Home | History | Annotate | Line # | Download | only in lfs
lfs_vnops.c revision 1.108
      1 /*	$NetBSD: lfs_vnops.c,v 1.108 2003/06/29 18:43:46 thorpej Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Konrad E. Schroder <perseant (at) hhhh.org>.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  * 3. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *	This product includes software developed by the NetBSD
     21  *	Foundation, Inc. and its contributors.
     22  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23  *    contributors may be used to endorse or promote products derived
     24  *    from this software without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36  * POSSIBILITY OF SUCH DAMAGE.
     37  */
     38 /*
     39  * Copyright (c) 1986, 1989, 1991, 1993, 1995
     40  *	The Regents of the University of California.  All rights reserved.
     41  *
     42  * Redistribution and use in source and binary forms, with or without
     43  * modification, are permitted provided that the following conditions
     44  * are met:
     45  * 1. Redistributions of source code must retain the above copyright
     46  *    notice, this list of conditions and the following disclaimer.
     47  * 2. Redistributions in binary form must reproduce the above copyright
     48  *    notice, this list of conditions and the following disclaimer in the
     49  *    documentation and/or other materials provided with the distribution.
     50  * 3. All advertising materials mentioning features or use of this software
     51  *    must display the following acknowledgement:
     52  *	This product includes software developed by the University of
     53  *	California, Berkeley and its contributors.
     54  * 4. Neither the name of the University nor the names of its contributors
     55  *    may be used to endorse or promote products derived from this software
     56  *    without specific prior written permission.
     57  *
     58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     68  * SUCH DAMAGE.
     69  *
     70  *	@(#)lfs_vnops.c	8.13 (Berkeley) 6/10/95
     71  */
     72 
     73 #include <sys/cdefs.h>
     74 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.108 2003/06/29 18:43:46 thorpej Exp $");
     75 
     76 #include <sys/param.h>
     77 #include <sys/systm.h>
     78 #include <sys/namei.h>
     79 #include <sys/resourcevar.h>
     80 #include <sys/kernel.h>
     81 #include <sys/file.h>
     82 #include <sys/stat.h>
     83 #include <sys/buf.h>
     84 #include <sys/proc.h>
     85 #include <sys/mount.h>
     86 #include <sys/vnode.h>
     87 #include <sys/malloc.h>
     88 #include <sys/pool.h>
     89 #include <sys/signalvar.h>
     90 
     91 #include <miscfs/fifofs/fifo.h>
     92 #include <miscfs/genfs/genfs.h>
     93 #include <miscfs/specfs/specdev.h>
     94 
     95 #include <ufs/ufs/inode.h>
     96 #include <ufs/ufs/dir.h>
     97 #include <ufs/ufs/ufsmount.h>
     98 #include <ufs/ufs/ufs_extern.h>
     99 
    100 #include <uvm/uvm.h>
    101 #include <uvm/uvm_pmap.h>
    102 #include <uvm/uvm_stat.h>
    103 #include <uvm/uvm_pager.h>
    104 
    105 #include <ufs/lfs/lfs.h>
    106 #include <ufs/lfs/lfs_extern.h>
    107 
    108 extern pid_t lfs_writer_daemon;
    109 extern int lfs_subsys_pages;
    110 extern int lfs_dirvcount;
    111 extern struct simplelock lfs_subsys_lock;
    112 
    113 /* Global vfs data structures for lfs. */
    114 int (**lfs_vnodeop_p)(void *);
    115 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = {
    116 	{ &vop_default_desc, vn_default_error },
    117 	{ &vop_lookup_desc, ufs_lookup },		/* lookup */
    118 	{ &vop_create_desc, lfs_create },		/* create */
    119 	{ &vop_whiteout_desc, ufs_whiteout },		/* whiteout */
    120 	{ &vop_mknod_desc, lfs_mknod },			/* mknod */
    121 	{ &vop_open_desc, ufs_open },			/* open */
    122 	{ &vop_close_desc, lfs_close },			/* close */
    123 	{ &vop_access_desc, ufs_access },		/* access */
    124 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
    125 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
    126 	{ &vop_read_desc, lfs_read },			/* read */
    127 	{ &vop_write_desc, lfs_write },			/* write */
    128 	{ &vop_lease_desc, ufs_lease_check },		/* lease */
    129 	{ &vop_ioctl_desc, ufs_ioctl },			/* ioctl */
    130 	{ &vop_fcntl_desc, lfs_fcntl },			/* fcntl */
    131 	{ &vop_poll_desc, ufs_poll },			/* poll */
    132 	{ &vop_kqfilter_desc, genfs_kqfilter },		/* kqfilter */
    133 	{ &vop_revoke_desc, ufs_revoke },		/* revoke */
    134 	{ &vop_mmap_desc, lfs_mmap },			/* mmap */
    135 	{ &vop_fsync_desc, lfs_fsync },			/* fsync */
    136 	{ &vop_seek_desc, ufs_seek },			/* seek */
    137 	{ &vop_remove_desc, lfs_remove },		/* remove */
    138 	{ &vop_link_desc, lfs_link },			/* link */
    139 	{ &vop_rename_desc, lfs_rename },		/* rename */
    140 	{ &vop_mkdir_desc, lfs_mkdir },			/* mkdir */
    141 	{ &vop_rmdir_desc, lfs_rmdir },			/* rmdir */
    142 	{ &vop_symlink_desc, lfs_symlink },		/* symlink */
    143 	{ &vop_readdir_desc, ufs_readdir },		/* readdir */
    144 	{ &vop_readlink_desc, ufs_readlink },		/* readlink */
    145 	{ &vop_abortop_desc, ufs_abortop },		/* abortop */
    146 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
    147 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
    148 	{ &vop_lock_desc, ufs_lock },			/* lock */
    149 	{ &vop_unlock_desc, ufs_unlock },		/* unlock */
    150 	{ &vop_bmap_desc, ufs_bmap },			/* bmap */
    151 	{ &vop_strategy_desc, lfs_strategy },		/* strategy */
    152 	{ &vop_print_desc, ufs_print },			/* print */
    153 	{ &vop_islocked_desc, ufs_islocked },		/* islocked */
    154 	{ &vop_pathconf_desc, ufs_pathconf },		/* pathconf */
    155 	{ &vop_advlock_desc, ufs_advlock },		/* advlock */
    156 	{ &vop_blkatoff_desc, lfs_blkatoff },		/* blkatoff */
    157 	{ &vop_valloc_desc, lfs_valloc },		/* valloc */
    158 	{ &vop_balloc_desc, lfs_balloc },		/* balloc */
    159 	{ &vop_vfree_desc, lfs_vfree },			/* vfree */
    160 	{ &vop_truncate_desc, lfs_truncate },		/* truncate */
    161 	{ &vop_update_desc, lfs_update },		/* update */
    162 	{ &vop_bwrite_desc, lfs_bwrite },		/* bwrite */
    163 	{ &vop_getpages_desc, lfs_getpages },		/* getpages */
    164 	{ &vop_putpages_desc, lfs_putpages },		/* putpages */
    165 	{ NULL, NULL }
    166 };
    167 const struct vnodeopv_desc lfs_vnodeop_opv_desc =
    168 	{ &lfs_vnodeop_p, lfs_vnodeop_entries };
    169 
    170 int (**lfs_specop_p)(void *);
    171 const struct vnodeopv_entry_desc lfs_specop_entries[] = {
    172 	{ &vop_default_desc, vn_default_error },
    173 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
    174 	{ &vop_create_desc, spec_create },		/* create */
    175 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
    176 	{ &vop_open_desc, spec_open },			/* open */
    177 	{ &vop_close_desc, lfsspec_close },		/* close */
    178 	{ &vop_access_desc, ufs_access },		/* access */
    179 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
    180 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
    181 	{ &vop_read_desc, ufsspec_read },		/* read */
    182 	{ &vop_write_desc, ufsspec_write },		/* write */
    183 	{ &vop_lease_desc, spec_lease_check },		/* lease */
    184 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
    185 	{ &vop_fcntl_desc, ufs_fcntl },			/* fcntl */
    186 	{ &vop_poll_desc, spec_poll },			/* poll */
    187 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
    188 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
    189 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
    190 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
    191 	{ &vop_seek_desc, spec_seek },			/* seek */
    192 	{ &vop_remove_desc, spec_remove },		/* remove */
    193 	{ &vop_link_desc, spec_link },			/* link */
    194 	{ &vop_rename_desc, spec_rename },		/* rename */
    195 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
    196 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
    197 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
    198 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
    199 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
    200 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
    201 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
    202 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
    203 	{ &vop_lock_desc, ufs_lock },			/* lock */
    204 	{ &vop_unlock_desc, ufs_unlock },		/* unlock */
    205 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
    206 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
    207 	{ &vop_print_desc, ufs_print },			/* print */
    208 	{ &vop_islocked_desc, ufs_islocked },		/* islocked */
    209 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
    210 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
    211 	{ &vop_blkatoff_desc, spec_blkatoff },		/* blkatoff */
    212 	{ &vop_valloc_desc, spec_valloc },		/* valloc */
    213 	{ &vop_vfree_desc, lfs_vfree },			/* vfree */
    214 	{ &vop_truncate_desc, spec_truncate },		/* truncate */
    215 	{ &vop_update_desc, lfs_update },		/* update */
    216 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
    217 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
    218 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
    219 	{ NULL, NULL }
    220 };
    221 const struct vnodeopv_desc lfs_specop_opv_desc =
    222 	{ &lfs_specop_p, lfs_specop_entries };
    223 
    224 int (**lfs_fifoop_p)(void *);
    225 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = {
    226 	{ &vop_default_desc, vn_default_error },
    227 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
    228 	{ &vop_create_desc, fifo_create },		/* create */
    229 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
    230 	{ &vop_open_desc, fifo_open },			/* open */
    231 	{ &vop_close_desc, lfsfifo_close },		/* close */
    232 	{ &vop_access_desc, ufs_access },		/* access */
    233 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
    234 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
    235 	{ &vop_read_desc, ufsfifo_read },		/* read */
    236 	{ &vop_write_desc, ufsfifo_write },		/* write */
    237 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
    238 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
    239 	{ &vop_fcntl_desc, ufs_fcntl },			/* fcntl */
    240 	{ &vop_poll_desc, fifo_poll },			/* poll */
    241 	{ &vop_kqfilter_desc, fifo_kqfilter },		/* kqfilter */
    242 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
    243 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
    244 	{ &vop_fsync_desc, fifo_fsync },		/* fsync */
    245 	{ &vop_seek_desc, fifo_seek },			/* seek */
    246 	{ &vop_remove_desc, fifo_remove },		/* remove */
    247 	{ &vop_link_desc, fifo_link },			/* link */
    248 	{ &vop_rename_desc, fifo_rename },		/* rename */
    249 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
    250 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
    251 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
    252 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
    253 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
    254 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
    255 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
    256 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
    257 	{ &vop_lock_desc, ufs_lock },			/* lock */
    258 	{ &vop_unlock_desc, ufs_unlock },		/* unlock */
    259 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
    260 	{ &vop_strategy_desc, fifo_strategy },		/* strategy */
    261 	{ &vop_print_desc, ufs_print },			/* print */
    262 	{ &vop_islocked_desc, ufs_islocked },		/* islocked */
    263 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
    264 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
    265 	{ &vop_blkatoff_desc, fifo_blkatoff },		/* blkatoff */
    266 	{ &vop_valloc_desc, fifo_valloc },		/* valloc */
    267 	{ &vop_vfree_desc, lfs_vfree },			/* vfree */
    268 	{ &vop_truncate_desc, fifo_truncate },		/* truncate */
    269 	{ &vop_update_desc, lfs_update },		/* update */
    270 	{ &vop_bwrite_desc, lfs_bwrite },		/* bwrite */
    271 	{ &vop_putpages_desc, fifo_putpages },		/* putpages */
    272 	{ NULL, NULL }
    273 };
    274 const struct vnodeopv_desc lfs_fifoop_opv_desc =
    275 	{ &lfs_fifoop_p, lfs_fifoop_entries };
    276 
    277 /*
    278  * A function version of LFS_ITIMES, for the UFS functions which call ITIMES
    279  */
    280 void
    281 lfs_itimes(struct inode *ip, struct timespec *acc, struct timespec *mod, struct timespec *cre)
    282 {
    283 	LFS_ITIMES(ip, acc, mod, cre);
    284 }
    285 
    286 #define	LFS_READWRITE
    287 #include <ufs/ufs/ufs_readwrite.c>
    288 #undef	LFS_READWRITE
    289 
    290 /*
    291  * Synch an open file.
