xref: /netbsd-src/sys/fs/puffs/puffs_vnops.c (revision cef8759bd76c1b621f8eab8faa6f208faabc2e15)
1 /*	$NetBSD: puffs_vnops.c,v 1.217 2020/05/16 18:31:49 christos Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006, 2007  Antti Kantee.  All Rights Reserved.
5  *
6  * Development of this software was supported by the
7  * Google Summer of Code program and the Ulla Tuominen Foundation.
8  * The Google SoC project was mentored by Bill Studenmund.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
20  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
21  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
25  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: puffs_vnops.c,v 1.217 2020/05/16 18:31:49 christos Exp $");
34 
35 #include <sys/param.h>
36 #include <sys/buf.h>
37 #include <sys/lockf.h>
38 #include <sys/malloc.h>
39 #include <sys/mount.h>
40 #include <sys/namei.h>
41 #include <sys/vnode.h>
42 #include <sys/proc.h>
43 #include <sys/kernel.h> /* For hz, getticks() */
44 
45 #include <uvm/uvm.h>
46 
47 #include <fs/puffs/puffs_msgif.h>
48 #include <fs/puffs/puffs_sys.h>
49 
50 #include <miscfs/fifofs/fifo.h>
51 #include <miscfs/genfs/genfs.h>
52 #include <miscfs/specfs/specdev.h>
53 
54 int	puffs_vnop_lookup(void *);
55 int	puffs_vnop_create(void *);
56 int	puffs_vnop_access(void *);
57 int	puffs_vnop_mknod(void *);
58 int	puffs_vnop_open(void *);
59 int	puffs_vnop_close(void *);
60 int	puffs_vnop_getattr(void *);
61 int	puffs_vnop_setattr(void *);
62 int	puffs_vnop_reclaim(void *);
63 int	puffs_vnop_readdir(void *);
64 int	puffs_vnop_poll(void *);
65 int	puffs_vnop_fsync(void *);
66 int	puffs_vnop_seek(void *);
67 int	puffs_vnop_remove(void *);
68 int	puffs_vnop_mkdir(void *);
69 int	puffs_vnop_rmdir(void *);
70 int	puffs_vnop_link(void *);
71 int	puffs_vnop_readlink(void *);
72 int	puffs_vnop_symlink(void *);
73 int	puffs_vnop_rename(void *);
74 int	puffs_vnop_read(void *);
75 int	puffs_vnop_write(void *);
76 int	puffs_vnop_fallocate(void *);
77 int	puffs_vnop_fdiscard(void *);
78 int	puffs_vnop_fcntl(void *);
79 int	puffs_vnop_ioctl(void *);
80 int	puffs_vnop_inactive(void *);
81 int	puffs_vnop_print(void *);
82 int	puffs_vnop_pathconf(void *);
83 int	puffs_vnop_advlock(void *);
84 int	puffs_vnop_strategy(void *);
85 int	puffs_vnop_bmap(void *);
86 int	puffs_vnop_mmap(void *);
87 int	puffs_vnop_getpages(void *);
88 int	puffs_vnop_abortop(void *);
89 int	puffs_vnop_getextattr(void *);
90 int	puffs_vnop_setextattr(void *);
91 int	puffs_vnop_listextattr(void *);
92 int	puffs_vnop_deleteextattr(void *);
93 
94 int	puffs_vnop_spec_read(void *);
95 int	puffs_vnop_spec_write(void *);
96 int	puffs_vnop_fifo_read(void *);
97 int	puffs_vnop_fifo_write(void *);
98 
99 int	puffs_vnop_checkop(void *);
100 
101 #define puffs_vnop_lock genfs_lock
102 #define puffs_vnop_unlock genfs_unlock
103 #define puffs_vnop_islocked genfs_islocked
104 
105 int (**puffs_vnodeop_p)(void *);
106 const struct vnodeopv_entry_desc puffs_vnodeop_entries[] = {
107 	{ &vop_default_desc, vn_default_error },
108 	{ &vop_lookup_desc, puffs_vnop_lookup },	/* REAL lookup */
109 	{ &vop_create_desc, puffs_vnop_checkop },	/* create */
110         { &vop_mknod_desc, puffs_vnop_checkop },	/* mknod */
111         { &vop_open_desc, puffs_vnop_open },		/* REAL open */
