
   1194      1.162  perseant 	for (j = 0; j < blksinblk; j++) {
   1195      1.104  perseant 		sp->fip->fi_blocks[sp->fip->fi_nblocks++] = bp->b_lblkno + j;
   1196      1.162  perseant 		/* This block's accounting moves from lfs_favail to lfs_avail */
   1197      1.162  perseant 		lfs_deregister_block(sp->vp, bp->b_lblkno + j);
   1198      1.162  perseant 	}
   1199      1.157     perry 
   1200      1.104  perseant 	sp->sum_bytes_left -= sizeof(int32_t) * blksinblk;
   1201       1.10      fvdl 	sp->seg_bytes_left -= bp->b_bcount;
   1202       1.73       chs 	return (0);
   1203        1.1   mycroft }
   1204        1.1   mycroft 
   1205       1.15  perseant int
   1206      1.128      yamt lfs_gather(struct lfs *fs, struct segment *sp, struct vnode *vp,
   1207      1.128      yamt     int (*match)(struct lfs *, struct buf *))
   1208        1.1   mycroft {
   1209       1.77  perseant 	struct buf *bp, *nbp;
   1210       1.73       chs 	int s, count = 0;
   1211      1.157     perry 
   1212      1.159  perseant 	ASSERT_SEGLOCK(fs);
   1213      1.172  perseant 	if (vp->v_type == VBLK)
   1214      1.172  perseant 		return 0;
   1215      1.141      yamt 	KASSERT(sp->vp == NULL);
   1216        1.1   mycroft 	sp->vp = vp;
   1217        1.1   mycroft 	s = splbio();
   1218       1.15  perseant 
   1219       1.15  perseant #ifndef LFS_NO_BACKBUF_HACK
   1220       1.10      fvdl /* This is a hack to see if ordering the blocks in LFS makes a difference. */
   1221      1.128      yamt # define	BUF_OFFSET	\
   1222      1.128      yamt 	(((caddr_t)&LIST_NEXT(bp, b_vnbufs)) - (caddr_t)bp)
   1223      1.128      yamt # define	BACK_BUF(BP)	\
   1224      1.128      yamt 	((struct buf *)(((caddr_t)(BP)->b_vnbufs.le_prev) - BUF_OFFSET))
   1225      1.128      yamt # define	BEG_OF_LIST	\
   1226      1.128      yamt 	((struct buf *)(((caddr_t)&LIST_FIRST(&vp->v_dirtyblkhd)) - BUF_OFFSET))
   1227      1.128      yamt 
   1228      1.128      yamt loop:
   1229      1.128      yamt 	/* Find last buffer. */
   1230      1.128      yamt 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd);
   1231      1.128      yamt 	     bp && LIST_NEXT(bp, b_vnbufs) != NULL;
   1232      1.128      yamt 	     bp = LIST_NEXT(bp, b_vnbufs))
   1233      1.128      yamt 		/* nothing */;
   1234       1.77  perseant 	for (; bp && bp != BEG_OF_LIST; bp = nbp) {
   1235       1.77  perseant 		nbp = BACK_BUF(bp);
   1236       1.77  perseant #else /* LFS_NO_BACKBUF_HACK */
   1237      1.128      yamt loop:
   1238      1.128      yamt 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
   1239       1.77  perseant 		nbp = LIST_NEXT(bp, b_vnbufs);
   1240       1.15  perseant #endif /* LFS_NO_BACKBUF_HACK */
   1241       1.74  perseant 		if ((bp->b_flags & (B_BUSY|B_GATHERED)) || !match(fs, bp)) {
   1242      1.158  perseant #ifdef DEBUG
   1243      1.128      yamt 			if (vp == fs->lfs_ivnode &&
   1244      1.128      yamt 			    (bp->b_flags & (B_BUSY|B_GATHERED)) == B_BUSY)
   1245      1.158  perseant 				DLOG((DLOG_SEG, "lfs_gather: ifile lbn %"
   1246      1.158  perseant 				      PRId64 " busy (%x)",
   1247      1.158  perseant 				      bp->b_lblkno, bp->b_flags));
   1248       1.74  perseant #endif
   1249        1.1   mycroft 			continue;
   1250       1.74  perseant 		}
   1251        1.1   mycroft #ifdef DIAGNOSTIC
   1252      1.109  perseant # ifdef LFS_USE_B_INVAL
   1253      1.172  perseant 		if ((bp->b_flags & (B_CALL|B_INVAL)) == B_INVAL) {
   1254      1.172  perseant 			DLOG((DLOG_SEG, "lfs_gather: lbn %" PRId64
   1255      1.172  perseant 			      " is B_INVAL\n", bp->b_lblkno));
   1256      1.172  perseant 			VOP_PRINT(bp->b_vp);
   1257      1.172  perseant 		}
   1258      1.109  perseant # endif /* LFS_USE_B_INVAL */
   1259      1.172  perseant 		if (!(bp->b_flags & B_DELWRI))
   1260      1.172  perseant 			panic("lfs_gather: bp not B_DELWRI");
   1261      1.172  perseant 		if (!(bp->b_flags & B_LOCKED)) {
   1262      1.172  perseant 			DLOG((DLOG_SEG, "lfs_gather: lbn %" PRId64
   1263      1.172  perseant 			      " blk %" PRId64 " not B_LOCKED\n",
   1264      1.172  perseant 			      bp->b_lblkno,
   1265      1.172  perseant 			      dbtofsb(fs, bp->b_blkno)));
   1266      1.172  perseant 			VOP_PRINT(bp->b_vp);
   1267      1.172  perseant 			panic("lfs_gather: bp not B_LOCKED");
   1268      1.172  perseant 		}
   1269        1.1   mycroft #endif
   1270      1.172  perseant 		if (lfs_gatherblock(sp, bp, &s)) {
   1271      1.172  perseant 			goto loop;
   1272       1.30  perseant 		}
   1273       1.15  perseant 		count++;
   1274        1.1   mycroft 	}
   1275        1.1   mycroft 	splx(s);
   1276        1.1   mycroft 	lfs_updatemeta(sp);
   1277      1.141      yamt 	KASSERT(sp->vp == vp);
   1278        1.1   mycroft 	sp->vp = NULL;
   1279       1.15  perseant 	return count;
   1280        1.1   mycroft }
   1281        1.1   mycroft 
   1282      1.101  perseant #if DEBUG
   1283      1.101  perseant # define DEBUG_OOFF(n) do {						\
   1284      1.101  perseant 	if (ooff == 0) {						\
   1285      1.158  perseant 		DLOG((DLOG_SEG, "lfs_updatemeta[%d]: warning: writing " \
   1286      1.105  perseant 			"ino %d lbn %" PRId64 " at 0x%" PRIx32		\
   1287      1.105  perseant 			", was 0x0 (or %" PRId64 ")\n",			\
   1288      1.158  perseant 			(n), ip->i_number, lbn, ndaddr, daddr));	\
   1289      1.101  perseant 	}								\
   1290      1.115  perseant } while (0)
   1291      1.101  perseant #else
   1292      1.101  perseant # define DEBUG_OOFF(n)
   1293      1.101  perseant #endif
   1294      1.101  perseant 
   1295      1.101  perseant /*
   1296      1.101  perseant  * Change the given block's address to ndaddr, finding its previous
   1297      1.101  perseant  * location using ufs_bmaparray().
   1298      1.101  perseant  *
   1299      1.101  perseant  * Account for this change in the segment table.
   1300      1.143      yamt  *
   1301      1.143      yamt  * called with sp == NULL by roll-forwarding code.
   1302      1.101  perseant  */
   1303      1.101  perseant void
   1304      1.143      yamt lfs_update_single(struct lfs *fs, struct segment *sp, struct vnode *vp,
   1305      1.143      yamt     daddr_t lbn, int32_t ndaddr, int size)
   1306      1.101  perseant {
   1307      1.101  perseant 	SEGUSE *sup;
   1308      1.101  perseant 	struct buf *bp;
   1309      1.101  perseant 	struct indir a[NIADDR + 2], *ap;
   1310      1.101  perseant 	struct inode *ip;
   1311      1.101  perseant 	daddr_t daddr, ooff;
   1312      1.104  perseant 	int num, error;
   1313      1.101  perseant 	int bb, osize, obb;
   1314      1.157     perry 
   1315      1.159  perseant 	ASSERT_SEGLOCK(fs);
   1316      1.143      yamt 	KASSERT(sp == NULL || sp->vp == vp);
   1317      1.101  perseant 	ip = VTOI(vp);
   1318      1.101  perseant 
   1319      1.121      yamt 	error = ufs_bmaparray(vp, lbn, &daddr, a, &num, NULL, NULL);
   1320      1.101  perseant 	if (error)
   1321      1.101  perseant 		panic("lfs_updatemeta: ufs_bmaparray returned %d", error);
   1322      1.116  perseant 
   1323      1.156  perseant 	daddr = (daddr_t)((int32_t)daddr); /* XXX ondisk32 */
   1324      1.116  perseant 	KASSERT(daddr <= LFS_MAX_DADDR);
   1325      1.101  perseant 	if (daddr > 0)
   1326      1.101  perseant 		daddr = dbtofsb(fs, daddr);
   1327      1.157     perry 
   1328      1.101  perseant 	bb = fragstofsb(fs, numfrags(fs, size));
   1329      1.101  perseant 	switch (num) {
   1330      1.101  perseant 	    case 0:
   1331      1.119      fvdl 		    ooff = ip->i_ffs1_db[lbn];
   1332      1.101  perseant 		    DEBUG_OOFF(0);
   1333      1.101  perseant 		    if (ooff == UNWRITTEN)
   1334      1.119      fvdl 			    ip->i_ffs1_blocks += bb;
   1335      1.101  perseant 		    else {
   1336      1.101  perseant 			    /* possible fragment truncation or extension */
   1337      1.101  perseant 			    obb = btofsb(fs, ip->i_lfs_fragsize[lbn]);
   1338      1.119      fvdl 			    ip->i_ffs1_blocks += (bb - obb);
   1339      1.101  perseant 		    }
   1340      1.119      fvdl 		    ip->i_ffs1_db[lbn] = ndaddr;
   1341      1.101  perseant 		    break;
   1342      1.101  perseant 	    case 1:
   1343      1.119      fvdl 		    ooff = ip->i_ffs1_ib[a[0].in_off];
   1344      1.101  perseant 		    DEBUG_OOFF(1);
   1345      1.101  perseant 		    if (ooff == UNWRITTEN)
   1346      1.119      fvdl 			    ip->i_ffs1_blocks += bb;
   1347      1.119      fvdl 		    ip->i_ffs1_ib[a[0].in_off] = ndaddr;
   1348      1.101  perseant 		    break;
   1349      1.101  perseant 	    default:
   1350      1.101  perseant 		    ap = &a[num - 1];
   1351      1.101  perseant 		    if (bread(vp, ap->in_lbn, fs->lfs_bsize, NOCRED, &bp))
   1352      1.101  perseant 			    panic("lfs_updatemeta: bread bno %" PRId64,
   1353      1.101  perseant 				  ap->in_lbn);
   1354      1.101  perseant 
   1355      1.101  perseant 		    /* XXX ondisk32 */
   1356      1.101  perseant 		    ooff = ((int32_t *)bp->b_data)[ap->in_off];
   1357      1.101  perseant 		    DEBUG_OOFF(num);
   1358      1.101  perseant 		    if (ooff == UNWRITTEN)
   1359      1.119      fvdl 			    ip->i_ffs1_blocks += bb;
   1360      1.101  perseant 		    /* XXX ondisk32 */
   1361      1.101  perseant 		    ((int32_t *)bp->b_data)[ap->in_off] = ndaddr;
   1362      1.101  perseant 		    (void) VOP_BWRITE(bp);
   1363      1.101  perseant 	}
   1364      1.106  perseant 
   1365      1.150      yamt 	KASSERT(ooff == 0 || ooff == UNWRITTEN || ooff == daddr);
   1366      1.150      yamt 
   1367      1.160  perseant 	/* Update hiblk when extending the file */
   1368      1.160  perseant 	if (lbn > ip->i_lfs_hiblk)
   1369      1.160  perseant 		ip->i_lfs_hiblk = lbn;
   1370      1.160  perseant 
   1371      1.106  perseant 	/*
   1372      1.106  perseant 	 * Though we'd rather it couldn't, this *can* happen right now
   1373      1.106  perseant 	 * if cleaning blocks and regular blocks coexist.