    292  */
    293 /* ARGSUSED */
    294 int
    295 lfs_fsync(void *v)
    296 {
    297 	struct vop_fsync_args /* {
    298 		struct vnode *a_vp;
    299 		struct ucred *a_cred;
    300 		int a_flags;
    301 		off_t offlo;
    302 		off_t offhi;
    303 		struct lwp *a_l;
    304 	} */ *ap = v;
    305 	struct vnode *vp = ap->a_vp;
    306 	int error, wait;
    307 
    308 	/*
    309 	 * Trickle sync checks for need to do a checkpoint after possible
    310 	 * activity from the pagedaemon.
    311 	 */
    312 	if (ap->a_flags & FSYNC_LAZY) {
    313 		wakeup(&lfs_writer_daemon);
    314 		return 0;
    315 	}
    316 
    317 	wait = (ap->a_flags & FSYNC_WAIT);
    318 	simple_lock(&vp->v_interlock);
    319 	error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo),
    320 			round_page(ap->a_offhi),
    321 			PGO_CLEANIT | (wait ? PGO_SYNCIO : 0));
    322 	if (error)
    323 		return error;
    324 	error = VOP_UPDATE(vp, NULL, NULL, wait ? UPDATE_WAIT : 0);
    325 	if (wait && !VPISEMPTY(vp))
    326 		LFS_SET_UINO(VTOI(vp), IN_MODIFIED);
    327 
    328 	return error;
    329 }
    330 
    331 /*
    332  * Take IN_ADIROP off, then call ufs_inactive.
    333  */
    334 int
    335 lfs_inactive(void *v)
    336 {
    337 	struct vop_inactive_args /* {
    338 		struct vnode *a_vp;
    339 		struct lwp *a_l;
    340 	} */ *ap = v;
    341 
    342 	KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink);
    343 
    344 	lfs_unmark_vnode(ap->a_vp);
    345 
    346 	/*
    347 	 * The Ifile is only ever inactivated on unmount.
    348 	 * Streamline this process by not giving it more dirty blocks.
    349 	 */
    350 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) {
    351 		LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD);
    352 		VOP_UNLOCK(ap->a_vp, 0);
    353 		return 0;
    354 	}
    355 
    356 	return ufs_inactive(v);
    357 }
    358 
    359 /*
    360  * These macros are used to bracket UFS directory ops, so that we can
    361  * identify all the pages touched during directory ops which need to
    362  * be ordered and flushed atomically, so that they may be recovered.
    363  */
    364 /*
    365  * XXX KS - Because we have to mark nodes VDIROP in order to prevent
    366  * the cache from reclaiming them while a dirop is in progress, we must
    367  * also manage the number of nodes so marked (otherwise we can run out).
    368  * We do this by setting lfs_dirvcount to the number of marked vnodes; it
    369  * is decremented during segment write, when VDIROP is taken off.
    370  */
    371 #define	SET_DIROP(vp)		SET_DIROP2((vp), NULL)
    372 #define	SET_DIROP2(vp, vp2)	lfs_set_dirop((vp), (vp2))
    373 static int lfs_set_dirop(struct vnode *, struct vnode *);
    374 extern int lfs_dirvcount;
    375 extern int lfs_do_flush;
    376 
    377 #define	NRESERVE(fs)	(btofsb(fs, (NIADDR + 3 + (2 * NIADDR + 3)) << fs->lfs_bshift))
    378 
    379 static int
    380 lfs_set_dirop(struct vnode *vp, struct vnode *vp2)
    381 {
    382 	struct lfs *fs;
    383 	int error;
    384 
    385 	KASSERT(VOP_ISLOCKED(vp));
    386 	KASSERT(vp2 == NULL || VOP_ISLOCKED(vp2));
    387 
    388 	fs = VTOI(vp)->i_lfs;
    389 	/*
    390 	 * We might need one directory block plus supporting indirect blocks,
    391 	 * plus an inode block and ifile page for the new vnode.
    392 	 */
    393 	if ((error = lfs_reserve(fs, vp, vp2, NRESERVE(fs))) != 0)
    394 		return (error);
    395 
    396 	if (fs->lfs_dirops == 0)
    397 		lfs_check(vp, LFS_UNUSED_LBN, 0);
    398 	while (fs->lfs_writer || lfs_dirvcount > LFS_MAX_DIROP) {
    399 		if (fs->lfs_writer)
    400 			tsleep(&fs->lfs_dirops, PRIBIO + 1, "lfs_sdirop", 0);
    401 		if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) {
    402 			wakeup(&lfs_writer_daemon);
    403 			preempt(NULL);
    404 		}
    405 
    406 		if (lfs_dirvcount > LFS_MAX_DIROP) {
    407 #ifdef DEBUG_LFS
    408 			printf("lfs_set_dirop: sleeping with dirops=%d, "
    409 			       "dirvcount=%d\n", fs->lfs_dirops,
    410 			       lfs_dirvcount);
    411 #endif
    412 			if ((error = tsleep(&lfs_dirvcount, PCATCH|PUSER,
    413 					   "lfs_maxdirop", 0)) != 0) {
    414 				goto unreserve;
    415 			}
    416 		}
    417 	}
    418 	++fs->lfs_dirops;
    419 	fs->lfs_doifile = 1;
    420 
    421 	/* Hold a reference so SET_ENDOP will be happy */
    422 	vref(vp);
    423 	if (vp2)
    424 		vref(vp2);
    425 
    426 	return 0;
    427 
    428 unreserve:
    429 	lfs_reserve(fs, vp, vp2, -NRESERVE(fs));
    430 	return error;
    431 }
    432 
    433 #define	SET_ENDOP(fs, vp, str)	SET_ENDOP2((fs), (vp), NULL, (str))
    434 #define	SET_ENDOP2(fs, vp, vp2, str) {					\
    435 	--(fs)->lfs_dirops;						\
    436 	if (!(fs)->lfs_dirops) {					\
    437 		if ((fs)->lfs_nadirop) {				\
    438 			panic("SET_ENDOP: %s: no dirops but nadirop=%d", \
    439 			      (str), (fs)->lfs_nadirop);		\
    440 		}							\
    441 		wakeup(&(fs)->lfs_writer);				\
    442 		lfs_check((vp),LFS_UNUSED_LBN,0);			\
    443 	}								\
    444 	lfs_reserve((fs), vp, vp2, -NRESERVE(fs)); /* XXX */		\
    445 	vrele(vp);							\
    446 	if (vp2)							\
    447 		vrele(vp2);						\
    448 }
    449 
    450 #define	MARK_VNODE(dvp)	 do {						\
    451 	struct inode *_ip = VTOI(dvp);					\
    452 	struct lfs *_fs = _ip->i_lfs;					\
    453 									\
    454 	if (!((dvp)->v_flag & VDIROP)) {				\
    455 		(void)lfs_vref(dvp);					\
    456 		++lfs_dirvcount;					\
    457 		TAILQ_INSERT_TAIL(&_fs->lfs_dchainhd, _ip, i_lfs_dchain); \
    458 	}								\
    459 	(dvp)->v_flag |= VDIROP;					\
    460 	if (!(_ip->i_flag & IN_ADIROP)) {				\
    461 		++_fs->lfs_nadirop;					\
    462 	}								\
    463 	_ip->i_flag |= IN_ADIROP;					\
    464 } while (0)
    465 
    466 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp)
    467 
    468 void lfs_unmark_vnode(struct vnode *vp)
    469 {
    470 	struct inode *ip;
    471 
    472 	ip = VTOI(vp);
    473 
    474 	if (ip->i_flag & IN_ADIROP)
    475 		--ip->i_lfs->lfs_nadirop;
    476 	ip->i_flag &= ~IN_ADIROP;
    477 }
    478 
    479 int
    480 lfs_symlink(void *v)
    481 {
    482 	struct vop_symlink_args /* {
    483 		struct vnode *a_dvp;
    484 		struct vnode **a_vpp;
    485 		struct componentname *a_cnp;
    486 		struct vattr *a_vap;
    487 		char *a_target;
    488 	} */ *ap = v;
    489 	int error;
    490 
    491 	if ((error = SET_DIROP(ap->a_dvp)) != 0) {
    492 		vput(ap->a_dvp);
    493 		return error;
    494 	}
    495 	MARK_VNODE(ap->a_dvp);
    496 	error = ufs_symlink(ap);
    497 	UNMARK_VNODE(ap->a_dvp);
    498 	if (*(ap->a_vpp))
    499 		UNMARK_VNODE(*(ap->a_vpp));
    500 	SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"symlink");
    501 	return (error);
    502 }
    503 
    504 int
    505 lfs_mknod(void *v)
    506 {
    507 	struct vop_mknod_args	/* {
    508 		struct vnode *a_dvp;
    509 		struct vnode **a_vpp;
    510 		struct componentname *a_cnp;
    511 		struct vattr *a_vap;
    512 		} */ *ap = v;
    513 	struct vattr *vap = ap->a_vap;
    514 	struct vnode **vpp = ap->a_vpp;
    515 	struct componentname *cnp = ap->a_cnp;
    516 	struct inode *ip;
    517 	int error;
    518 	struct mount	*mp;
    519 	ino_t		ino;
    520 
    521 	if ((error = SET_DIROP(ap->a_dvp)) != 0) {
    522 		vput(ap->a_dvp);
    523 		return error;
    524 	}
    525 	MARK_VNODE(ap->a_dvp);
    526 	error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
    527 	    ap->a_dvp, vpp, cnp);
    528 	UNMARK_VNODE(ap->a_dvp);
    529 	if (*(ap->a_vpp))
    530 		UNMARK_VNODE(*(ap->a_vpp));
    531 
    532 	/* Either way we're done with the dirop at this point */
    533 	SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"mknod");
    534 
    535 	if (error)
    536 		return (error);
    537 
    538 	ip = VTOI(*vpp);
    539 	mp  = (*vpp)->v_mount;
    540 	ino = ip->i_number;
    541 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
    542 	if (vap->va_rdev != VNOVAL) {
    543 		/*
    544 		 * Want to be able to use this to make badblock
    545 		 * inodes, so don't truncate the dev number.
