Home | History | Annotate | Line # | Download | only in lfs
lfs_bio.c revision 1.51
      1  1.50      yamt /*	$NetBSD: lfs_bio.c,v 1.51 2002/12/26 13:37:18 yamt Exp $	*/
      2   1.2       cgd 
      3   1.7  perseant /*-
      4  1.27  perseant  * Copyright (c) 1999, 2000 The NetBSD Foundation, Inc.
      5   1.7  perseant  * All rights reserved.
      6   1.7  perseant  *
      7   1.7  perseant  * This code is derived from software contributed to The NetBSD Foundation
      8   1.7  perseant  * by Konrad E. Schroder <perseant (at) hhhh.org>.
      9   1.7  perseant  *
     10   1.7  perseant  * Redistribution and use in source and binary forms, with or without
     11   1.7  perseant  * modification, are permitted provided that the following conditions
     12   1.7  perseant  * are met:
     13   1.7  perseant  * 1. Redistributions of source code must retain the above copyright
     14   1.7  perseant  *    notice, this list of conditions and the following disclaimer.
     15   1.7  perseant  * 2. Redistributions in binary form must reproduce the above copyright
     16   1.7  perseant  *    notice, this list of conditions and the following disclaimer in the
     17   1.7  perseant  *    documentation and/or other materials provided with the distribution.
     18   1.7  perseant  * 3. All advertising materials mentioning features or use of this software
     19   1.7  perseant  *    must display the following acknowledgement:
     20   1.7  perseant  *      This product includes software developed by the NetBSD
     21   1.7  perseant  *      Foundation, Inc. and its contributors.
     22   1.7  perseant  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23   1.7  perseant  *    contributors may be used to endorse or promote products derived
     24   1.7  perseant  *    from this software without specific prior written permission.
     25   1.7  perseant  *
     26   1.7  perseant  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27   1.7  perseant  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28   1.7  perseant  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29   1.7  perseant  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30   1.7  perseant  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31   1.7  perseant  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32   1.7  perseant  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33   1.7  perseant  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34   1.7  perseant  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35   1.7  perseant  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36   1.7  perseant  * POSSIBILITY OF SUCH DAMAGE.
     37   1.7  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   1.1   mycroft  * 3. All advertising materials mentioning features or use of this software
     51   1.1   mycroft  *    must display the following acknowledgement:
     52   1.1   mycroft  *	This product includes software developed by the University of
     53   1.1   mycroft  *	California, Berkeley and its contributors.
     54   1.1   mycroft  * 4. Neither the name of the University nor the names of its contributors
     55   1.1   mycroft  *    may be used to endorse or promote products derived from this software
     56   1.1   mycroft  *    without specific prior written permission.
     57   1.1   mycroft  *
     58   1.1   mycroft  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     59   1.1   mycroft  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     60   1.1   mycroft  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     61   1.1   mycroft  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     62   1.1   mycroft  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     63   1.1   mycroft  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     64   1.1   mycroft  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     65   1.1   mycroft  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     66   1.1   mycroft  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     67   1.1   mycroft  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     68   1.1   mycroft  * SUCH DAMAGE.
     69   1.1   mycroft  *
     70   1.6      fvdl  *	@(#)lfs_bio.c	8.10 (Berkeley) 6/10/95
     71   1.1   mycroft  */
     72  1.39     lukem 
     73  1.39     lukem #include <sys/cdefs.h>
     74  1.50      yamt __KERNEL_RCSID(0, "$NetBSD: lfs_bio.c,v 1.51 2002/12/26 13:37:18 yamt Exp $");
     75   1.1   mycroft 
     76   1.1   mycroft #include <sys/param.h>
     77   1.5  christos #include <sys/systm.h>
     78   1.1   mycroft #include <sys/proc.h>
     79   1.1   mycroft #include <sys/buf.h>
     80   1.1   mycroft #include <sys/vnode.h>
     81   1.1   mycroft #include <sys/resourcevar.h>
     82   1.1   mycroft #include <sys/mount.h>
     83   1.1   mycroft #include <sys/kernel.h>
     84   1.1   mycroft 
     85   1.1   mycroft #include <ufs/ufs/inode.h>
     86   1.1   mycroft #include <ufs/ufs/ufsmount.h>
     87   1.5  christos #include <ufs/ufs/ufs_extern.h>
     88   1.1   mycroft 
     89   1.7  perseant #include <sys/malloc.h>
     90   1.1   mycroft #include <ufs/lfs/lfs.h>
     91   1.1   mycroft #include <ufs/lfs/lfs_extern.h>
     92   1.1   mycroft 
     93   1.7  perseant /* Macros to clear/set/test flags. */
     94   1.7  perseant # define	SET(t, f)	(t) |= (f)
     95   1.7  perseant # define	CLR(t, f)	(t) &= ~(f)
     96   1.7  perseant # define	ISSET(t, f)	((t) & (f))
     97   1.7  perseant 
     98   1.1   mycroft /*
     99   1.1   mycroft  * LFS block write function.
    100   1.1   mycroft  *
    101   1.1   mycroft  * XXX
    102   1.1   mycroft  * No write cost accounting is done.
