Home | History | Annotate | Line # | Download | only in lfs
lfs_bio.c revision 1.52
      1  1.52      yamt /*	$NetBSD: lfs_bio.c,v 1.52 2002/12/28 14:39:08 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  1.52      yamt __KERNEL_RCSID(0, "$NetBSD: lfs_bio.c,v 1.52 2002/12/28 14:39:08 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  1.52      yamt int lfs_reservebuf(struct lfs *, struct vnode *vp, struct vnode *vp2,
    119  1.52      yamt     int, int);
    120  1.52      yamt int lfs_reserveavail(struct lfs *, struct vnode *vp, struct vnode *vp2, int);
    121  1.51      yamt 
    122  1.51      yamt int
    123  1.51      yamt lfs_fits_buf(struct lfs *fs, int n, int bytes)
    124  1.51      yamt {
    125  1.51      yamt 	int count_fit =
    126  1.51      yamt 	    (locked_queue_count + locked_queue_rcount + n < LFS_WAIT_BUFS);
    127  1.51      yamt 	int bytes_fit =
    128  1.51      yamt 	    (locked_queue_bytes + locked_queue_rbytes + bytes < LFS_WAIT_BYTES);
    129  1.51      yamt 
    130  1.51      yamt #ifdef DEBUG_LFS
    131  1.51      yamt 	if (!count_fit) {
    132  1.51      yamt 		printf("lfs_fits_buf: no fit count: %d + %d + %d >= %d\n",
    133  1.51      yamt 			locked_queue_count, locked_queue_rcount,
    134  1.51      yamt 			n, LFS_WAIT_BUFS);
    135  1.51      yamt 	}
    136  1.51      yamt 	if (!bytes_fit) {
    137  1.51      yamt 		printf("lfs_fits_buf: no fit bytes: %ld + %ld + %d >= %d\n",
    138  1.51      yamt 			locked_queue_bytes, locked_queue_rbytes,
    139  1.51      yamt 			bytes, LFS_WAIT_BYTES);
    140  1.51      yamt 	}
    141  1.51      yamt #endif /* DEBUG_LFS */
    142  1.51      yamt 
    143  1.51      yamt 	return (count_fit && bytes_fit);
    144  1.51      yamt }
    145  1.51      yamt 
    146  1.52      yamt /* ARGSUSED */
    147  1.51      yamt int
    148  1.52      yamt lfs_reservebuf(struct lfs *fs, struct vnode *vp, struct vnode *vp2,
    149  1.52      yamt     int n, int bytes)
    150  1.51      yamt {
    151  1.51      yamt 	KASSERT(locked_queue_rcount >= 0);
    152  1.51      yamt 	KASSERT(locked_queue_rbytes >= 0);
    153  1.51      yamt 
    154  1.51      yamt 	while (n > 0 && !lfs_fits_buf(fs, n, bytes)) {
    155  1.51      yamt 		int error;
    156  1.51      yamt 
    157  1.51      yamt 		++fs->lfs_writer;
    158  1.51      yamt 		lfs_flush(fs, 0);
    159  1.51      yamt 		if (--fs->lfs_writer == 0)
    160  1.51      yamt 			wakeup(&fs->lfs_dirops);
    161  1.51      yamt 
    162  1.51      yamt 		error = tsleep(&locked_queue_count, PCATCH | PUSER,
    163  1.51      yamt 		    "lfsresbuf", hz * LFS_BUFWAIT);
    164  1.51      yamt 		if (error && error != EWOULDBLOCK)
    165  1.51      yamt 			return error;
    166  1.51      yamt 	}
    167  1.51      yamt 
    168  1.51      yamt 	locked_queue_rcount += n;
    169  1.51      yamt 	locked_queue_rbytes += bytes;
    170  1.51      yamt 
    171  1.51      yamt 	KASSERT(locked_queue_rcount >= 0);
    172  1.51      yamt 	KASSERT(locked_queue_rbytes >= 0);
    173  1.51      yamt 
    174  1.51      yamt 	return 0;
    175  1.51      yamt }
    176  1.51      yamt 
    177  1.51      yamt /*
    178  1.27  perseant  * Try to reserve some blocks, prior to performing a sensitive operation that
    179  1.27  perseant  * requires the vnode lock to be honored.  If there is not enough space, give
    180  1.27  perseant  * up the vnode lock temporarily and wait for the space to become available.
    181  1.27  perseant  *
    182  1.44  perseant  * Called with vp locked.  (Note nowever that if fsb < 0, vp is ignored.)
    183  1.46      yamt  *
    184  1.47      yamt  * XXX YAMT - it isn't safe to unlock vp here
    185  1.47      yamt  * because the node might be modified while we sleep.