    546 		 */
    547 #if 0
    548 		ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev,
    549 		    UFS_MPNEEDSWAP((*vpp)->v_mount));
    550 #else
    551 		ip->i_ffs1_rdev = vap->va_rdev;
    552 #endif
    553 	}
    554 	/*
    555 	 * Call fsync to write the vnode so that we don't have to deal with
    556 	 * flushing it when it's marked VDIROP|VXLOCK.
    557 	 *
    558 	 * XXX KS - If we can't flush we also can't call vgone(), so must
    559 	 * return.  But, that leaves this vnode in limbo, also not good.
    560 	 * Can this ever happen (barring hardware failure)?
    561 	 */
    562 	if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0,
    563 	    cnp->cn_lwp)) != 0) {
    564 		printf("Couldn't fsync in mknod (ino %d)---what do I do?\n",
    565 		       VTOI(*vpp)->i_number);
    566 		return (error);
    567 	}
    568 	/*
    569 	 * Remove vnode so that it will be reloaded by VFS_VGET and
    570 	 * checked to see if it is an alias of an existing entry in
    571 	 * the inode cache.
    572 	 */
    573 	/* Used to be vput, but that causes us to call VOP_INACTIVE twice. */
    574 	VOP_UNLOCK(*vpp, 0);
    575 	lfs_vunref(*vpp);
    576 	(*vpp)->v_type = VNON;
    577 	vgone(*vpp);
    578 	error = VFS_VGET(mp, ino, vpp);
    579 	if (error != 0) {
    580 		*vpp = NULL;
    581 		return (error);
    582 	}
    583 	return (0);
    584 }
    585 
    586 int
    587 lfs_create(void *v)
    588 {
    589 	struct vop_create_args	/* {
    590 		struct vnode *a_dvp;
    591 		struct vnode **a_vpp;
    592 		struct componentname *a_cnp;
    593 		struct vattr *a_vap;
    594 	} */ *ap = v;
    595 	int error;
    596 
    597 	if ((error = SET_DIROP(ap->a_dvp)) != 0) {
    598 		vput(ap->a_dvp);
    599 		return error;
    600 	}
    601 	MARK_VNODE(ap->a_dvp);
    602 	error = ufs_create(ap);
    603 	UNMARK_VNODE(ap->a_dvp);
    604 	if (*(ap->a_vpp))
    605 		UNMARK_VNODE(*(ap->a_vpp));
    606 	SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"create");
    607 	return (error);
    608 }
    609 
    610 int
    611 lfs_mkdir(void *v)
    612 {
    613 	struct vop_mkdir_args	/* {
    614 		struct vnode *a_dvp;
    615 		struct vnode **a_vpp;
    616 		struct componentname *a_cnp;
    617 		struct vattr *a_vap;
    618 	} */ *ap = v;
    619 	int error;
    620 
    621 	if ((error = SET_DIROP(ap->a_dvp)) != 0) {
    622 		vput(ap->a_dvp);
    623 		return error;
    624 	}
    625 	MARK_VNODE(ap->a_dvp);
    626 	error = ufs_mkdir(ap);
    627 	UNMARK_VNODE(ap->a_dvp);
    628 	if (*(ap->a_vpp))
    629 		UNMARK_VNODE(*(ap->a_vpp));
    630 	SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"mkdir");
    631 	return (error);
    632 }
    633 
    634 int
    635 lfs_remove(void *v)
    636 {
    637 	struct vop_remove_args	/* {
    638 		struct vnode *a_dvp;
    639 		struct vnode *a_vp;
    640 		struct componentname *a_cnp;
    641 	} */ *ap = v;
    642 	struct vnode *dvp, *vp;
    643 	int error;
    644 
    645 	dvp = ap->a_dvp;
    646 	vp = ap->a_vp;
    647 	if ((error = SET_DIROP2(dvp, vp)) != 0) {
    648 		if (dvp == vp)
    649 			vrele(vp);
    650 		else
    651 			vput(vp);
    652 		vput(dvp);
    653 		return error;
    654 	}
    655 	MARK_VNODE(dvp);
    656 	MARK_VNODE(vp);
    657 	error = ufs_remove(ap);
    658 	UNMARK_VNODE(dvp);
    659 	UNMARK_VNODE(vp);
    660 
    661 	SET_ENDOP2(VTOI(dvp)->i_lfs, dvp, vp, "remove");
    662 	return (error);
    663 }
    664 
    665 int
    666 lfs_rmdir(void *v)
    667 {
    668 	struct vop_rmdir_args	/* {
    669 		struct vnodeop_desc *a_desc;
    670 		struct vnode *a_dvp;
    671 		struct vnode *a_vp;
    672 		struct componentname *a_cnp;
    673 	} */ *ap = v;
    674 	struct vnode *vp;
    675 	int error;
    676 
    677 	vp = ap->a_vp;
    678 	if ((error = SET_DIROP2(ap->a_dvp, ap->a_vp)) != 0) {
    679 		vrele(ap->a_dvp);
    680 		if (ap->a_vp != ap->a_dvp)
    681 			VOP_UNLOCK(ap->a_dvp, 0);
    682 		vput(vp);
    683 		return error;
    684 	}
    685 	MARK_VNODE(ap->a_dvp);
    686 	MARK_VNODE(vp);
    687 	error = ufs_rmdir(ap);
    688 	UNMARK_VNODE(ap->a_dvp);
    689 	UNMARK_VNODE(vp);
    690 
    691 	SET_ENDOP2(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vp, "rmdir");
    692 	return (error);
    693 }
    694 
    695 int
    696 lfs_link(void *v)
    697 {
    698 	struct vop_link_args	/* {
    699 		struct vnode *a_dvp;
    700 		struct vnode *a_vp;
    701 		struct componentname *a_cnp;
    702 	} */ *ap = v;
    703 	int error;
    704 
    705 	if ((error = SET_DIROP(ap->a_dvp)) != 0) {
    706 		vput(ap->a_dvp);
    707 		return error;
    708 	}
    709 	MARK_VNODE(ap->a_dvp);
    710 	error = ufs_link(ap);
    711 	UNMARK_VNODE(ap->a_dvp);
    712 	SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"link");
    713 	return (error);
    714 }
    715 
    716 int
    717 lfs_rename(void *v)
    718 {
    719 	struct vop_rename_args	/* {
    720 		struct vnode *a_fdvp;
    721 		struct vnode *a_fvp;
    722 		struct componentname *a_fcnp;
    723 		struct vnode *a_tdvp;
    724 		struct vnode *a_tvp;
    725 		struct componentname *a_tcnp;
    726 	} */ *ap = v;
    727 	struct vnode *tvp, *fvp, *tdvp, *fdvp;
    728 	struct componentname *tcnp, *fcnp;
    729 	int error;
    730 	struct lfs *fs;
    731 
    732 	fs = VTOI(ap->a_fdvp)->i_lfs;
    733 	tvp = ap->a_tvp;
    734 	tdvp = ap->a_tdvp;
    735 	tcnp = ap->a_tcnp;
    736 	fvp = ap->a_fvp;
    737 	fdvp = ap->a_fdvp;
    738 	fcnp = ap->a_fcnp;
    739 
    740 	/*
    741 	 * Check for cross-device rename.
    742 	 * If it is, we don't want to set dirops, just error out.
    743 	 * (In particular note that MARK_VNODE(tdvp) will DTWT on
    744 	 * a cross-device rename.)
    745 	 *
    746 	 * Copied from ufs_rename.