    103   1.1   mycroft  * This is almost certainly wrong for synchronous operations and NFS.
    104   1.1   mycroft  */
    105   1.7  perseant int	locked_queue_count   = 0;	/* XXX Count of locked-down buffers. */
    106   1.7  perseant long	locked_queue_bytes   = 0L;	/* XXX Total size of locked buffers. */
    107   1.7  perseant int	lfs_writing          = 0;	/* Set if already kicked off a writer
    108   1.1   mycroft 					   because of buffer space */
    109   1.7  perseant extern int lfs_dostats;
    110   1.7  perseant 
    111  1.27  perseant /*
    112  1.51      yamt  * reserved number/bytes of locked buffers
    113  1.51      yamt  */
    114  1.51      yamt int locked_queue_rcount = 0;
    115  1.51      yamt long locked_queue_rbytes = 0L;
    116  1.51      yamt 
    117  1.51      yamt int lfs_fits_buf(struct lfs *, int, int);
    118  1.51      yamt int lfs_reservebuf(struct lfs *, int, int);
    119  1.51      yamt int lfs_reserveavail(struct lfs *, struct vnode *, int);
    120  1.51      yamt 
    121  1.51      yamt int
    122  1.51      yamt lfs_fits_buf(struct lfs *fs, int n, int bytes)
    123  1.51      yamt {
    124  1.51      yamt 	int count_fit =
    125  1.51      yamt 	    (locked_queue_count + locked_queue_rcount + n < LFS_WAIT_BUFS);
    126  1.51      yamt 	int bytes_fit =
    127  1.51      yamt 	    (locked_queue_bytes + locked_queue_rbytes + bytes < LFS_WAIT_BYTES);
    128  1.51      yamt 
    129  1.51      yamt #ifdef DEBUG_LFS
    130  1.51      yamt 	if (!count_fit) {
    131  1.51      yamt 		printf("lfs_fits_buf: no fit count: %d + %d + %d >= %d\n",
    132  1.51      yamt 			locked_queue_count, locked_queue_rcount,
    133  1.51      yamt 			n, LFS_WAIT_BUFS);
    134  1.51      yamt 	}
    135  1.51      yamt 	if (!bytes_fit) {
    136  1.51      yamt 		printf("lfs_fits_buf: no fit bytes: %ld + %ld + %d >= %d\n",
    137  1.51      yamt 			locked_queue_bytes, locked_queue_rbytes,
    138  1.51      yamt 			bytes, LFS_WAIT_BYTES);
    139  1.51      yamt 	}
    140  1.51      yamt #endif /* DEBUG_LFS */
    141  1.51      yamt 
    142  1.51      yamt 	return (count_fit && bytes_fit);
    143  1.51      yamt }
    144  1.51      yamt 
    145  1.51      yamt int
    146  1.51      yamt lfs_reservebuf(struct lfs *fs, int n, int bytes)
    147  1.51      yamt {
    148  1.51      yamt 	KASSERT(locked_queue_rcount >= 0);
    149  1.51      yamt 	KASSERT(locked_queue_rbytes >= 0);
    150  1.51      yamt 
    151  1.51      yamt 	while (n > 0 && !lfs_fits_buf(fs, n, bytes)) {
    152  1.51      yamt 		int error;
    153  1.51      yamt 
    154  1.51      yamt 		++fs->lfs_writer;
    155  1.51      yamt 		lfs_flush(fs, 0);
    156  1.51      yamt 		if (--fs->lfs_writer == 0)
    157  1.51      yamt 			wakeup(&fs->lfs_dirops);
    158  1.51      yamt 
    159  1.51      yamt 		error = tsleep(&locked_queue_count, PCATCH | PUSER,
    160  1.51      yamt 		    "lfsresbuf", hz * LFS_BUFWAIT);
    161  1.51      yamt 		if (error && error != EWOULDBLOCK)
    162  1.51      yamt 			return error;
    163  1.51      yamt 	}
    164  1.51      yamt 
    165  1.51      yamt 	locked_queue_rcount += n;
    166  1.51      yamt 	locked_queue_rbytes += bytes;
    167  1.51      yamt 
    168  1.51      yamt 	KASSERT(locked_queue_rcount >= 0);
    169  1.51      yamt 	KASSERT(locked_queue_rbytes >= 0);
    170  1.51      yamt 
    171  1.51      yamt 	return 0;
    172  1.51      yamt }
    173  1.51      yamt 
    174  1.51      yamt /*
    175  1.27  perseant  * Try to reserve some blocks, prior to performing a sensitive operation that
    176  1.27  perseant  * requires the vnode lock to be honored.  If there is not enough space, give
    177  1.27  perseant  * up the vnode lock temporarily and wait for the space to become available.
    178  1.27  perseant  *
    179  1.44  perseant  * Called with vp locked.  (Note nowever that if fsb < 0, vp is ignored.)
    180  1.46      yamt  *
    181  1.47      yamt  * XXX YAMT - it isn't safe to unlock vp here
    182  1.47      yamt  * because the node might be modified while we sleep.
    183  1.47      yamt  * (eg. cached states like i_offset might be stale,
    184  1.47      yamt  *  the vnode might be truncated, etc..)