    186  1.47      yamt  * (eg. cached states like i_offset might be stale,
    187  1.47      yamt  *  the vnode might be truncated, etc..)
    188  1.47      yamt  * maybe we should have a way to restart the vnode op. (EVOPRESTART?)
    189  1.27  perseant  */
    190  1.27  perseant int
    191  1.52      yamt lfs_reserveavail(struct lfs *fs, struct vnode *vp, struct vnode *vp2, int fsb)
    192  1.27  perseant {
    193  1.27  perseant 	CLEANERINFO *cip;
    194  1.27  perseant 	struct buf *bp;
    195  1.27  perseant 	int error, slept;
    196  1.27  perseant 
    197  1.27  perseant 	slept = 0;
    198  1.45      yamt 	while (fsb > 0 && !lfs_fits(fs, fsb + fs->lfs_ravail) &&
    199  1.45      yamt 	    vp != fs->lfs_unlockvp) {
    200  1.49      yamt #if 0
    201  1.49      yamt 		/*
    202  1.49      yamt 		 * XXX ideally, we should unlock vnodes here
    203  1.49      yamt 		 * because we might sleep very long time.
    204  1.49      yamt 		 */
    205  1.27  perseant 		VOP_UNLOCK(vp, 0);
    206  1.52      yamt 		if (vp2 != NULL) {
    207  1.52      yamt 			VOP_UNLOCK(vp2, 0);
    208  1.52      yamt 		}
    209  1.50      yamt #else
    210  1.50      yamt 		/*
    211  1.50      yamt 		 * XXX since we'll sleep for cleaner with vnode lock holding,
    212  1.50      yamt 		 * deadlock will occur if cleaner wants to acquire the vnode
    213  1.50      yamt 		 * lock.
    214  1.50      yamt 		 * (eg. lfs_markv -> lfs_fastvget -> getnewvnode -> vclean)
    215  1.50      yamt 		 */
    216  1.49      yamt #endif
    217  1.27  perseant 
    218  1.30  perseant 		if (!slept) {
    219  1.30  perseant #ifdef DEBUG
    220  1.27  perseant 			printf("lfs_reserve: waiting for %ld (bfree = %d,"
    221  1.30  perseant 			       " est_bfree = %d)\n",
    222  1.44  perseant 			       fsb + fs->lfs_ravail, fs->lfs_bfree,
    223  1.30  perseant 			       LFS_EST_BFREE(fs));
    224  1.30  perseant #endif
    225  1.30  perseant 		}
    226  1.27  perseant 		++slept;
    227  1.27  perseant 
    228  1.27  perseant 		/* Wake up the cleaner */
    229  1.27  perseant 		LFS_CLEANERINFO(cip, fs, bp);
    230  1.31  perseant 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 0);
    231  1.27  perseant 		wakeup(&lfs_allclean_wakeup);
    232  1.27  perseant 		wakeup(&fs->lfs_nextseg);
    233  1.27  perseant 
    234  1.27  perseant 		error = tsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_reserve",
    235  1.27  perseant 			       0);
    236  1.49      yamt #if 0
    237  1.27  perseant 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX use lockstatus */
    238  1.52      yamt 		vn_lock(vp2, LK_EXCLUSIVE | LK_RETRY); /* XXX use lockstatus */
    239  1.49      yamt #endif
    240  1.27  perseant 		if (error)
    241  1.27  perseant 			return error;
    242  1.27  perseant 	}
    243  1.44  perseant #ifdef DEBUG
    244  1.27  perseant 	if (slept)
    245  1.27  perseant 		printf("lfs_reserve: woke up\n");
    246  1.44  perseant #endif
    247  1.44  perseant 	fs->lfs_ravail += fsb;
    248  1.51      yamt 
    249  1.27  perseant 	return 0;
    250  1.27  perseant }
    251  1.27  perseant 
    252  1.51      yamt #ifdef DIAGNOSTIC
    253  1.51      yamt int lfs_rescount;
    254  1.51      yamt int lfs_rescountdirop;
    255  1.51      yamt #endif
    256  1.51      yamt 
    257  1.51      yamt int
    258  1.52      yamt lfs_reserve(struct lfs *fs, struct vnode *vp, struct vnode *vp2, int fsb)
    259  1.51      yamt {
    260  1.51      yamt 	int error;
    261  1.51      yamt 	int cantwait;
    262  1.51      yamt 
    263  1.52      yamt 	KASSERT(fsb < 0 || VOP_ISLOCKED(vp));
    264  1.52      yamt 	KASSERT(vp2 == NULL || fsb < 0 || VOP_ISLOCKED(vp2));
    265  1.52      yamt 
    266  1.52      yamt 	/*
    267  1.52      yamt 	 * XXX
    268  1.52      yamt 	 * vref vnodes here so that cleaner doesn't try to reuse them.