    747 	 */
    748 	if ((fvp->v_mount != tdvp->v_mount) ||
    749 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
    750 		error = EXDEV;
    751 		goto errout;
    752 	}
    753 
    754 	/*
    755 	 * Check to make sure we're not renaming a vnode onto itself
    756 	 * (deleting a hard link by renaming one name onto another);
    757 	 * if we are we can't recursively call VOP_REMOVE since that
    758 	 * would leave us with an unaccounted-for number of live dirops.
    759 	 *
    760 	 * Inline the relevant section of ufs_rename here, *before*
    761 	 * calling SET_DIROP2.
    762 	 */
    763 	if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) ||
    764 	    (VTOI(tdvp)->i_flags & APPEND))) {
    765 		error = EPERM;
    766 		goto errout;
    767 	}
    768 	if (fvp == tvp) {
    769 		if (fvp->v_type == VDIR) {
    770 			error = EINVAL;
    771 			goto errout;
    772 		}
    773 
    774 		/* Release destination completely. */
    775 		VOP_ABORTOP(tdvp, tcnp);
    776 		vput(tdvp);
    777 		vput(tvp);
    778 
    779 		/* Delete source. */
    780 		vrele(fvp);
    781 		fcnp->cn_flags &= ~(MODMASK | SAVESTART);
    782 		fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
    783 		fcnp->cn_nameiop = DELETE;
    784 		if ((error = relookup(fdvp, &fvp, fcnp))){
    785 			/* relookup blew away fdvp */
    786 			return (error);
    787 		}
    788 		return (VOP_REMOVE(fdvp, fvp, fcnp));
    789 	}
    790 
    791 	if ((error = SET_DIROP2(tdvp, tvp)) != 0)
    792 		goto errout;
    793 	MARK_VNODE(fdvp);
    794 	MARK_VNODE(tdvp);
    795 	MARK_VNODE(fvp);
    796 	if (tvp) {
    797 		MARK_VNODE(tvp);
    798 	}
    799 
    800 	error = ufs_rename(ap);
    801 	UNMARK_VNODE(fdvp);
    802 	UNMARK_VNODE(tdvp);
    803 	UNMARK_VNODE(fvp);
    804 	if (tvp) {
    805 		UNMARK_VNODE(tvp);
    806 	}
    807 	SET_ENDOP2(fs, tdvp, tvp, "rename");
    808 	return (error);
    809 
    810     errout:
    811 	VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */
    812 	if (tdvp == tvp)
    813 		vrele(tdvp);
    814 	else
    815 		vput(tdvp);
    816 	if (tvp)
    817 		vput(tvp);
    818 	VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */
    819 	vrele(fdvp);
    820 	vrele(fvp);
    821 	return (error);
    822 }
    823 
    824 /* XXX hack to avoid calling ITIMES in getattr */
    825 int
    826 lfs_getattr(void *v)
    827 {
    828 	struct vop_getattr_args /* {
    829 		struct vnode *a_vp;
    830 		struct vattr *a_vap;
    831 		struct ucred *a_cred;
    832 		struct lwp *a_l;
    833 	} */ *ap = v;
    834 	struct vnode *vp = ap->a_vp;
    835 	struct inode *ip = VTOI(vp);
    836 	struct vattr *vap = ap->a_vap;
    837 	struct lfs *fs = ip->i_lfs;
    838 	/*
    839 	 * Copy from inode table
    840 	 */
    841 	vap->va_fsid = ip->i_dev;
    842 	vap->va_fileid = ip->i_number;
    843 	vap->va_mode = ip->i_mode & ~IFMT;
    844 	vap->va_nlink = ip->i_nlink;
    845 	vap->va_uid = ip->i_uid;
    846 	vap->va_gid = ip->i_gid;
    847 	vap->va_rdev = (dev_t)ip->i_ffs1_rdev;
    848 	vap->va_size = vp->v_size;
    849 	vap->va_atime.tv_sec = ip->i_ffs1_atime;
    850 	vap->va_atime.tv_nsec = ip->i_ffs1_atimensec;
    851 	vap->va_mtime.tv_sec = ip->i_ffs1_mtime;
    852 	vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec;
    853 	vap->va_ctime.tv_sec = ip->i_ffs1_ctime;
    854 	vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec;
    855 	vap->va_flags = ip->i_flags;
    856 	vap->va_gen = ip->i_gen;
    857 	/* this doesn't belong here */
    858 	if (vp->v_type == VBLK)
    859 		vap->va_blocksize = BLKDEV_IOSIZE;
    860 	else if (vp->v_type == VCHR)
    861 		vap->va_blocksize = MAXBSIZE;
    862 	else
    863 		vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
    864 	vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks);
    865 	vap->va_type = vp->v_type;
    866 	vap->va_filerev = ip->i_modrev;
    867 	return (0);
    868 }
    869 
    870 /*
    871  * Check to make sure the inode blocks won't choke the buffer
    872  * cache, then call ufs_setattr as usual.
    873  */
    874 int
    875 lfs_setattr(void *v)
    876 {
    877 	struct vop_getattr_args /* {
    878 		struct vnode *a_vp;
    879 		struct vattr *a_vap;
    880 		struct ucred *a_cred;
    881 		struct lwp *a_l;
    882 	} */ *ap = v;
    883 	struct vnode *vp = ap->a_vp;
    884 
    885 	lfs_check(vp, LFS_UNUSED_LBN, 0);
    886 	return ufs_setattr(v);
    887 }
    888 
    889 /*
    890  * Close called
    891  *
    892  * XXX -- we were using ufs_close, but since it updates the
    893  * times on the inode, we might need to bump the uinodes
    894  * count.
    895  */
    896 /* ARGSUSED */
    897 int
    898 lfs_close(void *v)
    899 {
    900 	struct vop_close_args /* {
    901 		struct vnode *a_vp;
    902 		int  a_fflag;
    903 		struct ucred *a_cred;
    904 		struct lwp *a_l;
    905 	} */ *ap = v;
    906 	struct vnode *vp = ap->a_vp;
    907 	struct inode *ip = VTOI(vp);
    908 	struct timespec ts;
    909 
    910 	if (vp == ip->i_lfs->lfs_ivnode &&
    911 	    vp->v_mount->mnt_flag & MNT_UNMOUNT)
    912 		return 0;
    913 
    914 	if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) {
    915 		TIMEVAL_TO_TIMESPEC(&time, &ts);
    916 		LFS_ITIMES(ip, &ts, &ts, &ts);
    917 	}
    918 	return (0);
    919 }
    920 
    921 /*
    922  * Close wrapper for special devices.
    923  *
    924  * Update the times on the inode then do device close.
    925  */
    926 int
    927 lfsspec_close(void *v)
    928 {
    929 	struct vop_close_args /* {
    930 		struct vnode	*a_vp;
    931 		int		a_fflag;
    932 		struct ucred	*a_cred;
    933 		struct lwp	*a_l;
    934 	} */ *ap = v;
    935 	struct vnode	*vp;
    936 	struct inode	*ip;
    937 	struct timespec	ts;
    938 
    939 	vp = ap->a_vp;
    940 	ip = VTOI(vp);
    941 	if (vp->v_usecount > 1) {
    942 		TIMEVAL_TO_TIMESPEC(&time, &ts);
    943 		LFS_ITIMES(ip, &ts, &ts, &ts);
    944 	}
    945 	return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap));
    946 }
    947 
    948 /*
    949  * Close wrapper for fifo's.
    950  *
    951  * Update the times on the inode then do device close.
    952  */
    953 int
    954 lfsfifo_close(void *v)
    955 {
    956 	struct vop_close_args /* {
    957 		struct vnode	*a_vp;
    958 		int		a_fflag;
    959 		struct ucred	*a_cred;
    960 		struct lwp	*a_l;
    961 	} */ *ap = v;
    962 	struct vnode	*vp;
    963 	struct inode	*ip;
    964 	struct timespec	ts;
    965 
    966 	vp = ap->a_vp;
    967 	ip = VTOI(vp);
    968 	if (ap->a_vp->v_usecount > 1) {
    969 		TIMEVAL_TO_TIMESPEC(&time, &ts);
    970 		LFS_ITIMES(ip, &ts, &ts, &ts);
    971 	}
    972 	return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap));
    973 }
    974 
    975 /*
    976  * Reclaim an inode so that it can be used for other purposes.
    977  */
    978 int lfs_no_inactive = 0;
    979 
    980 int
    981 lfs_reclaim(void *v)
    982 {
    983 	struct vop_reclaim_args /* {
    984 		struct vnode *a_vp;
    985 		struct lwp *a_l;
    986 	} */ *ap = v;
    987 	struct vnode *vp = ap->a_vp;
    988 	struct inode *ip = VTOI(vp);
    989 	int error;
    990 
    991 	KASSERT(ip->i_nlink == ip->i_ffs_effnlink);
    992 
    993 	LFS_CLR_UINO(ip, IN_ALLMOD);
    994 	if ((error = ufs_reclaim(vp, ap->a_l)))
    995 		return (error);
    996 	pool_put(&lfs_dinode_pool, VTOI(vp)->i_din.ffs1_din);
    997 	pool_put(&lfs_inoext_pool, ip->inode_ext.lfs);
    998 	ip->inode_ext.lfs = NULL;
    999 	pool_put(&lfs_inode_pool, vp->v_data);
   1000 	vp->v_data = NULL;
   1001 	return (0);
   1002 }
   1003 
   1004 /*
   1005  * Read a block from a storage device.
   1006  * In order to avoid reading blocks that are in the process of being
   1007  * written by the cleaner---and hence are not mutexed by the normal
   1008  * buffer cache / page cache mechanisms---check for collisions before
   1009  * reading.
   1010  *
   1011  * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before*
   1012  * the active cleaner test.
   1013  *
   1014  * XXX This code assumes that lfs_markv makes synchronous checkpoints.