   1374      1.106  perseant 	 */
   1375      1.106  perseant 	/* KASSERT(daddr < fs->lfs_lastpseg || daddr > ndaddr); */
   1376      1.101  perseant 
   1377      1.101  perseant 	/*
   1378      1.101  perseant 	 * Update segment usage information, based on old size
   1379      1.101  perseant 	 * and location.
   1380      1.101  perseant 	 */
   1381      1.101  perseant 	if (daddr > 0) {
   1382      1.101  perseant 		u_int32_t oldsn = dtosn(fs, daddr);
   1383      1.101  perseant #ifdef DIAGNOSTIC
   1384      1.143      yamt 		int ndupino;
   1385      1.143      yamt 
   1386      1.143      yamt 		if (sp && sp->seg_number == oldsn) {
   1387      1.143      yamt 			ndupino = sp->ndupino;
   1388      1.143      yamt 		} else {
   1389      1.143      yamt 			ndupino = 0;
   1390      1.143      yamt 		}
   1391      1.101  perseant #endif
   1392      1.116  perseant 		KASSERT(oldsn >= 0 && oldsn < fs->lfs_nseg);
   1393      1.101  perseant 		if (lbn >= 0 && lbn < NDADDR)
   1394      1.101  perseant 			osize = ip->i_lfs_fragsize[lbn];
   1395      1.101  perseant 		else
   1396      1.101  perseant 			osize = fs->lfs_bsize;
   1397      1.101  perseant 		LFS_SEGENTRY(sup, fs, oldsn, bp);
   1398      1.101  perseant #ifdef DIAGNOSTIC
   1399      1.119      fvdl 		if (sup->su_nbytes + sizeof (struct ufs1_dinode) * ndupino
   1400      1.119      fvdl 		    < osize) {
   1401      1.101  perseant 			printf("lfs_updatemeta: negative bytes "
   1402      1.101  perseant 			       "(segment %" PRIu32 " short by %" PRId64
   1403      1.101  perseant 			       ")\n", dtosn(fs, daddr),
   1404      1.101  perseant 			       (int64_t)osize -
   1405      1.143      yamt 			       (sizeof (struct ufs1_dinode) * ndupino +
   1406      1.101  perseant 				sup->su_nbytes));
   1407      1.165  christos 			printf("lfs_updatemeta: ino %llu, lbn %" PRId64
   1408      1.101  perseant 			       ", addr = 0x%" PRIx64 "\n",
   1409      1.165  christos 			       (unsigned long long)ip->i_number, lbn, daddr);
   1410      1.101  perseant 			printf("lfs_updatemeta: ndupino=%d\n", ndupino);
   1411      1.101  perseant 			panic("lfs_updatemeta: negative bytes");
   1412      1.119      fvdl 			sup->su_nbytes = osize -
   1413      1.143      yamt 			    sizeof (struct ufs1_dinode) * ndupino;
   1414      1.101  perseant 		}
   1415      1.101  perseant #endif
   1416      1.158  perseant 		DLOG((DLOG_SU, "seg %" PRIu32 " -= %d for ino %d lbn %" PRId64
   1417      1.158  perseant 		      " db 0x%" PRIx64 "\n",
   1418      1.158  perseant 		      dtosn(fs, daddr), osize,
   1419      1.158  perseant 		      ip->i_number, lbn, daddr));
   1420      1.101  perseant 		sup->su_nbytes -= osize;
   1421      1.159  perseant 		if (!(bp->b_flags & B_GATHERED)) {
   1422      1.159  perseant 			simple_lock(&fs->lfs_interlock);
   1423      1.101  perseant 			fs->lfs_flags |= LFS_IFDIRTY;
   1424      1.159  perseant 			simple_unlock(&fs->lfs_interlock);
   1425      1.159  perseant 		}
   1426      1.101  perseant 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp);
   1427      1.101  perseant 	}
   1428      1.101  perseant 	/*
   1429      1.101  perseant 	 * Now that this block has a new address, and its old
   1430      1.101  perseant 	 * segment no longer owns it, we can forget about its
   1431      1.101  perseant 	 * old size.
   1432      1.101  perseant 	 */
   1433      1.101  perseant 	if (lbn >= 0 && lbn < NDADDR)
   1434      1.101  perseant 		ip->i_lfs_fragsize[lbn] = size;
   1435      1.101  perseant }
   1436      1.101  perseant 
   1437        1.1   mycroft /*
   1438        1.1   mycroft  * Update the metadata that points to the blocks listed in the FINFO
   1439        1.1   mycroft  * array.
   1440        1.1   mycroft  */
   1441        1.1   mycroft void
   1442       1.69  perseant lfs_updatemeta(struct segment *sp)
   1443        1.1   mycroft {
   1444      1.101  perseant 	struct buf *sbp;
   1445        1.1   mycroft 	struct lfs *fs;
   1446        1.1   mycroft 	struct vnode *vp;
   1447      1.101  perseant 	daddr_t lbn;
   1448      1.101  perseant 	int i, nblocks, num;
   1449      1.101  perseant 	int bb;
   1450      1.101  perseant 	int bytesleft, size;
   1451      1.157     perry 
   1452      1.159  perseant 	ASSERT_SEGLOCK(sp->fs);
   1453        1.1   mycroft 	vp = sp->vp;
   1454        1.1   mycroft 	nblocks = &sp->fip->fi_blocks[sp->fip->fi_nblocks] - sp->start_lbp;
   1455      1.101  perseant 	KASSERT(nblocks >= 0);
   1456      1.141      yamt 	KASSERT(vp != NULL);
   1457      1.141      yamt 	if (nblocks == 0)
   1458        1.1   mycroft 		return;
   1459      1.104  perseant 
   1460      1.104  perseant 	/*
   1461      1.104  perseant 	 * This count may be high due to oversize blocks from lfs_gop_write.
   1462      1.104  perseant 	 * Correct for this. (XXX we should be able to keep track of these.)
   1463      1.104  perseant 	 */
   1464      1.104  perseant 	fs = sp->fs;
   1465      1.104  perseant 	for (i = 0; i < nblocks; i++) {
   1466      1.104  perseant 		if (sp->start_bpp[i] == NULL) {
   1467      1.158  perseant 			DLOG((DLOG_SEG, "lfs_updatemeta: nblocks = %d, not %d\n", i, nblocks));
   1468      1.104  perseant 			nblocks = i;
   1469      1.104  perseant 			break;
   1470      1.104  perseant 		}
   1471      1.104  perseant 		num = howmany(sp->start_bpp[i]->b_bcount, fs->lfs_bsize);
   1472      1.118      yamt 		KASSERT(sp->start_bpp[i]->b_lblkno >= 0 || num == 1);
   1473      1.104  perseant 		nblocks -= num - 1;
   1474      1.104  perseant 	}
   1475      1.118      yamt 
   1476      1.118      yamt 	KASSERT(vp->v_type == VREG ||
   1477      1.118      yamt 	   nblocks == &sp->fip->fi_blocks[sp->fip->fi_nblocks] - sp->start_lbp);
   1478      1.118      yamt 	KASSERT(nblocks == sp->cbpp - sp->start_bpp);
   1479      1.157     perry 
   1480       1.15  perseant 	/*
   1481      1.101  perseant 	 * Sort the blocks.
   1482      1.101  perseant 	 *
   1483      1.101  perseant 	 * We have to sort even if the blocks come from the
   1484       1.15  perseant 	 * cleaner, because there might be other pending blocks on the
   1485       1.15  perseant 	 * same inode...and if we don't sort, and there are fragments
   1486       1.15  perseant 	 * present, blocks may be written in the wrong place.
   1487       1.15  perseant 	 */
   1488      1.104  perseant 	lfs_shellsort(sp->start_bpp, sp->start_lbp, nblocks, fs->lfs_bsize);
   1489      1.157     perry 
   1490        1.1   mycroft 	/*
   1491       1.10      fvdl 	 * Record the length of the last block in case it's a fragment.
   1492       1.10      fvdl 	 * If there are indirect blocks present, they sort last.  An
   1493       1.10      fvdl 	 * indirect block will be lfs_bsize and its presence indicates
   1494       1.10      fvdl 	 * that you cannot have fragments.
   1495       1.80  perseant 	 *
   1496       1.80  perseant 	 * XXX This last is a lie.  A cleaned fragment can coexist with
   1497      1.103  perseant 	 * XXX a later indirect block.	This will continue to be
   1498       1.80  perseant 	 * XXX true until lfs_markv is fixed to do everything with
   1499       1.80  perseant 	 * XXX fake blocks (including fake inodes and fake indirect blocks).
   1500       1.10      fvdl 	 */
   1501      1.101  perseant 	sp->fip->fi_lastlength = ((sp->start_bpp[nblocks - 1]->b_bcount - 1) &
   1502      1.101  perseant 		fs->lfs_bmask) + 1;
   1503      1.157     perry 
   1504       1.10      fvdl 	/*
   1505        1.1   mycroft 	 * Assign disk addresses, and update references to the logical
   1506        1.1   mycroft 	 * block and the segment usage information.
   1507        1.1   mycroft 	 */
   1508        1.1   mycroft 	for (i = nblocks; i--; ++sp->start_bpp) {
   1509      1.101  perseant 		sbp = *sp->start_bpp;
   1510      1.104  perseant 		lbn = *sp->start_lbp;
   1511      1.118      yamt 		KASSERT(sbp->b_lblkno == lbn);
   1512      1.104  perseant 
   1513       1.80  perseant 		sbp->b_blkno = fsbtodb(fs, fs->lfs_offset);
   1514       1.80  perseant 
   1515       1.80  perseant 		/*
   1516       1.80  perseant 		 * If we write a frag in the wrong place, the cleaner won't
   1517       1.80  perseant 		 * be able to correctly identify its size later, and the
   1518      1.103  perseant 		 * segment will be uncleanable.	 (Even worse, it will assume
   1519       1.80  perseant 		 * that the indirect block that actually ends the list
   1520       1.80  perseant 		 * is of a smaller size!)
   1521       1.80  perseant 		 */
   1522      1.101  perseant 		if ((sbp->b_bcount & fs->lfs_bmask) && i != 0)
   1523       1.82    provos 			panic("lfs_updatemeta: fragment is not last block");
   1524      1.119      fvdl 
   1525       1.80  perseant 		/*
   1526      1.101  perseant 		 * For each subblock in this possibly oversized block,
   1527      1.101  perseant 		 * update its address on disk.
   1528       1.80  perseant 		 */
   1529      1.101  perseant 		KASSERT(lbn >= 0 || sbp->b_bcount == fs->lfs_bsize);
   1530      1.141      yamt 		KASSERT(vp == sbp->b_vp);
   1531      1.101  perseant 		for (bytesleft = sbp->b_bcount; bytesleft > 0;
   1532      1.101  perseant 		     bytesleft -= fs->lfs_bsize) {
   1533      1.101  perseant 			size = MIN(bytesleft, fs->lfs_bsize);
   1534      1.101  perseant 			bb = fragstofsb(fs, numfrags(fs, size));
   1535      1.104  perseant 			lbn = *sp->start_lbp++;
   1536      1.143      yamt 			lfs_update_single(fs, sp, sp->vp, lbn, fs->lfs_offset,
   1537      1.143      yamt 			    size);
   1538      1.101  perseant 			fs->lfs_offset += bb;
   1539        1.1   mycroft 		}
   1540      1.101  perseant 
   1541        1.1   mycroft 	}
   1542        1.1   mycroft }
   1543        1.1   mycroft 
   1544        1.1   mycroft /*
   1545      1.163  perseant  * Move lfs_offset to a segment earlier than sn.