    185  1.47      yamt  * maybe we should have a way to restart the vnode op. (EVOPRESTART?)
    186  1.47      yamt  *
    187  1.47      yamt  * XXX YAMT - we unlock the vnode so that cleaner can lock it.
    188  1.47      yamt  * but it isn't enough. eg. for VOP_REMOVE, we should unlock the vnode that
    189  1.47      yamt  * is going to be removed as well.
    190  1.27  perseant  */
    191  1.27  perseant int
    192  1.51      yamt lfs_reserveavail(struct lfs *fs, struct vnode *vp, int fsb)
    193  1.27  perseant {
    194  1.27  perseant 	CLEANERINFO *cip;
    195  1.27  perseant 	struct buf *bp;
    196  1.27  perseant 	int error, slept;
    197  1.27  perseant 
    198  1.27  perseant 	slept = 0;
    199  1.45      yamt 	while (fsb > 0 && !lfs_fits(fs, fsb + fs->lfs_ravail) &&
    200  1.45      yamt 	    vp != fs->lfs_unlockvp) {
    201  1.49      yamt #if 0
    202  1.49      yamt 		/*
    203  1.49      yamt 		 * XXX ideally, we should unlock vnodes here
    204  1.49      yamt 		 * because we might sleep very long time.
    205  1.49      yamt 		 */
    206  1.27  perseant 		VOP_UNLOCK(vp, 0);
    207  1.50      yamt #else
    208  1.50      yamt 		/*
    209  1.50      yamt 		 * XXX since we'll sleep for cleaner with vnode lock holding,
    210  1.50      yamt 		 * deadlock will occur if cleaner wants to acquire the vnode
    211  1.50      yamt 		 * lock.
    212  1.50      yamt 		 * (eg. lfs_markv -> lfs_fastvget -> getnewvnode -> vclean)
    213  1.50      yamt 		 */
    214  1.49      yamt #endif
    215  1.27  perseant 
    216  1.30  perseant 		if (!slept) {
    217  1.30  perseant #ifdef DEBUG
    218  1.27  perseant 			printf("lfs_reserve: waiting for %ld (bfree = %d,"
    219  1.30  perseant 			       " est_bfree = %d)\n",
    220  1.44  perseant 			       fsb + fs->lfs_ravail, fs->lfs_bfree,
    221  1.30  perseant 			       LFS_EST_BFREE(fs));
    222  1.30  perseant #endif
    223  1.30  perseant 		}
    224  1.27  perseant 		++slept;
    225  1.27  perseant 
    226  1.27  perseant 		/* Wake up the cleaner */
    227  1.27  perseant 		LFS_CLEANERINFO(cip, fs, bp);
    228  1.31  perseant 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 0);
    229  1.27  perseant 		wakeup(&lfs_allclean_wakeup);
    230  1.27  perseant 		wakeup(&fs->lfs_nextseg);
    231  1.27  perseant 
    232  1.27  perseant 		error = tsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_reserve",
    233  1.27  perseant 			       0);
    234  1.49      yamt #if 0
    235  1.27  perseant 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX use lockstatus */
    236  1.49      yamt #endif
    237  1.27  perseant 		if (error)
    238  1.27  perseant 			return error;
    239  1.27  perseant 	}
    240  1.44  perseant #ifdef DEBUG
    241  1.27  perseant 	if (slept)
    242  1.27  perseant 		printf("lfs_reserve: woke up\n");
    243  1.44  perseant #endif
    244  1.44  perseant 	fs->lfs_ravail += fsb;
    245  1.51      yamt 
    246  1.27  perseant 	return 0;
    247  1.27  perseant }
    248  1.27  perseant 
    249  1.51      yamt #ifdef DIAGNOSTIC
    250  1.51      yamt int lfs_rescount;
    251  1.51      yamt int lfs_rescountdirop;
    252  1.51      yamt #endif
    253  1.51      yamt 
    254  1.51      yamt int
    255  1.51      yamt lfs_reserve(struct lfs *fs, struct vnode *vp, int fsb)
    256  1.51      yamt {
    257  1.51      yamt 	int error;
    258  1.51      yamt 	int cantwait;
    259  1.51      yamt 
    260  1.51      yamt 	cantwait = (VTOI(vp)->i_flag & IN_ADIROP);
    261  1.51      yamt #ifdef DIAGNOSTIC
    262  1.51      yamt 	if (cantwait) {
    263  1.51      yamt 		if (fsb > 0)
    264  1.51      yamt 			lfs_rescountdirop++;
    265  1.51      yamt 		else if (fsb < 0)
    266  1.51      yamt 			lfs_rescountdirop--;
    267  1.51      yamt 		if (lfs_rescountdirop < 0)
    268  1.51      yamt 			panic("lfs_rescountdirop");
    269  1.51      yamt 	}
    270  1.51      yamt 	else {
    271  1.51      yamt 		if (fsb > 0)
    272  1.51      yamt 			lfs_rescount++;
    273  1.51      yamt 		else if (fsb < 0)
    274  1.51      yamt 			lfs_rescount--;
    275  1.51      yamt 		if (lfs_rescount < 0)
    276  1.51      yamt 			panic("lfs_rescount");
    277  1.51      yamt 	}
    278  1.51      yamt #endif
    279  1.51      yamt 	if (cantwait)
    280  1.51      yamt 		return 0;
    281  1.51      yamt 
    282  1.51      yamt 	error = lfs_reserveavail(fs, vp, fsb);
    283  1.51      yamt 	if (error)
    284  1.51      yamt 		return error;
    285  1.51      yamt 
    286  1.51      yamt 	/*
    287  1.51      yamt 	 * XXX just a guess. should be more precise.