   1015  */
   1016 int
   1017 lfs_strategy(void *v)
   1018 {
   1019 	struct vop_strategy_args /* {
   1020 		struct buf *a_bp;
   1021 	} */ *ap = v;
   1022 	struct buf	*bp;
   1023 	struct lfs	*fs;
   1024 	struct vnode	*vp;
   1025 	struct inode	*ip;
   1026 	daddr_t		tbn;
   1027 	int		i, sn, error, slept;
   1028 
   1029 	bp = ap->a_bp;
   1030 	vp = bp->b_vp;
   1031 	ip = VTOI(vp);
   1032 	fs = ip->i_lfs;
   1033 
   1034 	/* lfs uses its strategy routine only for read */
   1035 	KASSERT(bp->b_flags & B_READ);
   1036 
   1037 	if (vp->v_type == VBLK || vp->v_type == VCHR)
   1038 		panic("lfs_strategy: spec");
   1039 	KASSERT(bp->b_bcount != 0);
   1040 	if (bp->b_blkno == bp->b_lblkno) {
   1041 		error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno,
   1042 				 NULL);
   1043 		if (error) {
   1044 			bp->b_error = error;
   1045 			bp->b_flags |= B_ERROR;
   1046 			biodone(bp);
   1047 			return (error);
   1048 		}
   1049 		if ((long)bp->b_blkno == -1) /* no valid data */
   1050 			clrbuf(bp);
   1051 	}
   1052 	if ((long)bp->b_blkno < 0) { /* block is not on disk */
   1053 		biodone(bp);
   1054 		return (0);
   1055 	}
   1056 
   1057 	slept = 1;
   1058 	simple_lock(&fs->lfs_interlock);
   1059 	while (slept && fs->lfs_seglock) {
   1060 		simple_unlock(&fs->lfs_interlock);
   1061 		/*
   1062 		 * Look through list of intervals.
   1063 		 * There will only be intervals to look through
   1064 		 * if the cleaner holds the seglock.
   1065 		 * Since the cleaner is synchronous, we can trust
   1066 		 * the list of intervals to be current.
   1067 		 */
   1068 		tbn = dbtofsb(fs, bp->b_blkno);
   1069 		sn = dtosn(fs, tbn);
   1070 		slept = 0;
   1071 		for (i = 0; i < fs->lfs_cleanind; i++) {
   1072 			if (sn == dtosn(fs, fs->lfs_cleanint[i]) &&
   1073 			    tbn >= fs->lfs_cleanint[i]) {
   1074 #ifdef DEBUG_LFS
   1075 				printf("lfs_strategy: ino %d lbn %" PRId64
   1076 				       " ind %d sn %d fsb %" PRIx32
   1077 				       " given sn %d fsb %" PRIx64 "\n",
   1078 					ip->i_number, bp->b_lblkno, i,
   1079 					dtosn(fs, fs->lfs_cleanint[i]),
   1080 					fs->lfs_cleanint[i], sn, tbn);
   1081 				printf("lfs_strategy: sleeping on ino %d lbn %"
   1082 				       PRId64 "\n", ip->i_number, bp->b_lblkno);
   1083 #endif
   1084 				tsleep(&fs->lfs_seglock, PRIBIO+1,
   1085 					"lfs_strategy", 0);
   1086 				/* Things may be different now; start over. */
   1087 				slept = 1;
   1088 				break;
   1089 			}
   1090 		}
   1091 		simple_lock(&fs->lfs_interlock);
   1092 	}
   1093 	simple_unlock(&fs->lfs_interlock);
   1094 
   1095 	vp = ip->i_devvp;
   1096 	bp->b_dev = vp->v_rdev;
   1097 	VOCALL (vp->v_op, VOFFSET(vop_strategy), ap);
   1098 	return (0);
   1099 }
   1100 
   1101 static void
   1102 lfs_flush_dirops(struct lfs *fs)
   1103 {
   1104 	struct inode *ip, *nip;
   1105 	struct vnode *vp;
   1106 	extern int lfs_dostats;
   1107 	struct segment *sp;
   1108 	int needunlock;
   1109 
   1110 	if (fs->lfs_ronly)
   1111 		return;
   1112 
   1113 	if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL)
   1114 		return;
   1115 
   1116 	/* XXX simplelock fs->lfs_dirops */
   1117 	while (fs->lfs_dirops > 0) {
   1118 		++fs->lfs_diropwait;
   1119 		tsleep(&fs->lfs_writer, PRIBIO+1, "pndirop", 0);
   1120 		--fs->lfs_diropwait;
   1121 	}
   1122 	/* disallow dirops during flush */
   1123 	fs->lfs_writer++;
   1124 
   1125 	if (lfs_dostats)
   1126 		++lfs_stats.flush_invoked;
   1127 
   1128 	/*
   1129 	 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops.
   1130 	 * Technically this is a checkpoint (the on-disk state is valid)
   1131 	 * even though we are leaving out all the file data.
   1132 	 */
   1133 	lfs_imtime(fs);
   1134 	lfs_seglock(fs, SEGM_CKP);
   1135 	sp = fs->lfs_sp;
   1136 
   1137 	/*
   1138 	 * lfs_writevnodes, optimized to get dirops out of the way.
   1139 	 * Only write dirops, and don't flush files' pages, only
   1140 	 * blocks from the directories.
   1141 	 *
   1142 	 * We don't need to vref these files because they are
   1143 	 * dirops and so hold an extra reference until the
   1144 	 * segunlock clears them of that status.
   1145 	 *
   1146 	 * We don't need to check for IN_ADIROP because we know that
   1147 	 * no dirops are active.
   1148 	 *
   1149 	 */
   1150 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
   1151 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
   1152 		vp = ITOV(ip);
   1153 
   1154 		/*
   1155 		 * All writes to directories come from dirops; all
   1156 		 * writes to files' direct blocks go through the page
   1157 		 * cache, which we're not touching.  Reads to files
   1158 		 * and/or directories will not be affected by writing
   1159 		 * directory blocks inodes and file inodes.  So we don't
   1160 		 * really need to lock.  If we don't lock, though,
   1161 		 * make sure that we don't clear IN_MODIFIED
   1162 		 * unnecessarily.
   1163 		 */
   1164 		if (vp->v_flag & VXLOCK)
   1165 			continue;
   1166 		if (vn_lock(vp, LK_EXCLUSIVE | LK_CANRECURSE |
   1167 			    LK_NOWAIT) == 0) {
   1168 			needunlock = 1;
   1169 		} else {
   1170 			printf("lfs_flush_dirops: flushing locked ino %d\n",
   1171 			       VTOI(vp)->i_number);
   1172 			needunlock = 0;
   1173 		}
   1174 		if (vp->v_type != VREG &&
   1175 		    ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) {
   1176 			lfs_writefile(fs, sp, vp);
   1177 			if (!VPISEMPTY(vp) && !WRITEINPROG(vp) &&
   1178 			    !(ip->i_flag & IN_ALLMOD)) {
   1179 				LFS_SET_UINO(ip, IN_MODIFIED);
   1180 			}
   1181 		}
   1182 		(void) lfs_writeinode(fs, sp, ip);
   1183 		if (needunlock)
   1184 			VOP_UNLOCK(vp, 0);
   1185 		else
   1186 			LFS_SET_UINO(ip, IN_MODIFIED);
   1187 	}
   1188 	/* We've written all the dirops there are */
   1189 	((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT);
   1190 	(void) lfs_writeseg(fs, sp);
   1191 	lfs_segunlock(fs);
   1192 
   1193 	if (--fs->lfs_writer == 0)
   1194 		wakeup(&fs->lfs_dirops);
   1195 }
   1196 
   1197 /*
   1198  * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}.
   1199  */
   1200 int
   1201 lfs_fcntl(void *v)
   1202 {
   1203         struct vop_fcntl_args /* {
   1204                 struct vnode *a_vp;
   1205                 u_long a_command;
   1206                 caddr_t  a_data;
   1207                 int  a_fflag;
   1208                 struct ucred *a_cred;
   1209                 struct lwp *a_l;
   1210         } */ *ap = v;
   1211 	struct timeval *tvp;
   1212 	BLOCK_INFO *blkiov;
   1213 	CLEANERINFO *cip;
   1214 	int blkcnt, error, oclean;
   1215 	struct lfs_fcntl_markv blkvp;
   1216 	struct proc *p;
   1217 	fsid_t *fsidp;
   1218 	struct lfs *fs;
   1219 	struct buf *bp;
   1220 	daddr_t off;
   1221 
   1222 	/* Only respect LFS fcntls on fs root or Ifile */
   1223 	if (VTOI(ap->a_vp)->i_number != ROOTINO &&
   1224 	    VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) {
   1225 		return ufs_fcntl(v);
   1226 	}
   1227 
   1228 	/* Avoid locking a draining lock */
   1229 	if (ap->a_vp->v_mount->mnt_flag & MNT_UNMOUNT) {
   1230 		return ESHUTDOWN;
   1231 	}
   1232 
   1233 	p = ap->a_l->l_proc;
   1234 	fs = VTOI(ap->a_vp)->i_lfs;
   1235 	fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsid;
   1236 
   1237 	switch (ap->a_command) {
   1238 	    case LFCNSEGWAITALL:
   1239 		fsidp = NULL;
   1240 		/* FALLSTHROUGH */
   1241 	    case LFCNSEGWAIT:
   1242 		tvp = (struct timeval *)ap->a_data;
   1243 		simple_lock(&fs->lfs_interlock);
   1244 		++fs->lfs_sleepers;
   1245 		simple_unlock(&fs->lfs_interlock);
   1246 		VOP_UNLOCK(ap->a_vp, 0);
   1247 
   1248 		error = lfs_segwait(fsidp, tvp);
   1249 
   1250 		VOP_LOCK(ap->a_vp, LK_EXCLUSIVE);
   1251 		simple_lock(&fs->lfs_interlock);
   1252 		if (--fs->lfs_sleepers == 0)
   1253 			wakeup(&fs->lfs_sleepers);
   1254 		simple_unlock(&fs->lfs_interlock);
   1255 		return error;
   1256 
   1257 	    case LFCNBMAPV:
   1258 	    case LFCNMARKV:
   1259 		if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
   1260 			return (error);
   1261 		blkvp = *(struct lfs_fcntl_markv *)ap->a_data;
   1262 
   1263 		blkcnt = blkvp.blkcnt;
   1264 		if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT)
   1265 			return (EINVAL);
   1266 		blkiov = malloc(blkcnt * sizeof(BLOCK_INFO), M_SEGMENT, M_WAITOK);
   1267 		if ((error = copyin(blkvp.blkiov, blkiov,
   1268 		     blkcnt * sizeof(BLOCK_INFO))) != 0) {
   1269 			free(blkiov, M_SEGMENT);
   1270 			return error;
   1271 		}
   1272 
   1273 		simple_lock(&fs->lfs_interlock);
   1274 		++fs->lfs_sleepers;
   1275 		simple_unlock(&fs->lfs_interlock);
   1276 		VOP_UNLOCK(ap->a_vp, 0);
   1277 		if (ap->a_command == LFCNBMAPV)
   1278 			error = lfs_bmapv(ap->a_l, fsidp, blkiov, blkcnt);
   1279 		else /* LFCNMARKV */
   1280 			error = lfs_markv(p, fsidp, blkiov, blkcnt);
   1281 		if (error == 0)
   1282 			error = copyout(blkiov, blkvp.blkiov,
   1283 					blkcnt * sizeof(BLOCK_INFO));
   1284 		VOP_LOCK(ap->a_vp, LK_EXCLUSIVE);
   1285 		simple_lock(&fs->lfs_interlock);
   1286 		if (--fs->lfs_sleepers == 0)
   1287 			wakeup(&fs->lfs_sleepers);
   1288 		simple_unlock(&fs->lfs_interlock);
   1289 		free(blkiov, M_SEGMENT);
   1290 		return error;
   1291 
   1292 	    case LFCNRECLAIM:
   1293 		/*
   1294 		 * Flush dirops and write Ifile, allowing empty segments
   1295 		 * to be immediately reclaimed.
