Home | History | Annotate | Line # | Download | only in lfs
lfs_vnops.c revision 1.134
      1  1.134  perseant /*	$NetBSD: lfs_vnops.c,v 1.134 2005/02/26 05:40:42 perseant Exp $	*/
      2    1.2       cgd 
      3   1.22  perseant /*-
      4   1.84  perseant  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
      5   1.22  perseant  * All rights reserved.
      6   1.22  perseant  *
      7   1.22  perseant  * This code is derived from software contributed to The NetBSD Foundation
      8   1.22  perseant  * by Konrad E. Schroder <perseant (at) hhhh.org>.
      9   1.22  perseant  *
     10   1.22  perseant  * Redistribution and use in source and binary forms, with or without
     11   1.22  perseant  * modification, are permitted provided that the following conditions
     12   1.22  perseant  * are met:
     13   1.22  perseant  * 1. Redistributions of source code must retain the above copyright
     14   1.22  perseant  *    notice, this list of conditions and the following disclaimer.
     15   1.22  perseant  * 2. Redistributions in binary form must reproduce the above copyright
     16   1.22  perseant  *    notice, this list of conditions and the following disclaimer in the
     17   1.22  perseant  *    documentation and/or other materials provided with the distribution.
     18   1.22  perseant  * 3. All advertising materials mentioning features or use of this software
     19   1.22  perseant  *    must display the following acknowledgement:
     20   1.86  perseant  *	This product includes software developed by the NetBSD
     21   1.86  perseant  *	Foundation, Inc. and its contributors.
     22   1.22  perseant  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23   1.22  perseant  *    contributors may be used to endorse or promote products derived
     24   1.22  perseant  *    from this software without specific prior written permission.
     25   1.22  perseant  *
     26   1.22  perseant  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27   1.22  perseant  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28   1.22  perseant  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29   1.22  perseant  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30   1.22  perseant  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31   1.22  perseant  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32   1.22  perseant  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33   1.22  perseant  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34   1.22  perseant  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35   1.22  perseant  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36   1.22  perseant  * POSSIBILITY OF SUCH DAMAGE.
     37   1.22  perseant  */
     38    1.1   mycroft /*
     39   1.15      fvdl  * Copyright (c) 1986, 1989, 1991, 1993, 1995
     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  1.114       agc  * 3. Neither the name of the University nor the names of its contributors
     51    1.1   mycroft  *    may be used to endorse or promote products derived from this software
     52    1.1   mycroft  *    without specific prior written permission.
     53    1.1   mycroft  *
     54    1.1   mycroft  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     55    1.1   mycroft  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     56    1.1   mycroft  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     57    1.1   mycroft  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     58    1.1   mycroft  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     59    1.1   mycroft  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     60    1.1   mycroft  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     61    1.1   mycroft  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     62    1.1   mycroft  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     63    1.1   mycroft  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     64    1.1   mycroft  * SUCH DAMAGE.
     65    1.1   mycroft  *
     66   1.15      fvdl  *	@(#)lfs_vnops.c	8.13 (Berkeley) 6/10/95
     67    1.1   mycroft  */
     68   1.58     lukem 
     69   1.58     lukem #include <sys/cdefs.h>
     70  1.134  perseant __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.134 2005/02/26 05:40:42 perseant Exp $");
     71   1.17  sommerfe 
     72    1.1   mycroft #include <sys/param.h>
     73    1.1   mycroft #include <sys/systm.h>
     74    1.1   mycroft #include <sys/namei.h>
     75    1.1   mycroft #include <sys/resourcevar.h>
     76    1.1   mycroft #include <sys/kernel.h>
     77    1.1   mycroft #include <sys/file.h>
     78    1.1   mycroft #include <sys/stat.h>
     79    1.1   mycroft #include <sys/buf.h>
     80    1.1   mycroft #include <sys/proc.h>
     81    1.1   mycroft #include <sys/mount.h>
     82    1.1   mycroft #include <sys/vnode.h>
     83    1.1   mycroft #include <sys/malloc.h>
     84   1.19   thorpej #include <sys/pool.h>
     85   1.10  christos #include <sys/signalvar.h>
     86    1.1   mycroft 
     87   1.12   mycroft #include <miscfs/fifofs/fifo.h>
     88   1.12   mycroft #include <miscfs/genfs/genfs.h>
     89    1.1   mycroft #include <miscfs/specfs/specdev.h>
     90    1.1   mycroft 
     91    1.1   mycroft #include <ufs/ufs/inode.h>
     92    1.1   mycroft #include <ufs/ufs/dir.h>
     93    1.1   mycroft #include <ufs/ufs/ufsmount.h>
     94    1.1   mycroft #include <ufs/ufs/ufs_extern.h>
     95    1.1   mycroft 
     96   1.84  perseant #include <uvm/uvm.h>
     97   1.95  perseant #include <uvm/uvm_pmap.h>
     98   1.95  perseant #include <uvm/uvm_stat.h>
     99   1.95  perseant #include <uvm/uvm_pager.h>
    100   1.84  perseant 
    101    1.1   mycroft #include <ufs/lfs/lfs.h>
    102    1.1   mycroft #include <ufs/lfs/lfs_extern.h>
    103    1.1   mycroft 
    104   1.91      yamt extern pid_t lfs_writer_daemon;
    105   1.84  perseant 
    106    1.1   mycroft /* Global vfs data structures for lfs. */
    107   1.51  perseant int (**lfs_vnodeop_p)(void *);
    108   1.50  jdolecek const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = {
    109    1.1   mycroft 	{ &vop_default_desc, vn_default_error },
    110    1.1   mycroft 	{ &vop_lookup_desc, ufs_lookup },		/* lookup */
    111   1.22  perseant 	{ &vop_create_desc, lfs_create },		/* create */
    112   1.82      yamt 	{ &vop_whiteout_desc, ufs_whiteout },		/* whiteout */
    113   1.22  perseant 	{ &vop_mknod_desc, lfs_mknod },			/* mknod */
    114    1.1   mycroft 	{ &vop_open_desc, ufs_open },			/* open */
    115    1.1   mycroft 	{ &vop_close_desc, lfs_close },			/* close */
    116    1.1   mycroft 	{ &vop_access_desc, ufs_access },		/* access */
    117    1.1   mycroft 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
    118   1.61  perseant 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
    119    1.1   mycroft 	{ &vop_read_desc, lfs_read },			/* read */
    120    1.1   mycroft 	{ &vop_write_desc, lfs_write },			/* write */
    121    1.4   mycroft 	{ &vop_lease_desc, ufs_lease_check },		/* lease */
    122   1.90  perseant 	{ &vop_ioctl_desc, ufs_ioctl },			/* ioctl */
    123   1.90  perseant 	{ &vop_fcntl_desc, lfs_fcntl },			/* fcntl */
    124   1.13   mycroft 	{ &vop_poll_desc, ufs_poll },			/* poll */
    125   1.68  jdolecek 	{ &vop_kqfilter_desc, genfs_kqfilter },		/* kqfilter */
    126   1.15      fvdl 	{ &vop_revoke_desc, ufs_revoke },		/* revoke */
    127   1.84  perseant 	{ &vop_mmap_desc, lfs_mmap },			/* mmap */
    128    1.1   mycroft 	{ &vop_fsync_desc, lfs_fsync },			/* fsync */
    129    1.1   mycroft 	{ &vop_seek_desc, ufs_seek },			/* seek */
    130   1.22  perseant 	{ &vop_remove_desc, lfs_remove },		/* remove */
    131   1.22  perseant 	{ &vop_link_desc, lfs_link },			/* link */
    132   1.22  perseant 	{ &vop_rename_desc, lfs_rename },		/* rename */
    133   1.22  perseant 	{ &vop_mkdir_desc, lfs_mkdir },			/* mkdir */
    134   1.22  perseant 	{ &vop_rmdir_desc, lfs_rmdir },			/* rmdir */
    135   1.22  perseant 	{ &vop_symlink_desc, lfs_symlink },		/* symlink */
    136    1.1   mycroft 	{ &vop_readdir_desc, ufs_readdir },		/* readdir */
    137    1.1   mycroft 	{ &vop_readlink_desc, ufs_readlink },		/* readlink */
    138    1.1   mycroft 	{ &vop_abortop_desc, ufs_abortop },		/* abortop */
    139   1.40  perseant 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
    140    1.1   mycroft 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
    141    1.1   mycroft 	{ &vop_lock_desc, ufs_lock },			/* lock */
    142    1.1   mycroft 	{ &vop_unlock_desc, ufs_unlock },		/* unlock */
    143    1.1   mycroft 	{ &vop_bmap_desc, ufs_bmap },			/* bmap */
    144   1.94  perseant 	{ &vop_strategy_desc, lfs_strategy },		/* strategy */
    145    1.1   mycroft 	{ &vop_print_desc, ufs_print },			/* print */
    146    1.1   mycroft 	{ &vop_islocked_desc, ufs_islocked },		/* islocked */
    147    1.1   mycroft 	{ &vop_pathconf_desc, ufs_pathconf },		/* pathconf */
    148    1.1   mycroft 	{ &vop_advlock_desc, ufs_advlock },		/* advlock */
    149    1.1   mycroft 	{ &vop_blkatoff_desc, lfs_blkatoff },		/* blkatoff */
    150    1.1   mycroft 	{ &vop_valloc_desc, lfs_valloc },		/* valloc */
    151   1.32      fvdl 	{ &vop_balloc_desc, lfs_balloc },		/* balloc */
    152    1.1   mycroft 	{ &vop_vfree_desc, lfs_vfree },			/* vfree */
    153    1.1   mycroft 	{ &vop_truncate_desc, lfs_truncate },		/* truncate */
    154    1.1   mycroft 	{ &vop_update_desc, lfs_update },		/* update */
    155    1.1   mycroft 	{ &vop_bwrite_desc, lfs_bwrite },		/* bwrite */
    156   1.60       chs 	{ &vop_getpages_desc, lfs_getpages },		/* getpages */
    157   1.60       chs 	{ &vop_putpages_desc, lfs_putpages },		/* putpages */
    158   1.53       chs 	{ NULL, NULL }
    159    1.1   mycroft };
    160   1.50  jdolecek const struct vnodeopv_desc lfs_vnodeop_opv_desc =
    161    1.1   mycroft 	{ &lfs_vnodeop_p, lfs_vnodeop_entries };
    162    1.1   mycroft 
    163   1.51  perseant int (**lfs_specop_p)(void *);
    164   1.50  jdolecek const struct vnodeopv_entry_desc lfs_specop_entries[] = {
    165    1.1   mycroft 	{ &vop_default_desc, vn_default_error },
    166    1.1   mycroft 	{ &vop_lookup_desc, spec_lookup },		/* lookup */
    167    1.1   mycroft 	{ &vop_create_desc, spec_create },		/* create */
    168    1.1   mycroft 	{ &vop_mknod_desc, spec_mknod },		/* mknod */
    169    1.1   mycroft 	{ &vop_open_desc, spec_open },			/* open */
    170   1.65  perseant 	{ &vop_close_desc, lfsspec_close },		/* close */
    171    1.1   mycroft 	{ &vop_access_desc, ufs_access },		/* access */
    172    1.1   mycroft 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
    173   1.61  perseant 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
    174    1.1   mycroft 	{ &vop_read_desc, ufsspec_read },		/* read */
    175    1.1   mycroft 	{ &vop_write_desc, ufsspec_write },		/* write */
    176    1.4   mycroft 	{ &vop_lease_desc, spec_lease_check },		/* lease */
    177    1.1   mycroft 	{ &vop_ioctl_desc, spec_ioctl },		/* ioctl */
    178   1.27  wrstuden 	{ &vop_fcntl_desc, ufs_fcntl },			/* fcntl */
    179   1.13   mycroft 	{ &vop_poll_desc, spec_poll },			/* poll */
    180   1.68  jdolecek 	{ &vop_kqfilter_desc, spec_kqfilter },		/* kqfilter */
    181   1.15      fvdl 	{ &vop_revoke_desc, spec_revoke },		/* revoke */
    182    1.1   mycroft 	{ &vop_mmap_desc, spec_mmap },			/* mmap */
    183    1.1   mycroft 	{ &vop_fsync_desc, spec_fsync },		/* fsync */
    184    1.1   mycroft 	{ &vop_seek_desc, spec_seek },			/* seek */
    185    1.1   mycroft 	{ &vop_remove_desc, spec_remove },		/* remove */
    186    1.1   mycroft 	{ &vop_link_desc, spec_link },			/* link */
    187    1.1   mycroft 	{ &vop_rename_desc, spec_rename },		/* rename */