   1546      1.163  perseant  */
   1547      1.163  perseant int
   1548      1.163  perseant lfs_rewind(struct lfs *fs, int newsn)
   1549      1.163  perseant {
   1550      1.163  perseant 	int sn, osn, isdirty;
   1551      1.163  perseant 	struct buf *bp;
   1552      1.163  perseant 	SEGUSE *sup;
   1553      1.163  perseant 
   1554      1.163  perseant 	ASSERT_SEGLOCK(fs);
   1555      1.163  perseant 
   1556      1.163  perseant 	osn = dtosn(fs, fs->lfs_offset);
   1557      1.163  perseant 	if (osn < newsn)
   1558      1.163  perseant 		return 0;
   1559      1.163  perseant 
   1560      1.163  perseant 	/* lfs_avail eats the remaining space in this segment */
   1561      1.163  perseant 	fs->lfs_avail -= fs->lfs_fsbpseg - (fs->lfs_offset - fs->lfs_curseg);
   1562      1.163  perseant 
   1563      1.163  perseant 	/* Find a low-numbered segment */
   1564      1.163  perseant 	for (sn = 0; sn < fs->lfs_nseg; ++sn) {
   1565      1.163  perseant 		LFS_SEGENTRY(sup, fs, sn, bp);
   1566      1.163  perseant 		isdirty = sup->su_flags & SEGUSE_DIRTY;
   1567      1.163  perseant 		brelse(bp);
   1568      1.163  perseant 
   1569      1.163  perseant 		if (!isdirty)
   1570      1.163  perseant 			break;
   1571      1.163  perseant 	}
   1572      1.163  perseant 	if (sn == fs->lfs_nseg)
   1573      1.163  perseant 		panic("lfs_rewind: no clean segments");
   1574      1.163  perseant 	if (sn >= newsn)
   1575      1.163  perseant 		return ENOENT;
   1576      1.163  perseant 	fs->lfs_nextseg = sn;
   1577      1.163  perseant 	lfs_newseg(fs);
   1578      1.163  perseant 	fs->lfs_offset = fs->lfs_curseg;
   1579      1.163  perseant 
   1580      1.163  perseant 	return 0;
   1581      1.163  perseant }
   1582      1.163  perseant 
   1583      1.163  perseant /*
   1584      1.139      yamt  * Start a new partial segment.
   1585      1.139      yamt  *
   1586      1.139      yamt  * Return 1 when we entered to a new segment.
   1587      1.139      yamt  * Otherwise, return 0.
   1588        1.1   mycroft  */
   1589        1.1   mycroft int
   1590       1.69  perseant lfs_initseg(struct lfs *fs)
   1591        1.1   mycroft {
   1592      1.139      yamt 	struct segment *sp = fs->lfs_sp;
   1593        1.1   mycroft 	SEGSUM *ssp;
   1594      1.139      yamt 	struct buf *sbp;	/* buffer for SEGSUM */
   1595      1.139      yamt 	int repeat = 0;		/* return value */
   1596      1.107  perseant 
   1597      1.159  perseant 	ASSERT_SEGLOCK(fs);
   1598        1.1   mycroft 	/* Advance to the next segment. */
   1599        1.1   mycroft 	if (!LFS_PARTIAL_FITS(fs)) {
   1600      1.139      yamt 		SEGUSE *sup;
   1601      1.139      yamt 		struct buf *bp;
   1602      1.139      yamt 
   1603       1.55  perseant 		/* lfs_avail eats the remaining space */
   1604       1.69  perseant 		fs->lfs_avail -= fs->lfs_fsbpseg - (fs->lfs_offset -
   1605       1.55  perseant 						   fs->lfs_curseg);
   1606        1.1   mycroft 		/* Wake up any cleaning procs waiting on this file system. */
   1607        1.1   mycroft 		wakeup(&lfs_allclean_wakeup);
   1608       1.10      fvdl 		wakeup(&fs->lfs_nextseg);
   1609        1.1   mycroft 		lfs_newseg(fs);
   1610        1.1   mycroft 		repeat = 1;
   1611        1.1   mycroft 		fs->lfs_offset = fs->lfs_curseg;
   1612      1.157     perry 
   1613       1.69  perseant 		sp->seg_number = dtosn(fs, fs->lfs_curseg);
   1614       1.69  perseant 		sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg);
   1615      1.101  perseant 
   1616        1.1   mycroft 		/*
   1617        1.1   mycroft 		 * If the segment contains a superblock, update the offset
   1618        1.1   mycroft 		 * and summary address to skip over it.
   1619        1.1   mycroft 		 */
   1620        1.1   mycroft 		LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
   1621        1.1   mycroft 		if (sup->su_flags & SEGUSE_SUPERBLOCK) {
   1622       1.69  perseant 			fs->lfs_offset += btofsb(fs, LFS_SBPAD);
   1623        1.1   mycroft 			sp->seg_bytes_left -= LFS_SBPAD;
   1624        1.1   mycroft 		}
   1625        1.1   mycroft 		brelse(bp);
   1626       1.69  perseant 		/* Segment zero could also contain the labelpad */
   1627       1.69  perseant 		if (fs->lfs_version > 1 && sp->seg_number == 0 &&
   1628       1.69  perseant 		    fs->lfs_start < btofsb(fs, LFS_LABELPAD)) {
   1629      1.128      yamt 			fs->lfs_offset +=
   1630      1.128      yamt 			    btofsb(fs, LFS_LABELPAD) - fs->lfs_start;
   1631      1.128      yamt 			sp->seg_bytes_left -=
   1632      1.128      yamt 			    LFS_LABELPAD - fsbtob(fs, fs->lfs_start);
   1633       1.69  perseant 		}
   1634        1.1   mycroft 	} else {
   1635       1.69  perseant 		sp->seg_number = dtosn(fs, fs->lfs_curseg);
   1636       1.69  perseant 		sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg -
   1637       1.58  perseant 				      (fs->lfs_offset - fs->lfs_curseg));
   1638        1.1   mycroft 	}
   1639        1.1   mycroft 	fs->lfs_lastpseg = fs->lfs_offset;
   1640      1.157     perry 
   1641      1.107  perseant 	/* Record first address of this partial segment */
   1642      1.107  perseant 	if (sp->seg_flags & SEGM_CLEAN) {
   1643      1.107  perseant 		fs->lfs_cleanint[fs->lfs_cleanind] = fs->lfs_offset;
   1644      1.107  perseant 		if (++fs->lfs_cleanind >= LFS_MAX_CLEANIND) {
   1645      1.107  perseant 			/* "1" is the artificial inc in lfs_seglock */
   1646      1.159  perseant 			simple_lock(&fs->lfs_interlock);
   1647      1.107  perseant 			while (fs->lfs_iocount > 1) {
   1648      1.159  perseant 				ltsleep(&fs->lfs_iocount, PRIBIO + 1,
   1649      1.159  perseant 				    "lfs_initseg", 0, &fs->lfs_interlock);
   1650      1.107  perseant 			}
   1651      1.159  perseant 			simple_unlock(&fs->lfs_interlock);
   1652      1.107  perseant 			fs->lfs_cleanind = 0;
   1653      1.107  perseant 		}
   1654      1.107  perseant 	}
   1655      1.107  perseant 
   1656        1.1   mycroft 	sp->fs = fs;
   1657        1.1   mycroft 	sp->ibp = NULL;
   1658       1.27  perseant 	sp->idp = NULL;
   1659        1.1   mycroft 	sp->ninodes = 0;
   1660       1.80  perseant 	sp->ndupino = 0;
   1661       1.69  perseant 
   1662        1.1   mycroft 	sp->cbpp = sp->bpp;
   1663      1.139      yamt 
   1664      1.139      yamt 	/* Get a new buffer for SEGSUM */
   1665      1.145      yamt 	sbp = lfs_newbuf(fs, VTOI(fs->lfs_ivnode)->i_devvp,
   1666      1.128      yamt 	    fsbtodb(fs, fs->lfs_offset), fs->lfs_sumsize, LFS_NB_SUMMARY);
   1667      1.139      yamt 
   1668      1.139      yamt 	/* ... and enter it into the buffer list. */
   1669      1.139      yamt 	*sp->cbpp = sbp;
   1670      1.139      yamt 	sp->cbpp++;
   1671       1.69  perseant 	fs->lfs_offset += btofsb(fs, fs->lfs_sumsize);
   1672      1.139      yamt 
   1673      1.139      yamt 	sp->start_bpp = sp->cbpp;
   1674      1.157     perry 
   1675        1.1   mycroft 	/* Set point to SEGSUM, initialize it. */
   1676      1.139      yamt 	ssp = sp->segsum = sbp->b_data;
   1677      1.139      yamt 	memset(ssp, 0, fs->lfs_sumsize);
   1678        1.1   mycroft 	ssp->ss_next = fs->lfs_nextseg;
   1679        1.1   mycroft 	ssp->ss_nfinfo = ssp->ss_ninos = 0;
   1680       1.10      fvdl 	ssp->ss_magic = SS_MAGIC;
   1681        1.1   mycroft 
   1682        1.1   mycroft 	/* Set pointer to first FINFO, initialize it. */
   1683       1.69  perseant 	sp->fip = (struct finfo *)((caddr_t)sp->segsum + SEGSUM_SIZE(fs));
   1684        1.1   mycroft 	sp->fip->fi_nblocks = 0;
   1685        1.1   mycroft 	sp->start_lbp = &sp->fip->fi_blocks[0];
   1686       1.10      fvdl 	sp->fip->fi_lastlength = 0;
   1687      1.157     perry 
   1688       1.69  perseant 	sp->seg_bytes_left -= fs->lfs_sumsize;
   1689       1.69  perseant 	sp->sum_bytes_left = fs->lfs_sumsize - SEGSUM_SIZE(fs);
   1690      1.104  perseant 
   1691       1.73       chs 	return (repeat);
   1692        1.1   mycroft }
   1693        1.1   mycroft 
   1694        1.1   mycroft /*
   1695      1.163  perseant  * Remove SEGUSE_INVAL from all segments.
   1696      1.163  perseant  */
   1697      1.163  perseant void
   1698      1.163  perseant lfs_unset_inval_all(struct lfs *fs)
   1699      1.163  perseant {
   1700      1.163  perseant 	SEGUSE *sup;
   1701      1.163  perseant 	struct buf *bp;
   1702      1.163  perseant 	int i;
   1703      1.163  perseant 
   1704      1.163  perseant 	for (i = 0; i < fs->lfs_nseg; i++) {
   1705      1.163  perseant 		LFS_SEGENTRY(sup, fs, i, bp);
   1706      1.163  perseant 		if (sup->su_flags & SEGUSE_INVAL) {
   1707      1.163  perseant 			sup->su_flags &= ~SEGUSE_INVAL;
   1708      1.163  perseant 			VOP_BWRITE(bp);
   1709      1.163  perseant 		} else
   1710      1.163  perseant 			brelse(bp);
   1711      1.163  perseant 	}
   1712      1.163  perseant }
   1713      1.163  perseant 
   1714      1.163  perseant /*
   1715        1.1   mycroft  * Return the next segment to write.