   1296 		 */
   1297 		off = fs->lfs_offset;
   1298 		lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP);
   1299 		lfs_flush_dirops(fs);
   1300 		LFS_CLEANERINFO(cip, fs, bp);
   1301 		oclean = cip->clean;
   1302 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
   1303 		lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP);
   1304 		lfs_segunlock(fs);
   1305 
   1306 #ifdef DEBUG_LFS
   1307 		LFS_CLEANERINFO(cip, fs, bp);
   1308 		oclean = cip->clean;
   1309 		printf("lfs_fcntl: reclaim wrote %" PRId64 " blocks, cleaned "
   1310 			"%" PRId32 " segments (activesb %d)\n",
   1311 			fs->lfs_offset - off, cip->clean - oclean,
   1312 			fs->lfs_activesb);
   1313 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 0);
   1314 #endif
   1315 
   1316 		return 0;
   1317 
   1318 	    default:
   1319 		return ufs_fcntl(v);
   1320 	}
   1321 	return 0;
   1322 }
   1323 
   1324 int
   1325 lfs_getpages(void *v)
   1326 {
   1327 	struct vop_getpages_args /* {
   1328 		struct vnode *a_vp;
   1329 		voff_t a_offset;
   1330 		struct vm_page **a_m;
   1331 		int *a_count;
   1332 		int a_centeridx;
   1333 		vm_prot_t a_access_type;
   1334 		int a_advice;
   1335 		int a_flags;
   1336 	} */ *ap = v;
   1337 
   1338 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM &&
   1339 	    (ap->a_access_type & VM_PROT_WRITE) != 0) {
   1340 		return EPERM;
   1341 	}
   1342 	if ((ap->a_access_type & VM_PROT_WRITE) != 0) {
   1343 		LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED);
   1344 	}
   1345 	return genfs_getpages(v);
   1346 }
   1347 
   1348 /*
   1349  * Make sure that for all pages in every block in the given range,
   1350  * either all are dirty or all are clean.  If any of the pages
   1351  * we've seen so far are dirty, put the vnode on the paging chain,
   1352  * and mark it IN_PAGING.
   1353  *
   1354  * If checkfirst != 0, don't check all the pages but return at the
   1355  * first dirty page.
   1356  */
   1357 static int
   1358 check_dirty(struct lfs *fs, struct vnode *vp,
   1359 	    off_t startoffset, off_t endoffset, off_t blkeof,
   1360 	    int flags, int checkfirst)
   1361 {
   1362 	int by_list;
   1363 	struct vm_page *curpg, *pgs[MAXBSIZE / PAGE_SIZE], *pg;
   1364 	struct lwp *l = curlwp ? curlwp : &lwp0;
   1365 	off_t soff;
   1366 	voff_t off;
   1367 	int i, dirty, tdirty, nonexistent, any_dirty;
   1368 	int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT;
   1369 
   1370   top:
   1371 	by_list = (vp->v_uobj.uo_npages <=
   1372 		   ((endoffset - startoffset) >> PAGE_SHIFT) *
   1373 		   UVM_PAGE_HASH_PENALTY);
   1374 	any_dirty = 0;
   1375 
   1376 	if (by_list) {
   1377 		curpg = TAILQ_FIRST(&vp->v_uobj.memq);
   1378 		PHOLD(l);
   1379 	} else {
   1380 		soff = startoffset;
   1381 	}
   1382 	while (by_list || soff < MIN(blkeof, endoffset)) {
   1383 		if (by_list) {
   1384 			if (pages_per_block > 1) {
   1385 				while (curpg && (curpg->offset & fs->lfs_bmask))
   1386 					curpg = TAILQ_NEXT(curpg, listq);
   1387 			}
   1388 			if (curpg == NULL)
   1389 				break;
   1390 			soff = curpg->offset;
   1391 		}
   1392 
   1393 		/*
   1394 		 * Mark all pages in extended range busy; find out if any
   1395 		 * of them are dirty.
   1396 		 */
   1397 		nonexistent = dirty = 0;
   1398 		for (i = 0; i == 0 || i < pages_per_block; i++) {
   1399 			if (by_list && pages_per_block <= 1) {
   1400 				pgs[i] = pg = curpg;
   1401 			} else {
   1402 				off = soff + (i << PAGE_SHIFT);
   1403 				pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off);
   1404 				if (pg == NULL) {
   1405 					++nonexistent;
   1406 					continue;
   1407 				}
   1408 			}
   1409 			KASSERT(pg != NULL);
   1410 			while (pg->flags & PG_BUSY) {
   1411 				pg->flags |= PG_WANTED;
   1412 				UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0,
   1413 						    "lfsput", 0);
   1414 				simple_lock(&vp->v_interlock);
   1415 				if (by_list) {
   1416 					if (i > 0)
   1417 						uvm_page_unbusy(pgs, i);
   1418 					goto top;
   1419 				}
   1420 			}
   1421 			pg->flags |= PG_BUSY;
   1422 			UVM_PAGE_OWN(pg, "lfs_putpages");
   1423 
   1424 			pmap_page_protect(pg, VM_PROT_NONE);
   1425 			tdirty = (pmap_clear_modify(pg) ||
   1426 				  (pg->flags & PG_CLEAN) == 0);
   1427 			dirty += tdirty;
   1428 		}
   1429 		if (pages_per_block > 0 && nonexistent >= pages_per_block) {
   1430 			if (by_list) {
   1431 				curpg = TAILQ_NEXT(curpg, listq);
   1432 			} else {
   1433 				soff += fs->lfs_bsize;
   1434 			}
   1435 			continue;
   1436 		}
   1437 
   1438 		any_dirty += dirty;
   1439 		KASSERT(nonexistent == 0);
   1440 
   1441 		/*
   1442 		 * If any are dirty make all dirty; unbusy them,
   1443 		 * but if we were asked to clean, wire them so that
   1444 		 * the pagedaemon doesn't bother us about them while
   1445 		 * they're on their way to disk.
   1446 		 */
   1447 		for (i = 0; i == 0 || i < pages_per_block; i++) {
   1448 			pg = pgs[i];
   1449 			KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI)));
   1450 			if (dirty) {
   1451 				pg->flags &= ~PG_CLEAN;
   1452 				if (flags & PGO_FREE) {
   1453 					/* XXXUBC need better way to update */
   1454 					simple_lock(&lfs_subsys_lock);
   1455 					lfs_subsys_pages += MIN(1, pages_per_block);
   1456 					simple_unlock(&lfs_subsys_lock);
   1457 					/*
   1458 					 * Wire the page so that
   1459 					 * pdaemon doesn't see it again.
   1460 					 */
   1461 					uvm_lock_pageq();
   1462 					uvm_pagewire(pg);
   1463 					uvm_unlock_pageq();
   1464 
   1465 					/* Suspended write flag */
   1466 					pg->flags |= PG_DELWRI;
   1467 				}
   1468 			}
   1469 			if (pg->flags & PG_WANTED)
   1470 				wakeup(pg);
   1471 			pg->flags &= ~(PG_WANTED|PG_BUSY);
   1472 			UVM_PAGE_OWN(pg, NULL);
   1473 		}
   1474 
   1475 		if (checkfirst && any_dirty)
   1476 			return any_dirty;
   1477 
   1478 		if (by_list) {
   1479 			curpg = TAILQ_NEXT(curpg, listq);
   1480 		} else {
   1481 			soff += MAX(PAGE_SIZE, fs->lfs_bsize);
   1482 		}
   1483 	}
   1484 	if (by_list) {
   1485 		PRELE(l);
   1486 	}
   1487 
   1488 	/*
   1489 	 * If any pages were dirty, mark this inode as "pageout requested",
   1490 	 * and put it on the paging queue.
   1491 	 * XXXUBC locking (check locking on dchainhd too)
   1492 	 */
   1493 #ifdef notyet
   1494 	if (any_dirty) {
   1495 		if (!(ip->i_flags & IN_PAGING)) {
   1496 			ip->i_flags |= IN_PAGING;
   1497 			TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain);
   1498 		}
   1499 	}
   1500 #endif
   1501 	return any_dirty;
   1502 }
   1503 
   1504 /*
   1505  * lfs_putpages functions like genfs_putpages except that
   1506  *
   1507  * (1) It needs to bounds-check the incoming requests to ensure that
   1508  *     they are block-aligned; if they are not, expand the range and
   1509  *     do the right thing in case, e.g., the requested range is clean
   1510  *     but the expanded range is dirty.