    269  1.52      yamt 	 */
    270  1.52      yamt 	lfs_vref(vp);
    271  1.52      yamt 	if (vp2 != NULL) {
    272  1.52      yamt 		lfs_vref(vp2);
    273  1.52      yamt 	}
    274  1.52      yamt 
    275  1.51      yamt 	cantwait = (VTOI(vp)->i_flag & IN_ADIROP);
    276  1.51      yamt #ifdef DIAGNOSTIC
    277  1.51      yamt 	if (cantwait) {
    278  1.51      yamt 		if (fsb > 0)
    279  1.51      yamt 			lfs_rescountdirop++;
    280  1.51      yamt 		else if (fsb < 0)
    281  1.51      yamt 			lfs_rescountdirop--;
    282  1.51      yamt 		if (lfs_rescountdirop < 0)
    283  1.51      yamt 			panic("lfs_rescountdirop");
    284  1.51      yamt 	}
    285  1.51      yamt 	else {
    286  1.51      yamt 		if (fsb > 0)
    287  1.51      yamt 			lfs_rescount++;
    288  1.51      yamt 		else if (fsb < 0)
    289  1.51      yamt 			lfs_rescount--;
    290  1.51      yamt 		if (lfs_rescount < 0)
    291  1.51      yamt 			panic("lfs_rescount");
    292  1.51      yamt 	}
    293  1.51      yamt #endif
    294  1.51      yamt 	if (cantwait)
    295  1.51      yamt 		return 0;
    296  1.51      yamt 
    297  1.52      yamt 	error = lfs_reserveavail(fs, vp, vp2, fsb);
    298  1.51      yamt 	if (error)
    299  1.52      yamt 		goto done;
    300  1.51      yamt 
    301  1.51      yamt 	/*
    302  1.51      yamt 	 * XXX just a guess. should be more precise.
    303  1.51      yamt 	 */
    304  1.52      yamt 	error = lfs_reservebuf(fs, vp, vp2,
    305  1.52      yamt 	    fragstoblks(fs, fsb), fsbtob(fs, fsb));
    306  1.51      yamt 	if (error)
    307  1.52      yamt 		lfs_reserveavail(fs, vp, vp2, -fsb);
    308  1.52      yamt 
    309  1.52      yamt done:
    310  1.52      yamt 	lfs_vunref(vp);
    311  1.52      yamt 	if (vp2 != NULL) {
    312  1.52      yamt 		lfs_vunref(vp2);
    313  1.52      yamt 	}
    314  1.51      yamt 
    315  1.51      yamt 	return error;
    316  1.51      yamt }
    317   1.1   mycroft 
    318   1.1   mycroft int
    319  1.36  perseant lfs_bwrite(void *v)
    320   1.5  christos {
    321   1.1   mycroft 	struct vop_bwrite_args /* {
    322   1.1   mycroft 		struct buf *a_bp;
    323   1.5  christos 	} */ *ap = v;
    324  1.17  augustss 	struct buf *bp = ap->a_bp;
    325   1.7  perseant 
    326   1.7  perseant #ifdef DIAGNOSTIC
    327  1.23  perseant         if (VTOI(bp->b_vp)->i_lfs->lfs_ronly == 0 && (bp->b_flags & B_ASYNC)) {
    328   1.7  perseant 		panic("bawrite LFS buffer");
    329  1.16  perseant 	}
    330   1.7  perseant #endif /* DIAGNOSTIC */
    331   1.7  perseant 	return lfs_bwrite_ext(bp,0);
    332   1.7  perseant }
    333   1.7  perseant 
    334   1.7  perseant /*
    335  1.44  perseant  * Determine if there is enough room currently available to write fsb
    336  1.44  perseant  * blocks.  We need enough blocks for the new blocks, the current
    337  1.44  perseant  * inode blocks (including potentially the ifile inode), a summary block,
    338  1.44  perseant  * and the segment usage table, plus an ifile block.