   1716        1.1   mycroft  */
   1717        1.1   mycroft void
   1718       1.69  perseant lfs_newseg(struct lfs *fs)
   1719        1.1   mycroft {
   1720        1.1   mycroft 	CLEANERINFO *cip;
   1721        1.1   mycroft 	SEGUSE *sup;
   1722        1.1   mycroft 	struct buf *bp;
   1723      1.163  perseant 	int curseg, isdirty, sn, skip_inval;
   1724      1.157     perry 
   1725      1.159  perseant 	ASSERT_SEGLOCK(fs);
   1726      1.175  perseant 
   1727      1.175  perseant 	/* Honor LFCNWRAPSTOP */
   1728      1.175  perseant 	simple_lock(&fs->lfs_interlock);
   1729      1.175  perseant 	if (fs->lfs_nowrap && fs->lfs_nextseg < fs->lfs_curseg) {
   1730      1.175  perseant 		wakeup(&fs->lfs_nowrap);
   1731      1.175  perseant 		ltsleep(&fs->lfs_nowrap, PVFS, "newseg", 0,
   1732      1.175  perseant 			&fs->lfs_interlock);
   1733      1.175  perseant 	}
   1734      1.175  perseant 	simple_unlock(&fs->lfs_interlock);
   1735      1.175  perseant 
   1736       1.69  perseant 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp);
   1737      1.158  perseant 	DLOG((DLOG_SU, "lfs_newseg: seg %d := 0 in newseg\n",
   1738      1.158  perseant 	      dtosn(fs, fs->lfs_nextseg)));
   1739       1.15  perseant 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
   1740        1.1   mycroft 	sup->su_nbytes = 0;
   1741        1.1   mycroft 	sup->su_nsums = 0;
   1742        1.1   mycroft 	sup->su_ninos = 0;
   1743      1.101  perseant 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp);
   1744        1.1   mycroft 
   1745        1.1   mycroft 	LFS_CLEANERINFO(cip, fs, bp);
   1746        1.1   mycroft 	--cip->clean;
   1747        1.1   mycroft 	++cip->dirty;
   1748       1.15  perseant 	fs->lfs_nclean = cip->clean;
   1749       1.61  perseant 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
   1750      1.101  perseant 
   1751        1.1   mycroft 	fs->lfs_lastseg = fs->lfs_curseg;
   1752        1.1   mycroft 	fs->lfs_curseg = fs->lfs_nextseg;
   1753      1.163  perseant 	skip_inval = 1;
   1754       1.69  perseant 	for (sn = curseg = dtosn(fs, fs->lfs_curseg) + fs->lfs_interleave;;) {
   1755        1.1   mycroft 		sn = (sn + 1) % fs->lfs_nseg;
   1756      1.174  perseant 
   1757      1.163  perseant 		if (sn == curseg) {
   1758      1.163  perseant 			if (skip_inval)
   1759      1.163  perseant 				skip_inval = 0;
   1760      1.163  perseant 			else
   1761      1.163  perseant 				panic("lfs_nextseg: no clean segments");
   1762      1.163  perseant 		}
   1763        1.1   mycroft 		LFS_SEGENTRY(sup, fs, sn, bp);
   1764      1.163  perseant 		isdirty = sup->su_flags & (SEGUSE_DIRTY | (skip_inval ? SEGUSE_INVAL : 0));
   1765      1.101  perseant 		/* Check SEGUSE_EMPTY as we go along */
   1766      1.128      yamt 		if (isdirty && sup->su_nbytes == 0 &&
   1767      1.128      yamt 		    !(sup->su_flags & SEGUSE_EMPTY))
   1768      1.101  perseant 			LFS_WRITESEGENTRY(sup, fs, sn, bp);
   1769      1.101  perseant 		else
   1770      1.101  perseant 			brelse(bp);
   1771      1.101  perseant 
   1772        1.1   mycroft 		if (!isdirty)
   1773        1.1   mycroft 			break;
   1774        1.1   mycroft 	}
   1775      1.163  perseant 	if (skip_inval == 0)
   1776      1.163  perseant 		lfs_unset_inval_all(fs);
   1777      1.157     perry 
   1778        1.1   mycroft 	++fs->lfs_nactive;
   1779       1.69  perseant 	fs->lfs_nextseg = sntod(fs, sn);
   1780       1.73       chs 	if (lfs_dostats) {
   1781       1.15  perseant 		++lfs_stats.segsused;
   1782       1.15  perseant 	}
   1783        1.1   mycroft }
   1784        1.1   mycroft 
   1785       1.74  perseant static struct buf *
   1786       1.74  perseant lfs_newclusterbuf(struct lfs *fs, struct vnode *vp, daddr_t addr, int n)
   1787       1.74  perseant {
   1788       1.74  perseant 	struct lfs_cluster *cl;
   1789       1.74  perseant 	struct buf **bpp, *bp;
   1790       1.74  perseant 
   1791      1.159  perseant 	ASSERT_SEGLOCK(fs);
   1792      1.101  perseant 	cl = (struct lfs_cluster *)pool_get(&fs->lfs_clpool, PR_WAITOK);
   1793      1.101  perseant 	bpp = (struct buf **)pool_get(&fs->lfs_bpppool, PR_WAITOK);
   1794       1.79  perseant 	memset(cl, 0, sizeof(*cl));
   1795       1.74  perseant 	cl->fs = fs;
   1796       1.74  perseant 	cl->bpp = bpp;
   1797       1.74  perseant 	cl->bufcount = 0;
   1798       1.74  perseant 	cl->bufsize = 0;
   1799       1.74  perseant 
   1800       1.79  perseant 	/* If this segment is being written synchronously, note that */
   1801       1.79  perseant 	if (fs->lfs_sp->seg_flags & SEGM_SYNC) {
   1802       1.79  perseant 		cl->flags |= LFS_CL_SYNC;
   1803       1.79  perseant 		cl->seg = fs->lfs_sp;
   1804       1.79  perseant 		++cl->seg->seg_iocount;
   1805       1.79  perseant 	}
   1806       1.79  perseant 
   1807       1.74  perseant 	/* Get an empty buffer header, or maybe one with something on it */
   1808      1.169      yamt 	bp = getiobuf();
   1809      1.130      yamt 	bp->b_flags = B_BUSY | B_CALL;
   1810      1.130      yamt 	bp->b_dev = NODEV;
   1811       1.74  perseant 	bp->b_blkno = bp->b_lblkno = addr;
   1812       1.74  perseant 	bp->b_iodone = lfs_cluster_callback;
   1813      1.144      yamt 	bp->b_private = cl;
   1814      1.130      yamt 	bp->b_vp = vp;
   1815       1.74  perseant 
   1816       1.74  perseant 	return bp;
   1817       1.74  perseant }
   1818       1.74  perseant 
   1819        1.1   mycroft int
   1820       1.69  perseant lfs_writeseg(struct lfs *fs, struct segment *sp)
   1821        1.1   mycroft {
   1822      1.109  perseant 	struct buf **bpp, *bp, *cbp, *newbp;
   1823        1.1   mycroft 	SEGUSE *sup;
   1824        1.1   mycroft 	SEGSUM *ssp;
   1825      1.101  perseant 	int i, s;
   1826      1.101  perseant 	int do_again, nblocks, byteoffset;
   1827       1.70  jdolecek 	size_t el_size;
   1828      1.103  perseant 	struct lfs_cluster *cl;
   1829        1.1   mycroft 	u_short ninos;
   1830       1.15  perseant 	struct vnode *devvp;
   1831      1.142  christos 	char *p = NULL;
   1832       1.69  perseant 	struct vnode *vp;
   1833       1.91      fvdl 	int32_t *daddrp;	/* XXX ondisk32 */
   1834       1.55  perseant 	int changed;
   1835      1.152      yamt 	u_int32_t sum;
   1836      1.180  perseant #ifdef DEBUG
   1837      1.180  perseant 	FINFO *fip;
   1838      1.180  perseant 	int findex;
   1839      1.180  perseant #endif
   1840       1.74  perseant 
   1841      1.159  perseant 	ASSERT_SEGLOCK(fs);
   1842        1.1   mycroft 	/*
   1843        1.1   mycroft 	 * If there are no buffers other than the segment summary to write
   1844        1.1   mycroft 	 * and it is not a checkpoint, don't do anything.  On a checkpoint,
   1845        1.1   mycroft 	 * even if there aren't any buffers, you need to write the superblock.
   1846        1.1   mycroft 	 */
   1847        1.1   mycroft 	if ((nblocks = sp->cbpp - sp->bpp) == 1)
   1848        1.1   mycroft 		return (0);
   1849      1.157     perry 
   1850       1.27  perseant 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
   1851       1.27  perseant 
   1852       1.10      fvdl 	/* Update the segment usage information. */
   1853       1.10      fvdl 	LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
   1854      1.157     perry 
   1855       1.10      fvdl 	/* Loop through all blocks, except the segment summary. */
   1856       1.27  perseant 	for (bpp = sp->bpp; ++bpp < sp->cbpp; ) {
   1857       1.73       chs 		if ((*bpp)->b_vp != devvp) {
   1858       1.27  perseant 			sup->su_nbytes += (*bpp)->b_bcount;
   1859      1.158  perseant 			DLOG((DLOG_SU, "seg %" PRIu32 " += %ld for ino %d"
   1860      1.158  perseant 			      " lbn %" PRId64 " db 0x%" PRIx64 "\n",
   1861      1.158  perseant 			      sp->seg_number, (*bpp)->b_bcount,
   1862      1.158  perseant 			      VTOI((*bpp)->b_vp)->i_number, (*bpp)->b_lblkno,
   1863      1.158  perseant 			      (*bpp)->b_blkno));
   1864       1.69  perseant 		}
   1865       1.27  perseant 	}
   1866      1.157     perry 
   1867        1.1   mycroft 	ssp = (SEGSUM *)sp->segsum;
   1868      1.157     perry 
   1869      1.180  perseant #ifdef DEBUG
   1870  1.180.2.1      chap 	/* Check for zero-length and zero-version FINFO entries. */
   1871      1.180  perseant 	fip = (struct finfo *)((caddr_t)ssp + SEGSUM_SIZE(fs));
   1872      1.180  perseant 	for (findex = 0; findex < ssp->ss_nfinfo; findex++) {
   1873      1.180  perseant 		KDASSERT(fip->fi_nblocks > 0);
   1874  1.180.2.1      chap 		KDASSERT(fip->fi_version > 0);
   1875      1.180  perseant 		fip = (FINFO *)((caddr_t)fip + FINFOSIZE +
   1876      1.180  perseant 			sizeof(int32_t) * fip->fi_nblocks);
   1877      1.180  perseant 	}
   1878      1.180  perseant #endif /* DEBUG */
   1879      1.180  perseant 
   1880        1.1   mycroft 	ninos = (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs);
   1881      1.158  perseant 	DLOG((DLOG_SU, "seg %d += %d for %d inodes\n",
   1882      1.158  perseant 	      sp->seg_number, ssp->ss_ninos * sizeof (struct ufs1_dinode),
   1883      1.158  perseant 	      ssp->ss_ninos));
   1884      1.119      fvdl 	sup->su_nbytes += ssp->ss_ninos * sizeof (struct ufs1_dinode);
   1885       1.69  perseant 	/* sup->su_nbytes += fs->lfs_sumsize; */
   1886       1.69  perseant 	if (fs->lfs_version == 1)
   1887  1.180.2.1      chap 		sup->su_olastmod = time_second;
   1888       1.69  perseant 	else
   1889  1.180.2.1      chap 		sup->su_lastmod = time_second;
   1890        1.1   mycroft 	sup->su_ninos += ninos;
   1891        1.1   mycroft 	++sup->su_nsums;
   1892       1.69  perseant 	fs->lfs_avail -= btofsb(fs, fs->lfs_sumsize);
   1893       1.15  perseant 
   1894        1.1   mycroft 	do_again = !(bp->b_flags & B_GATHERED);
   1895      1.101  perseant 	LFS_WRITESEGENTRY(sup, fs, sp->seg_number, bp); /* Ifile */
   1896      1.101  perseant 
   1897        1.1   mycroft 	/*
   1898       1.53  perseant 	 * Mark blocks B_BUSY, to prevent then from being changed between
   1899       1.53  perseant 	 * the checksum computation and the actual write.
   1900       1.53  perseant 	 *
   1901       1.53  perseant 	 * If we are cleaning, check indirect blocks for UNWRITTEN, and if
   1902       1.53  perseant 	 * there are any, replace them with copies that have UNASSIGNED
   1903       1.53  perseant 	 * instead.
   1904       1.53  perseant 	 */
   1905       1.53  perseant 	for (bpp = sp->bpp, i = nblocks - 1; i--;) {
   1906       1.53  perseant 		++bpp;
   1907      1.101  perseant 		bp = *bpp;
   1908      1.101  perseant 		if (bp->b_flags & B_CALL) { /* UBC or malloced buffer */
   1909      1.101  perseant 			bp->b_flags |= B_BUSY;
   1910       1.53  perseant 			continue;
   1911      1.101  perseant 		}
   1912      1.159  perseant 
   1913      1.159  perseant 		simple_lock(&bp->b_interlock);
   1914       1.53  perseant 		s = splbio();
   1915      1.159  perseant 		while (bp->b_flags & B_BUSY) {
   1916      1.158  perseant 			DLOG((DLOG_SEG, "lfs_writeseg: avoiding potential"
   1917      1.158  perseant 			      " data summary corruption for ino %d, lbn %"
   1918      1.158  perseant 			      PRId64 "\n",
   1919      1.158  perseant 			      VTOI(bp->b_vp)->i_number, bp->b_lblkno));
   1920       1.53  perseant 			bp->b_flags |= B_WANTED;
   1921      1.159  perseant 			ltsleep(bp, (PRIBIO + 1), "lfs_writeseg", 0,
   1922      1.159  perseant 				&bp->b_interlock);
   1923       1.53  perseant 			splx(s);
   1924      1.159  perseant 			s = splbio();
   1925       1.53  perseant 		}
   1926       1.53  perseant 		bp->b_flags |= B_BUSY;
   1927       1.53  perseant 		splx(s);
   1928      1.159  perseant 		simple_unlock(&bp->b_interlock);
   1929      1.159  perseant 
   1930      1.101  perseant 		/*
   1931      1.101  perseant 		 * Check and replace indirect block UNWRITTEN bogosity.