    188    1.1   mycroft 	{ &vop_mkdir_desc, spec_mkdir },		/* mkdir */
    189    1.1   mycroft 	{ &vop_rmdir_desc, spec_rmdir },		/* rmdir */
    190    1.1   mycroft 	{ &vop_symlink_desc, spec_symlink },		/* symlink */
    191    1.1   mycroft 	{ &vop_readdir_desc, spec_readdir },		/* readdir */
    192    1.1   mycroft 	{ &vop_readlink_desc, spec_readlink },		/* readlink */
    193    1.1   mycroft 	{ &vop_abortop_desc, spec_abortop },		/* abortop */
    194   1.40  perseant 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
    195    1.1   mycroft 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
    196    1.1   mycroft 	{ &vop_lock_desc, ufs_lock },			/* lock */
    197    1.1   mycroft 	{ &vop_unlock_desc, ufs_unlock },		/* unlock */
    198    1.1   mycroft 	{ &vop_bmap_desc, spec_bmap },			/* bmap */
    199    1.1   mycroft 	{ &vop_strategy_desc, spec_strategy },		/* strategy */
    200    1.1   mycroft 	{ &vop_print_desc, ufs_print },			/* print */
    201    1.1   mycroft 	{ &vop_islocked_desc, ufs_islocked },		/* islocked */
    202    1.1   mycroft 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
    203    1.1   mycroft 	{ &vop_advlock_desc, spec_advlock },		/* advlock */
    204    1.1   mycroft 	{ &vop_blkatoff_desc, spec_blkatoff },		/* blkatoff */
    205    1.1   mycroft 	{ &vop_valloc_desc, spec_valloc },		/* valloc */
    206    1.1   mycroft 	{ &vop_vfree_desc, lfs_vfree },			/* vfree */
    207    1.1   mycroft 	{ &vop_truncate_desc, spec_truncate },		/* truncate */
    208    1.1   mycroft 	{ &vop_update_desc, lfs_update },		/* update */
    209   1.28  perseant 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
    210   1.53       chs 	{ &vop_getpages_desc, spec_getpages },		/* getpages */
    211   1.53       chs 	{ &vop_putpages_desc, spec_putpages },		/* putpages */
    212   1.53       chs 	{ NULL, NULL }
    213    1.1   mycroft };
    214   1.50  jdolecek const struct vnodeopv_desc lfs_specop_opv_desc =
    215    1.1   mycroft 	{ &lfs_specop_p, lfs_specop_entries };
    216    1.1   mycroft 
    217   1.51  perseant int (**lfs_fifoop_p)(void *);
    218   1.50  jdolecek const struct vnodeopv_entry_desc lfs_fifoop_entries[] = {
    219    1.1   mycroft 	{ &vop_default_desc, vn_default_error },
    220    1.1   mycroft 	{ &vop_lookup_desc, fifo_lookup },		/* lookup */
    221    1.1   mycroft 	{ &vop_create_desc, fifo_create },		/* create */
    222    1.1   mycroft 	{ &vop_mknod_desc, fifo_mknod },		/* mknod */
    223    1.1   mycroft 	{ &vop_open_desc, fifo_open },			/* open */
    224   1.65  perseant 	{ &vop_close_desc, lfsfifo_close },		/* close */
    225    1.1   mycroft 	{ &vop_access_desc, ufs_access },		/* access */
    226    1.1   mycroft 	{ &vop_getattr_desc, lfs_getattr },		/* getattr */
    227   1.61  perseant 	{ &vop_setattr_desc, lfs_setattr },		/* setattr */
    228    1.1   mycroft 	{ &vop_read_desc, ufsfifo_read },		/* read */
    229    1.1   mycroft 	{ &vop_write_desc, ufsfifo_write },		/* write */
    230    1.4   mycroft 	{ &vop_lease_desc, fifo_lease_check },		/* lease */
    231    1.1   mycroft 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
    232   1.27  wrstuden 	{ &vop_fcntl_desc, ufs_fcntl },			/* fcntl */
    233   1.13   mycroft 	{ &vop_poll_desc, fifo_poll },			/* poll */
    234   1.68  jdolecek 	{ &vop_kqfilter_desc, fifo_kqfilter },		/* kqfilter */
    235   1.15      fvdl 	{ &vop_revoke_desc, fifo_revoke },		/* revoke */
    236    1.1   mycroft 	{ &vop_mmap_desc, fifo_mmap },			/* mmap */
    237    1.1   mycroft 	{ &vop_fsync_desc, fifo_fsync },		/* fsync */
    238    1.1   mycroft 	{ &vop_seek_desc, fifo_seek },			/* seek */
    239    1.1   mycroft 	{ &vop_remove_desc, fifo_remove },		/* remove */
    240    1.1   mycroft 	{ &vop_link_desc, fifo_link },			/* link */
    241    1.1   mycroft 	{ &vop_rename_desc, fifo_rename },		/* rename */
    242    1.1   mycroft 	{ &vop_mkdir_desc, fifo_mkdir },		/* mkdir */
    243    1.1   mycroft 	{ &vop_rmdir_desc, fifo_rmdir },		/* rmdir */
    244    1.1   mycroft 	{ &vop_symlink_desc, fifo_symlink },		/* symlink */
    245    1.1   mycroft 	{ &vop_readdir_desc, fifo_readdir },		/* readdir */
    246    1.1   mycroft 	{ &vop_readlink_desc, fifo_readlink },		/* readlink */
    247    1.1   mycroft 	{ &vop_abortop_desc, fifo_abortop },		/* abortop */
    248   1.40  perseant 	{ &vop_inactive_desc, lfs_inactive },		/* inactive */
    249    1.1   mycroft 	{ &vop_reclaim_desc, lfs_reclaim },		/* reclaim */
    250    1.1   mycroft 	{ &vop_lock_desc, ufs_lock },			/* lock */
    251    1.1   mycroft 	{ &vop_unlock_desc, ufs_unlock },		/* unlock */
    252    1.1   mycroft 	{ &vop_bmap_desc, fifo_bmap },			/* bmap */
    253    1.1   mycroft 	{ &vop_strategy_desc, fifo_strategy },		/* strategy */
    254    1.1   mycroft 	{ &vop_print_desc, ufs_print },			/* print */
    255    1.1   mycroft 	{ &vop_islocked_desc, ufs_islocked },		/* islocked */
    256    1.1   mycroft 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
    257    1.1   mycroft 	{ &vop_advlock_desc, fifo_advlock },		/* advlock */
    258    1.1   mycroft 	{ &vop_blkatoff_desc, fifo_blkatoff },		/* blkatoff */
    259    1.1   mycroft 	{ &vop_valloc_desc, fifo_valloc },		/* valloc */
    260    1.1   mycroft 	{ &vop_vfree_desc, lfs_vfree },			/* vfree */
    261    1.1   mycroft 	{ &vop_truncate_desc, fifo_truncate },		/* truncate */
    262    1.1   mycroft 	{ &vop_update_desc, lfs_update },		/* update */
    263    1.1   mycroft 	{ &vop_bwrite_desc, lfs_bwrite },		/* bwrite */
    264   1.86  perseant 	{ &vop_putpages_desc, fifo_putpages },		/* putpages */
    265   1.53       chs 	{ NULL, NULL }
    266    1.1   mycroft };
    267   1.50  jdolecek const struct vnodeopv_desc lfs_fifoop_opv_desc =
    268    1.1   mycroft 	{ &lfs_fifoop_p, lfs_fifoop_entries };
    269    1.1   mycroft 
    270  1.134  perseant static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int);
    271  1.134  perseant 
    272   1.43  perseant /*
    273   1.43  perseant  * A function version of LFS_ITIMES, for the UFS functions which call ITIMES
    274   1.43  perseant  */
    275   1.43  perseant void
    276   1.51  perseant lfs_itimes(struct inode *ip, struct timespec *acc, struct timespec *mod, struct timespec *cre)
    277   1.43  perseant {
    278   1.43  perseant 	LFS_ITIMES(ip, acc, mod, cre);
    279   1.43  perseant }
    280   1.43  perseant 
    281    1.1   mycroft #define	LFS_READWRITE
    282    1.1   mycroft #include <ufs/ufs/ufs_readwrite.c>
    283    1.1   mycroft #undef	LFS_READWRITE
    284    1.1   mycroft 
    285    1.1   mycroft /*
    286    1.1   mycroft  * Synch an open file.
    287    1.1   mycroft  */
    288    1.1   mycroft /* ARGSUSED */
    289   1.10  christos int
    290   1.51  perseant lfs_fsync(void *v)
    291   1.10  christos {
    292    1.1   mycroft 	struct vop_fsync_args /* {
    293    1.1   mycroft 		struct vnode *a_vp;
    294    1.1   mycroft 		struct ucred *a_cred;
    295   1.22  perseant 		int a_flags;
    296   1.49    toshii 		off_t offlo;
    297   1.49    toshii 		off_t offhi;
    298  1.109      fvdl 		struct proc *a_p;
    299   1.10  christos 	} */ *ap = v;
    300   1.60       chs 	struct vnode *vp = ap->a_vp;
    301   1.84  perseant 	int error, wait;
    302   1.84  perseant 
    303   1.86  perseant 	/*
    304   1.84  perseant 	 * Trickle sync checks for need to do a checkpoint after possible
    305   1.84  perseant 	 * activity from the pagedaemon.
    306   1.86  perseant 	 */
    307   1.84  perseant 	if (ap->a_flags & FSYNC_LAZY) {
    308  1.113      yamt 		simple_lock(&lfs_subsys_lock);
    309   1.84  perseant 		wakeup(&lfs_writer_daemon);
    310  1.113      yamt 		simple_unlock(&lfs_subsys_lock);
    311   1.47  perseant 		return 0;
    312   1.84  perseant 	}
    313   1.47  perseant 
    314   1.84  perseant 	wait = (ap->a_flags & FSYNC_WAIT);
    315  1.103  perseant 	simple_lock(&vp->v_interlock);
    316  1.103  perseant 	error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo),
    317  1.103  perseant 			round_page(ap->a_offhi),
    318  1.103  perseant 			PGO_CLEANIT | (wait ? PGO_SYNCIO : 0));
    319  1.103  perseant 	if (error)
    320  1.103  perseant 		return error;
    321  1.103  perseant 	error = VOP_UPDATE(vp, NULL, NULL, wait ? UPDATE_WAIT : 0);
    322  1.133  wrstuden 	if (error == 0 && ap->a_flags & FSYNC_CACHE) {
    323  1.133  wrstuden 		int l = 0;
    324  1.133  wrstuden 		error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE,
    325  1.133  wrstuden 				  ap->a_p->p_ucred, ap->a_p);
    326  1.133  wrstuden 	}
    327  1.103  perseant 	if (wait && !VPISEMPTY(vp))
    328  1.103  perseant 		LFS_SET_UINO(VTOI(vp), IN_MODIFIED);
    329   1.84  perseant 
    330   1.63  perseant 	return error;
    331    1.1   mycroft }
    332    1.1   mycroft 
    333    1.1   mycroft /*
    334   1.40  perseant  * Take IN_ADIROP off, then call ufs_inactive.
    335   1.40  perseant  */
    336   1.40  perseant int
    337   1.51  perseant lfs_inactive(void *v)
    338   1.40  perseant {
    339   1.40  perseant 	struct vop_inactive_args /* {
    340   1.40  perseant 		struct vnode *a_vp;
    341  1.109      fvdl 		struct proc *a_p;
    342   1.40  perseant 	} */ *ap = v;
    343   1.72      yamt 
    344  1.102      fvdl 	KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink);
    345   1.77      yamt 
    346   1.76      yamt 	lfs_unmark_vnode(ap->a_vp);
    347   1.76      yamt 
    348   1.97  perseant 	/*
    349   1.97  perseant 	 * The Ifile is only ever inactivated on unmount.
    350   1.97  perseant 	 * Streamline this process by not giving it more dirty blocks.
    351   1.97  perseant 	 */
    352   1.97  perseant 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) {
    353   1.97  perseant 		LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD);
    354   1.99  perseant 		VOP_UNLOCK(ap->a_vp, 0);
    355   1.97  perseant 		return 0;
    356   1.97  perseant 	}
    357   1.97  perseant 
    358   1.75      yamt 	return ufs_inactive(v);
    359   1.40  perseant }
    360   1.40  perseant 
    361   1.40  perseant /*
    362    1.1   mycroft  * These macros are used to bracket UFS directory ops, so that we can
    363    1.1   mycroft  * identify all the pages touched during directory ops which need to
    364    1.1   mycroft  * be ordered and flushed atomically, so that they may be recovered.
    365    1.1   mycroft  */
    366   1.22  perseant /*
    367   1.22  perseant  * XXX KS - Because we have to mark nodes VDIROP in order to prevent
    368   1.22  perseant  * the cache from reclaiming them while a dirop is in progress, we must
    369   1.22  perseant  * also manage the number of nodes so marked (otherwise we can run out).
    370   1.22  perseant  * We do this by setting lfs_dirvcount to the number of marked vnodes; it
    371   1.22  perseant  * is decremented during segment write, when VDIROP is taken off.
    372   1.22  perseant  */
    373   1.71      yamt #define	SET_DIROP(vp)		SET_DIROP2((vp), NULL)
    374   1.71      yamt #define	SET_DIROP2(vp, vp2)	lfs_set_dirop((vp), (vp2))
    375   1.71      yamt static int lfs_set_dirop(struct vnode *, struct vnode *);
    376   1.24  perseant 
    377   1.46  perseant static int
    378   1.71      yamt lfs_set_dirop(struct vnode *vp, struct vnode *vp2)
    379   1.40  perseant {
    380   1.24  perseant 	struct lfs *fs;
    381   1.24  perseant 	int error;
    382   1.24  perseant 
    383   1.71      yamt 	KASSERT(VOP_ISLOCKED(vp));
    384   1.71      yamt 	KASSERT(vp2 == NULL || VOP_ISLOCKED(vp2));
    385   1.71      yamt 
    386   1.40  perseant 	fs = VTOI(vp)->i_lfs;
    387   1.44  perseant 	/*
    388  1.134  perseant 	 * LFS_NRESERVE calculates direct and indirect blocks as well
    389  1.134  perseant 	 * as an inode block; an overestimate in most cases.