    339   1.7  perseant  */
    340  1.32  perseant int
    341  1.36  perseant lfs_fits(struct lfs *fs, int fsb)
    342   1.7  perseant {
    343  1.26  perseant 	int needed;
    344  1.26  perseant 
    345  1.36  perseant 	needed = fsb + btofsb(fs, fs->lfs_sumsize) +
    346  1.44  perseant 		 ((howmany(fs->lfs_uinodes + 1, INOPB(fs)) + fs->lfs_segtabsz +
    347  1.44  perseant 		   1) << (fs->lfs_blktodb - fs->lfs_fsbtodb));
    348  1.27  perseant 
    349  1.26  perseant 	if (needed >= fs->lfs_avail) {
    350  1.27  perseant #ifdef DEBUG
    351  1.36  perseant 		printf("lfs_fits: no fit: fsb = %d, uinodes = %d, "
    352  1.26  perseant 		       "needed = %d, avail = %d\n",
    353  1.36  perseant 		       fsb, fs->lfs_uinodes, needed, fs->lfs_avail);
    354  1.26  perseant #endif
    355   1.7  perseant 		return 0;
    356   1.7  perseant 	}
    357   1.7  perseant 	return 1;
    358   1.7  perseant }
    359   1.7  perseant 
    360   1.7  perseant int
    361  1.44  perseant lfs_availwait(struct lfs *fs, int fsb)
    362  1.32  perseant {
    363  1.32  perseant 	int error;
    364  1.32  perseant 	CLEANERINFO *cip;
    365  1.32  perseant 	struct buf *cbp;
    366  1.32  perseant 
    367  1.44  perseant 	while (!lfs_fits(fs, fsb)) {
    368  1.32  perseant 		/*
    369  1.32  perseant 		 * Out of space, need cleaner to run.
    370  1.32  perseant 		 * Update the cleaner info, then wake it up.
    371  1.32  perseant 		 * Note the cleanerinfo block is on the ifile
    372  1.32  perseant 		 * so it CANT_WAIT.
    373  1.32  perseant 		 */
    374  1.32  perseant 		LFS_CLEANERINFO(cip, fs, cbp);
    375  1.32  perseant 		LFS_SYNC_CLEANERINFO(cip, fs, cbp, 0);
    376  1.32  perseant 
    377  1.32  perseant 		printf("lfs_availwait: out of available space, "
    378  1.32  perseant 		       "waiting on cleaner\n");
    379  1.32  perseant 
    380  1.32  perseant 		wakeup(&lfs_allclean_wakeup);
    381  1.32  perseant 		wakeup(&fs->lfs_nextseg);
    382  1.32  perseant #ifdef DIAGNOSTIC
    383  1.32  perseant 		if (fs->lfs_seglock && fs->lfs_lockpid == curproc->p_pid)
    384  1.32  perseant 			panic("lfs_availwait: deadlock");
    385  1.32  perseant #endif
    386  1.32  perseant 		error = tsleep(&fs->lfs_avail, PCATCH | PUSER, "cleaner", 0);
    387  1.32  perseant 		if (error)
    388  1.32  perseant 			return (error);
    389  1.32  perseant 	}
    390  1.32  perseant 	return 0;
    391  1.32  perseant }
    392  1.32  perseant 
    393  1.32  perseant int
    394  1.36  perseant lfs_bwrite_ext(struct buf *bp, int flags)
    395   1.7  perseant {
    396   1.1   mycroft 	struct lfs *fs;
    397   1.1   mycroft 	struct inode *ip;
    398  1.51      yamt 	int fsb, s;
    399  1.48      yamt 
    400  1.48      yamt 	KASSERT(bp->b_flags & B_BUSY);
    401  1.48      yamt 	KASSERT(flags & BW_CLEAN || !(bp->b_flags & B_CALL));
    402  1.48      yamt 
    403   1.1   mycroft 	/*
    404  1.16  perseant 	 * Don't write *any* blocks if we're mounted read-only.
    405  1.16  perseant 	 * In particular the cleaner can't write blocks either.
    406  1.16  perseant 	 */
    407  1.40       chs         if (VTOI(bp->b_vp)->i_lfs->lfs_ronly) {
    408  1.32  perseant 		bp->b_flags &= ~(B_DELWRI | B_READ | B_ERROR);
    409  1.32  perseant 		LFS_UNLOCK_BUF(bp);
    410  1.40       chs 		if (bp->b_flags & B_CALL)
    411  1.16  perseant 			bp->b_flags &= ~B_BUSY;
    412  1.16  perseant 		else
    413  1.16  perseant 			brelse(bp);
    414  1.16  perseant 		return EROFS;
    415  1.16  perseant 	}
    416  1.16  perseant 
    417  1.16  perseant 	/*
    418   1.1   mycroft 	 * Set the delayed write flag and use reassignbuf to move the buffer
    419   1.1   mycroft 	 * from the clean list to the dirty one.
    420   1.1   mycroft 	 *
    421   1.1   mycroft 	 * Set the B_LOCKED flag and unlock the buffer, causing brelse to move
    422   1.1   mycroft 	 * the buffer onto the LOCKED free list.  This is necessary, otherwise
    423   1.1   mycroft 	 * getnewbuf() would try to reclaim the buffers using bawrite, which
    424   1.1   mycroft 	 * isn't going to work.