   1932      1.101  perseant 		 * XXX See comment in lfs_writefile.
   1933      1.101  perseant 		 */
   1934       1.73       chs 		if (bp->b_lblkno < 0 && bp->b_vp != devvp && bp->b_vp &&
   1935      1.119      fvdl 		   VTOI(bp->b_vp)->i_ffs1_blocks !=
   1936       1.53  perseant 		   VTOI(bp->b_vp)->i_lfs_effnblks) {
   1937      1.158  perseant 			DLOG((DLOG_VNODE, "lfs_writeseg: cleansing ino %d (%d != %d)\n",
   1938      1.158  perseant 			      VTOI(bp->b_vp)->i_number,
   1939      1.158  perseant 			      VTOI(bp->b_vp)->i_lfs_effnblks,
   1940      1.158  perseant 			      VTOI(bp->b_vp)->i_ffs1_blocks));
   1941       1.53  perseant 			/* Make a copy we'll make changes to */
   1942       1.69  perseant 			newbp = lfs_newbuf(fs, bp->b_vp, bp->b_lblkno,
   1943      1.101  perseant 					   bp->b_bcount, LFS_NB_IBLOCK);
   1944       1.53  perseant 			newbp->b_blkno = bp->b_blkno;
   1945       1.53  perseant 			memcpy(newbp->b_data, bp->b_data,
   1946       1.53  perseant 			       newbp->b_bcount);
   1947       1.53  perseant 
   1948       1.55  perseant 			changed = 0;
   1949       1.91      fvdl 			/* XXX ondisk32 */
   1950       1.91      fvdl 			for (daddrp = (int32_t *)(newbp->b_data);
   1951       1.91      fvdl 			     daddrp < (int32_t *)(newbp->b_data +
   1952       1.53  perseant 						  newbp->b_bcount); daddrp++) {
   1953       1.53  perseant 				if (*daddrp == UNWRITTEN) {
   1954       1.55  perseant 					++changed;
   1955       1.53  perseant 					*daddrp = 0;
   1956       1.53  perseant 				}
   1957       1.53  perseant 			}
   1958       1.55  perseant 			/*
   1959       1.55  perseant 			 * Get rid of the old buffer.  Don't mark it clean,
   1960       1.55  perseant 			 * though, if it still has dirty data on it.
   1961       1.55  perseant 			 */
   1962       1.55  perseant 			if (changed) {
   1963      1.158  perseant 				DLOG((DLOG_SEG, "lfs_writeseg: replacing UNWRITTEN(%d):"
   1964      1.158  perseant 				      " bp = %p newbp = %p\n", changed, bp,
   1965      1.158  perseant 				      newbp));
   1966      1.101  perseant 				*bpp = newbp;
   1967      1.120  perseant 				bp->b_flags &= ~(B_ERROR | B_GATHERED);
   1968       1.69  perseant 				if (bp->b_flags & B_CALL) {
   1969      1.158  perseant 					DLOG((DLOG_SEG, "lfs_writeseg: "
   1970      1.158  perseant 					      "indir bp should not be B_CALL\n"));
   1971      1.101  perseant 					s = splbio();
   1972      1.101  perseant 					biodone(bp);
   1973      1.101  perseant 					splx(s);
   1974       1.69  perseant 					bp = NULL;
   1975       1.69  perseant 				} else {
   1976       1.57  perseant 					/* Still on free list, leave it there */
   1977       1.57  perseant 					s = splbio();
   1978       1.57  perseant 					bp->b_flags &= ~B_BUSY;
   1979       1.57  perseant 					if (bp->b_flags & B_WANTED)
   1980       1.57  perseant 						wakeup(bp);
   1981      1.103  perseant 					splx(s);
   1982       1.62  perseant 					/*
   1983       1.62  perseant 					 * We have to re-decrement lfs_avail
   1984       1.62  perseant 					 * since this block is going to come
   1985       1.62  perseant 					 * back around to us in the next
   1986       1.62  perseant 					 * segment.
   1987       1.62  perseant 					 */
   1988      1.128      yamt 					fs->lfs_avail -=
   1989      1.128      yamt 					    btofsb(fs, bp->b_bcount);
   1990       1.57  perseant 				}
   1991       1.55  perseant 			} else {
   1992      1.101  perseant 				lfs_freebuf(fs, newbp);
   1993       1.55  perseant 			}
   1994       1.53  perseant 		}
   1995       1.53  perseant 	}
   1996       1.53  perseant 	/*
   1997        1.1   mycroft 	 * Compute checksum across data and then across summary; the first
   1998        1.1   mycroft 	 * block (the summary block) is skipped.  Set the create time here
   1999        1.1   mycroft 	 * so that it's guaranteed to be later than the inode mod times.
   2000        1.1   mycroft 	 */
   2001      1.152      yamt 	sum = 0;
   2002       1.69  perseant 	if (fs->lfs_version == 1)
   2003       1.69  perseant 		el_size = sizeof(u_long);
   2004       1.69  perseant 	else
   2005       1.69  perseant 		el_size = sizeof(u_int32_t);
   2006      1.101  perseant 	for (bpp = sp->bpp, i = nblocks - 1; i--; ) {
   2007      1.101  perseant 		++bpp;
   2008      1.101  perseant 		/* Loop through gop_write cluster blocks */
   2009      1.101  perseant 		for (byteoffset = 0; byteoffset < (*bpp)->b_bcount;
   2010      1.101  perseant 		     byteoffset += fs->lfs_bsize) {
   2011      1.109  perseant #ifdef LFS_USE_B_INVAL
   2012      1.101  perseant 			if (((*bpp)->b_flags & (B_CALL | B_INVAL)) ==
   2013      1.101  perseant 			    (B_CALL | B_INVAL)) {
   2014      1.101  perseant 				if (copyin((caddr_t)(*bpp)->b_saveaddr +
   2015      1.101  perseant 					   byteoffset, dp, el_size)) {
   2016      1.128      yamt 					panic("lfs_writeseg: copyin failed [1]:"
   2017      1.128      yamt 						" ino %d blk %" PRId64,
   2018      1.101  perseant 						VTOI((*bpp)->b_vp)->i_number,
   2019      1.101  perseant 						(*bpp)->b_lblkno);
   2020      1.101  perseant 				}
   2021      1.109  perseant 			} else
   2022      1.109  perseant #endif /* LFS_USE_B_INVAL */
   2023      1.109  perseant 			{
   2024      1.152      yamt 				sum = lfs_cksum_part(
   2025      1.152      yamt 				    (*bpp)->b_data + byteoffset, el_size, sum);
   2026      1.101  perseant 			}
   2027      1.101  perseant 		}
   2028       1.69  perseant 	}
   2029       1.69  perseant 	if (fs->lfs_version == 1)
   2030  1.180.2.1      chap 		ssp->ss_ocreate = time_second;
   2031       1.69  perseant 	else {
   2032  1.180.2.1      chap 		ssp->ss_create = time_second;
   2033       1.69  perseant 		ssp->ss_serial = ++fs->lfs_serial;
   2034       1.69  perseant 		ssp->ss_ident  = fs->lfs_ident;
   2035        1.1   mycroft 	}
   2036      1.152      yamt 	ssp->ss_datasum = lfs_cksum_fold(sum);
   2037      1.152      yamt 	ssp->ss_sumsum = cksum(&ssp->ss_datasum,
   2038      1.152      yamt 	    fs->lfs_sumsize - sizeof(ssp->ss_sumsum));
   2039      1.156  perseant 
   2040      1.159  perseant 	simple_lock(&fs->lfs_interlock);
   2041       1.69  perseant 	fs->lfs_bfree -= (btofsb(fs, ninos * fs->lfs_ibsize) +
   2042       1.69  perseant 			  btofsb(fs, fs->lfs_sumsize));
   2043      1.170       tls 	fs->lfs_dmeta += (btofsb(fs, ninos * fs->lfs_ibsize) +
   2044      1.170       tls 			  btofsb(fs, fs->lfs_sumsize));
   2045      1.159  perseant 	simple_unlock(&fs->lfs_interlock);
   2046        1.1   mycroft 
   2047        1.1   mycroft 	/*
   2048      1.103  perseant 	 * When we simply write the blocks we lose a rotation for every block
   2049      1.109  perseant 	 * written.  To avoid this problem, we cluster the buffers into a
   2050      1.109  perseant 	 * chunk and write the chunk.  MAXPHYS is the largest size I/O
   2051      1.109  perseant 	 * devices can handle, use that for the size of the chunks.
   2052      1.103  perseant 	 *
   2053      1.109  perseant 	 * Blocks that are already clusters (from GOP_WRITE), however, we
   2054      1.109  perseant 	 * don't bother to copy into other clusters.
   2055        1.1   mycroft 	 */
   2056       1.15  perseant 
   2057       1.15  perseant #define CHUNKSIZE MAXPHYS
   2058       1.15  perseant 
   2059       1.73       chs 	if (devvp == NULL)
   2060       1.15  perseant 		panic("devvp is NULL");
   2061       1.74  perseant 	for (bpp = sp->bpp, i = nblocks; i;) {
   2062       1.74  perseant 		cbp = lfs_newclusterbuf(fs, devvp, (*bpp)->b_blkno, i);
   2063      1.144      yamt 		cl = cbp->b_private;
   2064       1.74  perseant 
   2065        1.1   mycroft 		cbp->b_flags |= B_ASYNC | B_BUSY;
   2066       1.10      fvdl 		cbp->b_bcount = 0;
   2067        1.1   mycroft 
   2068       1.74  perseant #if defined(DEBUG) && defined(DIAGNOSTIC)
   2069      1.104  perseant 		if (bpp - sp->bpp > (fs->lfs_sumsize - SEGSUM_SIZE(fs))
   2070      1.104  perseant 		    / sizeof(int32_t)) {
   2071      1.104  perseant 			panic("lfs_writeseg: real bpp overwrite");
   2072      1.104  perseant 		}
   2073      1.151      yamt 		if (bpp - sp->bpp > segsize(fs) / fs->lfs_fsize) {
   2074      1.104  perseant 			panic("lfs_writeseg: theoretical bpp overwrite");
   2075      1.104  perseant 		}
   2076       1.17  perseant #endif
   2077       1.17  perseant 
   2078       1.74  perseant 		/*
   2079       1.74  perseant 		 * Construct the cluster.
   2080       1.74  perseant 		 */
   2081      1.159  perseant 		simple_lock(&fs->lfs_interlock);
   2082        1.1   mycroft 		++fs->lfs_iocount;
   2083      1.159  perseant 		simple_unlock(&fs->lfs_interlock);
   2084      1.115  perseant 		while (i && cbp->b_bcount < CHUNKSIZE) {
   2085       1.10      fvdl 			bp = *bpp;
   2086       1.15  perseant 
   2087       1.15  perseant 			if (bp->b_bcount > (CHUNKSIZE - cbp->b_bcount))
   2088       1.10      fvdl 				break;
   2089      1.109  perseant 			if (cbp->b_bcount > 0 && !(cl->flags & LFS_CL_MALLOC))
   2090      1.109  perseant 				break;
   2091       1.10      fvdl 
   2092      1.109  perseant 			/* Clusters from GOP_WRITE are expedited */
   2093      1.109  perseant 			if (bp->b_bcount > fs->lfs_bsize) {
   2094      1.109  perseant 				if (cbp->b_bcount > 0)
   2095      1.109  perseant 					/* Put in its own buffer */
   2096      1.109  perseant 					break;
   2097      1.109  perseant 				else {
   2098      1.109  perseant 					cbp->b_data = bp->b_data;
   2099      1.109  perseant 				}
   2100      1.109  perseant 			} else if (cbp->b_bcount == 0) {
   2101      1.109  perseant 				p = cbp->b_data = lfs_malloc(fs, CHUNKSIZE,
   2102      1.109  perseant 							     LFS_NB_CLUSTER);
   2103      1.109  perseant 				cl->flags |= LFS_CL_MALLOC;
   2104      1.109  perseant 			}
   2105      1.101  perseant #ifdef DIAGNOSTIC
   2106      1.101  perseant 			if (dtosn(fs, dbtofsb(fs, bp->b_blkno +
   2107      1.101  perseant 					      btodb(bp->b_bcount - 1))) !=
   2108      1.101  perseant 			    sp->seg_number) {
   2109      1.155      yamt 				printf("blk size %d daddr %" PRIx64
   2110      1.128      yamt 				    " not in seg %d\n",
   2111      1.128      yamt 				    bp->b_bcount, bp->b_blkno,
   2112      1.128      yamt 				    sp->seg_number);
   2113      1.101  perseant 				panic("segment overwrite");
   2114      1.101  perseant 			}
   2115      1.101  perseant #endif
   2116      1.101  perseant 
   2117      1.109  perseant #ifdef LFS_USE_B_INVAL
   2118        1.1   mycroft 			/*
   2119        1.1   mycroft 			 * Fake buffers from the cleaner are marked as B_INVAL.