    390   1.44  perseant 	 */
    391  1.134  perseant 	if ((error = lfs_reserve(fs, vp, vp2, LFS_NRESERVE(fs))) != 0)
    392   1.44  perseant 		return (error);
    393   1.70      yamt 
    394   1.40  perseant 	if (fs->lfs_dirops == 0)
    395   1.40  perseant 		lfs_check(vp, LFS_UNUSED_LBN, 0);
    396  1.113      yamt restart:
    397  1.113      yamt 	simple_lock(&fs->lfs_interlock);
    398  1.113      yamt 	if (fs->lfs_writer) {
    399  1.113      yamt 		ltsleep(&fs->lfs_dirops, (PRIBIO + 1) | PNORELOCK,
    400  1.113      yamt 		    "lfs_sdirop", 0, &fs->lfs_interlock);
    401  1.113      yamt 		goto restart;
    402  1.113      yamt 	}
    403  1.113      yamt 	simple_lock(&lfs_subsys_lock);
    404  1.113      yamt 	if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) {
    405  1.113      yamt 		wakeup(&lfs_writer_daemon);
    406  1.113      yamt 		simple_unlock(&lfs_subsys_lock);
    407  1.113      yamt 		simple_unlock(&fs->lfs_interlock);
    408  1.121      fvdl 		preempt(1);
    409  1.113      yamt 		goto restart;
    410  1.113      yamt 	}
    411   1.33  perseant 
    412  1.113      yamt 	if (lfs_dirvcount > LFS_MAX_DIROP) {
    413  1.113      yamt 		simple_unlock(&fs->lfs_interlock);
    414   1.24  perseant #ifdef DEBUG_LFS
    415  1.113      yamt 		printf("lfs_set_dirop: sleeping with dirops=%d, "
    416  1.113      yamt 		       "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount);
    417   1.24  perseant #endif
    418  1.113      yamt 		if ((error = ltsleep(&lfs_dirvcount,
    419  1.113      yamt 		    PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0,
    420  1.113      yamt 		    &lfs_subsys_lock)) != 0) {
    421  1.113      yamt 			goto unreserve;
    422  1.113      yamt 		}
    423  1.113      yamt 		goto restart;
    424  1.113      yamt 	}
    425  1.113      yamt 	simple_unlock(&lfs_subsys_lock);
    426  1.113      yamt 
    427   1.24  perseant 	++fs->lfs_dirops;
    428   1.24  perseant 	fs->lfs_doifile = 1;
    429  1.113      yamt 	simple_unlock(&fs->lfs_interlock);
    430   1.24  perseant 
    431   1.46  perseant 	/* Hold a reference so SET_ENDOP will be happy */
    432   1.78      yamt 	vref(vp);
    433   1.78      yamt 	if (vp2)
    434   1.78      yamt 		vref(vp2);
    435   1.46  perseant 
    436   1.24  perseant 	return 0;
    437   1.70      yamt 
    438   1.70      yamt unreserve:
    439  1.134  perseant 	lfs_reserve(fs, vp, vp2, -LFS_NRESERVE(fs));
    440   1.70      yamt 	return error;
    441    1.1   mycroft }
    442    1.1   mycroft 
    443   1.71      yamt #define	SET_ENDOP(fs, vp, str)	SET_ENDOP2((fs), (vp), NULL, (str))
    444   1.71      yamt #define	SET_ENDOP2(fs, vp, vp2, str) {					\
    445    1.1   mycroft 	--(fs)->lfs_dirops;						\
    446   1.22  perseant 	if (!(fs)->lfs_dirops) {					\
    447   1.40  perseant 		if ((fs)->lfs_nadirop) {				\
    448   1.67    provos 			panic("SET_ENDOP: %s: no dirops but nadirop=%d", \
    449   1.40  perseant 			      (str), (fs)->lfs_nadirop);		\
    450   1.40  perseant 		}							\
    451    1.1   mycroft 		wakeup(&(fs)->lfs_writer);				\
    452   1.26  perseant 		lfs_check((vp),LFS_UNUSED_LBN,0);			\
    453   1.22  perseant 	}								\
    454  1.134  perseant 	lfs_reserve((fs), vp, vp2, -LFS_NRESERVE(fs)); /* XXX */	\
    455   1.78      yamt 	vrele(vp);							\
    456   1.78      yamt 	if (vp2)							\
    457   1.78      yamt 		vrele(vp2);						\
    458    1.1   mycroft }
    459    1.1   mycroft 
    460  1.117      yamt #define	MARK_VNODE(vp)		lfs_mark_vnode(vp)
    461  1.117      yamt #define	UNMARK_VNODE(vp)	lfs_unmark_vnode(vp)
    462  1.117      yamt 
    463  1.117      yamt void
    464  1.117      yamt lfs_mark_vnode(struct vnode *vp)
    465  1.117      yamt {
    466  1.117      yamt 	struct inode *ip = VTOI(vp);
    467  1.117      yamt 	struct lfs *fs = ip->i_lfs;
    468   1.37  perseant 
    469  1.117      yamt 	if (!(ip->i_flag & IN_ADIROP)) {
    470  1.117      yamt 		if (!(vp->v_flag & VDIROP)) {
    471  1.117      yamt 			(void)lfs_vref(vp);
    472  1.117      yamt 			++lfs_dirvcount;
    473  1.117      yamt 			TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain);
    474  1.117      yamt 			vp->v_flag |= VDIROP;
    475  1.117      yamt 		}
    476  1.117      yamt 		++fs->lfs_nadirop;
    477  1.117      yamt 		ip->i_flag |= IN_ADIROP;
    478  1.117      yamt 	} else
    479  1.117      yamt 		KASSERT(vp->v_flag & VDIROP);
    480  1.117      yamt }
    481   1.40  perseant 
    482  1.117      yamt void
    483  1.117      yamt lfs_unmark_vnode(struct vnode *vp)
    484   1.40  perseant {
    485  1.117      yamt 	struct inode *ip = VTOI(vp);
    486   1.40  perseant 
    487  1.117      yamt 	if (ip->i_flag & IN_ADIROP) {
    488  1.117      yamt 		KASSERT(vp->v_flag & VDIROP);
    489   1.40  perseant 		--ip->i_lfs->lfs_nadirop;
    490  1.117      yamt 		ip->i_flag &= ~IN_ADIROP;
    491  1.117      yamt 	}
    492   1.40  perseant }
    493   1.15      fvdl 
    494    1.1   mycroft int
    495   1.51  perseant lfs_symlink(void *v)
    496   1.10  christos {
    497    1.1   mycroft 	struct vop_symlink_args /* {
    498    1.1   mycroft 		struct vnode *a_dvp;
    499    1.1   mycroft 		struct vnode **a_vpp;
    500    1.1   mycroft 		struct componentname *a_cnp;
    501    1.1   mycroft 		struct vattr *a_vap;
    502    1.1   mycroft 		char *a_target;
    503   1.10  christos 	} */ *ap = v;
    504   1.37  perseant 	int error;
    505    1.1   mycroft 
    506   1.40  perseant 	if ((error = SET_DIROP(ap->a_dvp)) != 0) {
    507   1.34  perseant 		vput(ap->a_dvp);
    508   1.37  perseant 		return error;
    509   1.34  perseant 	}
    510    1.1   mycroft 	MARK_VNODE(ap->a_dvp);
    511   1.37  perseant 	error = ufs_symlink(ap);
    512   1.37  perseant 	UNMARK_VNODE(ap->a_dvp);
    513   1.40  perseant 	if (*(ap->a_vpp))
    514   1.37  perseant 		UNMARK_VNODE(*(ap->a_vpp));
    515   1.37  perseant 	SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"symlink");
    516   1.37  perseant 	return (error);
    517    1.1   mycroft }
    518    1.1   mycroft 
    519    1.1   mycroft int
    520   1.51  perseant lfs_mknod(void *v)
    521   1.10  christos {
    522   1.22  perseant 	struct vop_mknod_args	/* {
    523    1.1   mycroft 		struct vnode *a_dvp;
    524    1.1   mycroft 		struct vnode **a_vpp;
    525    1.1   mycroft 		struct componentname *a_cnp;
    526    1.1   mycroft 		struct vattr *a_vap;
    527   1.22  perseant 		} */ *ap = v;
    528   1.86  perseant 	struct vattr *vap = ap->a_vap;
    529   1.86  perseant 	struct vnode **vpp = ap->a_vpp;
    530   1.86  perseant 	struct inode *ip;
    531   1.86  perseant 	int error;
    532   1.52     assar 	struct mount	*mp;
    533   1.52     assar 	ino_t		ino;
    534    1.1   mycroft 
    535   1.40  perseant 	if ((error = SET_DIROP(ap->a_dvp)) != 0) {
    536   1.34  perseant 		vput(ap->a_dvp);
    537   1.28  perseant 		return error;
    538   1.34  perseant 	}
    539    1.1   mycroft 	MARK_VNODE(ap->a_dvp);
    540   1.28  perseant 	error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
    541  1.109      fvdl 	    ap->a_dvp, vpp, ap->a_cnp);
    542   1.37  perseant 	UNMARK_VNODE(ap->a_dvp);
    543   1.86  perseant 	if (*(ap->a_vpp))
    544   1.86  perseant 		UNMARK_VNODE(*(ap->a_vpp));
    545   1.28  perseant 
    546   1.28  perseant 	/* Either way we're done with the dirop at this point */
    547   1.26  perseant 	SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"mknod");
    548   1.28  perseant 
    549   1.86  perseant 	if (error)
    550   1.28  perseant 		return (error);
    551   1.28  perseant 
    552   1.86  perseant 	ip = VTOI(*vpp);
    553   1.52     assar 	mp  = (*vpp)->v_mount;
    554   1.52     assar 	ino = ip->i_number;
    555   1.86  perseant 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
    556   1.86  perseant 	if (vap->va_rdev != VNOVAL) {
    557   1.86  perseant 		/*
    558   1.86  perseant 		 * Want to be able to use this to make badblock
    559   1.86  perseant 		 * inodes, so don't truncate the dev number.
    560   1.86  perseant 		 */
    561   1.28  perseant #if 0
    562  1.102      fvdl 		ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev,
    563   1.86  perseant 		    UFS_MPNEEDSWAP((*vpp)->v_mount));
    564   1.28  perseant #else
    565  1.102      fvdl 		ip->i_ffs1_rdev = vap->va_rdev;
    566   1.28  perseant #endif
    567   1.86  perseant 	}
    568  1.134  perseant 
    569   1.28  perseant 	/*
    570   1.28  perseant 	 * Call fsync to write the vnode so that we don't have to deal with
    571   1.28  perseant 	 * flushing it when it's marked VDIROP|VXLOCK.
    572   1.28  perseant 	 *
    573   1.28  perseant 	 * XXX KS - If we can't flush we also can't call vgone(), so must
    574   1.28  perseant 	 * return.  But, that leaves this vnode in limbo, also not good.
    575   1.28  perseant 	 * Can this ever happen (barring hardware failure)?
    576   1.28  perseant 	 */
    577   1.80   thorpej 	if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0,
    578  1.109      fvdl 	    curproc)) != 0) {
    579   1.40  perseant 		printf("Couldn't fsync in mknod (ino %d)---what do I do?\n",
    580   1.40  perseant 		       VTOI(*vpp)->i_number);
    581   1.28  perseant 		return (error);
    582   1.40  perseant 	}
    583   1.86  perseant 	/*
    584   1.86  perseant 	 * Remove vnode so that it will be reloaded by VFS_VGET and
    585   1.86  perseant 	 * checked to see if it is an alias of an existing entry in
    586   1.86  perseant 	 * the inode cache.
    587   1.86  perseant 	 */
    588   1.28  perseant 	/* Used to be vput, but that causes us to call VOP_INACTIVE twice. */
    589  1.134  perseant 
    590   1.40  perseant 	VOP_UNLOCK(*vpp, 0);
    591   1.28  perseant 	lfs_vunref(*vpp);
    592   1.86  perseant 	(*vpp)->v_type = VNON;
    593   1.86  perseant 	vgone(*vpp);
    594  1.108   thorpej 	error = VFS_VGET(mp, ino, vpp);
    595  1.134  perseant 
    596   1.52     assar 	if (error != 0) {
    597   1.52     assar 		*vpp = NULL;
    598   1.52     assar 		return (error);
    599   1.52     assar 	}
    600   1.86  perseant 	return (0);
    601    1.1   mycroft }
    602    1.1   mycroft 
    603    1.1   mycroft int
    604   1.51  perseant lfs_create(void *v)
    605   1.10  christos {
    606   1.22  perseant 	struct vop_create_args	/* {
    607    1.1   mycroft 		struct vnode *a_dvp;
    608    1.1   mycroft 		struct vnode **a_vpp;
    609    1.1   mycroft 		struct componentname *a_cnp;
    610    1.1   mycroft 		struct vattr *a_vap;
    611   1.10  christos 	} */ *ap = v;
    612   1.37  perseant 	int error;
    613    1.1   mycroft 
    614   1.59       chs 	if ((error = SET_DIROP(ap->a_dvp)) != 0) {
    615   1.34  perseant 		vput(ap->a_dvp);
    616   1.37  perseant 		return error;
    617   1.34  perseant 	}
    618    1.1   mycroft 	MARK_VNODE(ap->a_dvp);
    619   1.37  perseant 	error = ufs_create(ap);
    620   1.37  perseant 	UNMARK_VNODE(ap->a_dvp);
    621   1.86  perseant 	if (*(ap->a_vpp))
    622   1.86  perseant 		UNMARK_VNODE(*(ap->a_vpp));
    623   1.26  perseant 	SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"create");
    624   1.37  perseant 	return (error);
    625   1.22  perseant }
    626   1.22  perseant 
    627   1.22  perseant int
    628   1.51  perseant lfs_mkdir(void *v)
    629   1.10  christos {
    630   1.22  perseant 	struct vop_mkdir_args	/* {
    631    1.1   mycroft 		struct vnode *a_dvp;
    632    1.1   mycroft 		struct vnode **a_vpp;
    633    1.1   mycroft 		struct componentname *a_cnp;
    634    1.1   mycroft 		struct vattr *a_vap;
    635   1.10  christos 	} */ *ap = v;
    636   1.37  perseant 	int error;
    637    1.1   mycroft 
    638   1.59       chs 	if ((error = SET_DIROP(ap->a_dvp)) != 0) {
    639   1.34  perseant 		vput(ap->a_dvp);
    640   1.37  perseant 		return error;
    641   1.34  perseant 	}
    642    1.1   mycroft 	MARK_VNODE(ap->a_dvp);
    643   1.37  perseant 	error = ufs_mkdir(ap);
    644   1.37  perseant 	UNMARK_VNODE(ap->a_dvp);
    645   1.86  perseant 	if (*(ap->a_vpp))
    646   1.86  perseant 		UNMARK_VNODE(*(ap->a_vpp));
    647   1.26  perseant 	SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"mkdir");
    648   1.37  perseant 	return (error);
    649    1.1   mycroft }
    650    1.1   mycroft 
    651    1.1   mycroft int
    652   1.51  perseant lfs_remove(void *v)
    653   1.10  christos {
    654   1.22  perseant 	struct vop_remove_args	/* {
    655    1.1   mycroft 		struct vnode *a_dvp;
    656    1.1   mycroft 		struct vnode *a_vp;
    657    1.1   mycroft 		struct componentname *a_cnp;
    658   1.10  christos 	} */ *ap = v;
    659   1.34  perseant 	struct vnode *dvp, *vp;
    660   1.37  perseant 	int error;
    661   1.34  perseant 
    662   1.34  perseant 	dvp = ap->a_dvp;
    663   1.34  perseant 	vp = ap->a_vp;
    664   1.71      yamt 	if ((error = SET_DIROP2(dvp, vp)) != 0) {
    665   1.34  perseant 		if (dvp == vp)
    666   1.34  perseant 			vrele(vp);
    667   1.34  perseant 		else
    668   1.34  perseant 			vput(vp);
    669   1.34  perseant 		vput(dvp);
    670   1.37  perseant 		return error;
    671   1.34  perseant 	}
    672   1.34  perseant 	MARK_VNODE(dvp);
    673   1.34  perseant 	MARK_VNODE(vp);
    674   1.37  perseant 	error = ufs_remove(ap);
    675   1.37  perseant 	UNMARK_VNODE(dvp);
    676   1.37  perseant 	UNMARK_VNODE(vp);
    677   1.40  perseant 
    678   1.71      yamt 	SET_ENDOP2(VTOI(dvp)->i_lfs, dvp, vp, "remove");
    679   1.37  perseant 	return (error);
    680    1.1   mycroft }
    681    1.1   mycroft 
    682    1.1   mycroft int
    683   1.51  perseant lfs_rmdir(void *v)
    684   1.10  christos {
    685   1.22  perseant 	struct vop_rmdir_args	/* {
    686    1.1   mycroft 		struct vnodeop_desc *a_desc;
    687    1.1   mycroft 		struct vnode *a_dvp;
    688    1.1   mycroft 		struct vnode *a_vp;
    689    1.1   mycroft 		struct componentname *a_cnp;
    690   1.10  christos 	} */ *ap = v;
    691   1.84  perseant 	struct vnode *vp;
    692   1.37  perseant 	int error;
    693    1.1   mycroft 
    694   1.84  perseant 	vp = ap->a_vp;
    695   1.71      yamt 	if ((error = SET_DIROP2(ap->a_dvp, ap->a_vp)) != 0) {
    696   1.34  perseant 		vrele(ap->a_dvp);
    697   1.69      yamt 		if (ap->a_vp != ap->a_dvp)
    698   1.34  perseant 			VOP_UNLOCK(ap->a_dvp, 0);
    699   1.84  perseant 		vput(vp);
    700   1.37  perseant 		return error;
    701   1.34  perseant 	}
    702    1.1   mycroft 	MARK_VNODE(ap->a_dvp);
    703   1.84  perseant 	MARK_VNODE(vp);
    704   1.37  perseant 	error = ufs_rmdir(ap);
    705   1.37  perseant 	UNMARK_VNODE(ap->a_dvp);
    706   1.84  perseant 	UNMARK_VNODE(vp);
    707   1.40  perseant 
    708   1.84  perseant 	SET_ENDOP2(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vp, "rmdir");
    709   1.37  perseant 	return (error);
    710    1.1   mycroft }
    711    1.1   mycroft 
    712    1.1   mycroft int
    713   1.51  perseant lfs_link(void *v)
    714   1.10  christos {
    715   1.22  perseant 	struct vop_link_args	/* {
    716    1.9   mycroft 		struct vnode *a_dvp;
    717    1.1   mycroft 		struct vnode *a_vp;
    718    1.1   mycroft 		struct componentname *a_cnp;
    719   1.10  christos 	} */ *ap = v;
    720   1.37  perseant 	int error;
    721    1.1   mycroft 
    722   1.40  perseant 	if ((error = SET_DIROP(ap->a_dvp)) != 0) {
    723   1.34  perseant 		vput(ap->a_dvp);
    724   1.37  perseant 		return error;
    725   1.34  perseant 	}
    726    1.9   mycroft 	MARK_VNODE(ap->a_dvp);
    727   1.37  perseant 	error = ufs_link(ap);
    728   1.37  perseant 	UNMARK_VNODE(ap->a_dvp);
    729   1.26  perseant 	SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"link");
    730   1.37  perseant 	return (error);
    731    1.1   mycroft }
    732   1.22  perseant 
    733    1.1   mycroft int
    734   1.51  perseant lfs_rename(void *v)
    735   1.10  christos {
    736   1.22  perseant 	struct vop_rename_args	/* {
    737    1.1   mycroft 		struct vnode *a_fdvp;
    738    1.1   mycroft 		struct vnode *a_fvp;
    739    1.1   mycroft 		struct componentname *a_fcnp;
    740    1.1   mycroft 		struct vnode *a_tdvp;
    741    1.1   mycroft 		struct vnode *a_tvp;
    742    1.1   mycroft 		struct componentname *a_tcnp;
    743   1.10  christos 	} */ *ap = v;
    744   1.30  perseant 	struct vnode *tvp, *fvp, *tdvp, *fdvp;
    745   1.83  perseant 	struct componentname *tcnp, *fcnp;
    746   1.30  perseant 	int error;
    747   1.29  perseant 	struct lfs *fs;
    748   1.29  perseant 
    749   1.29  perseant 	fs = VTOI(ap->a_fdvp)->i_lfs;
    750   1.30  perseant 	tvp = ap->a_tvp;
    751   1.30  perseant 	tdvp = ap->a_tdvp;
    752   1.83  perseant 	tcnp = ap->a_tcnp;
    753   1.30  perseant 	fvp = ap->a_fvp;
    754   1.30  perseant 	fdvp = ap->a_fdvp;
    755   1.83  perseant 	fcnp = ap->a_fcnp;
    756   1.30  perseant 
    757   1.30  perseant 	/*
    758   1.30  perseant 	 * Check for cross-device rename.