    425   1.1   mycroft 	 *
    426   1.1   mycroft 	 * XXX we don't let meta-data writes run out of space because they can
    427   1.1   mycroft 	 * come from the segment writer.  We need to make sure that there is
    428   1.1   mycroft 	 * enough space reserved so that there's room to write meta-data
    429   1.1   mycroft 	 * blocks.
    430   1.1   mycroft 	 */
    431   1.1   mycroft 	if (!(bp->b_flags & B_LOCKED)) {
    432   1.1   mycroft 		fs = VFSTOUFS(bp->b_vp->v_mount)->um_lfs;
    433  1.36  perseant 		fsb = fragstofsb(fs, numfrags(fs, bp->b_bcount));
    434   1.7  perseant 
    435   1.7  perseant 		ip = VTOI(bp->b_vp);
    436  1.48      yamt 		if (flags & BW_CLEAN) {
    437  1.26  perseant 			LFS_SET_UINO(ip, IN_CLEANING);
    438   1.7  perseant 		} else {
    439  1.36  perseant 			LFS_SET_UINO(ip, IN_MODIFIED);
    440  1.36  perseant 			if (bp->b_lblkno >= 0)
    441  1.36  perseant 				LFS_SET_UINO(ip, IN_UPDATE);
    442   1.7  perseant 		}
    443  1.36  perseant 		fs->lfs_avail -= fsb;
    444  1.32  perseant 		bp->b_flags |= B_DELWRI;
    445  1.32  perseant 
    446  1.32  perseant 		LFS_LOCK_BUF(bp);
    447  1.43  perseant 		bp->b_flags &= ~(B_READ | B_DONE | B_ERROR);
    448  1.14      fvdl 		s = splbio();
    449   1.1   mycroft 		reassignbuf(bp, bp->b_vp);
    450   1.1   mycroft 		splx(s);
    451   1.1   mycroft 	}
    452   1.7  perseant 
    453  1.40       chs 	if (bp->b_flags & B_CALL)
    454   1.7  perseant 		bp->b_flags &= ~B_BUSY;
    455   1.7  perseant 	else
    456   1.7  perseant 		brelse(bp);
    457   1.7  perseant 
    458   1.1   mycroft 	return (0);
    459   1.1   mycroft }
    460   1.1   mycroft 
    461  1.24  perseant void
    462  1.36  perseant lfs_flush_fs(struct lfs *fs, int flags)
    463  1.11  perseant {
    464  1.40       chs 	if (fs->lfs_ronly == 0 && fs->lfs_dirops == 0)
    465  1.11  perseant 	{
    466  1.11  perseant 		/* disallow dirops during flush */
    467  1.24  perseant 		fs->lfs_writer++;
    468  1.11  perseant 
    469  1.11  perseant 		/*
    470  1.11  perseant 		 * We set the queue to 0 here because we
    471  1.11  perseant 		 * are about to write all the dirty
    472  1.11  perseant 		 * buffers we have.  If more come in
    473  1.11  perseant 		 * while we're writing the segment, they
    474  1.11  perseant 		 * may not get written, so we want the
    475  1.11  perseant 		 * count to reflect these new writes
    476  1.11  perseant 		 * after the segwrite completes.
    477  1.11  perseant 		 */
    478  1.40       chs 		if (lfs_dostats)
    479  1.11  perseant 			++lfs_stats.flush_invoked;
    480  1.24  perseant 		lfs_segwrite(fs->lfs_ivnode->v_mount, flags);
    481  1.11  perseant 
    482  1.11  perseant 		/* XXX KS - allow dirops again */
    483  1.40       chs 		if (--fs->lfs_writer == 0)
    484  1.24  perseant 			wakeup(&fs->lfs_dirops);
    485  1.11  perseant 	}
    486  1.11  perseant }
    487  1.11  perseant 
    488   1.1   mycroft /*
    489   1.1   mycroft  * XXX
    490   1.1   mycroft  * This routine flushes buffers out of the B_LOCKED queue when LFS has too
    491   1.1   mycroft  * many locked down.  Eventually the pageout daemon will simply call LFS
    492   1.1   mycroft  * when pages need to be reclaimed.  Note, we have one static count of locked
    493   1.1   mycroft  * buffers, so we can't have more than a single file system.  To make this
    494   1.1   mycroft  * work for multiple file systems, put the count into the mount structure.
