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