   1511  * (2) It needs to explicitly send blocks to be written when it is done.
   1512  *     VOP_PUTPAGES is not ever called with the seglock held, so
   1513  *     we simply take the seglock and let lfs_segunlock wait for us.
   1514  *     XXX Actually we can be called with the seglock held, if we have
   1515  *     XXX to flush a vnode while lfs_markv is in operation.  As of this
   1516  *     XXX writing we panic in this case.
   1517  *
   1518  * Assumptions:
   1519  *
   1520  * (1) The caller does not hold any pages in this vnode busy.  If it does,
   1521  *     there is a danger that when we expand the page range and busy the
   1522  *     pages we will deadlock.
   1523  * (2) We are called with vp->v_interlock held; we must return with it
   1524  *     released.
   1525  * (3) We don't absolutely have to free pages right away, provided that
   1526  *     the request does not have PGO_SYNCIO.  When the pagedaemon gives
   1527  *     us a request with PGO_FREE, we take the pages out of the paging
   1528  *     queue and wake up the writer, which will handle freeing them for us.
   1529  *
   1530  *     We ensure that for any filesystem block, all pages for that
   1531  *     block are either resident or not, even if those pages are higher
   1532  *     than EOF; that means that we will be getting requests to free
   1533  *     "unused" pages above EOF all the time, and should ignore them.
   1534  */
   1535 
   1536 int
   1537 lfs_putpages(void *v)
   1538 {
   1539 	int error;
   1540 	struct vop_putpages_args /* {
   1541 		struct vnode *a_vp;
   1542 		voff_t a_offlo;
   1543 		voff_t a_offhi;
   1544 		int a_flags;
   1545 	} */ *ap = v;
   1546 	struct vnode *vp;
   1547 	struct inode *ip;
   1548 	struct lfs *fs;
   1549 	struct segment *sp;
   1550 	off_t origoffset, startoffset, endoffset, origendoffset, blkeof;
   1551 	off_t off, max_endoffset;
   1552 	int pages_per_block;
   1553 	int s, sync, dirty, pagedaemon;
   1554 	struct vm_page *pg;
   1555 	UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist);
   1556 
   1557 	vp = ap->a_vp;
   1558 	ip = VTOI(vp);
   1559 	fs = ip->i_lfs;
   1560 	sync = (ap->a_flags & PGO_SYNCIO);
   1561 	pagedaemon = (curproc == uvm.pagedaemon_proc);
   1562 
   1563 	/* Putpages does nothing for metadata. */
   1564 	if (vp == fs->lfs_ivnode || vp->v_type != VREG) {
   1565 		simple_unlock(&vp->v_interlock);
   1566 		return 0;
   1567 	}
   1568 
   1569 	/*
   1570 	 * If there are no pages, don't do anything.
   1571 	 */
   1572 	if (vp->v_uobj.uo_npages == 0) {
   1573 		s = splbio();
   1574 		if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL &&
   1575 		    (vp->v_flag & VONWORKLST)) {
   1576 			vp->v_flag &= ~VONWORKLST;
   1577 			LIST_REMOVE(vp, v_synclist);
   1578 		}
   1579 		splx(s);
   1580 		simple_unlock(&vp->v_interlock);
   1581 		return 0;
   1582 	}
   1583 
   1584 	blkeof = blkroundup(fs, ip->i_size);
   1585 
   1586 	/*
   1587 	 * Ignore requests to free pages past EOF but in the same block
   1588 	 * as EOF, unless the request is synchronous. (XXX why sync?)
   1589 	 * XXXUBC Make these pages look "active" so the pagedaemon won't
   1590 	 * XXXUBC bother us with them again.
   1591 	 */
   1592 	if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) {
   1593 		origoffset = ap->a_offlo;
   1594 		for (off = origoffset; off < blkeof; off += fs->lfs_bsize) {
   1595 			pg = uvm_pagelookup(&vp->v_uobj, off);
   1596 			KASSERT(pg != NULL);
   1597 			while (pg->flags & PG_BUSY) {
   1598 				pg->flags |= PG_WANTED;
   1599 				UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0,
   1600 						    "lfsput2", 0);
   1601 				simple_lock(&vp->v_interlock);
   1602 			}
   1603 			uvm_lock_pageq();
   1604 			uvm_pageactivate(pg);
   1605 			uvm_unlock_pageq();
   1606 		}
   1607 		ap->a_offlo = blkeof;
   1608 		if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) {
   1609 			simple_unlock(&vp->v_interlock);
   1610 			return 0;
   1611 		}
   1612 	}
   1613 
   1614 	/*
   1615 	 * Extend page range to start and end at block boundaries.
   1616 	 * (For the purposes of VOP_PUTPAGES, fragments don't exist.)
   1617 	 */
   1618 	pages_per_block = fs->lfs_bsize >> PAGE_SHIFT;
   1619 	origoffset = ap->a_offlo;
   1620 	origendoffset = ap->a_offhi;
   1621 	startoffset = origoffset & ~(fs->lfs_bmask);
   1622 	max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift)
   1623 					       << fs->lfs_bshift;
   1624 
   1625 	if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) {
   1626 		endoffset = max_endoffset;
   1627 		origendoffset = endoffset;
   1628 	} else {
   1629 		origendoffset = round_page(ap->a_offhi);
   1630 		endoffset = round_page(blkroundup(fs, origendoffset));
   1631 	}
   1632 
   1633 	KASSERT(startoffset > 0 || endoffset >= startoffset);
   1634 	if (startoffset == endoffset) {
   1635 		/* Nothing to do, why were we called? */
   1636 		simple_unlock(&vp->v_interlock);
   1637 #ifdef DEBUG
   1638 		printf("lfs_putpages: startoffset = endoffset = %" PRId64 "\n",
   1639 			startoffset);
   1640 #endif
   1641 		return 0;
   1642 	}
   1643 
   1644 	ap->a_offlo = startoffset;
   1645 	ap->a_offhi = endoffset;
   1646 
   1647 	if (!(ap->a_flags & PGO_CLEANIT))
   1648 		return genfs_putpages(v);
   1649 
   1650 	/*
   1651 	 * If there are more than one page per block, we don't want
   1652 	 * to get caught locking them backwards; so set PGO_BUSYFAIL
   1653 	 * to avoid deadlocks.
   1654 	 */
   1655 	ap->a_flags |= PGO_BUSYFAIL;
   1656 
   1657 	do {
   1658 		int r;
   1659 
   1660 		/* If no pages are dirty, we can just use genfs_putpages. */
   1661 		if ((dirty = check_dirty(fs, vp, startoffset, endoffset, blkeof,
   1662 					 ap->a_flags, 1)) != 0)
   1663 			break;
   1664 
   1665 		if ((r = genfs_putpages(v)) != EDEADLK)
   1666 			return r;
   1667 
   1668 		/* Start over. */
   1669 		preempt(NULL);
   1670 		simple_lock(&vp->v_interlock);
   1671 	} while(1);
   1672 
   1673 	/*
   1674 	 * Dirty and asked to clean.
   1675 	 *
   1676 	 * Pagedaemon can't actually write LFS pages; wake up
   1677 	 * the writer to take care of that.  The writer will
   1678 	 * notice the pager inode queue and act on that.
   1679 	 */
   1680 	if (pagedaemon) {
   1681 		++fs->lfs_pdflush;
   1682 		wakeup(&lfs_writer_daemon);
   1683 		simple_unlock(&vp->v_interlock);
   1684 		return EWOULDBLOCK;
   1685 	}
   1686 
   1687 	/*
   1688 	 * If this is a file created in a recent dirop, we can't flush its
   1689 	 * inode until the dirop is complete.  Drain dirops, then flush the
   1690 	 * filesystem (taking care of any other pending dirops while we're
   1691 	 * at it).
   1692 	 */
   1693 	if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT &&
   1694 	    (vp->v_flag & VDIROP)) {
   1695 		int locked;
   1696 
   1697 		/* printf("putpages to clean VDIROP, flushing\n"); */
   1698 		while (fs->lfs_dirops > 0) {
   1699 			++fs->lfs_diropwait;
   1700 			tsleep(&fs->lfs_writer, PRIBIO+1, "ppdirop", 0);
   1701 			--fs->lfs_diropwait;
   1702 		}
   1703 		++fs->lfs_writer;
   1704 		locked = VOP_ISLOCKED(vp) && /* XXX */
   1705 			vp->v_lock.lk_lockholder == curproc->p_pid;
   1706 		if (locked)
   1707 			VOP_UNLOCK(vp, 0);
   1708 		simple_unlock(&vp->v_interlock);
   1709 
   1710 		lfs_flush_fs(fs, sync ? SEGM_SYNC : 0);
   1711 
   1712 		simple_lock(&vp->v_interlock);
   1713 		if (locked)
   1714 			VOP_LOCK(vp, LK_EXCLUSIVE);
   1715 		if (--fs->lfs_writer == 0)
   1716 			wakeup(&fs->lfs_dirops);
   1717 
   1718 		/* XXX the flush should have taken care of this one too! */
   1719 	}
   1720 
   1721 	/*
   1722 	 * This is it.	We are going to write some pages.  From here on
   1723 	 * down it's all just mechanics.
   1724 	 *
   1725 	 * Don't let genfs_putpages wait; lfs_segunlock will wait for us.
   1726 	 */
   1727 	ap->a_flags &= ~PGO_SYNCIO;
   1728 
   1729 	/*
   1730 	 * If we've already got the seglock, flush the node and return.
   1731 	 * The FIP has already been set up for us by lfs_writefile,
   1732 	 * and FIP cleanup and lfs_updatemeta will also be done there,
   1733 	 * unless genfs_putpages returns EDEADLK; then we must flush
   1734 	 * what we have, and correct FIP and segment header accounting.
   1735 	 */
   1736 	if (ap->a_flags & PGO_LOCKED) {
   1737 		sp = fs->lfs_sp;
   1738 		sp->vp = vp;
   1739 
   1740 		/*
   1741 		 * Make sure that all pages in any given block are dirty, or
   1742 		 * none of them are.