    759   1.30  perseant 	 * If it is, we don't want to set dirops, just error out.
    760   1.30  perseant 	 * (In particular note that MARK_VNODE(tdvp) will DTWT on
    761   1.30  perseant 	 * a cross-device rename.)
    762   1.30  perseant 	 *
    763   1.30  perseant 	 * Copied from ufs_rename.
    764   1.30  perseant 	 */
    765   1.30  perseant 	if ((fvp->v_mount != tdvp->v_mount) ||
    766   1.30  perseant 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
    767   1.30  perseant 		error = EXDEV;
    768   1.34  perseant 		goto errout;
    769   1.30  perseant 	}
    770   1.83  perseant 
    771   1.83  perseant 	/*
    772   1.83  perseant 	 * Check to make sure we're not renaming a vnode onto itself
    773   1.83  perseant 	 * (deleting a hard link by renaming one name onto another);
    774   1.83  perseant 	 * if we are we can't recursively call VOP_REMOVE since that
    775   1.83  perseant 	 * would leave us with an unaccounted-for number of live dirops.
    776   1.83  perseant 	 *
    777   1.83  perseant 	 * Inline the relevant section of ufs_rename here, *before*
    778   1.83  perseant 	 * calling SET_DIROP2.
    779   1.83  perseant 	 */
    780  1.102      fvdl 	if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) ||
    781  1.102      fvdl 	    (VTOI(tdvp)->i_flags & APPEND))) {
    782   1.83  perseant 		error = EPERM;
    783   1.83  perseant 		goto errout;
    784   1.83  perseant 	}
    785   1.86  perseant 	if (fvp == tvp) {
    786   1.86  perseant 		if (fvp->v_type == VDIR) {
    787   1.86  perseant 			error = EINVAL;
    788   1.86  perseant 			goto errout;
    789   1.86  perseant 		}
    790   1.86  perseant 
    791   1.86  perseant 		/* Release destination completely. */
    792   1.86  perseant 		VOP_ABORTOP(tdvp, tcnp);
    793   1.86  perseant 		vput(tdvp);
    794   1.86  perseant 		vput(tvp);
    795   1.86  perseant 
    796   1.86  perseant 		/* Delete source. */
    797   1.86  perseant 		vrele(fvp);
    798   1.86  perseant 		fcnp->cn_flags &= ~(MODMASK | SAVESTART);
    799   1.86  perseant 		fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
    800   1.86  perseant 		fcnp->cn_nameiop = DELETE;
    801   1.86  perseant 		if ((error = relookup(fdvp, &fvp, fcnp))){
    802   1.86  perseant 			/* relookup blew away fdvp */
    803   1.86  perseant 			return (error);
    804   1.86  perseant 		}
    805   1.86  perseant 		return (VOP_REMOVE(fdvp, fvp, fcnp));
    806   1.86  perseant 	}
    807   1.83  perseant 
    808   1.71      yamt 	if ((error = SET_DIROP2(tdvp, tvp)) != 0)
    809   1.34  perseant 		goto errout;
    810   1.30  perseant 	MARK_VNODE(fdvp);
    811   1.30  perseant 	MARK_VNODE(tdvp);
    812   1.71      yamt 	MARK_VNODE(fvp);
    813   1.71      yamt 	if (tvp) {
    814   1.71      yamt 		MARK_VNODE(tvp);
    815   1.71      yamt 	}
    816   1.83  perseant 
    817   1.30  perseant 	error = ufs_rename(ap);
    818   1.37  perseant 	UNMARK_VNODE(fdvp);
    819   1.37  perseant 	UNMARK_VNODE(tdvp);
    820   1.71      yamt 	UNMARK_VNODE(fvp);
    821   1.71      yamt 	if (tvp) {
    822   1.71      yamt 		UNMARK_VNODE(tvp);
    823   1.71      yamt 	}
    824   1.71      yamt 	SET_ENDOP2(fs, tdvp, tvp, "rename");
    825   1.34  perseant 	return (error);
    826   1.34  perseant 
    827   1.34  perseant     errout:
    828   1.34  perseant 	VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */
    829   1.34  perseant 	if (tdvp == tvp)
    830   1.34  perseant 		vrele(tdvp);
    831   1.34  perseant 	else
    832   1.34  perseant 		vput(tdvp);
    833   1.34  perseant 	if (tvp)
    834   1.34  perseant 		vput(tvp);
    835   1.34  perseant 	VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */
    836   1.34  perseant 	vrele(fdvp);
    837   1.34  perseant 	vrele(fvp);
    838   1.30  perseant 	return (error);
    839    1.1   mycroft }
    840   1.22  perseant 
    841    1.1   mycroft /* XXX hack to avoid calling ITIMES in getattr */
    842    1.1   mycroft int
    843   1.51  perseant lfs_getattr(void *v)
    844   1.10  christos {
    845    1.1   mycroft 	struct vop_getattr_args /* {
    846    1.1   mycroft 		struct vnode *a_vp;
    847    1.1   mycroft 		struct vattr *a_vap;
    848    1.1   mycroft 		struct ucred *a_cred;
    849  1.109      fvdl 		struct proc *a_p;
    850   1.10  christos 	} */ *ap = v;
    851   1.35  augustss 	struct vnode *vp = ap->a_vp;
    852   1.35  augustss 	struct inode *ip = VTOI(vp);
    853   1.35  augustss 	struct vattr *vap = ap->a_vap;
    854   1.51  perseant 	struct lfs *fs = ip->i_lfs;
    855    1.1   mycroft 	/*
    856    1.1   mycroft 	 * Copy from inode table
    857    1.1   mycroft 	 */
    858    1.1   mycroft 	vap->va_fsid = ip->i_dev;
    859    1.1   mycroft 	vap->va_fileid = ip->i_number;
    860  1.102      fvdl 	vap->va_mode = ip->i_mode & ~IFMT;
    861  1.102      fvdl 	vap->va_nlink = ip->i_nlink;
    862  1.102      fvdl 	vap->va_uid = ip->i_uid;
    863  1.102      fvdl 	vap->va_gid = ip->i_gid;
    864  1.102      fvdl 	vap->va_rdev = (dev_t)ip->i_ffs1_rdev;
    865   1.55       chs 	vap->va_size = vp->v_size;
    866  1.102      fvdl 	vap->va_atime.tv_sec = ip->i_ffs1_atime;
    867  1.102      fvdl 	vap->va_atime.tv_nsec = ip->i_ffs1_atimensec;
    868  1.102      fvdl 	vap->va_mtime.tv_sec = ip->i_ffs1_mtime;
    869  1.102      fvdl 	vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec;
    870  1.102      fvdl 	vap->va_ctime.tv_sec = ip->i_ffs1_ctime;
    871  1.102      fvdl 	vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec;
    872  1.102      fvdl 	vap->va_flags = ip->i_flags;
    873  1.102      fvdl 	vap->va_gen = ip->i_gen;
    874    1.1   mycroft 	/* this doesn't belong here */
    875    1.1   mycroft 	if (vp->v_type == VBLK)
    876    1.1   mycroft 		vap->va_blocksize = BLKDEV_IOSIZE;
    877    1.1   mycroft 	else if (vp->v_type == VCHR)
    878    1.1   mycroft 		vap->va_blocksize = MAXBSIZE;
    879    1.1   mycroft 	else
    880    1.1   mycroft 		vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
    881   1.84  perseant 	vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks);
    882    1.1   mycroft 	vap->va_type = vp->v_type;
    883    1.1   mycroft 	vap->va_filerev = ip->i_modrev;
    884    1.1   mycroft 	return (0);
    885   1.61  perseant }
    886   1.61  perseant 
    887   1.61  perseant /*
    888   1.61  perseant  * Check to make sure the inode blocks won't choke the buffer
    889   1.61  perseant  * cache, then call ufs_setattr as usual.
    890   1.61  perseant  */
    891   1.61  perseant int
    892   1.61  perseant lfs_setattr(void *v)
    893   1.61  perseant {
    894   1.61  perseant 	struct vop_getattr_args /* {
    895   1.61  perseant 		struct vnode *a_vp;
    896   1.61  perseant 		struct vattr *a_vap;
    897   1.61  perseant 		struct ucred *a_cred;
    898  1.109      fvdl 		struct proc *a_p;
    899   1.61  perseant 	} */ *ap = v;
    900   1.61  perseant 	struct vnode *vp = ap->a_vp;
    901   1.61  perseant 
    902   1.61  perseant 	lfs_check(vp, LFS_UNUSED_LBN, 0);
    903   1.61  perseant 	return ufs_setattr(v);
    904    1.1   mycroft }
    905   1.22  perseant 
    906    1.1   mycroft /*
    907    1.1   mycroft  * Close called
    908    1.1   mycroft  *
    909    1.1   mycroft  * XXX -- we were using ufs_close, but since it updates the
    910    1.1   mycroft  * times on the inode, we might need to bump the uinodes
    911    1.1   mycroft  * count.
    912    1.1   mycroft  */
    913    1.1   mycroft /* ARGSUSED */
    914    1.1   mycroft int
    915   1.51  perseant lfs_close(void *v)
    916   1.10  christos {
    917    1.1   mycroft 	struct vop_close_args /* {
    918    1.1   mycroft 		struct vnode *a_vp;
    919    1.1   mycroft 		int  a_fflag;
    920    1.1   mycroft 		struct ucred *a_cred;
    921  1.109      fvdl 		struct proc *a_p;
    922   1.10  christos 	} */ *ap = v;
    923   1.35  augustss 	struct vnode *vp = ap->a_vp;
    924   1.35  augustss 	struct inode *ip = VTOI(vp);
    925   1.12   mycroft 	struct timespec ts;
    926    1.1   mycroft 
    927   1.97  perseant 	if (vp == ip->i_lfs->lfs_ivnode &&
    928  1.119       dbj 	    vp->v_mount->mnt_iflag & IMNT_UNMOUNT)
    929   1.97  perseant 		return 0;
    930   1.97  perseant 
    931   1.97  perseant 	if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) {
    932   1.12   mycroft 		TIMEVAL_TO_TIMESPEC(&time, &ts);
    933   1.22  perseant 		LFS_ITIMES(ip, &ts, &ts, &ts);
    934    1.1   mycroft 	}
    935    1.1   mycroft 	return (0);
    936   1.65  perseant }
    937   1.65  perseant 
    938   1.65  perseant /*
    939   1.65  perseant  * Close wrapper for special devices.
    940   1.65  perseant  *
    941   1.65  perseant  * Update the times on the inode then do device close.
    942   1.65  perseant  */
    943   1.65  perseant int
    944   1.65  perseant lfsspec_close(void *v)
    945   1.65  perseant {
    946   1.65  perseant 	struct vop_close_args /* {
    947   1.65  perseant 		struct vnode	*a_vp;
    948   1.65  perseant 		int		a_fflag;
    949   1.65  perseant 		struct ucred	*a_cred;
    950  1.109      fvdl 		struct proc	*a_p;
    951   1.65  perseant 	} */ *ap = v;
    952   1.65  perseant 	struct vnode	*vp;
    953   1.65  perseant 	struct inode	*ip;
    954   1.65  perseant 	struct timespec	ts;
    955   1.65  perseant 
    956   1.65  perseant 	vp = ap->a_vp;
    957   1.65  perseant 	ip = VTOI(vp);
    958   1.65  perseant 	if (vp->v_usecount > 1) {
    959   1.65  perseant 		TIMEVAL_TO_TIMESPEC(&time, &ts);
    960   1.65  perseant 		LFS_ITIMES(ip, &ts, &ts, &ts);
    961   1.65  perseant 	}
    962   1.65  perseant 	return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap));
    963   1.65  perseant }
    964   1.65  perseant 
    965   1.65  perseant /*
    966   1.65  perseant  * Close wrapper for fifo's.
    967   1.65  perseant  *
    968   1.65  perseant  * Update the times on the inode then do device close.
    969   1.65  perseant  */
    970   1.65  perseant int
    971   1.65  perseant lfsfifo_close(void *v)
    972   1.65  perseant {
    973   1.65  perseant 	struct vop_close_args /* {
    974   1.65  perseant 		struct vnode	*a_vp;
    975   1.65  perseant 		int		a_fflag;
    976   1.65  perseant 		struct ucred	*a_cred;
    977  1.109      fvdl 		struct proc	*a_p;
    978   1.65  perseant 	} */ *ap = v;
    979   1.65  perseant 	struct vnode	*vp;
    980   1.65  perseant 	struct inode	*ip;
    981   1.65  perseant 	struct timespec	ts;
    982   1.65  perseant 
    983   1.65  perseant 	vp = ap->a_vp;
    984   1.65  perseant 	ip = VTOI(vp);
    985   1.65  perseant 	if (ap->a_vp->v_usecount > 1) {
    986   1.65  perseant 		TIMEVAL_TO_TIMESPEC(&time, &ts);
    987   1.65  perseant 		LFS_ITIMES(ip, &ts, &ts, &ts);
    988   1.65  perseant 	}
    989   1.65  perseant 	return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap));
    990    1.1   mycroft }
    991    1.1   mycroft 
    992    1.1   mycroft /*
    993   1.15      fvdl  * Reclaim an inode so that it can be used for other purposes.