   2120        1.1   mycroft 			 * We need to copy the data from user space rather than
   2121        1.1   mycroft 			 * from the buffer indicated.
   2122        1.1   mycroft 			 * XXX == what do I do on an error?
   2123        1.1   mycroft 			 */
   2124      1.128      yamt 			if ((bp->b_flags & (B_CALL|B_INVAL)) ==
   2125      1.128      yamt 			    (B_CALL|B_INVAL)) {
   2126        1.1   mycroft 				if (copyin(bp->b_saveaddr, p, bp->b_bcount))
   2127      1.128      yamt 					panic("lfs_writeseg: "
   2128      1.128      yamt 					    "copyin failed [2]");
   2129      1.109  perseant 			} else
   2130      1.109  perseant #endif /* LFS_USE_B_INVAL */
   2131      1.109  perseant 			if (cl->flags & LFS_CL_MALLOC) {
   2132      1.137      yamt 				/* copy data into our cluster. */
   2133      1.137      yamt 				memcpy(p, bp->b_data, bp->b_bcount);
   2134      1.137      yamt 				p += bp->b_bcount;
   2135      1.103  perseant 			}
   2136      1.157     perry 
   2137      1.109  perseant 			cbp->b_bcount += bp->b_bcount;
   2138      1.109  perseant 			cl->bufsize += bp->b_bcount;
   2139      1.109  perseant 
   2140       1.74  perseant 			bp->b_flags &= ~(B_ERROR | B_READ | B_DELWRI | B_DONE);
   2141       1.74  perseant 			cl->bpp[cl->bufcount++] = bp;
   2142       1.74  perseant 			vp = bp->b_vp;
   2143       1.79  perseant 			s = splbio();
   2144      1.135      yamt 			reassignbuf(bp, vp);
   2145      1.100        pk 			V_INCR_NUMOUTPUT(vp);
   2146       1.79  perseant 			splx(s);
   2147       1.26  perseant 
   2148       1.26  perseant 			bpp++;
   2149      1.109  perseant 			i--;
   2150        1.1   mycroft 		}
   2151      1.147      yamt 		if (fs->lfs_sp->seg_flags & SEGM_SYNC)
   2152      1.147      yamt 			BIO_SETPRIO(cbp, BPRIO_TIMECRITICAL);
   2153      1.147      yamt 		else
   2154      1.147      yamt 			BIO_SETPRIO(cbp, BPRIO_TIMELIMITED);
   2155       1.79  perseant 		s = splbio();
   2156      1.109  perseant 		V_INCR_NUMOUTPUT(devvp);
   2157        1.1   mycroft 		splx(s);
   2158      1.149      yamt 		VOP_STRATEGY(devvp, cbp);
   2159      1.109  perseant 		curproc->p_stats->p_ru.ru_oublock++;
   2160        1.1   mycroft 	}
   2161       1.74  perseant 
   2162       1.73       chs 	if (lfs_dostats) {
   2163       1.15  perseant 		++lfs_stats.psegwrites;
   2164       1.15  perseant 		lfs_stats.blocktot += nblocks - 1;
   2165       1.15  perseant 		if (fs->lfs_sp->seg_flags & SEGM_SYNC)
   2166       1.15  perseant 			++lfs_stats.psyncwrites;
   2167       1.15  perseant 		if (fs->lfs_sp->seg_flags & SEGM_CLEAN) {
   2168       1.15  perseant 			++lfs_stats.pcleanwrites;
   2169       1.15  perseant 			lfs_stats.cleanblocks += nblocks - 1;
   2170       1.15  perseant 		}
   2171        1.1   mycroft 	}
   2172        1.1   mycroft 	return (lfs_initseg(fs) || do_again);
   2173        1.1   mycroft }
   2174        1.1   mycroft 
   2175        1.1   mycroft void
   2176       1.69  perseant lfs_writesuper(struct lfs *fs, daddr_t daddr)
   2177        1.1   mycroft {
   2178        1.1   mycroft 	struct buf *bp;
   2179        1.1   mycroft 	int s;
   2180      1.149      yamt 	struct vnode *devvp = VTOI(fs->lfs_ivnode)->i_devvp;
   2181        1.1   mycroft 
   2182      1.159  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
   2183      1.156  perseant #ifdef DIAGNOSTIC
   2184      1.156  perseant 	KASSERT(fs->lfs_magic == LFS_MAGIC);
   2185      1.156  perseant #endif
   2186       1.15  perseant 	/*
   2187       1.15  perseant 	 * If we can write one superblock while another is in
   2188       1.15  perseant 	 * progress, we risk not having a complete checkpoint if we crash.
   2189       1.15  perseant 	 * So, block here if a superblock write is in progress.
   2190       1.15  perseant 	 */
   2191      1.159  perseant 	simple_lock(&fs->lfs_interlock);
   2192       1.36  perseant 	s = splbio();
   2193       1.73       chs 	while (fs->lfs_sbactive) {
   2194      1.159  perseant 		ltsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs sb", 0,
   2195      1.159  perseant 			&fs->lfs_interlock);
   2196       1.15  perseant 	}
   2197       1.15  perseant 	fs->lfs_sbactive = daddr;
   2198       1.36  perseant 	splx(s);
   2199      1.159  perseant 	simple_unlock(&fs->lfs_interlock);
   2200        1.1   mycroft 
   2201       1.15  perseant 	/* Set timestamp of this version of the superblock */
   2202       1.69  perseant 	if (fs->lfs_version == 1)
   2203  1.180.2.1      chap 		fs->lfs_otstamp = time_second;
   2204  1.180.2.1      chap 	fs->lfs_tstamp = time_second;
   2205       1.15  perseant 
   2206        1.1   mycroft 	/* Checksum the superblock and copy it into a buffer. */
   2207       1.12        pk 	fs->lfs_cksum = lfs_sb_cksum(&(fs->lfs_dlfs));
   2208      1.149      yamt 	bp = lfs_newbuf(fs, devvp,
   2209      1.128      yamt 	    fsbtodb(fs, daddr), LFS_SBPAD, LFS_NB_SBLOCK);
   2210      1.128      yamt 	memset(bp->b_data + sizeof(struct dlfs), 0,
   2211      1.128      yamt 	    LFS_SBPAD - sizeof(struct dlfs));
   2212       1.12        pk 	*(struct dlfs *)bp->b_data = fs->lfs_dlfs;
   2213      1.157     perry 
   2214        1.1   mycroft 	bp->b_flags |= B_BUSY | B_CALL | B_ASYNC;
   2215        1.1   mycroft 	bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI);
   2216        1.1   mycroft 	bp->b_iodone = lfs_supercallback;
   2217       1.15  perseant 
   2218      1.147      yamt 	if (fs->lfs_sp != NULL && fs->lfs_sp->seg_flags & SEGM_SYNC)
   2219      1.147      yamt 		BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
   2220      1.147      yamt 	else
   2221      1.147      yamt 		BIO_SETPRIO(bp, BPRIO_TIMELIMITED);
   2222      1.109  perseant 	curproc->p_stats->p_ru.ru_oublock++;
   2223        1.1   mycroft 	s = splbio();
   2224      1.100        pk 	V_INCR_NUMOUTPUT(bp->b_vp);
   2225       1.79  perseant 	splx(s);
   2226      1.159  perseant 	simple_lock(&fs->lfs_interlock);
   2227       1.52  perseant 	++fs->lfs_iocount;
   2228      1.159  perseant 	simple_unlock(&fs->lfs_interlock);
   2229      1.149      yamt 	VOP_STRATEGY(devvp, bp);
   2230        1.1   mycroft }
   2231        1.1   mycroft 
   2232        1.1   mycroft /*
   2233        1.1   mycroft  * Logical block number match routines used when traversing the dirty block
   2234        1.1   mycroft  * chain.
   2235        1.1   mycroft  */
   2236        1.1   mycroft int
   2237       1.69  perseant lfs_match_fake(struct lfs *fs, struct buf *bp)
   2238       1.15  perseant {
   2239      1.129      yamt 
   2240      1.159  perseant 	ASSERT_SEGLOCK(fs);
   2241      1.101  perseant 	return LFS_IS_MALLOC_BUF(bp);
   2242       1.15  perseant }
   2243       1.15  perseant 
   2244      1.101  perseant #if 0
   2245      1.101  perseant int
   2246      1.101  perseant lfs_match_real(struct lfs *fs, struct buf *bp)
   2247      1.101  perseant {
   2248      1.129      yamt 
   2249      1.159  perseant 	ASSERT_SEGLOCK(fs);
   2250      1.101  perseant 	return (lfs_match_data(fs, bp) && !lfs_match_fake(fs, bp));
   2251      1.101  perseant }
   2252      1.101  perseant #endif
   2253      1.101  perseant 
   2254       1.15  perseant int
   2255       1.69  perseant lfs_match_data(struct lfs *fs, struct buf *bp)
   2256        1.1   mycroft {
   2257      1.129      yamt 
   2258      1.159  perseant 	ASSERT_SEGLOCK(fs);
   2259        1.1   mycroft 	return (bp->b_lblkno >= 0);
   2260        1.1   mycroft }
   2261        1.1   mycroft 
   2262        1.1   mycroft int
   2263       1.69  perseant lfs_match_indir(struct lfs *fs, struct buf *bp)
   2264        1.1   mycroft {
   2265       1.91      fvdl 	daddr_t lbn;
   2266        1.1   mycroft 
   2267      1.159  perseant 	ASSERT_SEGLOCK(fs);
   2268        1.1   mycroft 	lbn = bp->b_lblkno;
   2269        1.1   mycroft 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 0);
   2270        1.1   mycroft }
   2271        1.1   mycroft 
   2272        1.1   mycroft int
   2273       1.69  perseant lfs_match_dindir(struct lfs *fs, struct buf *bp)
   2274        1.1   mycroft {
   2275       1.91      fvdl 	daddr_t lbn;
   2276        1.1   mycroft 
   2277      1.159  perseant 	ASSERT_SEGLOCK(fs);
   2278        1.1   mycroft 	lbn = bp->b_lblkno;
   2279        1.1   mycroft 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 1);
   2280        1.1   mycroft }
   2281        1.1   mycroft 
   2282        1.1   mycroft int
   2283       1.69  perseant lfs_match_tindir(struct lfs *fs, struct buf *bp)
   2284        1.1   mycroft {
   2285       1.91      fvdl 	daddr_t lbn;
   2286        1.1   mycroft 
   2287      1.159  perseant 	ASSERT_SEGLOCK(fs);
   2288        1.1   mycroft 	lbn = bp->b_lblkno;
   2289        1.1   mycroft 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 2);
   2290        1.1   mycroft }
   2291        1.1   mycroft 
   2292        1.1   mycroft /*
   2293       1.15  perseant  * XXX - The only buffers that are going to hit these functions are the
   2294       1.15  perseant  * segment write blocks, or the segment summaries, or the superblocks.
   2295      1.157     perry  *
   2296       1.15  perseant  * All of the above are created by lfs_newbuf, and so do not need to be
   2297       1.15  perseant  * released via brelse.