    288  1.51      yamt 	 */
    289  1.51      yamt 	error = lfs_reservebuf(fs, fragstoblks(fs, fsb), fsbtob(fs, fsb));
    290  1.51      yamt 	if (error)
    291  1.51      yamt 		lfs_reserveavail(fs, vp, -fsb);
    292  1.51      yamt 
    293  1.51      yamt 	return error;
    294  1.51      yamt }
    295   1.1   mycroft 
    296   1.1   mycroft int
    297  1.36  perseant lfs_bwrite(void *v)
    298   1.5  christos {
    299   1.1   mycroft 	struct vop_bwrite_args /* {
    300   1.1   mycroft 		struct buf *a_bp;
    301   1.5  christos 	} */ *ap = v;
    302  1.17  augustss 	struct buf *bp = ap->a_bp;
    303   1.7  perseant 
    304   1.7  perseant #ifdef DIAGNOSTIC
    305  1.23  perseant         if (VTOI(bp->b_vp)->i_lfs->lfs_ronly == 0 && (bp->b_flags & B_ASYNC)) {
    306   1.7  perseant 		panic("bawrite LFS buffer");
    307  1.16  perseant 	}
    308   1.7  perseant #endif /* DIAGNOSTIC */
    309   1.7  perseant 	return lfs_bwrite_ext(bp,0);
    310   1.7  perseant }
    311   1.7  perseant 
    312   1.7  perseant /*
    313  1.44  perseant  * Determine if there is enough room currently available to write fsb
    314  1.44  perseant  * blocks.  We need enough blocks for the new blocks, the current
    315  1.44  perseant  * inode blocks (including potentially the ifile inode), a summary block,
    316  1.44  perseant  * and the segment usage table, plus an ifile block.
    317   1.7  perseant  */
    318  1.32  perseant int
    319  1.36  perseant lfs_fits(struct lfs *fs, int fsb)
    320   1.7  perseant {
    321  1.26  perseant 	int needed;
    322  1.26  perseant 
    323  1.36  perseant 	needed = fsb + btofsb(fs, fs->lfs_sumsize) +
    324  1.44  perseant 		 ((howmany(fs->lfs_uinodes + 1, INOPB(fs)) + fs->lfs_segtabsz +
    325  1.44  perseant 		   1) << (fs->lfs_blktodb - fs->lfs_fsbtodb));
    326  1.27  perseant 
    327  1.26  perseant 	if (needed >= fs->lfs_avail) {
    328  1.27  perseant #ifdef DEBUG
    329  1.36  perseant 		printf("lfs_fits: no fit: fsb = %d, uinodes = %d, "
    330  1.26  perseant 		       "needed = %d, avail = %d\n",
    331  1.36  perseant 		       fsb, fs->lfs_uinodes, needed, fs->lfs_avail);
    332  1.26  perseant #endif
    333   1.7  perseant 		return 0;
    334   1.7  perseant 	}
    335   1.7  perseant 	return 1;
    336   1.7  perseant }
    337   1.7  perseant 
    338   1.7  perseant int
    339  1.44  perseant lfs_availwait(struct lfs *fs, int fsb)
    340  1.32  perseant {
    341  1.32  perseant 	int error;
    342  1.32  perseant 	CLEANERINFO *cip;
    343  1.32  perseant 	struct buf *cbp;
    344  1.32  perseant 
    345  1.44  perseant 	while (!lfs_fits(fs, fsb)) {
    346  1.32  perseant 		/*
    347  1.32  perseant 		 * Out of space, need cleaner to run.
    348  1.32  perseant 		 * Update the cleaner info, then wake it up.
    349  1.32  perseant 		 * Note the cleanerinfo block is on the ifile
    350  1.32  perseant 		 * so it CANT_WAIT.
    351  1.32  perseant 		 */
    352  1.32  perseant 		LFS_CLEANERINFO(cip, fs, cbp);
    353  1.32  perseant 		LFS_SYNC_CLEANERINFO(cip, fs, cbp, 0);
    354  1.32  perseant 
    355  1.32  perseant 		printf("lfs_availwait: out of available space, "
    356  1.32  perseant 		       "waiting on cleaner\n");
    357  1.32  perseant 
    358  1.32  perseant 		wakeup(&lfs_allclean_wakeup);
    359  1.32  perseant 		wakeup(&fs->lfs_nextseg);
    360  1.32  perseant #ifdef DIAGNOSTIC
    361  1.32  perseant 		if (fs->lfs_seglock && fs->lfs_lockpid == curproc->p_pid)
    362  1.32  perseant 			panic("lfs_availwait: deadlock");
    363  1.32  perseant #endif
    364  1.32  perseant 		error = tsleep(&fs->lfs_avail, PCATCH | PUSER, "cleaner", 0);
    365  1.32  perseant 		if (error)
    366  1.32  perseant 			return (error);
    367  1.32  perseant 	}
    368  1.32  perseant 	return 0;
    369  1.32  perseant }
    370  1.32  perseant 
    371  1.32  perseant int
    372  1.36  perseant lfs_bwrite_ext(struct buf *bp, int flags)
    373   1.7  perseant {
    374   1.1   mycroft 	struct lfs *fs;
    375   1.1   mycroft 	struct inode *ip;
    376  1.51      yamt 	int fsb, s;
    377  1.48      yamt 
    378  1.48      yamt 	KASSERT(bp->b_flags & B_BUSY);
    379  1.48      yamt 	KASSERT(flags & BW_CLEAN || !(bp->b_flags & B_CALL));
    380  1.48      yamt 
    381   1.1   mycroft 	/*
    382  1.16  perseant 	 * Don't write *any* blocks if we're mounted read-only.