    994    1.1   mycroft  */
    995    1.1   mycroft 
    996    1.1   mycroft int
    997   1.51  perseant lfs_reclaim(void *v)
    998   1.10  christos {
    999    1.1   mycroft 	struct vop_reclaim_args /* {
   1000    1.1   mycroft 		struct vnode *a_vp;
   1001  1.109      fvdl 		struct proc *a_p;
   1002   1.10  christos 	} */ *ap = v;
   1003   1.15      fvdl 	struct vnode *vp = ap->a_vp;
   1004   1.84  perseant 	struct inode *ip = VTOI(vp);
   1005    1.1   mycroft 	int error;
   1006   1.77      yamt 
   1007  1.102      fvdl 	KASSERT(ip->i_nlink == ip->i_ffs_effnlink);
   1008    1.1   mycroft 
   1009   1.84  perseant 	LFS_CLR_UINO(ip, IN_ALLMOD);
   1010  1.109      fvdl 	if ((error = ufs_reclaim(vp, ap->a_p)))
   1011    1.1   mycroft 		return (error);
   1012  1.134  perseant 	lfs_deregister_all(vp);
   1013  1.102      fvdl 	pool_put(&lfs_dinode_pool, VTOI(vp)->i_din.ffs1_din);
   1014   1.84  perseant 	pool_put(&lfs_inoext_pool, ip->inode_ext.lfs);
   1015   1.84  perseant 	ip->inode_ext.lfs = NULL;
   1016   1.19   thorpej 	pool_put(&lfs_inode_pool, vp->v_data);
   1017    1.1   mycroft 	vp->v_data = NULL;
   1018   1.94  perseant 	return (0);
   1019   1.94  perseant }
   1020   1.94  perseant 
   1021   1.94  perseant /*
   1022  1.101      yamt  * Read a block from a storage device.
   1023   1.94  perseant  * In order to avoid reading blocks that are in the process of being
   1024   1.94  perseant  * written by the cleaner---and hence are not mutexed by the normal
   1025   1.94  perseant  * buffer cache / page cache mechanisms---check for collisions before
   1026   1.94  perseant  * reading.
   1027   1.94  perseant  *
   1028   1.94  perseant  * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before*
   1029   1.94  perseant  * the active cleaner test.
   1030   1.94  perseant  *
   1031   1.94  perseant  * XXX This code assumes that lfs_markv makes synchronous checkpoints.
   1032   1.94  perseant  */
   1033   1.94  perseant int
   1034   1.94  perseant lfs_strategy(void *v)
   1035   1.94  perseant {
   1036   1.94  perseant 	struct vop_strategy_args /* {
   1037  1.128   hannken 		struct vnode *a_vp;
   1038   1.94  perseant 		struct buf *a_bp;
   1039   1.94  perseant 	} */ *ap = v;
   1040   1.94  perseant 	struct buf	*bp;
   1041   1.94  perseant 	struct lfs	*fs;
   1042   1.94  perseant 	struct vnode	*vp;
   1043   1.94  perseant 	struct inode	*ip;
   1044   1.94  perseant 	daddr_t		tbn;
   1045   1.94  perseant 	int		i, sn, error, slept;
   1046   1.94  perseant 
   1047   1.94  perseant 	bp = ap->a_bp;
   1048  1.128   hannken 	vp = ap->a_vp;
   1049   1.94  perseant 	ip = VTOI(vp);
   1050   1.94  perseant 	fs = ip->i_lfs;
   1051   1.94  perseant 
   1052  1.101      yamt 	/* lfs uses its strategy routine only for read */
   1053  1.101      yamt 	KASSERT(bp->b_flags & B_READ);
   1054  1.101      yamt 
   1055   1.94  perseant 	if (vp->v_type == VBLK || vp->v_type == VCHR)
   1056   1.94  perseant 		panic("lfs_strategy: spec");
   1057   1.94  perseant 	KASSERT(bp->b_bcount != 0);
   1058   1.94  perseant 	if (bp->b_blkno == bp->b_lblkno) {
   1059   1.94  perseant 		error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno,
   1060   1.94  perseant 				 NULL);
   1061   1.94  perseant 		if (error) {
   1062   1.94  perseant 			bp->b_error = error;
   1063   1.94  perseant 			bp->b_flags |= B_ERROR;
   1064   1.94  perseant 			biodone(bp);
   1065   1.94  perseant 			return (error);
   1066   1.94  perseant 		}
   1067   1.94  perseant 		if ((long)bp->b_blkno == -1) /* no valid data */
   1068   1.94  perseant 			clrbuf(bp);
   1069   1.94  perseant 	}
   1070   1.94  perseant 	if ((long)bp->b_blkno < 0) { /* block is not on disk */
   1071   1.94  perseant 		biodone(bp);
   1072   1.94  perseant 		return (0);
   1073   1.94  perseant 	}
   1074   1.94  perseant 
   1075   1.94  perseant 	slept = 1;
   1076   1.96  perseant 	simple_lock(&fs->lfs_interlock);
   1077  1.101      yamt 	while (slept && fs->lfs_seglock) {
   1078   1.96  perseant 		simple_unlock(&fs->lfs_interlock);
   1079   1.94  perseant 		/*
   1080   1.94  perseant 		 * Look through list of intervals.
   1081   1.94  perseant 		 * There will only be intervals to look through
   1082   1.94  perseant 		 * if the cleaner holds the seglock.
   1083   1.94  perseant 		 * Since the cleaner is synchronous, we can trust
   1084   1.94  perseant 		 * the list of intervals to be current.
   1085   1.94  perseant 		 */
   1086   1.94  perseant 		tbn = dbtofsb(fs, bp->b_blkno);
   1087   1.94  perseant 		sn = dtosn(fs, tbn);
   1088   1.94  perseant 		slept = 0;
   1089   1.94  perseant 		for (i = 0; i < fs->lfs_cleanind; i++) {
   1090   1.94  perseant 			if (sn == dtosn(fs, fs->lfs_cleanint[i]) &&
   1091   1.94  perseant 			    tbn >= fs->lfs_cleanint[i]) {
   1092   1.94  perseant #ifdef DEBUG_LFS
   1093   1.94  perseant 				printf("lfs_strategy: ino %d lbn %" PRId64
   1094   1.94  perseant 				       " ind %d sn %d fsb %" PRIx32
   1095   1.94  perseant 				       " given sn %d fsb %" PRIx64 "\n",
   1096   1.94  perseant 					ip->i_number, bp->b_lblkno, i,
   1097   1.94  perseant 					dtosn(fs, fs->lfs_cleanint[i]),
   1098   1.94  perseant 					fs->lfs_cleanint[i], sn, tbn);
   1099   1.94  perseant 				printf("lfs_strategy: sleeping on ino %d lbn %"
   1100   1.94  perseant 				       PRId64 "\n", ip->i_number, bp->b_lblkno);
   1101   1.94  perseant #endif
   1102   1.94  perseant 				tsleep(&fs->lfs_seglock, PRIBIO+1,
   1103   1.94  perseant 					"lfs_strategy", 0);
   1104   1.94  perseant 				/* Things may be different now; start over. */
   1105   1.94  perseant 				slept = 1;
   1106   1.94  perseant 				break;
   1107   1.94  perseant 			}
   1108   1.94  perseant 		}
   1109   1.96  perseant 		simple_lock(&fs->lfs_interlock);
   1110   1.94  perseant 	}
   1111   1.96  perseant 	simple_unlock(&fs->lfs_interlock);
   1112   1.94  perseant 
   1113   1.94  perseant 	vp = ip->i_devvp;
   1114  1.127   hannken 	VOP_STRATEGY(vp, bp);
   1115    1.1   mycroft 	return (0);
   1116   1.89  perseant }
   1117   1.89  perseant 
   1118   1.92  perseant static void
   1119   1.92  perseant lfs_flush_dirops(struct lfs *fs)
   1120   1.92  perseant {
   1121   1.92  perseant 	struct inode *ip, *nip;
   1122   1.92  perseant 	struct vnode *vp;
   1123   1.92  perseant 	extern int lfs_dostats;
   1124   1.92  perseant 	struct segment *sp;
   1125   1.92  perseant 	int needunlock;
   1126   1.92  perseant 
   1127   1.92  perseant 	if (fs->lfs_ronly)
   1128   1.92  perseant 		return;
   1129   1.92  perseant 
   1130   1.92  perseant 	if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL)
   1131   1.92  perseant 		return;
   1132   1.92  perseant 
   1133   1.92  perseant 	if (lfs_dostats)
   1134   1.92  perseant 		++lfs_stats.flush_invoked;
   1135   1.92  perseant 
   1136   1.92  perseant 	/*
   1137   1.92  perseant 	 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops.
   1138   1.92  perseant 	 * Technically this is a checkpoint (the on-disk state is valid)
   1139   1.92  perseant 	 * even though we are leaving out all the file data.
   1140   1.92  perseant 	 */
   1141   1.92  perseant 	lfs_imtime(fs);
   1142   1.92  perseant 	lfs_seglock(fs, SEGM_CKP);
   1143   1.92  perseant 	sp = fs->lfs_sp;
   1144   1.92  perseant 
   1145   1.92  perseant 	/*
   1146   1.92  perseant 	 * lfs_writevnodes, optimized to get dirops out of the way.
   1147   1.92  perseant 	 * Only write dirops, and don't flush files' pages, only
   1148   1.92  perseant 	 * blocks from the directories.
   1149   1.92  perseant 	 *
   1150   1.92  perseant 	 * We don't need to vref these files because they are
   1151   1.92  perseant 	 * dirops and so hold an extra reference until the
   1152   1.92  perseant 	 * segunlock clears them of that status.
   1153   1.92  perseant 	 *
   1154   1.92  perseant 	 * We don't need to check for IN_ADIROP because we know that
   1155   1.92  perseant 	 * no dirops are active.
   1156   1.92  perseant 	 *
   1157   1.92  perseant 	 */
   1158   1.92  perseant 	for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) {
   1159   1.92  perseant 		nip = TAILQ_NEXT(ip, i_lfs_dchain);
   1160   1.92  perseant 		vp = ITOV(ip);
   1161   1.92  perseant 
   1162   1.92  perseant 		/*
   1163   1.92  perseant 		 * All writes to directories come from dirops; all
   1164   1.92  perseant 		 * writes to files' direct blocks go through the page
   1165   1.92  perseant 		 * cache, which we're not touching.  Reads to files
   1166   1.92  perseant 		 * and/or directories will not be affected by writing
   1167   1.92  perseant 		 * directory blocks inodes and file inodes.  So we don't
   1168   1.92  perseant 		 * really need to lock.  If we don't lock, though,
   1169   1.92  perseant 		 * make sure that we don't clear IN_MODIFIED
   1170   1.92  perseant 		 * unnecessarily.
   1171   1.92  perseant 		 */
   1172   1.92  perseant 		if (vp->v_flag & VXLOCK)
   1173   1.92  perseant 			continue;
   1174   1.92  perseant 		if (vn_lock(vp, LK_EXCLUSIVE | LK_CANRECURSE |
   1175   1.92  perseant 			    LK_NOWAIT) == 0) {
   1176   1.92  perseant 			needunlock = 1;
   1177   1.92  perseant 		} else {
   1178   1.92  perseant 			printf("lfs_flush_dirops: flushing locked ino %d\n",
   1179   1.92  perseant 			       VTOI(vp)->i_number);
   1180   1.92  perseant 			needunlock = 0;
   1181   1.92  perseant 		}
   1182   1.92  perseant 		if (vp->v_type != VREG &&
   1183   1.92  perseant 		    ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) {
   1184   1.92  perseant 			lfs_writefile(fs, sp, vp);
   1185   1.92  perseant 			if (!VPISEMPTY(vp) && !WRITEINPROG(vp) &&
   1186   1.92  perseant 			    !(ip->i_flag & IN_ALLMOD)) {
   1187   1.92  perseant 				LFS_SET_UINO(ip, IN_MODIFIED);
   1188   1.92  perseant 			}
   1189   1.92  perseant 		}
   1190   1.92  perseant 		(void) lfs_writeinode(fs, sp, ip);
   1191   1.92  perseant 		if (needunlock)
   1192   1.92  perseant 			VOP_UNLOCK(vp, 0);
   1193   1.92  perseant 		else
   1194   1.92  perseant 			LFS_SET_UINO(ip, IN_MODIFIED);
   1195   1.92  perseant 	}
   1196   1.92  perseant 	/* We've written all the dirops there are */
   1197   1.92  perseant 	((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT);
   1198   1.92  perseant 	(void) lfs_writeseg(fs, sp);
   1199   1.92  perseant 	lfs_segunlock(fs);
   1200   1.92  perseant }
   1201   1.92  perseant 
   1202   1.89  perseant /*
   1203   1.90  perseant  * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}.