   1743 		 */
   1744 	    again:
   1745 		check_dirty(fs, vp, startoffset, endoffset, blkeof,
   1746 			    ap->a_flags, 0);
   1747 
   1748 		if ((error = genfs_putpages(v)) == EDEADLK) {
   1749 #ifdef DEBUG_LFS
   1750 			printf("lfs_putpages: genfs_putpages returned EDEADLK"
   1751 			       " ino %d off %x (seg %d)\n",
   1752 			       ip->i_number, fs->lfs_offset,
   1753 			       dtosn(fs, fs->lfs_offset));
   1754 #endif
   1755 			/* If nothing to write, short-circuit */
   1756 			if (sp->cbpp - sp->bpp == 1) {
   1757 				preempt(NULL);
   1758 				simple_lock(&vp->v_interlock);
   1759 				goto again;
   1760 			}
   1761 			/* Write gathered pages */
   1762 			lfs_updatemeta(sp);
   1763 			(void) lfs_writeseg(fs, sp);
   1764 
   1765 			/* Reinitialize brand new FIP and add us to it */
   1766 			sp->vp = vp;
   1767 			sp->fip->fi_version = ip->i_gen;
   1768 			sp->fip->fi_ino = ip->i_number;
   1769 			/* Add us to the new segment summary. */
   1770 			++((SEGSUM *)(sp->segsum))->ss_nfinfo;
   1771 			sp->sum_bytes_left -=
   1772 				sizeof(struct finfo) - sizeof(int32_t);
   1773 
   1774 			/* Give the write a chance to complete */
   1775 			preempt(NULL);
   1776 
   1777 			/* We've lost the interlock.  Start over. */
   1778 			simple_lock(&vp->v_interlock);
   1779 			goto again;
   1780 		}
   1781 		lfs_updatemeta(sp);
   1782 		return error;
   1783 	}
   1784 
   1785 	simple_unlock(&vp->v_interlock);
   1786 	/*
   1787 	 * Take the seglock, because we are going to be writing pages.
   1788 	 */
   1789 	if ((error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0))) != 0)
   1790 		return error;
   1791 
   1792 	/*
   1793 	 * VOP_PUTPAGES should not be called while holding the seglock.
   1794 	 * XXXUBC fix lfs_markv, or do this properly.
   1795 	 */
   1796 	/* KASSERT(fs->lfs_seglock == 1); */
   1797 
   1798 	/*
   1799 	 * We assume we're being called with sp->fip pointing at blank space.
   1800 	 * Account for a new FIP in the segment header, and set sp->vp.
   1801 	 * (This should duplicate the setup at the top of lfs_writefile().)
   1802 	 */
   1803 	sp = fs->lfs_sp;
   1804 	if (sp->seg_bytes_left < fs->lfs_bsize ||
   1805 	    sp->sum_bytes_left < sizeof(struct finfo))
   1806 		(void) lfs_writeseg(fs, fs->lfs_sp);
   1807 
   1808 	sp->sum_bytes_left -= sizeof(struct finfo) - sizeof(int32_t);
   1809 	++((SEGSUM *)(sp->segsum))->ss_nfinfo;
   1810 	sp->vp = vp;
   1811 
   1812 	if (vp->v_flag & VDIROP)
   1813 		((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT);
   1814 
   1815 	sp->fip->fi_nblocks = 0;
   1816 	sp->fip->fi_ino = ip->i_number;
   1817 	sp->fip->fi_version = ip->i_gen;
   1818 
   1819 	/*
   1820 	 * Loop through genfs_putpages until all pages are gathered.
   1821 	 * genfs_putpages() drops the interlock, so reacquire it if necessary.
   1822 	 * Whenever we lose the interlock we have to rerun check_dirty, as
   1823 	 * well.
   1824 	 */
   1825     again2:
   1826 	simple_lock(&vp->v_interlock);
   1827 	check_dirty(fs, vp, startoffset, endoffset, blkeof, ap->a_flags, 0);
   1828 
   1829 	if ((error = genfs_putpages(v)) == EDEADLK) {
   1830 #ifdef DEBUG_LFS
   1831 		printf("lfs_putpages: genfs_putpages returned EDEADLK [2]"
   1832 		       " ino %d off %x (seg %d)\n",
   1833 		       ip->i_number, fs->lfs_offset,
   1834 		       dtosn(fs, fs->lfs_offset));
   1835 #endif
   1836 		/* If nothing to write, short-circuit */
   1837 		if (sp->cbpp - sp->bpp == 1) {
   1838 			preempt(NULL);
   1839 			goto again2;
   1840 		}
   1841 		/* Write gathered pages */
   1842 		lfs_updatemeta(sp);
   1843 		(void) lfs_writeseg(fs, sp);
   1844 
   1845 		/*
   1846 		 * Reinitialize brand new FIP and add us to it.
   1847 		 * (This should duplicate the fixup in lfs_gatherpages().)
   1848 		 */
   1849 		sp->vp = vp;
   1850 		sp->fip->fi_version = ip->i_gen;
   1851 		sp->fip->fi_ino = ip->i_number;
   1852 		/* Add us to the new segment summary. */
   1853 		++((SEGSUM *)(sp->segsum))->ss_nfinfo;
   1854 		sp->sum_bytes_left -=
   1855 			sizeof(struct finfo) - sizeof(int32_t);
   1856 
   1857 		/* Give the write a chance to complete */
   1858 		preempt(NULL);
   1859 
   1860 		/* We've lost the interlock.  Start over. */
   1861 		goto again2;
   1862 	}
   1863 
   1864 	/* Write indirect blocks as well */
   1865 	lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir);
   1866 	lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir);
   1867 	lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir);
   1868 
   1869 	/*
   1870 	 * Blocks are now gathered into a segment waiting to be written.
   1871 	 * All that's left to do is update metadata, and write them.
   1872 	 */
   1873 	lfs_updatemeta(fs->lfs_sp);
   1874 	fs->lfs_sp->vp = NULL;
   1875 	/*
   1876 	 * Clean up FIP, since we're done writing this file.
   1877 	 * This should duplicate cleanup at the end of lfs_writefile().
   1878 	 */
   1879 	if (sp->fip->fi_nblocks != 0) {
   1880 		sp->fip = (FINFO*)((caddr_t)sp->fip + sizeof(struct finfo) +
   1881 			sizeof(int32_t) * (sp->fip->fi_nblocks - 1));
   1882 		sp->start_lbp = &sp->fip->fi_blocks[0];
   1883 	} else {
   1884 		sp->sum_bytes_left += sizeof(FINFO) - sizeof(int32_t);
   1885 		--((SEGSUM *)(sp->segsum))->ss_nfinfo;
   1886 	}
   1887 	lfs_writeseg(fs, fs->lfs_sp);
   1888 
   1889 	/*
   1890 	 * XXX - with the malloc/copy writeseg, the pages are freed by now
   1891 	 * even if we don't wait (e.g. if we hold a nested lock).  This
   1892 	 * will not be true if we stop using malloc/copy.
   1893 	 */
   1894 	KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT);
   1895 	lfs_segunlock(fs);
   1896 
   1897 	/*
   1898 	 * Wait for v_numoutput to drop to zero.  The seglock should
   1899 	 * take care of this, but there is a slight possibility that
   1900 	 * aiodoned might not have got around to our buffers yet.
   1901 	 */
   1902 	if (sync) {
   1903 		int s;
   1904 
   1905 		s = splbio();
   1906 		simple_lock(&global_v_numoutput_slock);
   1907 		while (vp->v_numoutput > 0) {
   1908 #ifdef DEBUG
   1909 			printf("ino %d sleeping on num %d\n",
   1910 				ip->i_number, vp->v_numoutput);
   1911 #endif
   1912 			vp->v_flag |= VBWAIT;
   1913 			ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0,
   1914 			    &global_v_numoutput_slock);
   1915 		}
   1916 		simple_unlock(&global_v_numoutput_slock);
   1917 		splx(s);
   1918 	}
   1919 	return error;
   1920 }
   1921 
   1922 /*
   1923  * Return the last logical file offset that should be written for this file
   1924  * if we're doing a write that ends at "size".	If writing, we need to know
   1925  * about sizes on disk, i.e. fragments if there are any; if reading, we need
   1926  * to know about entire blocks.
   1927  */
   1928 void
   1929 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags)
   1930 {
   1931 	struct inode *ip = VTOI(vp);
   1932 	struct lfs *fs = ip->i_lfs;
   1933 	daddr_t olbn, nlbn;
   1934 
   1935 	KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE));
   1936 	KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE))
   1937 		!= (GOP_SIZE_READ | GOP_SIZE_WRITE));
   1938 
   1939 	olbn = lblkno(fs, ip->i_size);
   1940 	nlbn = lblkno(fs, size);
   1941 	if ((flags & GOP_SIZE_WRITE) && nlbn < NDADDR && olbn <= nlbn) {
   1942 		*eobp = fragroundup(fs, size);
   1943 	} else {
   1944 		*eobp = blkroundup(fs, size);
   1945 	}
   1946 }
   1947 
   1948 #ifdef DEBUG
   1949 void lfs_dump_vop(void *);
   1950 
   1951 void
   1952 lfs_dump_vop(void *v)
   1953 {
   1954 	struct vop_putpages_args /* {
   1955 		struct vnode *a_vp;
   1956 		voff_t a_offlo;
   1957 		voff_t a_offhi;
   1958 		int a_flags;
   1959 	} */ *ap = v;
   1960 
   1961 #ifdef DDB
   1962 	vfs_vnode_print(ap->a_vp, 0, printf);
   1963 #endif
   1964 	lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din);
   1965 }
   1966 #endif
   1967 
   1968 int
   1969 lfs_mmap(void *v)
   1970 {
   1971 	struct vop_mmap_args /* {
   1972 		const struct vnodeop_desc *a_desc;
   1973 		struct vnode *a_vp;
   1974 		int a_fflags;
   1975 		struct ucred *a_cred;
   1976 		struct lwp *a_l;
   1977 	} */ *ap = v;
   1978 
   1979 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM)
   1980 		return EOPNOTSUPP;
   1981 	return ufs_mmap(v);
   1982 }
   1983