    383  1.16  perseant 	 * In particular the cleaner can't write blocks either.
    384  1.16  perseant 	 */
    385  1.40       chs         if (VTOI(bp->b_vp)->i_lfs->lfs_ronly) {
    386  1.32  perseant 		bp->b_flags &= ~(B_DELWRI | B_READ | B_ERROR);
    387  1.32  perseant 		LFS_UNLOCK_BUF(bp);
    388  1.40       chs 		if (bp->b_flags & B_CALL)
    389  1.16  perseant 			bp->b_flags &= ~B_BUSY;
    390  1.16  perseant 		else
    391  1.16  perseant 			brelse(bp);
    392  1.16  perseant 		return EROFS;
    393  1.16  perseant 	}
    394  1.16  perseant 
    395  1.16  perseant 	/*
    396   1.1   mycroft 	 * Set the delayed write flag and use reassignbuf to move the buffer
    397   1.1   mycroft 	 * from the clean list to the dirty one.
    398   1.1   mycroft 	 *
    399   1.1   mycroft 	 * Set the B_LOCKED flag and unlock the buffer, causing brelse to move
    400   1.1   mycroft 	 * the buffer onto the LOCKED free list.  This is necessary, otherwise
    401   1.1   mycroft 	 * getnewbuf() would try to reclaim the buffers using bawrite, which
    402   1.1   mycroft 	 * isn't going to work.
    403   1.1   mycroft 	 *
    404   1.1   mycroft 	 * XXX we don't let meta-data writes run out of space because they can
    405   1.1   mycroft 	 * come from the segment writer.  We need to make sure that there is
    406   1.1   mycroft 	 * enough space reserved so that there's room to write meta-data
    407   1.1   mycroft 	 * blocks.
    408   1.1   mycroft 	 */
    409   1.1   mycroft 	if (!(bp->b_flags & B_LOCKED)) {
    410   1.1   mycroft 		fs = VFSTOUFS(bp->b_vp->v_mount)->um_lfs;
    411  1.36  perseant 		fsb = fragstofsb(fs, numfrags(fs, bp->b_bcount));
    412   1.7  perseant 
    413   1.7  perseant 		ip = VTOI(bp->b_vp);
    414  1.48      yamt 		if (flags & BW_CLEAN) {
    415  1.26  perseant 			LFS_SET_UINO(ip, IN_CLEANING);
    416   1.7  perseant 		} else {
    417  1.36  perseant 			LFS_SET_UINO(ip, IN_MODIFIED);
    418  1.36  perseant 			if (bp->b_lblkno >= 0)
    419  1.36  perseant 				LFS_SET_UINO(ip, IN_UPDATE);
    420   1.7  perseant 		}
    421  1.36  perseant 		fs->lfs_avail -= fsb;
    422  1.32  perseant 		bp->b_flags |= B_DELWRI;
    423  1.32  perseant 
    424  1.32  perseant 		LFS_LOCK_BUF(bp);
    425  1.43  perseant 		bp->b_flags &= ~(B_READ | B_DONE | B_ERROR);
    426  1.14      fvdl 		s = splbio();
    427   1.1   mycroft 		reassignbuf(bp, bp->b_vp);
    428   1.1   mycroft 		splx(s);
    429   1.1   mycroft 	}
    430   1.7  perseant 
    431  1.40       chs 	if (bp->b_flags & B_CALL)
    432   1.7  perseant 		bp->b_flags &= ~B_BUSY;
    433   1.7  perseant 	else
    434   1.7  perseant 		brelse(bp);
    435   1.7  perseant 
    436   1.1   mycroft 	return (0);
    437   1.1   mycroft }
    438   1.1   mycroft 
    439  1.24  perseant void
    440  1.36  perseant lfs_flush_fs(struct lfs *fs, int flags)
    441  1.11  perseant {
    442  1.40       chs 	if (fs->lfs_ronly == 0 && fs->lfs_dirops == 0)
    443  1.11  perseant 	{
    444  1.11  perseant 		/* disallow dirops during flush */
    445  1.24  perseant 		fs->lfs_writer++;
    446  1.11  perseant 
    447  1.11  perseant 		/*
    448  1.11  perseant 		 * We set the queue to 0 here because we
    449  1.11  perseant 		 * are about to write all the dirty
    450  1.11  perseant 		 * buffers we have.  If more come in
    451  1.11  perseant 		 * while we're writing the segment, they
    452  1.11  perseant 		 * may not get written, so we want the
    453  1.11  perseant 		 * count to reflect these new writes
    454  1.11  perseant 		 * after the segwrite completes.