   1204   1.89  perseant  */
   1205   1.89  perseant int
   1206   1.90  perseant lfs_fcntl(void *v)
   1207   1.89  perseant {
   1208   1.90  perseant         struct vop_fcntl_args /* {
   1209   1.89  perseant                 struct vnode *a_vp;
   1210   1.89  perseant                 u_long a_command;
   1211   1.89  perseant                 caddr_t  a_data;
   1212   1.89  perseant                 int  a_fflag;
   1213   1.89  perseant                 struct ucred *a_cred;
   1214  1.109      fvdl                 struct proc *a_p;
   1215   1.89  perseant         } */ *ap = v;
   1216   1.89  perseant 	struct timeval *tvp;
   1217   1.89  perseant 	BLOCK_INFO *blkiov;
   1218   1.92  perseant 	CLEANERINFO *cip;
   1219   1.92  perseant 	int blkcnt, error, oclean;
   1220   1.90  perseant 	struct lfs_fcntl_markv blkvp;
   1221   1.89  perseant 	fsid_t *fsidp;
   1222   1.92  perseant 	struct lfs *fs;
   1223   1.92  perseant 	struct buf *bp;
   1224  1.134  perseant 	fhandle_t *fhp;
   1225   1.92  perseant 	daddr_t off;
   1226   1.89  perseant 
   1227   1.90  perseant 	/* Only respect LFS fcntls on fs root or Ifile */
   1228   1.89  perseant 	if (VTOI(ap->a_vp)->i_number != ROOTINO &&
   1229   1.89  perseant 	    VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) {
   1230   1.90  perseant 		return ufs_fcntl(v);
   1231   1.89  perseant 	}
   1232   1.89  perseant 
   1233  1.100  perseant 	/* Avoid locking a draining lock */
   1234  1.119       dbj 	if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) {
   1235  1.100  perseant 		return ESHUTDOWN;
   1236  1.100  perseant 	}
   1237  1.100  perseant 
   1238  1.100  perseant 	fs = VTOI(ap->a_vp)->i_lfs;
   1239  1.131  christos 	fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx;
   1240   1.89  perseant 
   1241   1.98  perseant 	switch (ap->a_command) {
   1242   1.90  perseant 	    case LFCNSEGWAITALL:
   1243  1.134  perseant 	    case LFCNSEGWAITALL_COMPAT:
   1244   1.89  perseant 		fsidp = NULL;
   1245   1.89  perseant 		/* FALLSTHROUGH */
   1246   1.90  perseant 	    case LFCNSEGWAIT:
   1247  1.134  perseant 	    case LFCNSEGWAIT_COMPAT:
   1248   1.89  perseant 		tvp = (struct timeval *)ap->a_data;
   1249  1.100  perseant 		simple_lock(&fs->lfs_interlock);
   1250  1.100  perseant 		++fs->lfs_sleepers;
   1251  1.100  perseant 		simple_unlock(&fs->lfs_interlock);
   1252   1.90  perseant 		VOP_UNLOCK(ap->a_vp, 0);
   1253  1.100  perseant 
   1254   1.90  perseant 		error = lfs_segwait(fsidp, tvp);
   1255  1.100  perseant 
   1256   1.90  perseant 		VOP_LOCK(ap->a_vp, LK_EXCLUSIVE);
   1257  1.100  perseant 		simple_lock(&fs->lfs_interlock);
   1258  1.100  perseant 		if (--fs->lfs_sleepers == 0)
   1259  1.100  perseant 			wakeup(&fs->lfs_sleepers);
   1260  1.100  perseant 		simple_unlock(&fs->lfs_interlock);
   1261   1.90  perseant 		return error;
   1262   1.89  perseant 
   1263   1.90  perseant 	    case LFCNBMAPV:
   1264   1.90  perseant 	    case LFCNMARKV:
   1265  1.109      fvdl 		if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0)
   1266   1.89  perseant 			return (error);
   1267   1.90  perseant 		blkvp = *(struct lfs_fcntl_markv *)ap->a_data;
   1268   1.89  perseant 
   1269   1.89  perseant 		blkcnt = blkvp.blkcnt;
   1270   1.89  perseant 		if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT)
   1271   1.89  perseant 			return (EINVAL);
   1272   1.89  perseant 		blkiov = malloc(blkcnt * sizeof(BLOCK_INFO), M_SEGMENT, M_WAITOK);
   1273   1.89  perseant 		if ((error = copyin(blkvp.blkiov, blkiov,
   1274   1.89  perseant 		     blkcnt * sizeof(BLOCK_INFO))) != 0) {
   1275   1.89  perseant 			free(blkiov, M_SEGMENT);
   1276   1.89  perseant 			return error;
   1277   1.89  perseant 		}
   1278   1.89  perseant 
   1279  1.100  perseant 		simple_lock(&fs->lfs_interlock);
   1280  1.100  perseant 		++fs->lfs_sleepers;
   1281  1.100  perseant 		simple_unlock(&fs->lfs_interlock);
   1282   1.90  perseant 		VOP_UNLOCK(ap->a_vp, 0);
   1283   1.90  perseant 		if (ap->a_command == LFCNBMAPV)
   1284  1.109      fvdl 			error = lfs_bmapv(ap->a_p, fsidp, blkiov, blkcnt);
   1285   1.90  perseant 		else /* LFCNMARKV */
   1286  1.109      fvdl 			error = lfs_markv(ap->a_p, fsidp, blkiov, blkcnt);
   1287   1.89  perseant 		if (error == 0)
   1288   1.89  perseant 			error = copyout(blkiov, blkvp.blkiov,
   1289   1.89  perseant 					blkcnt * sizeof(BLOCK_INFO));
   1290   1.90  perseant 		VOP_LOCK(ap->a_vp, LK_EXCLUSIVE);
   1291  1.100  perseant 		simple_lock(&fs->lfs_interlock);
   1292  1.100  perseant 		if (--fs->lfs_sleepers == 0)
   1293  1.100  perseant 			wakeup(&fs->lfs_sleepers);
   1294  1.100  perseant 		simple_unlock(&fs->lfs_interlock);
   1295   1.89  perseant 		free(blkiov, M_SEGMENT);
   1296   1.89  perseant 		return error;
   1297   1.92  perseant 
   1298   1.92  perseant 	    case LFCNRECLAIM:
   1299   1.92  perseant 		/*
   1300   1.92  perseant 		 * Flush dirops and write Ifile, allowing empty segments
   1301   1.92  perseant 		 * to be immediately reclaimed.
   1302   1.92  perseant 		 */
   1303  1.111      yamt 		lfs_writer_enter(fs, "pndirop");
   1304   1.92  perseant 		off = fs->lfs_offset;
   1305   1.92  perseant 		lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP);
   1306   1.92  perseant 		lfs_flush_dirops(fs);
   1307   1.92  perseant 		LFS_CLEANERINFO(cip, fs, bp);
   1308   1.92  perseant 		oclean = cip->clean;
   1309   1.92  perseant 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
   1310   1.92  perseant 		lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP);
   1311   1.92  perseant 		lfs_segunlock(fs);
   1312  1.111      yamt 		lfs_writer_leave(fs);
   1313   1.92  perseant 
   1314   1.92  perseant #ifdef DEBUG_LFS
   1315   1.92  perseant 		LFS_CLEANERINFO(cip, fs, bp);
   1316   1.92  perseant 		oclean = cip->clean;
   1317   1.92  perseant 		printf("lfs_fcntl: reclaim wrote %" PRId64 " blocks, cleaned "
   1318   1.92  perseant 			"%" PRId32 " segments (activesb %d)\n",
   1319   1.92  perseant 			fs->lfs_offset - off, cip->clean - oclean,
   1320   1.92  perseant 			fs->lfs_activesb);
   1321   1.92  perseant 		LFS_SYNC_CLEANERINFO(cip, fs, bp, 0);
   1322   1.92  perseant #endif
   1323   1.92  perseant 
   1324   1.92  perseant 		return 0;
   1325   1.89  perseant 
   1326  1.134  perseant 	    case LFCNIFILEFH:
   1327  1.134  perseant 		/* Return the filehandle of the Ifile */
   1328  1.134  perseant 		if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0)
   1329  1.134  perseant 			return (error);
   1330  1.134  perseant 		fhp = (struct fhandle *)ap->a_data;
   1331  1.134  perseant 		fhp->fh_fsid = *fsidp;
   1332  1.134  perseant 		return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid));
   1333  1.134  perseant 
   1334   1.89  perseant 	    default:
   1335   1.90  perseant 		return ufs_fcntl(v);
   1336   1.89  perseant 	}
   1337   1.89  perseant 	return 0;
   1338   1.60       chs }
   1339   1.60       chs 
   1340   1.60       chs int
   1341   1.60       chs lfs_getpages(void *v)
   1342   1.60       chs {
   1343   1.60       chs 	struct vop_getpages_args /* {
   1344   1.60       chs 		struct vnode *a_vp;
   1345   1.60       chs 		voff_t a_offset;
   1346   1.60       chs 		struct vm_page **a_m;
   1347   1.60       chs 		int *a_count;
   1348   1.60       chs 		int a_centeridx;
   1349   1.60       chs 		vm_prot_t a_access_type;
   1350   1.60       chs 		int a_advice;
   1351   1.60       chs 		int a_flags;
   1352   1.60       chs 	} */ *ap = v;
   1353   1.60       chs 
   1354   1.97  perseant 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM &&
   1355   1.97  perseant 	    (ap->a_access_type & VM_PROT_WRITE) != 0) {
   1356   1.97  perseant 		return EPERM;
   1357   1.97  perseant 	}
   1358   1.60       chs 	if ((ap->a_access_type & VM_PROT_WRITE) != 0) {
   1359   1.60       chs 		LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED);
   1360   1.60       chs 	}
   1361  1.115      yamt 
   1362  1.115      yamt 	/*
   1363  1.115      yamt 	 * we're relying on the fact that genfs_getpages() always read in
   1364  1.115      yamt 	 * entire filesystem blocks.
   1365  1.115      yamt 	 */
   1366   1.95  perseant 	return genfs_getpages(v);
   1367    1.1   mycroft }
   1368   1.84  perseant 
   1369   1.84  perseant /*
   1370   1.84  perseant  * Make sure that for all pages in every block in the given range,
   1371   1.84  perseant  * either all are dirty or all are clean.  If any of the pages
   1372   1.84  perseant  * we've seen so far are dirty, put the vnode on the paging chain,
   1373   1.84  perseant  * and mark it IN_PAGING.
   1374  1.105  perseant  *
   1375  1.105  perseant  * If checkfirst != 0, don't check all the pages but return at the
   1376  1.105  perseant  * first dirty page.
   1377   1.84  perseant  */
   1378   1.84  perseant static int
   1379   1.84  perseant check_dirty(struct lfs *fs, struct vnode *vp,
   1380   1.84  perseant 	    off_t startoffset, off_t endoffset, off_t blkeof,
   1381  1.103  perseant 	    int flags, int checkfirst)
   1382   1.84  perseant {
   1383   1.86  perseant 	int by_list;
   1384  1.122  christos 	struct vm_page *curpg = NULL; /* XXX: gcc */
   1385  1.122  christos 	struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg;
   1386  1.122  christos 	off_t soff = 0; /* XXX: gcc */
   1387   1.84  perseant 	voff_t off;
   1388  1.115      yamt 	int i;
   1389  1.115      yamt 	int nonexistent;
   1390  1.115      yamt 	int any_dirty;	/* number of dirty pages */
   1391  1.115      yamt 	int dirty;	/* number of dirty pages in a block */
   1392  1.115      yamt 	int tdirty;
   1393   1.84  perseant 	int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT;
   1394   1.84  perseant 
   1395   1.84  perseant   top:
   1396   1.84  perseant 	by_list = (vp->v_uobj.uo_npages <=
   1397   1.84  perseant 		   ((endoffset - startoffset) >> PAGE_SHIFT) *
   1398   1.84  perseant 		   UVM_PAGE_HASH_PENALTY);
   1399   1.84  perseant 	any_dirty = 0;
   1400   1.84  perseant 
   1401   1.84  perseant 	if (by_list) {
   1402   1.84  perseant 		curpg = TAILQ_FIRST(&vp->v_uobj.memq);
   1403   1.84  perseant 	} else {
   1404   1.84  perseant 		soff = startoffset;
   1405   1.84  perseant 	}
   1406   1.84  perseant 	while (by_list || soff < MIN(blkeof, endoffset)) {
   1407   1.84  perseant 		if (by_list) {
   1408  1.115      yamt 			/*
   1409  1.115      yamt 			 * find the first page in a block.
   1410  1.115      yamt 			 */
   1411   1.84  perseant 			if (pages_per_block > 1) {
   1412   1.84  perseant 				while (curpg && (curpg->offset & fs->lfs_bmask))
   1413   1.84  perseant 					curpg = TAILQ_NEXT(curpg, listq);
   1414   1.84  perseant 			}
   1415   1.84  perseant 			if (curpg == NULL)
   1416   1.84  perseant 				break;
   1417   1.84  perseant 			soff = curpg->offset;
   1418   1.84  perseant 		}
   1419   1.84  perseant 
   1420   1.84  perseant 		/*
   1421   1.84  perseant 		 * Mark all pages in extended range busy; find out if any
   1422   1.84  perseant 		 * of them are dirty.
   1423   1.84  perseant 		 */
   1424   1.84  perseant 		nonexistent = dirty = 0;
   1425   1.84  perseant 		for (i = 0; i == 0 || i < pages_per_block; i++) {
   1426   1.84  perseant 			if (by_list && pages_per_block <= 1) {
   1427   1.84  perseant 				pgs[i] = pg = curpg;
   1428   1.84  perseant 			} else {
   1429   1.84  perseant 				off = soff + (i << PAGE_SHIFT);
   1430   1.84  perseant 				pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off);
   1431   1.84  perseant 				if (pg == NULL) {
   1432   1.84  perseant 					++nonexistent;
   1433   1.84  perseant 					continue;
   1434   1.84  perseant 				}
   1435   1.84  perseant 			}
   1436   1.84  perseant 			KASSERT(pg != NULL);
   1437   1.84  perseant 			while (pg->flags & PG_BUSY) {
   1438   1.84  perseant 				pg->flags |= PG_WANTED;
   1439   1.84  perseant 				UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0,
   1440   1.84  perseant 						    "lfsput", 0);
   1441   1.84  perseant 				simple_lock(&vp->v_interlock);
   1442   1.96  perseant 				if (by_list) {
   1443   1.96  perseant 					if (i > 0)
   1444   1.96  perseant 						uvm_page_unbusy(pgs, i);
   1445   1.84  perseant 					goto top;
   1446   1.96  perseant 				}
   1447   1.84  perseant 			}
   1448   1.84  perseant 			pg->flags |= PG_BUSY;
   1449   1.84  perseant 			UVM_PAGE_OWN(pg, "lfs_putpages");
   1450   1.84  perseant 
   1451   1.84  perseant 			pmap_page_protect(pg, VM_PROT_NONE);
   1452   1.84  perseant 			tdirty = (pmap_clear_modify(pg) ||
   1453   1.84  perseant 				  (pg->flags & PG_CLEAN) == 0);
   1454   1.84  perseant 			dirty += tdirty;
   1455   1.84  perseant 		}
   1456   1.84  perseant 		if (pages_per_block > 0 && nonexistent >= pages_per_block) {
   1457   1.84  perseant 			if (by_list) {
   1458   1.84  perseant 				curpg = TAILQ_NEXT(curpg, listq);
   1459   1.84  perseant 			} else {
   1460   1.84  perseant 				soff += fs->lfs_bsize;
   1461   1.84  perseant 			}
   1462   1.84  perseant 			continue;
   1463   1.84  perseant 		}
   1464   1.84  perseant 
   1465   1.84  perseant 		any_dirty += dirty;
   1466   1.84  perseant 		KASSERT(nonexistent == 0);
   1467   1.84  perseant 
   1468   1.84  perseant 		/*
   1469   1.84  perseant 		 * If any are dirty make all dirty; unbusy them,
   1470   1.88  perseant 		 * but if we were asked to clean, wire them so that
   1471   1.88  perseant 		 * the pagedaemon doesn't bother us about them while
   1472   1.88  perseant 		 * they're on their way to disk.
   1473   1.84  perseant 		 */
   1474   1.84  perseant 		for (i = 0; i == 0 || i < pages_per_block; i++) {
   1475   1.84  perseant 			pg = pgs[i];
   1476   1.84  perseant 			KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI)));
   1477   1.84  perseant 			if (dirty) {
   1478   1.84  perseant 				pg->flags &= ~PG_CLEAN;
   1479   1.84  perseant 				if (flags & PGO_FREE) {
   1480   1.84  perseant 					/* XXXUBC need better way to update */
   1481   1.96  perseant 					simple_lock(&lfs_subsys_lock);
   1482   1.84  perseant 					lfs_subsys_pages += MIN(1, pages_per_block);
   1483   1.96  perseant 					simple_unlock(&lfs_subsys_lock);
   1484   1.85      yamt 					/*
   1485   1.96  perseant 					 * Wire the page so that
   1486   1.96  perseant 					 * pdaemon doesn't see it again.