    495   1.1   mycroft  */
    496   1.1   mycroft void
    497  1.36  perseant lfs_flush(struct lfs *fs, int flags)
    498   1.1   mycroft {
    499  1.17  augustss 	struct mount *mp, *nmp;
    500   1.7  perseant 
    501  1.40       chs 	if (lfs_dostats)
    502   1.7  perseant 		++lfs_stats.write_exceeded;
    503  1.40       chs 	if (lfs_writing && flags == 0) {/* XXX flags */
    504  1.16  perseant #ifdef DEBUG_LFS
    505  1.16  perseant 		printf("lfs_flush: not flushing because another flush is active\n");
    506  1.16  perseant #endif
    507   1.1   mycroft 		return;
    508  1.16  perseant 	}
    509   1.1   mycroft 	lfs_writing = 1;
    510   1.7  perseant 
    511   1.6      fvdl 	simple_lock(&mountlist_slock);
    512   1.6      fvdl 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
    513   1.6      fvdl 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
    514   1.6      fvdl 			nmp = mp->mnt_list.cqe_next;
    515   1.6      fvdl 			continue;
    516   1.6      fvdl 		}
    517  1.40       chs 		if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS, MFSNAMELEN) == 0)
    518  1.24  perseant 			lfs_flush_fs(((struct ufsmount *)mp->mnt_data)->ufsmount_u.lfs, flags);
    519   1.6      fvdl 		simple_lock(&mountlist_slock);
    520   1.6      fvdl 		nmp = mp->mnt_list.cqe_next;
    521   1.6      fvdl 		vfs_unbusy(mp);
    522   1.1   mycroft 	}
    523   1.6      fvdl 	simple_unlock(&mountlist_slock);
    524   1.7  perseant 
    525  1.43  perseant 	LFS_DEBUG_COUNTLOCKED("flush");
    526   1.7  perseant 
    527   1.1   mycroft 	lfs_writing = 0;
    528   1.1   mycroft }
    529   1.1   mycroft 
    530  1.41  perseant #define INOCOUNT(fs) howmany((fs)->lfs_uinodes, INOPB(fs))
    531  1.41  perseant #define INOBYTES(fs) ((fs)->lfs_uinodes * DINODE_SIZE)
    532  1.41  perseant 
    533   1.1   mycroft int
    534  1.36  perseant lfs_check(struct vnode *vp, ufs_daddr_t blkno, int flags)
    535   1.1   mycroft {
    536   1.1   mycroft 	int error;
    537  1.10  perseant 	struct lfs *fs;
    538  1.20  perseant 	struct inode *ip;
    539  1.13  perseant 	extern int lfs_dirvcount;
    540  1.10  perseant 
    541   1.1   mycroft 	error = 0;
    542  1.20  perseant 	ip = VTOI(vp);
    543   1.7  perseant 
    544   1.7  perseant 	/* If out of buffers, wait on writer */
    545   1.7  perseant 	/* XXX KS - if it's the Ifile, we're probably the cleaner! */
    546  1.20  perseant 	if (ip->i_number == LFS_IFILE_INUM)
    547  1.20  perseant 		return 0;
    548  1.20  perseant 	/* If we're being called from inside a dirop, don't sleep */
    549  1.20  perseant 	if (ip->i_flag & IN_ADIROP)
    550   1.7  perseant 		return 0;
    551   1.7  perseant 
    552  1.20  perseant 	fs = ip->i_lfs;
    553  1.20  perseant 
    554  1.20  perseant 	/*
    555  1.20  perseant 	 * If we would flush below, but dirops are active, sleep.
    556  1.20  perseant 	 * Note that a dirop cannot ever reach this code!