    455  1.11  perseant 		 */
    456  1.40       chs 		if (lfs_dostats)
    457  1.11  perseant 			++lfs_stats.flush_invoked;
    458  1.24  perseant 		lfs_segwrite(fs->lfs_ivnode->v_mount, flags);
    459  1.11  perseant 
    460  1.11  perseant 		/* XXX KS - allow dirops again */
    461  1.40       chs 		if (--fs->lfs_writer == 0)
    462  1.24  perseant 			wakeup(&fs->lfs_dirops);
    463  1.11  perseant 	}
    464  1.11  perseant }
    465  1.11  perseant 
    466   1.1   mycroft /*
    467   1.1   mycroft  * XXX
    468   1.1   mycroft  * This routine flushes buffers out of the B_LOCKED queue when LFS has too
    469   1.1   mycroft  * many locked down.  Eventually the pageout daemon will simply call LFS
    470   1.1   mycroft  * when pages need to be reclaimed.  Note, we have one static count of locked
    471   1.1   mycroft  * buffers, so we can't have more than a single file system.  To make this
    472   1.1   mycroft  * work for multiple file systems, put the count into the mount structure.
    473   1.1   mycroft  */
    474   1.1   mycroft void
    475  1.36  perseant lfs_flush(struct lfs *fs, int flags)
    476   1.1   mycroft {
    477  1.17  augustss 	struct mount *mp, *nmp;
    478   1.7  perseant 
    479  1.40       chs 	if (lfs_dostats)
    480   1.7  perseant 		++lfs_stats.write_exceeded;
    481  1.40       chs 	if (lfs_writing && flags == 0) {/* XXX flags */
    482  1.16  perseant #ifdef DEBUG_LFS
    483  1.16  perseant 		printf("lfs_flush: not flushing because another flush is active\n");
    484  1.16  perseant #endif
    485   1.1   mycroft 		return;
    486  1.16  perseant 	}
    487   1.1   mycroft 	lfs_writing = 1;
    488   1.7  perseant 
    489   1.6      fvdl 	simple_lock(&mountlist_slock);
    490   1.6      fvdl 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
    491   1.6      fvdl 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
    492   1.6      fvdl 			nmp = mp->mnt_list.cqe_next;
    493   1.6      fvdl 			continue;
    494   1.6      fvdl 		}
    495  1.40       chs 		if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS, MFSNAMELEN) == 0)
    496  1.24  perseant 			lfs_flush_fs(((struct ufsmount *)mp->mnt_data)->ufsmount_u.lfs, flags);
    497   1.6      fvdl 		simple_lock(&mountlist_slock);
    498   1.6      fvdl 		nmp = mp->mnt_list.cqe_next;
    499   1.6      fvdl 		vfs_unbusy(mp);
    500   1.1   mycroft 	}
    501   1.6      fvdl 	simple_unlock(&mountlist_slock);
    502   1.7  perseant 
    503  1.43  perseant 	LFS_DEBUG_COUNTLOCKED("flush");
    504   1.7  perseant 
    505   1.1   mycroft 	lfs_writing = 0;
    506   1.1   mycroft }
    507   1.1   mycroft 
    508  1.41  perseant #define INOCOUNT(fs) howmany((fs)->lfs_uinodes, INOPB(fs))
    509  1.41  perseant #define INOBYTES(fs) ((fs)->lfs_uinodes * DINODE_SIZE)
    510  1.41  perseant 
    511   1.1   mycroft int
    512  1.36  perseant lfs_check(struct vnode *vp, ufs_daddr_t blkno, int flags)
    513   1.1   mycroft {
    514   1.1   mycroft 	int error;
    515  1.10  perseant 	struct lfs *fs;
    516  1.20  perseant 	struct inode *ip;
    517  1.13  perseant 	extern int lfs_dirvcount;
    518  1.10  perseant 
    519   1.1   mycroft 	error = 0;
    520  1.20  perseant 	ip = VTOI(vp);
    521   1.7  perseant 
    522   1.7  perseant 	/* If out of buffers, wait on writer */
    523   1.7  perseant 	/* XXX KS - if it's the Ifile, we're probably the cleaner! */
    524  1.20  perseant 	if (ip->i_number == LFS_IFILE_INUM)
    525  1.20  perseant 		return 0;
    526  1.20  perseant 	/* If we're being called from inside a dirop, don't sleep */
    527  1.20  perseant 	if (ip->i_flag & IN_ADIROP)
    528   1.7  perseant 		return 0;
    529   1.7  perseant 
    530  1.20  perseant 	fs = ip->i_lfs;
    531  1.20  perseant 
    532  1.20  perseant 	/*
    533  1.20  perseant 	 * If we would flush below, but dirops are active, sleep.
    534  1.20  perseant 	 * Note that a dirop cannot ever reach this code!