   1487   1.85      yamt 					 */
   1488   1.84  perseant 					uvm_lock_pageq();
   1489   1.85      yamt 					uvm_pagewire(pg);
   1490   1.85      yamt 					uvm_unlock_pageq();
   1491   1.88  perseant 
   1492   1.84  perseant 					/* Suspended write flag */
   1493   1.84  perseant 					pg->flags |= PG_DELWRI;
   1494   1.84  perseant 				}
   1495   1.84  perseant 			}
   1496   1.84  perseant 			if (pg->flags & PG_WANTED)
   1497   1.84  perseant 				wakeup(pg);
   1498   1.84  perseant 			pg->flags &= ~(PG_WANTED|PG_BUSY);
   1499   1.85      yamt 			UVM_PAGE_OWN(pg, NULL);
   1500   1.84  perseant 		}
   1501   1.84  perseant 
   1502  1.103  perseant 		if (checkfirst && any_dirty)
   1503  1.130      yamt 			break;
   1504  1.103  perseant 
   1505   1.84  perseant 		if (by_list) {
   1506   1.84  perseant 			curpg = TAILQ_NEXT(curpg, listq);
   1507   1.84  perseant 		} else {
   1508   1.84  perseant 			soff += MAX(PAGE_SIZE, fs->lfs_bsize);
   1509   1.84  perseant 		}
   1510   1.84  perseant 	}
   1511   1.84  perseant 
   1512   1.84  perseant 	/*
   1513   1.84  perseant 	 * If any pages were dirty, mark this inode as "pageout requested",
   1514   1.84  perseant 	 * and put it on the paging queue.
   1515   1.84  perseant 	 * XXXUBC locking (check locking on dchainhd too)
   1516   1.84  perseant 	 */
   1517   1.84  perseant #ifdef notyet
   1518   1.84  perseant 	if (any_dirty) {
   1519   1.84  perseant 		if (!(ip->i_flags & IN_PAGING)) {
   1520   1.84  perseant 			ip->i_flags |= IN_PAGING;
   1521   1.84  perseant 			TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain);
   1522   1.84  perseant 		}
   1523   1.84  perseant 	}
   1524   1.84  perseant #endif
   1525   1.84  perseant 	return any_dirty;
   1526   1.84  perseant }
   1527   1.84  perseant 
   1528   1.84  perseant /*
   1529   1.84  perseant  * lfs_putpages functions like genfs_putpages except that
   1530   1.84  perseant  *
   1531   1.84  perseant  * (1) It needs to bounds-check the incoming requests to ensure that
   1532   1.84  perseant  *     they are block-aligned; if they are not, expand the range and
   1533   1.84  perseant  *     do the right thing in case, e.g., the requested range is clean
   1534   1.84  perseant  *     but the expanded range is dirty.
   1535   1.84  perseant  * (2) It needs to explicitly send blocks to be written when it is done.
   1536   1.84  perseant  *     VOP_PUTPAGES is not ever called with the seglock held, so
   1537   1.84  perseant  *     we simply take the seglock and let lfs_segunlock wait for us.
   1538   1.84  perseant  *     XXX Actually we can be called with the seglock held, if we have
   1539   1.84  perseant  *     XXX to flush a vnode while lfs_markv is in operation.  As of this
   1540   1.84  perseant  *     XXX writing we panic in this case.
   1541   1.84  perseant  *
   1542   1.84  perseant  * Assumptions:
   1543   1.84  perseant  *
   1544   1.84  perseant  * (1) The caller does not hold any pages in this vnode busy.  If it does,
   1545   1.84  perseant  *     there is a danger that when we expand the page range and busy the
   1546   1.84  perseant  *     pages we will deadlock.
   1547   1.84  perseant  * (2) We are called with vp->v_interlock held; we must return with it
   1548   1.84  perseant  *     released.
   1549   1.84  perseant  * (3) We don't absolutely have to free pages right away, provided that
   1550   1.84  perseant  *     the request does not have PGO_SYNCIO.  When the pagedaemon gives
   1551   1.84  perseant  *     us a request with PGO_FREE, we take the pages out of the paging
   1552   1.84  perseant  *     queue and wake up the writer, which will handle freeing them for us.
   1553   1.84  perseant  *
   1554   1.84  perseant  *     We ensure that for any filesystem block, all pages for that
   1555   1.84  perseant  *     block are either resident or not, even if those pages are higher
   1556   1.84  perseant  *     than EOF; that means that we will be getting requests to free
   1557   1.84  perseant  *     "unused" pages above EOF all the time, and should ignore them.
   1558  1.115      yamt  *
   1559  1.115      yamt  * XXX note that we're (ab)using PGO_LOCKED as "seglock held".
   1560   1.84  perseant  */
   1561   1.84  perseant 
   1562   1.84  perseant int
   1563   1.84  perseant lfs_putpages(void *v)
   1564   1.84  perseant {
   1565   1.84  perseant 	int error;
   1566   1.84  perseant 	struct vop_putpages_args /* {
   1567   1.84  perseant 		struct vnode *a_vp;
   1568   1.84  perseant 		voff_t a_offlo;
   1569   1.84  perseant 		voff_t a_offhi;
   1570   1.84  perseant 		int a_flags;
   1571   1.84  perseant 	} */ *ap = v;
   1572   1.84  perseant 	struct vnode *vp;
   1573   1.84  perseant 	struct inode *ip;
   1574   1.84  perseant 	struct lfs *fs;
   1575   1.84  perseant 	struct segment *sp;
   1576   1.84  perseant 	off_t origoffset, startoffset, endoffset, origendoffset, blkeof;
   1577   1.95  perseant 	off_t off, max_endoffset;
   1578  1.126      yamt 	int s;
   1579  1.126      yamt 	boolean_t seglocked, sync, pagedaemon;
   1580   1.95  perseant 	struct vm_page *pg;
   1581   1.84  perseant 	UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist);
   1582   1.84  perseant 
   1583   1.84  perseant 	vp = ap->a_vp;
   1584   1.84  perseant 	ip = VTOI(vp);
   1585   1.84  perseant 	fs = ip->i_lfs;
   1586  1.126      yamt 	sync = (ap->a_flags & PGO_SYNCIO) != 0;
   1587   1.84  perseant 	pagedaemon = (curproc == uvm.pagedaemon_proc);
   1588   1.84  perseant 
   1589   1.84  perseant 	/* Putpages does nothing for metadata. */
   1590   1.84  perseant 	if (vp == fs->lfs_ivnode || vp->v_type != VREG) {
   1591   1.84  perseant 		simple_unlock(&vp->v_interlock);
   1592   1.84  perseant 		return 0;
   1593   1.84  perseant 	}
   1594   1.84  perseant 
   1595   1.84  perseant 	/*
   1596   1.84  perseant 	 * If there are no pages, don't do anything.
   1597   1.84  perseant 	 */
   1598   1.84  perseant 	if (vp->v_uobj.uo_npages == 0) {
   1599   1.84  perseant 		s = splbio();
   1600   1.84  perseant 		if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL &&
   1601   1.84  perseant 		    (vp->v_flag & VONWORKLST)) {
   1602   1.84  perseant 			vp->v_flag &= ~VONWORKLST;
   1603   1.84  perseant 			LIST_REMOVE(vp, v_synclist);
   1604   1.84  perseant 		}
   1605   1.84  perseant 		splx(s);
   1606   1.84  perseant 		simple_unlock(&vp->v_interlock);
   1607   1.84  perseant 		return 0;
   1608   1.84  perseant 	}
   1609   1.84  perseant 
   1610  1.102      fvdl 	blkeof = blkroundup(fs, ip->i_size);
   1611   1.84  perseant 
   1612   1.84  perseant 	/*
   1613   1.84  perseant 	 * Ignore requests to free pages past EOF but in the same block
   1614   1.84  perseant 	 * as EOF, unless the request is synchronous. (XXX why sync?)
   1615   1.84  perseant 	 * XXXUBC Make these pages look "active" so the pagedaemon won't
   1616   1.84  perseant 	 * XXXUBC bother us with them again.
   1617   1.84  perseant 	 */
   1618  1.102      fvdl 	if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) {
   1619   1.84  perseant 		origoffset = ap->a_offlo;
   1620   1.95  perseant 		for (off = origoffset; off < blkeof; off += fs->lfs_bsize) {
   1621   1.95  perseant 			pg = uvm_pagelookup(&vp->v_uobj, off);
   1622   1.95  perseant 			KASSERT(pg != NULL);
   1623   1.95  perseant 			while (pg->flags & PG_BUSY) {
   1624   1.95  perseant 				pg->flags |= PG_WANTED;
   1625   1.95  perseant 				UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0,
   1626   1.95  perseant 						    "lfsput2", 0);
   1627   1.95  perseant 				simple_lock(&vp->v_interlock);
   1628   1.95  perseant 			}
   1629   1.95  perseant 			uvm_lock_pageq();
   1630   1.95  perseant 			uvm_pageactivate(pg);
   1631   1.95  perseant 			uvm_unlock_pageq();
   1632   1.95  perseant 		}
   1633   1.84  perseant 		ap->a_offlo = blkeof;
   1634   1.84  perseant 		if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) {
   1635   1.84  perseant 			simple_unlock(&vp->v_interlock);
   1636   1.84  perseant 			return 0;
   1637   1.84  perseant 		}
   1638   1.84  perseant 	}
   1639   1.84  perseant 
   1640   1.84  perseant 	/*
   1641   1.84  perseant 	 * Extend page range to start and end at block boundaries.
   1642   1.84  perseant 	 * (For the purposes of VOP_PUTPAGES, fragments don't exist.)
   1643   1.84  perseant 	 */
   1644   1.86  perseant 	origoffset = ap->a_offlo;
   1645   1.84  perseant 	origendoffset = ap->a_offhi;
   1646   1.86  perseant 	startoffset = origoffset & ~(fs->lfs_bmask);
   1647   1.84  perseant 	max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift)
   1648   1.84  perseant 					       << fs->lfs_bshift;
   1649   1.84  perseant 
   1650   1.84  perseant 	if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) {
   1651   1.86  perseant 		endoffset = max_endoffset;
   1652   1.84  perseant 		origendoffset = endoffset;
   1653   1.86  perseant 	} else {
   1654   1.84  perseant 		origendoffset = round_page(ap->a_offhi);
   1655   1.84  perseant 		endoffset = round_page(blkroundup(fs, origendoffset));
   1656   1.84  perseant 	}
   1657   1.84  perseant 
   1658   1.84  perseant 	KASSERT(startoffset > 0 || endoffset >= startoffset);
   1659   1.84  perseant 	if (startoffset == endoffset) {
   1660   1.84  perseant 		/* Nothing to do, why were we called? */
   1661   1.84  perseant 		simple_unlock(&vp->v_interlock);
   1662   1.84  perseant #ifdef DEBUG
   1663   1.84  perseant 		printf("lfs_putpages: startoffset = endoffset = %" PRId64 "\n",
   1664   1.84  perseant 			startoffset);
   1665   1.84  perseant #endif
   1666   1.84  perseant 		return 0;
   1667   1.84  perseant 	}
   1668   1.84  perseant 
   1669   1.84  perseant 	ap->a_offlo = startoffset;
   1670   1.84  perseant 	ap->a_offhi = endoffset;
   1671   1.84  perseant 
   1672   1.84  perseant 	if (!(ap->a_flags & PGO_CLEANIT))
   1673   1.84  perseant 		return genfs_putpages(v);
   1674   1.84  perseant 
   1675   1.84  perseant 	/*
   1676  1.103  perseant 	 * If there are more than one page per block, we don't want
   1677  1.103  perseant 	 * to get caught locking them backwards; so set PGO_BUSYFAIL
   1678  1.103  perseant 	 * to avoid deadlocks.
   1679   1.84  perseant 	 */
   1680  1.103  perseant 	ap->a_flags |= PGO_BUSYFAIL;
   1681  1.103  perseant 
   1682  1.103  perseant 	do {
   1683  1.103  perseant 		int r;
   1684  1.103  perseant 
   1685  1.104      yamt 		/* If no pages are dirty, we can just use genfs_putpages. */
   1686  1.123    simonb 		if (check_dirty(fs, vp, startoffset, endoffset, blkeof,
   1687  1.123    simonb 				ap->a_flags, 1) != 0)
   1688  1.103  perseant 			break;
   1689  1.103  perseant 
   1690  1.134  perseant 		/*
   1691  1.134  perseant 		 * Sometimes pages are dirtied between the time that
   1692  1.134  perseant 		 * we check and the time we try to clean them.
   1693  1.134  perseant 		 * Instruct lfs_gop_write to return EDEADLK in this case
   1694  1.134  perseant 		 * so we can write them properly.
   1695  1.134  perseant 		 */
   1696  1.134  perseant 		ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE;
   1697  1.134  perseant 		r = genfs_putpages(v);
   1698  1.134  perseant 		ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE;
   1699  1.134  perseant 		if (r != EDEADLK)
   1700  1.103  perseant 			return r;
   1701  1.103  perseant 
   1702  1.103  perseant 		/* Start over. */
   1703  1.121      fvdl 		preempt(1);
   1704  1.103  perseant 		simple_lock(&vp->v_interlock);
   1705  1.103  perseant 	} while(1);
   1706   1.84  perseant 
   1707   1.84  perseant 	/*
   1708   1.84  perseant 	 * Dirty and asked to clean.
   1709   1.84  perseant 	 *
   1710   1.84  perseant 	 * Pagedaemon can't actually write LFS pages; wake up
   1711   1.84  perseant 	 * the writer to take care of that.  The writer will
   1712   1.84  perseant 	 * notice the pager inode queue and act on that.
   1713   1.84  perseant 	 */
   1714   1.84  perseant 	if (pagedaemon) {
   1715   1.84  perseant 		++fs->lfs_pdflush;
   1716   1.84  perseant 		wakeup(&lfs_writer_daemon);
   1717   1.87      yamt 		simple_unlock(&vp->v_interlock);
   1718   1.84  perseant 		return EWOULDBLOCK;
   1719   1.84  perseant 	}
   1720   1.84  perseant 
   1721   1.84  perseant 	/*
   1722   1.84  perseant 	 * If this is a file created in a recent dirop, we can't flush its
   1723   1.84  perseant 	 * inode until the dirop is complete.  Drain dirops, then flush the
   1724   1.84  perseant 	 * filesystem (taking care of any other pending dirops while we're
   1725   1.84  perseant 	 * at it).