112         { &vop_close_desc, puffs_vnop_checkop },	/* close */
113         { &vop_access_desc, puffs_vnop_access },	/* REAL access */
114         { &vop_accessx_desc, genfs_accessx },		/* accessx */
115         { &vop_getattr_desc, puffs_vnop_checkop },	/* getattr */
116         { &vop_setattr_desc, puffs_vnop_checkop },	/* setattr */
117         { &vop_read_desc, puffs_vnop_checkop },		/* read */
118         { &vop_write_desc, puffs_vnop_checkop },	/* write */
119 	{ &vop_fallocate_desc, puffs_vnop_fallocate },	/* fallocate */
120 	{ &vop_fdiscard_desc, puffs_vnop_fdiscard },	/* fdiscard */
121         { &vop_fsync_desc, puffs_vnop_fsync },		/* REAL fsync */
122         { &vop_seek_desc, puffs_vnop_checkop },		/* seek */
123         { &vop_remove_desc, puffs_vnop_checkop },	/* remove */
124         { &vop_link_desc, puffs_vnop_checkop },		/* link */
125         { &vop_rename_desc, puffs_vnop_checkop },	/* rename */
126         { &vop_mkdir_desc, puffs_vnop_checkop },	/* mkdir */
127         { &vop_rmdir_desc, puffs_vnop_checkop },	/* rmdir */
128         { &vop_symlink_desc, puffs_vnop_checkop },	/* symlink */
129         { &vop_readdir_desc, puffs_vnop_checkop },	/* readdir */
130         { &vop_readlink_desc, puffs_vnop_checkop },	/* readlink */
131         { &vop_getpages_desc, puffs_vnop_checkop },	/* getpages */
132         { &vop_putpages_desc, genfs_putpages },		/* REAL putpages */
133         { &vop_pathconf_desc, puffs_vnop_checkop },	/* pathconf */
134         { &vop_advlock_desc, puffs_vnop_advlock },	/* advlock */
135         { &vop_strategy_desc, puffs_vnop_strategy },	/* REAL strategy */
136         { &vop_revoke_desc, genfs_revoke },		/* REAL revoke */
137         { &vop_abortop_desc, puffs_vnop_abortop },	/* REAL abortop */
138         { &vop_inactive_desc, puffs_vnop_inactive },	/* REAL inactive */
139         { &vop_reclaim_desc, puffs_vnop_reclaim },	/* REAL reclaim */
140         { &vop_lock_desc, puffs_vnop_lock },		/* REAL lock */
141         { &vop_unlock_desc, puffs_vnop_unlock },	/* REAL unlock */
142         { &vop_bmap_desc, puffs_vnop_bmap },		/* REAL bmap */
143         { &vop_print_desc, puffs_vnop_print },		/* REAL print */
144         { &vop_islocked_desc, puffs_vnop_islocked },	/* REAL islocked */
145         { &vop_bwrite_desc, genfs_nullop },		/* REAL bwrite */
146         { &vop_mmap_desc, puffs_vnop_mmap },		/* REAL mmap */
147         { &vop_poll_desc, puffs_vnop_poll },		/* REAL poll */
148 	{ &vop_getextattr_desc, puffs_vnop_getextattr },	/* getextattr */
149 	{ &vop_setextattr_desc, puffs_vnop_setextattr },	/* setextattr */
150 	{ &vop_listextattr_desc, puffs_vnop_listextattr },	/* listextattr */
151 	{ &vop_deleteextattr_desc, puffs_vnop_deleteextattr },/* deleteextattr */
152 #if 0
153 	{ &vop_openextattr_desc, puffs_vnop_checkop },	/* openextattr */
154 	{ &vop_closeextattr_desc, puffs_vnop_checkop },	/* closeextattr */
155 #endif
156         { &vop_kqfilter_desc, genfs_eopnotsupp },	/* kqfilter XXX */
157 	{ NULL, NULL }
158 };
159 const struct vnodeopv_desc puffs_vnodeop_opv_desc =
160 	{ &puffs_vnodeop_p, puffs_vnodeop_entries };
161 
162 
163 int (**puffs_specop_p)(void *);
164 const struct vnodeopv_entry_desc puffs_specop_entries[] = {
165 	{ &vop_default_desc, vn_default_error },