    557  1.20  perseant 	 */
    558  1.20  perseant 	while (fs->lfs_dirops > 0 &&
    559  1.41  perseant 	       (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS ||
    560  1.41  perseant                 locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES ||
    561  1.20  perseant                 lfs_dirvcount > LFS_MAXDIROP || fs->lfs_diropwait > 0))
    562  1.20  perseant 	{
    563  1.20  perseant 		++fs->lfs_diropwait;
    564  1.20  perseant 		tsleep(&fs->lfs_writer, PRIBIO+1, "bufdirop", 0);
    565  1.20  perseant 		--fs->lfs_diropwait;
    566  1.20  perseant 	}
    567  1.10  perseant 
    568  1.41  perseant 	if (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS ||
    569  1.41  perseant 	    locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES ||
    570  1.20  perseant 	    lfs_dirvcount > LFS_MAXDIROP || fs->lfs_diropwait > 0)
    571   1.7  perseant 	{
    572  1.10  perseant 		++fs->lfs_writer;
    573  1.10  perseant 		lfs_flush(fs, flags);
    574  1.40       chs 		if (--fs->lfs_writer == 0)
    575  1.10  perseant 			wakeup(&fs->lfs_dirops);
    576   1.7  perseant 	}
    577  1.11  perseant 
    578  1.41  perseant 	while (locked_queue_count + INOCOUNT(fs) > LFS_WAIT_BUFS
    579  1.41  perseant 	       || locked_queue_bytes + INOBYTES(fs) > LFS_WAIT_BYTES)
    580   1.7  perseant 	{
    581  1.40       chs 		if (lfs_dostats)
    582   1.7  perseant 			++lfs_stats.wait_exceeded;
    583  1.36  perseant #ifdef DEBUG_LFS
    584  1.16  perseant 		printf("lfs_check: waiting: count=%d, bytes=%ld\n",
    585  1.16  perseant 			locked_queue_count, locked_queue_bytes);
    586  1.16  perseant #endif
    587   1.7  perseant 		error = tsleep(&locked_queue_count, PCATCH | PUSER,
    588   1.7  perseant 			       "buffers", hz * LFS_BUFWAIT);
    589  1.33  perseant 		if (error != EWOULDBLOCK)
    590  1.33  perseant 			break;
    591  1.18  perseant 		/*
    592  1.18  perseant 		 * lfs_flush might not flush all the buffers, if some of the
    593  1.33  perseant 		 * inodes were locked or if most of them were Ifile blocks
    594  1.33  perseant 		 * and we weren't asked to checkpoint.  Try flushing again
    595  1.33  perseant 		 * to keep us from blocking indefinitely.
    596  1.18  perseant 		 */
    597  1.41  perseant 		if (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS ||
    598  1.41  perseant 		    locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES)
    599  1.18  perseant 		{
    600  1.18  perseant 			++fs->lfs_writer;
    601  1.33  perseant 			lfs_flush(fs, flags | SEGM_CKP);
    602  1.40       chs 			if (--fs->lfs_writer == 0)
    603  1.18  perseant 				wakeup(&fs->lfs_dirops);
    604  1.18  perseant 		}
    605   1.7  perseant 	}
    606   1.7  perseant 	return (error);
    607   1.7  perseant }
    608   1.1   mycroft 
    609   1.7  perseant /*
    610   1.7  perseant  * Allocate a new buffer header.
    611   1.7  perseant  */
    612  1.34  perseant #ifdef MALLOCLOG
    613  1.34  perseant # define DOMALLOC(S, T, F) _malloc((S), (T), (F), file, line)
    614  1.34  perseant struct buf *
    615  1.36  perseant lfs_newbuf_malloclog(struct lfs *fs, struct vnode *vp, ufs_daddr_t daddr, size_t size, char *file, int line)
    616  1.34  perseant #else
    617  1.35  perseant # define DOMALLOC(S, T, F) malloc((S), (T), (F))
    618   1.7  perseant struct buf *
    619  1.36  perseant lfs_newbuf(struct lfs *fs, struct vnode *vp, ufs_daddr_t daddr, size_t size)
    620  1.34  perseant #endif
    621   1.7  perseant {
    622   1.7  perseant 	struct buf *bp;
    623   1.7  perseant 	size_t nbytes;
    624   1.7  perseant 	int s;
    625   1.7  perseant 
    626  1.36  perseant 	nbytes = roundup(size, fsbtob(fs, 1));
    627   1.7  perseant 
    628  1.34  perseant 	bp = DOMALLOC(sizeof(struct buf), M_SEGMENT, M_WAITOK);
    629   1.7  perseant 	bzero(bp, sizeof(struct buf));
    630  1.43  perseant 	if (nbytes) {
    631  1.34  perseant 		bp->b_data = DOMALLOC(nbytes, M_SEGMENT, M_WAITOK);
    632   1.7  perseant 		bzero(bp->b_data, nbytes);
    633   1.7  perseant 	}
    634   1.7  perseant #ifdef DIAGNOSTIC
    635  1.40       chs 	if (vp == NULL)
    636   1.7  perseant 		panic("vp is NULL in lfs_newbuf");