   1726   1.84  perseant 	 */
   1727   1.84  perseant 	if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT &&
   1728   1.84  perseant 	    (vp->v_flag & VDIROP)) {
   1729   1.84  perseant 		int locked;
   1730   1.84  perseant 
   1731   1.84  perseant 		/* printf("putpages to clean VDIROP, flushing\n"); */
   1732  1.111      yamt 		lfs_writer_enter(fs, "ppdirop");
   1733   1.84  perseant 		locked = VOP_ISLOCKED(vp) && /* XXX */
   1734   1.84  perseant 			vp->v_lock.lk_lockholder == curproc->p_pid;
   1735   1.84  perseant 		if (locked)
   1736   1.84  perseant 			VOP_UNLOCK(vp, 0);
   1737   1.84  perseant 		simple_unlock(&vp->v_interlock);
   1738   1.84  perseant 
   1739   1.84  perseant 		lfs_flush_fs(fs, sync ? SEGM_SYNC : 0);
   1740   1.84  perseant 
   1741   1.84  perseant 		simple_lock(&vp->v_interlock);
   1742   1.84  perseant 		if (locked)
   1743   1.84  perseant 			VOP_LOCK(vp, LK_EXCLUSIVE);
   1744  1.111      yamt 		lfs_writer_leave(fs);
   1745   1.84  perseant 
   1746   1.84  perseant 		/* XXX the flush should have taken care of this one too! */
   1747   1.84  perseant 	}
   1748   1.84  perseant 
   1749   1.84  perseant 	/*
   1750   1.86  perseant 	 * This is it.	We are going to write some pages.  From here on
   1751   1.84  perseant 	 * down it's all just mechanics.
   1752   1.84  perseant 	 *
   1753  1.103  perseant 	 * Don't let genfs_putpages wait; lfs_segunlock will wait for us.
   1754   1.84  perseant 	 */
   1755   1.84  perseant 	ap->a_flags &= ~PGO_SYNCIO;
   1756   1.84  perseant 
   1757   1.84  perseant 	/*
   1758   1.84  perseant 	 * If we've already got the seglock, flush the node and return.
   1759   1.84  perseant 	 * The FIP has already been set up for us by lfs_writefile,
   1760   1.84  perseant 	 * and FIP cleanup and lfs_updatemeta will also be done there,
   1761   1.84  perseant 	 * unless genfs_putpages returns EDEADLK; then we must flush
   1762   1.84  perseant 	 * what we have, and correct FIP and segment header accounting.
   1763   1.84  perseant 	 */
   1764   1.84  perseant 
   1765  1.126      yamt 	seglocked = (ap->a_flags & PGO_LOCKED) != 0;
   1766  1.126      yamt 	if (!seglocked) {
   1767  1.126      yamt 		simple_unlock(&vp->v_interlock);
   1768  1.103  perseant 		/*
   1769  1.126      yamt 		 * Take the seglock, because we are going to be writing pages.
   1770  1.103  perseant 		 */
   1771  1.126      yamt 		error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0));
   1772  1.126      yamt 		if (error != 0)
   1773  1.126      yamt 			return error;
   1774  1.126      yamt 		simple_lock(&vp->v_interlock);
   1775   1.84  perseant 	}
   1776   1.84  perseant 
   1777   1.84  perseant 	/*
   1778   1.84  perseant 	 * VOP_PUTPAGES should not be called while holding the seglock.
   1779   1.93  perseant 	 * XXXUBC fix lfs_markv, or do this properly.
   1780   1.84  perseant 	 */
   1781   1.93  perseant 	/* KASSERT(fs->lfs_seglock == 1); */
   1782   1.84  perseant 
   1783   1.84  perseant 	/*
   1784   1.84  perseant 	 * We assume we're being called with sp->fip pointing at blank space.
   1785   1.84  perseant 	 * Account for a new FIP in the segment header, and set sp->vp.
   1786   1.84  perseant 	 * (This should duplicate the setup at the top of lfs_writefile().)
   1787   1.84  perseant 	 */
   1788   1.84  perseant 	sp = fs->lfs_sp;
   1789  1.126      yamt 	if (!seglocked) {
   1790  1.126      yamt 		if (sp->seg_bytes_left < fs->lfs_bsize ||
   1791  1.126      yamt 		    sp->sum_bytes_left < sizeof(struct finfo))
   1792  1.126      yamt 			(void) lfs_writeseg(fs, fs->lfs_sp);
   1793  1.126      yamt 
   1794  1.126      yamt 		sp->sum_bytes_left -= FINFOSIZE;
   1795  1.126      yamt 		++((SEGSUM *)(sp->segsum))->ss_nfinfo;
   1796  1.126      yamt 	}
   1797  1.120      yamt 	KASSERT(sp->vp == NULL);
   1798   1.84  perseant 	sp->vp = vp;
   1799   1.84  perseant 
   1800  1.126      yamt 	if (!seglocked) {
   1801  1.126      yamt 		if (vp->v_flag & VDIROP)
   1802  1.126      yamt 			((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT);
   1803  1.126      yamt 	}
   1804   1.84  perseant 
   1805   1.86  perseant 	sp->fip->fi_nblocks = 0;
   1806   1.86  perseant 	sp->fip->fi_ino = ip->i_number;
   1807  1.102      fvdl 	sp->fip->fi_version = ip->i_gen;
   1808   1.84  perseant 
   1809   1.84  perseant 	/*
   1810   1.84  perseant 	 * Loop through genfs_putpages until all pages are gathered.
   1811   1.88  perseant 	 * genfs_putpages() drops the interlock, so reacquire it if necessary.
   1812  1.103  perseant 	 * Whenever we lose the interlock we have to rerun check_dirty, as
   1813  1.103  perseant 	 * well.
   1814   1.84  perseant 	 */
   1815  1.126      yamt again:
   1816  1.105  perseant 	check_dirty(fs, vp, startoffset, endoffset, blkeof, ap->a_flags, 0);
   1817  1.103  perseant 
   1818  1.103  perseant 	if ((error = genfs_putpages(v)) == EDEADLK) {
   1819   1.84  perseant #ifdef DEBUG_LFS
   1820   1.84  perseant 		printf("lfs_putpages: genfs_putpages returned EDEADLK [2]"
   1821   1.84  perseant 		       " ino %d off %x (seg %d)\n",
   1822   1.84  perseant 		       ip->i_number, fs->lfs_offset,
   1823   1.84  perseant 		       dtosn(fs, fs->lfs_offset));
   1824   1.84  perseant #endif
   1825   1.88  perseant 		/* If nothing to write, short-circuit */
   1826  1.129      yamt 		if (sp->cbpp - sp->bpp > 1) {
   1827  1.129      yamt 			/* Write gathered pages */
   1828  1.129      yamt 			lfs_updatemeta(sp);
   1829  1.129      yamt 			(void) lfs_writeseg(fs, sp);
   1830  1.129      yamt 
   1831  1.129      yamt 			/*
   1832  1.129      yamt 			 * Reinitialize brand new FIP and add us to it.
   1833  1.129      yamt 			 * (This should duplicate the fixup in
   1834  1.129      yamt 			 * lfs_gatherpages().)
   1835  1.129      yamt 			 */
   1836  1.129      yamt 			KASSERT(sp->vp == vp);
   1837  1.129      yamt 			sp->fip->fi_version = ip->i_gen;
   1838  1.129      yamt 			sp->fip->fi_ino = ip->i_number;
   1839  1.129      yamt 			/* Add us to the new segment summary. */
   1840  1.129      yamt 			++((SEGSUM *)(sp->segsum))->ss_nfinfo;
   1841  1.129      yamt 			sp->sum_bytes_left -= FINFOSIZE;
   1842   1.88  perseant 		}
   1843   1.84  perseant 
   1844   1.84  perseant 		/* Give the write a chance to complete */
   1845  1.121      fvdl 		preempt(1);
   1846  1.103  perseant 
   1847  1.103  perseant 		/* We've lost the interlock.  Start over. */
   1848  1.126      yamt 		simple_lock(&vp->v_interlock);
   1849  1.126      yamt 		goto again;
   1850   1.84  perseant 	}
   1851  1.103  perseant 
   1852  1.120      yamt 	KASSERT(sp->vp == vp);
   1853  1.126      yamt 	if (!seglocked) {
   1854  1.126      yamt 		sp->vp = NULL; /* XXX lfs_gather below will set this */
   1855  1.126      yamt 
   1856  1.126      yamt 		/* Write indirect blocks as well */
   1857  1.126      yamt 		lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir);
   1858  1.126      yamt 		lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir);
   1859  1.126      yamt 		lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir);
   1860  1.120      yamt 
   1861  1.126      yamt 		KASSERT(sp->vp == NULL);
   1862  1.126      yamt 		sp->vp = vp;
   1863  1.126      yamt 	}
   1864   1.84  perseant 
   1865   1.84  perseant 	/*
   1866   1.84  perseant 	 * Blocks are now gathered into a segment waiting to be written.
   1867   1.84  perseant 	 * All that's left to do is update metadata, and write them.
   1868   1.84  perseant 	 */
   1869  1.120      yamt 	lfs_updatemeta(sp);
   1870  1.120      yamt 	KASSERT(sp->vp == vp);
   1871  1.120      yamt 	sp->vp = NULL;
   1872  1.126      yamt 
   1873  1.126      yamt 	if (seglocked) {
   1874  1.126      yamt 		/* we're called by lfs_writefile. */
   1875  1.126      yamt 		return error;
   1876  1.126      yamt 	}
   1877  1.120      yamt 
   1878   1.84  perseant 	/*
   1879   1.88  perseant 	 * Clean up FIP, since we're done writing this file.
   1880   1.88  perseant 	 * This should duplicate cleanup at the end of lfs_writefile().
   1881   1.84  perseant 	 */
   1882   1.86  perseant 	if (sp->fip->fi_nblocks != 0) {
   1883  1.124      yamt 		sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE +
   1884  1.124      yamt 			sizeof(int32_t) * sp->fip->fi_nblocks);
   1885   1.86  perseant 		sp->start_lbp = &sp->fip->fi_blocks[0];
   1886   1.86  perseant 	} else {
   1887  1.124      yamt 		sp->sum_bytes_left += FINFOSIZE;
   1888   1.86  perseant 		--((SEGSUM *)(sp->segsum))->ss_nfinfo;
   1889   1.86  perseant 	}
   1890   1.88  perseant 	lfs_writeseg(fs, fs->lfs_sp);
   1891   1.88  perseant 
   1892   1.84  perseant 	/*
   1893   1.84  perseant 	 * XXX - with the malloc/copy writeseg, the pages are freed by now
   1894   1.84  perseant 	 * even if we don't wait (e.g. if we hold a nested lock).  This
   1895   1.84  perseant 	 * will not be true if we stop using malloc/copy.
   1896   1.84  perseant 	 */
   1897   1.84  perseant 	KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT);
   1898   1.84  perseant 	lfs_segunlock(fs);
   1899   1.84  perseant 
   1900   1.84  perseant 	/*
   1901   1.84  perseant 	 * Wait for v_numoutput to drop to zero.  The seglock should
   1902   1.84  perseant 	 * take care of this, but there is a slight possibility that
   1903   1.84  perseant 	 * aiodoned might not have got around to our buffers yet.
   1904   1.84  perseant 	 */
   1905   1.84  perseant 	if (sync) {
   1906   1.84  perseant 		int s;
   1907   1.84  perseant 
   1908   1.84  perseant 		s = splbio();
   1909   1.84  perseant 		simple_lock(&global_v_numoutput_slock);
   1910   1.98  perseant 		while (vp->v_numoutput > 0) {
   1911   1.84  perseant #ifdef DEBUG
   1912   1.84  perseant 			printf("ino %d sleeping on num %d\n",
   1913   1.84  perseant 				ip->i_number, vp->v_numoutput);
   1914   1.84  perseant #endif
   1915   1.84  perseant 			vp->v_flag |= VBWAIT;
   1916   1.87      yamt 			ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0,
   1917   1.87      yamt 			    &global_v_numoutput_slock);
   1918   1.84  perseant 		}
   1919   1.84  perseant 		simple_unlock(&global_v_numoutput_slock);
   1920   1.84  perseant 		splx(s);
   1921   1.84  perseant 	}
   1922   1.84  perseant 	return error;
   1923   1.84  perseant }
   1924   1.84  perseant 
   1925   1.84  perseant /*
   1926   1.84  perseant  * Return the last logical file offset that should be written for this file
   1927   1.86  perseant  * if we're doing a write that ends at "size".	If writing, we need to know
   1928   1.84  perseant  * about sizes on disk, i.e. fragments if there are any; if reading, we need
   1929   1.84  perseant  * to know about entire blocks.
   1930   1.84  perseant  */
   1931   1.84  perseant void
   1932   1.84  perseant lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags)
   1933   1.84  perseant {
   1934   1.84  perseant 	struct inode *ip = VTOI(vp);
   1935   1.84  perseant 	struct lfs *fs = ip->i_lfs;
   1936   1.84  perseant 	daddr_t olbn, nlbn;
   1937   1.84  perseant 
   1938   1.84  perseant 	KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE));
   1939   1.84  perseant 	KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE))
   1940   1.84  perseant 		!= (GOP_SIZE_READ | GOP_SIZE_WRITE));
   1941   1.84  perseant 
   1942  1.102      fvdl 	olbn = lblkno(fs, ip->i_size);
   1943   1.84  perseant 	nlbn = lblkno(fs, size);
   1944  1.118      yamt 	if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) {
   1945   1.86  perseant 		*eobp = fragroundup(fs, size);
   1946   1.86  perseant 	} else {
   1947   1.86  perseant 		*eobp = blkroundup(fs, size);
   1948   1.86  perseant 	}
   1949   1.84  perseant }
   1950   1.84  perseant 
   1951   1.84  perseant #ifdef DEBUG
   1952   1.84  perseant void lfs_dump_vop(void *);
   1953   1.84  perseant 
   1954   1.84  perseant void
   1955   1.84  perseant lfs_dump_vop(void *v)
   1956   1.84  perseant {
   1957   1.86  perseant 	struct vop_putpages_args /* {
   1958   1.86  perseant 		struct vnode *a_vp;
   1959   1.86  perseant 		voff_t a_offlo;
   1960   1.86  perseant 		voff_t a_offhi;
   1961   1.86  perseant 		int a_flags;
   1962   1.86  perseant 	} */ *ap = v;
   1963   1.84  perseant 
   1964  1.106     ragge #ifdef DDB
   1965   1.84  perseant 	vfs_vnode_print(ap->a_vp, 0, printf);
   1966  1.106     ragge #endif
   1967  1.102      fvdl 	lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din);
   1968   1.84  perseant }
   1969   1.84  perseant #endif
   1970   1.84  perseant 
   1971   1.84  perseant int
   1972   1.84  perseant lfs_mmap(void *v)
   1973   1.84  perseant {
   1974   1.84  perseant 	struct vop_mmap_args /* {
   1975   1.86  perseant 		const struct vnodeop_desc *a_desc;
   1976   1.86  perseant 		struct vnode *a_vp;
   1977   1.86  perseant 		int a_fflags;
   1978   1.86  perseant 		struct ucred *a_cred;
   1979  1.109      fvdl 		struct proc *a_p;
   1980   1.84  perseant 	} */ *ap = v;
   1981   1.84  perseant 
   1982   1.84  perseant 	if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM)
   1983   1.84  perseant 		return EOPNOTSUPP;
   1984   1.84  perseant 	return ufs_mmap(v);
   1985   1.84  perseant }
   1986