   2298        1.1   mycroft  */
   2299        1.1   mycroft void
   2300       1.69  perseant lfs_callback(struct buf *bp)
   2301        1.1   mycroft {
   2302      1.101  perseant 	struct lfs *fs;
   2303      1.101  perseant 
   2304      1.144      yamt 	fs = bp->b_private;
   2305      1.159  perseant 	ASSERT_NO_SEGLOCK(fs);
   2306      1.101  perseant 	lfs_freebuf(fs, bp);
   2307        1.1   mycroft }
   2308        1.1   mycroft 
   2309       1.79  perseant static void
   2310       1.79  perseant lfs_super_aiodone(struct buf *bp)
   2311        1.1   mycroft {
   2312       1.15  perseant 	struct lfs *fs;
   2313       1.15  perseant 
   2314      1.144      yamt 	fs = bp->b_private;
   2315      1.159  perseant 	ASSERT_NO_SEGLOCK(fs);
   2316      1.159  perseant 	simple_lock(&fs->lfs_interlock);
   2317       1.45   thorpej 	fs->lfs_sbactive = 0;
   2318      1.107  perseant 	if (--fs->lfs_iocount <= 1)
   2319       1.52  perseant 		wakeup(&fs->lfs_iocount);
   2320      1.159  perseant 	simple_unlock(&fs->lfs_interlock);
   2321      1.159  perseant 	wakeup(&fs->lfs_sbactive);
   2322      1.101  perseant 	lfs_freebuf(fs, bp);
   2323       1.74  perseant }
   2324       1.74  perseant 
   2325       1.74  perseant static void
   2326       1.79  perseant lfs_cluster_aiodone(struct buf *bp)
   2327       1.74  perseant {
   2328       1.74  perseant 	struct lfs_cluster *cl;
   2329       1.74  perseant 	struct lfs *fs;
   2330      1.109  perseant 	struct buf *tbp, *fbp;
   2331      1.101  perseant 	struct vnode *vp, *devvp;
   2332      1.109  perseant 	struct inode *ip;
   2333       1.79  perseant 	int s, error=0;
   2334       1.74  perseant 
   2335      1.115  perseant 	if (bp->b_flags & B_ERROR)
   2336       1.74  perseant 		error = bp->b_error;
   2337       1.74  perseant 
   2338      1.144      yamt 	cl = bp->b_private;
   2339       1.74  perseant 	fs = cl->fs;
   2340      1.101  perseant 	devvp = VTOI(fs->lfs_ivnode)->i_devvp;
   2341      1.159  perseant 	ASSERT_NO_SEGLOCK(fs);
   2342       1.74  perseant 
   2343       1.74  perseant 	/* Put the pages back, and release the buffer */
   2344      1.115  perseant 	while (cl->bufcount--) {
   2345       1.74  perseant 		tbp = cl->bpp[cl->bufcount];
   2346      1.153      yamt 		KASSERT(tbp->b_flags & B_BUSY);
   2347      1.115  perseant 		if (error) {
   2348       1.74  perseant 			tbp->b_flags |= B_ERROR;
   2349       1.74  perseant 			tbp->b_error = error;
   2350       1.74  perseant 		}
   2351       1.74  perseant 
   2352       1.74  perseant 		/*
   2353      1.103  perseant 		 * We're done with tbp.	 If it has not been re-dirtied since
   2354       1.74  perseant 		 * the cluster was written, free it.  Otherwise, keep it on
   2355       1.74  perseant 		 * the locked list to be written again.
   2356       1.74  perseant 		 */
   2357      1.101  perseant 		vp = tbp->b_vp;
   2358      1.114  perseant 
   2359       1.74  perseant 		tbp->b_flags &= ~B_GATHERED;
   2360       1.74  perseant 
   2361       1.74  perseant 		LFS_BCLEAN_LOG(fs, tbp);
   2362       1.74  perseant 
   2363      1.115  perseant 		if (!(tbp->b_flags & B_CALL)) {
   2364      1.136      yamt 			KASSERT(tbp->b_flags & B_LOCKED);
   2365      1.132      yamt 			s = splbio();
   2366      1.132      yamt 			simple_lock(&bqueue_slock);
   2367       1.74  perseant 			bremfree(tbp);
   2368      1.132      yamt 			simple_unlock(&bqueue_slock);
   2369      1.115  perseant 			if (vp)
   2370       1.74  perseant 				reassignbuf(tbp, vp);
   2371       1.79  perseant 			splx(s);
   2372       1.74  perseant 			tbp->b_flags |= B_ASYNC; /* for biodone */
   2373       1.74  perseant 		}
   2374      1.132      yamt 
   2375      1.132      yamt 		if ((tbp->b_flags & (B_LOCKED | B_DELWRI)) == B_LOCKED)
   2376      1.132      yamt 			LFS_UNLOCK_BUF(tbp);
   2377      1.132      yamt 
   2378       1.74  perseant 		if (tbp->b_flags & B_DONE) {
   2379      1.158  perseant 			DLOG((DLOG_SEG, "blk %d biodone already (flags %lx)\n",
   2380      1.158  perseant 				cl->bufcount, (long)tbp->b_flags));
   2381       1.74  perseant 		}
   2382      1.109  perseant 
   2383      1.153      yamt 		if ((tbp->b_flags & B_CALL) && !LFS_IS_MALLOC_BUF(tbp)) {
   2384      1.109  perseant 			/*
   2385      1.153      yamt 			 * A buffer from the page daemon.
   2386      1.153      yamt 			 * We use the same iodone as it does,
   2387      1.153      yamt 			 * so we must manually disassociate its
   2388      1.153      yamt 			 * buffers from the vp.
   2389      1.109  perseant 			 */
   2390      1.153      yamt 			if (tbp->b_vp) {
   2391      1.153      yamt 				/* This is just silly */
   2392      1.153      yamt 				s = splbio();
   2393      1.153      yamt 				brelvp(tbp);
   2394      1.153      yamt 				tbp->b_vp = vp;
   2395      1.153      yamt 				splx(s);
   2396      1.153      yamt 			}
   2397      1.153      yamt 			/* Put it back the way it was */
   2398      1.153      yamt 			tbp->b_flags |= B_ASYNC;
   2399      1.153      yamt 			/* Master buffers have B_AGE */
   2400      1.153      yamt 			if (tbp->b_private == tbp)
   2401      1.153      yamt 				tbp->b_flags |= B_AGE;
   2402      1.153      yamt 		}
   2403      1.153      yamt 		s = splbio();
   2404      1.153      yamt 		biodone(tbp);
   2405      1.153      yamt 
   2406      1.153      yamt 		/*
   2407      1.153      yamt 		 * If this is the last block for this vnode, but
   2408      1.153      yamt 		 * there are other blocks on its dirty list,
   2409      1.153      yamt 		 * set IN_MODIFIED/IN_CLEANING depending on what
   2410      1.153      yamt 		 * sort of block.  Only do this for our mount point,
   2411      1.153      yamt 		 * not for, e.g., inode blocks that are attached to
   2412      1.153      yamt 		 * the devvp.
   2413      1.153      yamt 		 * XXX KS - Shouldn't we set *both* if both types
   2414      1.153      yamt 		 * of blocks are present (traverse the dirty list?)
   2415      1.153      yamt 		 */
   2416      1.153      yamt 		simple_lock(&global_v_numoutput_slock);
   2417      1.153      yamt 		if (vp != devvp && vp->v_numoutput == 0 &&
   2418      1.153      yamt 		    (fbp = LIST_FIRST(&vp->v_dirtyblkhd)) != NULL) {
   2419      1.153      yamt 			ip = VTOI(vp);
   2420      1.158  perseant 			DLOG((DLOG_SEG, "lfs_cluster_aiodone: mark ino %d\n",
   2421      1.158  perseant 			       ip->i_number));
   2422      1.153      yamt 			if (LFS_IS_MALLOC_BUF(fbp))
   2423      1.153      yamt 				LFS_SET_UINO(ip, IN_CLEANING);
   2424      1.153      yamt 			else
   2425      1.153      yamt 				LFS_SET_UINO(ip, IN_MODIFIED);
   2426       1.74  perseant 		}
   2427      1.153      yamt 		simple_unlock(&global_v_numoutput_slock);
   2428      1.153      yamt 		splx(s);
   2429      1.153      yamt 		wakeup(vp);
   2430       1.74  perseant 	}
   2431       1.74  perseant 
   2432       1.74  perseant 	/* Fix up the cluster buffer, and release it */
   2433      1.109  perseant 	if (cl->flags & LFS_CL_MALLOC)
   2434      1.101  perseant 		lfs_free(fs, bp->b_data, LFS_NB_CLUSTER);
   2435      1.169      yamt 	putiobuf(bp);
   2436       1.74  perseant 
   2437       1.79  perseant 	/* Note i/o done */
   2438       1.79  perseant 	if (cl->flags & LFS_CL_SYNC) {
   2439      1.157     perry 		if (--cl->seg->seg_iocount == 0)
   2440       1.79  perseant 			wakeup(&cl->seg->seg_iocount);
   2441       1.79  perseant 	}
   2442      1.159  perseant 	simple_lock(&fs->lfs_interlock);
   2443       1.74  perseant #ifdef DIAGNOSTIC
   2444       1.74  perseant 	if (fs->lfs_iocount == 0)
   2445       1.82    provos 		panic("lfs_cluster_aiodone: zero iocount");
   2446       1.74  perseant #endif
   2447      1.107  perseant 	if (--fs->lfs_iocount <= 1)
   2448       1.74  perseant 		wakeup(&fs->lfs_iocount);
   2449      1.159  perseant 	simple_unlock(&fs->lfs_interlock);
   2450       1.79  perseant 
   2451      1.101  perseant 	pool_put(&fs->lfs_bpppool, cl->bpp);
   2452      1.101  perseant 	cl->bpp = NULL;
   2453      1.101  perseant 	pool_put(&fs->lfs_clpool, cl);
   2454       1.79  perseant }
   2455       1.79  perseant 
   2456       1.79  perseant static void
   2457       1.79  perseant lfs_generic_callback(struct buf *bp, void (*aiodone)(struct buf *))
   2458       1.79  perseant {
   2459       1.79  perseant 	/* reset b_iodone for when this is a single-buf i/o. */
   2460       1.79  perseant 	bp->b_iodone = aiodone;
   2461       1.79  perseant 
   2462      1.103  perseant 	simple_lock(&uvm.aiodoned_lock);	/* locks uvm.aio_done */
   2463       1.79  perseant 	TAILQ_INSERT_TAIL(&uvm.aio_done, bp, b_freelist);
   2464       1.79  perseant 	wakeup(&uvm.aiodoned);
   2465       1.79  perseant 	simple_unlock(&uvm.aiodoned_lock);
   2466       1.79  perseant }
   2467       1.79  perseant 
   2468       1.79  perseant static void
   2469       1.79  perseant lfs_cluster_callback(struct buf *bp)
   2470       1.79  perseant {
   2471      1.129      yamt 
   2472       1.79  perseant 	lfs_generic_callback(bp, lfs_cluster_aiodone);
   2473       1.79  perseant }
   2474       1.79  perseant 
   2475       1.79  perseant void
   2476       1.79  perseant lfs_supercallback(struct buf *bp)
   2477       1.79  perseant {
   2478      1.129      yamt 
   2479       1.79  perseant 	lfs_generic_callback(bp, lfs_super_aiodone);
   2480        1.1   mycroft }
   2481        1.1   mycroft 
   2482        1.1   mycroft /*
   2483        1.1   mycroft  * Shellsort (diminishing increment sort) from Data Structures and
   2484        1.1   mycroft  * Algorithms, Aho, Hopcraft and Ullman, 1983 Edition, page 290;
   2485        1.1   mycroft  * see also Knuth Vol. 3, page 84.  The increments are selected from
   2486        1.1   mycroft  * formula (8), page 95.  Roughly O(N^3/2).
   2487        1.1   mycroft  */
   2488        1.1   mycroft /*
   2489        1.1   mycroft  * This is our own private copy of shellsort because we want to sort
   2490        1.1   mycroft  * two parallel arrays (the array of buffer pointers and the array of
   2491        1.1   mycroft  * logical block numbers) simultaneously.  Note that we cast the array
   2492        1.1   mycroft  * of logical block numbers to a unsigned in this routine so that the
   2493        1.1   mycroft  * negative block numbers (meta data blocks) sort AFTER the data blocks.