    637  1.40       chs 	if (bp == NULL)
    638   1.7  perseant 		panic("bp is NULL after malloc in lfs_newbuf");
    639   1.1   mycroft #endif
    640   1.7  perseant 	s = splbio();
    641   1.7  perseant 	bgetvp(vp, bp);
    642   1.7  perseant 	splx(s);
    643   1.7  perseant 
    644  1.43  perseant 	bp->b_saveaddr = (caddr_t)fs;
    645   1.7  perseant 	bp->b_bufsize = size;
    646   1.7  perseant 	bp->b_bcount = size;
    647   1.7  perseant 	bp->b_lblkno = daddr;
    648   1.7  perseant 	bp->b_blkno = daddr;
    649   1.7  perseant 	bp->b_error = 0;
    650   1.7  perseant 	bp->b_resid = 0;
    651   1.7  perseant 	bp->b_iodone = lfs_callback;
    652   1.7  perseant 	bp->b_flags |= B_BUSY | B_CALL | B_NOCACHE;
    653   1.7  perseant 
    654   1.7  perseant 	return (bp);
    655   1.7  perseant }
    656   1.7  perseant 
    657  1.34  perseant #ifdef MALLOCLOG
    658  1.34  perseant # define DOFREE(A, T) _free((A), (T), file, line)
    659  1.34  perseant void
    660  1.36  perseant lfs_freebuf_malloclog(struct buf *bp, char *file, int line)
    661  1.34  perseant #else
    662  1.34  perseant # define DOFREE(A, T) free((A), (T))
    663   1.7  perseant void
    664  1.36  perseant lfs_freebuf(struct buf *bp)
    665  1.34  perseant #endif
    666   1.7  perseant {
    667   1.7  perseant 	int s;
    668   1.7  perseant 
    669   1.7  perseant 	s = splbio();
    670  1.40       chs 	if (bp->b_vp)
    671   1.7  perseant 		brelvp(bp);
    672   1.7  perseant 	splx(s);
    673   1.7  perseant 	if (!(bp->b_flags & B_INVAL)) { /* B_INVAL indicates a "fake" buffer */
    674  1.34  perseant 		DOFREE(bp->b_data, M_SEGMENT);
    675   1.7  perseant 		bp->b_data = NULL;
    676   1.1   mycroft 	}
    677  1.34  perseant 	DOFREE(bp, M_SEGMENT);
    678   1.7  perseant }
    679   1.1   mycroft 
    680   1.7  perseant /*
    681   1.7  perseant  * Definitions for the buffer free lists.
    682   1.7  perseant  */
    683   1.7  perseant #define BQUEUES		4		/* number of free buffer queues */
    684   1.7  perseant 
    685   1.7  perseant #define BQ_LOCKED	0		/* super-blocks &c */
    686   1.7  perseant #define BQ_LRU		1		/* lru, useful buffers */
    687   1.7  perseant #define BQ_AGE		2		/* rubbish */
    688   1.7  perseant #define BQ_EMPTY	3		/* buffer headers with no memory */
    689   1.7  perseant 
    690   1.7  perseant extern TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES];
    691   1.7  perseant 
    692   1.7  perseant /*
    693   1.7  perseant  * Return a count of buffers on the "locked" queue.
    694  1.27  perseant  * Don't count malloced buffers, since they don't detract from the total.
    695   1.7  perseant  */
    696   1.7  perseant void
    697  1.43  perseant lfs_countlocked(int *count, long *bytes, char *msg)
    698   1.7  perseant {
    699  1.17  augustss 	struct buf *bp;
    700  1.17  augustss 	int n = 0;
    701  1.17  augustss 	long int size = 0L;
    702   1.7  perseant 
    703   1.7  perseant 	for (bp = bufqueues[BQ_LOCKED].tqh_first; bp;
    704   1.7  perseant 	    bp = bp->b_freelist.tqe_next) {
    705  1.27  perseant 		if (bp->b_flags & B_CALL) /* Malloced buffer */
    706  1.27  perseant 			continue;
    707   1.7  perseant 		n++;
    708   1.7  perseant 		size += bp->b_bufsize;
    709  1.27  perseant #ifdef DEBUG_LOCKED_LIST
    710  1.27  perseant 		if (n > nbuf)
    711  1.27  perseant 			panic("lfs_countlocked: this can't happen: more"
    712  1.27  perseant 			      " buffers locked than exist");
    713  1.27  perseant #endif
    714   1.7  perseant 	}
    715  1.43  perseant #ifdef DEBUG_LOCKED_LIST
    716  1.27  perseant 	/* Theoretically this function never really does anything */
    717  1.27  perseant 	if (n != *count)
    718  1.43  perseant 		printf("lfs_countlocked: %s: adjusted buf count from %d to %d\n",
    719  1.43  perseant 		       msg, *count, n);
    720  1.27  perseant 	if (size != *bytes)
    721  1.43  perseant 		printf("lfs_countlocked: %s: adjusted byte count from %ld to %ld\n",
    722  1.43  perseant 		       msg, *bytes, size);
    723  1.27  perseant #endif
    724   1.7  perseant 	*count = n;
    725   1.7  perseant 	*bytes = size;
    726   1.7  perseant 	return;
    727   1.1   mycroft }
    728