    535  1.20  perseant 	 */
    536  1.20  perseant 	while (fs->lfs_dirops > 0 &&
    537  1.41  perseant 	       (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS ||
    538  1.41  perseant                 locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES ||
    539  1.20  perseant                 lfs_dirvcount > LFS_MAXDIROP || fs->lfs_diropwait > 0))
    540  1.20  perseant 	{
    541  1.20  perseant 		++fs->lfs_diropwait;
    542  1.20  perseant 		tsleep(&fs->lfs_writer, PRIBIO+1, "bufdirop", 0);
    543  1.20  perseant 		--fs->lfs_diropwait;
    544  1.20  perseant 	}
    545  1.10  perseant 
    546  1.41  perseant 	if (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS ||
    547  1.41  perseant 	    locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES ||
    548  1.20  perseant 	    lfs_dirvcount > LFS_MAXDIROP || fs->lfs_diropwait > 0)
    549   1.7  perseant 	{
    550  1.10  perseant 		++fs->lfs_writer;
    551  1.10  perseant 		lfs_flush(fs, flags);
    552  1.40       chs 		if (--fs->lfs_writer == 0)
    553  1.10  perseant 			wakeup(&fs->lfs_dirops);
    554   1.7  perseant 	}
    555  1.11  perseant 
    556  1.41  perseant 	while (locked_queue_count + INOCOUNT(fs) > LFS_WAIT_BUFS
    557  1.41  perseant 	       || locked_queue_bytes + INOBYTES(fs) > LFS_WAIT_BYTES)
    558   1.7  perseant 	{
    559  1.40       chs 		if (lfs_dostats)
    560   1.7  perseant 			++lfs_stats.wait_exceeded;
    561  1.36  perseant #ifdef DEBUG_LFS
    562  1.16  perseant 		printf("lfs_check: waiting: count=%d, bytes=%ld\n",
    563  1.16  perseant 			locked_queue_count, locked_queue_bytes);
    564  1.16  perseant #endif
    565   1.7  perseant 		error = tsleep(&locked_queue_count, PCATCH | PUSER,
    566   1.7  perseant 			       "buffers", hz * LFS_BUFWAIT);
    567  1.33  perseant 		if (error != EWOULDBLOCK)
    568  1.33  perseant 			break;
    569  1.18  perseant 		/*
    570  1.18  perseant 		 * lfs_flush might not flush all the buffers, if some of the
    571  1.33  perseant 		 * inodes were locked or if most of them were Ifile blocks
    572  1.33  perseant 		 * and we weren't asked to checkpoint.  Try flushing again
    573  1.33  perseant 		 * to keep us from blocking indefinitely.
    574  1.18  perseant 		 */
    575  1.41  perseant 		if (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS ||
    576  1.41  perseant 		    locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES)
    577  1.18  perseant 		{
    578  1.18  perseant 			++fs->lfs_writer;
    579  1.33  perseant 			lfs_flush(fs, flags | SEGM_CKP);
    580  1.40       chs 			if (--fs->lfs_writer == 0)
    581  1.18  perseant 				wakeup(&fs->lfs_dirops);
    582  1.18  perseant 		}
    583   1.7  perseant 	}
    584   1.7  perseant 	return (error);
    585   1.7  perseant }
    586   1.1   mycroft 
    587   1.7  perseant /*
    588   1.7  perseant  * Allocate a new buffer header.
    589   1.7  perseant  */
    590  1.34  perseant #ifdef MALLOCLOG
    591  1.34  perseant # define DOMALLOC(S, T, F) _malloc((S), (T), (F), file, line)
    592  1.34  perseant struct buf *
    593  1.36  perseant lfs_newbuf_malloclog(struct lfs *fs, struct vnode *vp, ufs_daddr_t daddr, size_t size, char *file, int line)
    594  1.34  perseant #else
    595  1.35  perseant # define DOMALLOC(S, T, F) malloc((S), (T), (F))
    596   1.7  perseant struct buf *
    597  1.36  perseant lfs_newbuf(struct lfs *fs, struct vnode *vp, ufs_daddr_t daddr, size_t size)
    598  1.34  perseant #endif
    599   1.7  perseant {
    600   1.7  perseant 	struct buf *bp;
    601   1.7  perseant 	size_t nbytes;
    602   1.7  perseant 	int s;
    603   1.7  perseant 
    604  1.36  perseant 	nbytes = roundup(size, fsbtob(fs, 1));
    605   1.7  perseant 
    606  1.34  perseant 	bp = DOMALLOC(sizeof(struct buf), M_SEGMENT, M_WAITOK);
    607   1.7  perseant 	bzero(bp, sizeof(struct buf));
    608  1.43  perseant 	if (nbytes) {
    609  1.34  perseant 		bp->b_data = DOMALLOC(nbytes, M_SEGMENT, M_WAITOK);
    610   1.7  perseant 		bzero(bp->b_data, nbytes);
    611   1.7  perseant 	}
    612   1.7  perseant #ifdef DIAGNOSTIC
    613  1.40       chs 	if (vp == NULL)
    614   1.7  perseant 		panic("vp is NULL in lfs_newbuf");