   2494        1.1   mycroft  */
   2495       1.15  perseant 
   2496        1.1   mycroft void
   2497      1.104  perseant lfs_shellsort(struct buf **bp_array, int32_t *lb_array, int nmemb, int size)
   2498        1.1   mycroft {
   2499        1.1   mycroft 	static int __rsshell_increments[] = { 4, 1, 0 };
   2500       1.42  augustss 	int incr, *incrp, t1, t2;
   2501        1.1   mycroft 	struct buf *bp_temp;
   2502      1.118      yamt 
   2503      1.118      yamt #ifdef DEBUG
   2504      1.118      yamt 	incr = 0;
   2505      1.118      yamt 	for (t1 = 0; t1 < nmemb; t1++) {
   2506      1.118      yamt 		for (t2 = 0; t2 * size < bp_array[t1]->b_bcount; t2++) {
   2507      1.118      yamt 			if (lb_array[incr++] != bp_array[t1]->b_lblkno + t2) {
   2508      1.118      yamt 				/* dump before panic */
   2509      1.118      yamt 				printf("lfs_shellsort: nmemb=%d, size=%d\n",
   2510      1.118      yamt 				    nmemb, size);
   2511      1.118      yamt 				incr = 0;
   2512      1.118      yamt 				for (t1 = 0; t1 < nmemb; t1++) {
   2513      1.118      yamt 					const struct buf *bp = bp_array[t1];
   2514      1.118      yamt 
   2515      1.118      yamt 					printf("bp[%d]: lbn=%" PRIu64 ", size=%"
   2516      1.118      yamt 					    PRIu64 "\n", t1,
   2517      1.118      yamt 					    (uint64_t)bp->b_bcount,
   2518      1.118      yamt 					    (uint64_t)bp->b_lblkno);
   2519      1.118      yamt 					printf("lbns:");
   2520      1.118      yamt 					for (t2 = 0; t2 * size < bp->b_bcount;
   2521      1.118      yamt 					    t2++) {
   2522      1.118      yamt 						printf(" %" PRId32,
   2523      1.118      yamt 						    lb_array[incr++]);
   2524      1.118      yamt 					}
   2525      1.118      yamt 					printf("\n");
   2526      1.118      yamt 				}
   2527      1.118      yamt 				panic("lfs_shellsort: inconsistent input");
   2528      1.118      yamt 			}
   2529      1.118      yamt 		}
   2530      1.118      yamt 	}
   2531      1.118      yamt #endif
   2532        1.1   mycroft 
   2533        1.4  christos 	for (incrp = __rsshell_increments; (incr = *incrp++) != 0;)
   2534        1.1   mycroft 		for (t1 = incr; t1 < nmemb; ++t1)
   2535        1.1   mycroft 			for (t2 = t1 - incr; t2 >= 0;)
   2536      1.104  perseant 				if ((u_int32_t)bp_array[t2]->b_lblkno >
   2537      1.104  perseant 				    (u_int32_t)bp_array[t2 + incr]->b_lblkno) {
   2538        1.1   mycroft 					bp_temp = bp_array[t2];
   2539        1.1   mycroft 					bp_array[t2] = bp_array[t2 + incr];
   2540        1.1   mycroft 					bp_array[t2 + incr] = bp_temp;
   2541        1.1   mycroft 					t2 -= incr;
   2542        1.1   mycroft 				} else
   2543        1.1   mycroft 					break;
   2544      1.104  perseant 
   2545      1.104  perseant 	/* Reform the list of logical blocks */
   2546      1.104  perseant 	incr = 0;
   2547      1.104  perseant 	for (t1 = 0; t1 < nmemb; t1++) {
   2548      1.104  perseant 		for (t2 = 0; t2 * size < bp_array[t1]->b_bcount; t2++) {
   2549      1.104  perseant 			lb_array[incr++] = bp_array[t1]->b_lblkno + t2;
   2550      1.104  perseant 		}
   2551      1.104  perseant 	}
   2552        1.1   mycroft }
   2553        1.1   mycroft 
   2554        1.1   mycroft /*
   2555      1.173  perseant  * Call vget with LK_NOWAIT.  If we are the one who holds VXLOCK/VFREEING,
   2556      1.173  perseant  * however, we must press on.  Just fake success in that case.
   2557        1.1   mycroft  */
   2558        1.4  christos int
   2559       1.69  perseant lfs_vref(struct vnode *vp)
   2560        1.1   mycroft {
   2561      1.173  perseant 	int error;
   2562      1.173  perseant 	struct lfs *fs;
   2563      1.173  perseant 
   2564      1.173  perseant 	fs = VTOI(vp)->i_lfs;
   2565      1.173  perseant 
   2566      1.173  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
   2567      1.173  perseant 
   2568       1.15  perseant 	/*
   2569       1.15  perseant 	 * If we return 1 here during a flush, we risk vinvalbuf() not
   2570       1.15  perseant 	 * being able to flush all of the pages from this vnode, which
   2571       1.15  perseant 	 * will cause it to panic.  So, return 0 if a flush is in progress.
   2572       1.15  perseant 	 */
   2573      1.173  perseant 	error = vget(vp, LK_NOWAIT);
   2574      1.173  perseant 	if (error == EBUSY && IS_FLUSHING(VTOI(vp)->i_lfs, vp)) {
   2575      1.173  perseant 		++fs->lfs_flushvp_fakevref;
   2576      1.173  perseant 		return 0;
   2577       1.15  perseant 	}
   2578      1.173  perseant 	return error;
   2579        1.1   mycroft }
   2580        1.1   mycroft 
   2581       1.10      fvdl /*
   2582       1.10      fvdl  * This is vrele except that we do not want to VOP_INACTIVE this vnode. We
   2583       1.10      fvdl  * inline vrele here to avoid the vn_lock and VOP_INACTIVE call at the end.
   2584       1.10      fvdl  */
   2585        1.1   mycroft void
   2586       1.69  perseant lfs_vunref(struct vnode *vp)
   2587        1.1   mycroft {
   2588      1.173  perseant 	struct lfs *fs;
   2589      1.173  perseant 
   2590      1.173  perseant 	fs = VTOI(vp)->i_lfs;
   2591      1.173  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
   2592      1.173  perseant 
   2593       1.17  perseant 	/*
   2594       1.17  perseant 	 * Analogous to lfs_vref, if the node is flushing, fake it.
   2595       1.17  perseant 	 */
   2596      1.173  perseant 	if (IS_FLUSHING(fs, vp) && fs->lfs_flushvp_fakevref) {
   2597      1.173  perseant 		--fs->lfs_flushvp_fakevref;
   2598       1.17  perseant 		return;
   2599       1.17  perseant 	}
   2600       1.17  perseant 
   2601       1.10      fvdl 	simple_lock(&vp->v_interlock);
   2602       1.15  perseant #ifdef DIAGNOSTIC
   2603       1.73       chs 	if (vp->v_usecount <= 0) {
   2604      1.165  christos 		printf("lfs_vunref: inum is %llu\n", (unsigned long long)
   2605      1.165  christos 		    VTOI(vp)->i_number);
   2606       1.69  perseant 		printf("lfs_vunref: flags are 0x%lx\n", (u_long)vp->v_flag);
   2607       1.69  perseant 		printf("lfs_vunref: usecount = %ld\n", (long)vp->v_usecount);
   2608      1.158  perseant 		panic("lfs_vunref: v_usecount < 0");
   2609       1.15  perseant 	}
   2610       1.15  perseant #endif
   2611       1.10      fvdl 	vp->v_usecount--;
   2612       1.10      fvdl 	if (vp->v_usecount > 0) {
   2613       1.15  perseant 		simple_unlock(&vp->v_interlock);
   2614       1.15  perseant 		return;
   2615       1.15  perseant 	}
   2616       1.15  perseant 	/*
   2617       1.10      fvdl 	 * insert at tail of LRU list
   2618        1.1   mycroft 	 */
   2619       1.10      fvdl 	simple_lock(&vnode_free_list_slock);
   2620       1.40  perseant 	if (vp->v_holdcnt > 0)
   2621       1.40  perseant 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
   2622       1.40  perseant 	else
   2623       1.40  perseant 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
   2624       1.10      fvdl 	simple_unlock(&vnode_free_list_slock);
   2625       1.10      fvdl 	simple_unlock(&vp->v_interlock);
   2626        1.1   mycroft }
   2627       1.15  perseant 
   2628       1.15  perseant /*
   2629       1.15  perseant  * We use this when we have vnodes that were loaded in solely for cleaning.
   2630       1.15  perseant  * There is no reason to believe that these vnodes will be referenced again
   2631       1.15  perseant  * soon, since the cleaning process is unrelated to normal filesystem
   2632       1.15  perseant  * activity.  Putting cleaned vnodes at the tail of the list has the effect
   2633       1.15  perseant  * of flushing the vnode LRU.  So, put vnodes that were loaded only for
   2634       1.15  perseant  * cleaning at the head of the list, instead.
   2635       1.15  perseant  */
   2636       1.15  perseant void
   2637       1.69  perseant lfs_vunref_head(struct vnode *vp)
   2638       1.15  perseant {
   2639      1.129      yamt 
   2640      1.159  perseant 	ASSERT_SEGLOCK(VTOI(vp)->i_lfs);
   2641       1.15  perseant 	simple_lock(&vp->v_interlock);
   2642       1.15  perseant #ifdef DIAGNOSTIC
   2643       1.73       chs 	if (vp->v_usecount == 0) {
   2644       1.15  perseant 		panic("lfs_vunref: v_usecount<0");
   2645       1.15  perseant 	}
   2646       1.15  perseant #endif
   2647       1.15  perseant 	vp->v_usecount--;
   2648       1.15  perseant 	if (vp->v_usecount > 0) {
   2649       1.15  perseant 		simple_unlock(&vp->v_interlock);
   2650       1.15  perseant 		return;
   2651       1.15  perseant 	}
   2652       1.15  perseant 	/*
   2653       1.15  perseant 	 * insert at head of LRU list
   2654       1.15  perseant 	 */
   2655       1.15  perseant 	simple_lock(&vnode_free_list_slock);
   2656       1.77  perseant 	if (vp->v_holdcnt > 0)
   2657       1.77  perseant 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
   2658       1.77  perseant 	else
   2659       1.77  perseant 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
   2660       1.15  perseant 	simple_unlock(&vnode_free_list_slock);
   2661       1.15  perseant 	simple_unlock(&vp->v_interlock);
   2662       1.15  perseant }
   2663       1.15  perseant 
   2664      1.180  perseant 
   2665      1.180  perseant /*
   2666      1.180  perseant  * Set up an FINFO entry for a new file.  The fip pointer is assumed to
   2667      1.180  perseant  * point at uninitialized space.
   2668      1.180  perseant  */
   2669      1.180  perseant void
   2670      1.180  perseant lfs_acquire_finfo(struct lfs *fs, ino_t ino, int vers)
   2671      1.180  perseant {
   2672      1.180  perseant 	struct segment *sp = fs->lfs_sp;
   2673      1.180  perseant 
   2674  1.180.2.1      chap 	KASSERT(vers > 0);
   2675  1.180.2.1      chap 
   2676      1.180  perseant 	if (sp->seg_bytes_left < fs->lfs_bsize ||
   2677      1.180  perseant 	    sp->sum_bytes_left < sizeof(struct finfo))
   2678      1.180  perseant 		(void) lfs_writeseg(fs, fs->lfs_sp);
   2679      1.180  perseant 
   2680      1.180  perseant 	sp->sum_bytes_left -= FINFOSIZE;
   2681      1.180  perseant 	++((SEGSUM *)(sp->segsum))->ss_nfinfo;
   2682      1.180  perseant 	sp->fip->fi_nblocks = 0;
   2683      1.180  perseant 	sp->fip->fi_ino = ino;
   2684      1.180  perseant 	sp->fip->fi_version = vers;
   2685      1.180  perseant }
   2686      1.180  perseant 
   2687      1.180  perseant /*
   2688      1.180  perseant  * Release the FINFO entry, either clearing out an unused entry or
   2689      1.180  perseant  * advancing us to the next available entry.
   2690      1.180  perseant  */
   2691      1.180  perseant void
   2692      1.180  perseant lfs_release_finfo(struct lfs *fs)
   2693      1.180  perseant {
   2694      1.180  perseant 	struct segment *sp = fs->lfs_sp;
   2695      1.180  perseant 
   2696      1.180  perseant 	if (sp->fip->fi_nblocks != 0) {
   2697      1.180  perseant 		sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE +
   2698      1.180  perseant 			sizeof(int32_t) * sp->fip->fi_nblocks);
   2699      1.180  perseant 		sp->start_lbp = &sp->fip->fi_blocks[0];
   2700      1.180  perseant 	} else {
   2701      1.180  perseant 		sp->sum_bytes_left += FINFOSIZE;
   2702      1.180  perseant 		--((SEGSUM *)(sp->segsum))->ss_nfinfo;
   2703      1.180  perseant 	}
   2704      1.180  perseant }
   2705