    615  1.40       chs 	if (bp == NULL)
    616   1.7  perseant 		panic("bp is NULL after malloc in lfs_newbuf");
    617   1.1   mycroft #endif
    618   1.7  perseant 	s = splbio();
    619   1.7  perseant 	bgetvp(vp, bp);
    620   1.7  perseant 	splx(s);
    621   1.7  perseant 
    622  1.43  perseant 	bp->b_saveaddr = (caddr_t)fs;
    623   1.7  perseant 	bp->b_bufsize = size;
    624   1.7  perseant 	bp->b_bcount = size;
    625   1.7  perseant 	bp->b_lblkno = daddr;
    626   1.7  perseant 	bp->b_blkno = daddr;
    627   1.7  perseant 	bp->b_error = 0;
    628   1.7  perseant 	bp->b_resid = 0;
    629   1.7  perseant 	bp->b_iodone = lfs_callback;
    630   1.7  perseant 	bp->b_flags |= B_BUSY | B_CALL | B_NOCACHE;
    631   1.7  perseant 
    632   1.7  perseant 	return (bp);
    633   1.7  perseant }
    634   1.7  perseant 
    635  1.34  perseant #ifdef MALLOCLOG
    636  1.34  perseant # define DOFREE(A, T) _free((A), (T), file, line)
    637  1.34  perseant void
    638  1.36  perseant lfs_freebuf_malloclog(struct buf *bp, char *file, int line)
    639  1.34  perseant #else
    640  1.34  perseant # define DOFREE(A, T) free((A), (T))
    641   1.7  perseant void
    642  1.36  perseant lfs_freebuf(struct buf *bp)
    643  1.34  perseant #endif
    644   1.7  perseant {
    645   1.7  perseant 	int s;
    646   1.7  perseant 
    647   1.7  perseant 	s = splbio();
    648  1.40       chs 	if (bp->b_vp)
    649   1.7  perseant 		brelvp(bp);
    650   1.7  perseant 	splx(s);
    651   1.7  perseant 	if (!(bp->b_flags & B_INVAL)) { /* B_INVAL indicates a "fake" buffer */
    652  1.34  perseant 		DOFREE(bp->b_data, M_SEGMENT);
    653   1.7  perseant 		bp->b_data = NULL;
    654   1.1   mycroft 	}
    655  1.34  perseant 	DOFREE(bp, M_SEGMENT);
    656   1.7  perseant }
    657   1.1   mycroft 
    658   1.7  perseant /*
    659   1.7  perseant  * Definitions for the buffer free lists.
    660   1.7  perseant  */
    661   1.7  perseant #define BQUEUES		4		/* number of free buffer queues */
    662   1.7  perseant 
    663   1.7  perseant #define BQ_LOCKED	0		/* super-blocks &c */
    664   1.7  perseant #define BQ_LRU		1		/* lru, useful buffers */
    665   1.7  perseant #define BQ_AGE		2		/* rubbish */
    666   1.7  perseant #define BQ_EMPTY	3		/* buffer headers with no memory */
    667   1.7  perseant 
    668   1.7  perseant extern TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES];
    669   1.7  perseant 
    670   1.7  perseant /*
    671   1.7  perseant  * Return a count of buffers on the "locked" queue.
    672  1.27  perseant  * Don't count malloced buffers, since they don't detract from the total.
    673   1.7  perseant  */
    674   1.7  perseant void
    675  1.43  perseant lfs_countlocked(int *count, long *bytes, char *msg)
    676   1.7  perseant {
    677  1.17  augustss 	struct buf *bp;
    678  1.17  augustss 	int n = 0;
    679  1.17  augustss 	long int size = 0L;
    680   1.7  perseant 
    681   1.7  perseant 	for (bp = bufqueues[BQ_LOCKED].tqh_first; bp;
    682   1.7  perseant 	    bp = bp->b_freelist.tqe_next) {
    683  1.27  perseant 		if (bp->b_flags & B_CALL) /* Malloced buffer */
    684  1.27  perseant 			continue;
    685   1.7  perseant 		n++;
    686   1.7  perseant 		size += bp->b_bufsize;
    687  1.27  perseant #ifdef DEBUG_LOCKED_LIST
    688  1.27  perseant 		if (n > nbuf)
    689  1.27  perseant 			panic("lfs_countlocked: this can't happen: more"
    690  1.27  perseant 			      " buffers locked than exist");
    691  1.27  perseant #endif
    692   1.7  perseant 	}
    693  1.43  perseant #ifdef DEBUG_LOCKED_LIST
    694  1.27  perseant 	/* Theoretically this function never really does anything */
    695  1.27  perseant 	if (n != *count)
    696  1.43  perseant 		printf("lfs_countlocked: %s: adjusted buf count from %d to %d\n",
    697  1.43  perseant 		       msg, *count, n);
    698  1.27  perseant 	if (size != *bytes)
    699  1.43  perseant 		printf("lfs_countlocked: %s: adjusted byte count from %ld to %ld\n",
    700  1.43  perseant 		       msg, *bytes, size);
    701  1.27  perseant #endif
    702   1.7  perseant 	*count = n;
    703   1.7  perseant 	*bytes = size;
    704   1.7  perseant 	return;
    705   1.1   mycroft }
    706