166 	{ &vop_lookup_desc, spec_lookup },		/* lookup, ENOTDIR */
167 	{ &vop_create_desc, spec_create },		/* genfs_badop */
168 	{ &vop_mknod_desc, spec_mknod },		/* genfs_badop */
169 	{ &vop_open_desc, spec_open },			/* spec_open */
170 	{ &vop_close_desc, spec_close },		/* spec_close */
171 	{ &vop_access_desc, puffs_vnop_checkop },	/* access */
172 	{ &vop_accessx_desc, genfs_accessx },		/* accessx */
173 	{ &vop_getattr_desc, puffs_vnop_checkop },	/* getattr */
174 	{ &vop_setattr_desc, puffs_vnop_checkop },	/* setattr */
175 	{ &vop_read_desc, puffs_vnop_spec_read },	/* update, read */
176 	{ &vop_write_desc, puffs_vnop_spec_write },	/* update, write */
177 	{ &vop_fallocate_desc, spec_fallocate },	/* fallocate */
178 	{ &vop_fdiscard_desc, spec_fdiscard },		/* fdiscard */
179 	{ &vop_ioctl_desc, spec_ioctl },		/* spec_ioctl */
180 	{ &vop_fcntl_desc, genfs_fcntl },		/* dummy */
181 	{ &vop_poll_desc, spec_poll },			/* spec_poll */
182 	{ &vop_kqfilter_desc, spec_kqfilter },		/* spec_kqfilter */
183 	{ &vop_revoke_desc, spec_revoke },		/* genfs_revoke */
184 	{ &vop_mmap_desc, spec_mmap },			/* spec_mmap */
185 	{ &vop_fsync_desc, spec_fsync },		/* vflushbuf */
186 	{ &vop_seek_desc, spec_seek },			/* genfs_nullop */
187 	{ &vop_remove_desc, spec_remove },		/* genfs_badop */
188 	{ &vop_link_desc, spec_link },			/* genfs_badop */
189 	{ &vop_rename_desc, spec_rename },		/* genfs_badop */
190 	{ &vop_mkdir_desc, spec_mkdir },		/* genfs_badop */
191 	{ &vop_rmdir_desc, spec_rmdir },		/* genfs_badop */
192 	{ &vop_symlink_desc, spec_symlink },		/* genfs_badop */
193 	{ &vop_readdir_desc, spec_readdir },		/* genfs_badop */
194 	{ &vop_readlink_desc, spec_readlink },		/* genfs_badop */
195 	{ &vop_abortop_desc, spec_abortop },		/* genfs_badop */
196 	{ &vop_inactive_desc, puffs_vnop_inactive },	/* REAL inactive */
197 	{ &vop_reclaim_desc, puffs_vnop_reclaim },	/* REAL reclaim */
198 	{ &vop_lock_desc, puffs_vnop_lock },		/* REAL lock */
199 	{ &vop_unlock_desc, puffs_vnop_unlock },	/* REAL unlock */
200 	{ &vop_bmap_desc, spec_bmap },			/* dummy */
201 	{ &vop_strategy_desc, spec_strategy },		/* dev strategy */
202 	{ &vop_print_desc, puffs_vnop_print },		/* REAL print */
203 	{ &vop_islocked_desc, puffs_vnop_islocked },	/* REAL islocked */
204 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
205 	{ &vop_advlock_desc, spec_advlock },		/* lf_advlock */
206 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
207 	{ &vop_getpages_desc, spec_getpages },		/* genfs_getpages */
208 	{ &vop_putpages_desc, spec_putpages },		/* genfs_putpages */
209 	{ &vop_getextattr_desc, puffs_vnop_checkop },	/* getextattr */
210 	{ &vop_setextattr_desc, puffs_vnop_checkop },	/* setextattr */
211 	{ &vop_listextattr_desc, puffs_vnop_checkop },	/* listextattr */
212 	{ &vop_deleteextattr_desc, puffs_vnop_checkop },/* deleteextattr */
213 #if 0
214 	{ &vop_openextattr_desc, _openextattr },	/* openextattr */
215 	{ &vop_closeextattr_desc, _closeextattr },	/* closeextattr */
216 #endif
217 	{ NULL, NULL }
218 };
219 const struct vnodeopv_desc puffs_specop_opv_desc =
220 	{ &puffs_specop_p, puffs_specop_entries };
221 
222 
223 int (**puffs_fifoop_p)(void *);
224 const struct vnodeopv_entry_desc puffs_fifoop_entries[] = {
225 	{ &vop_default_desc, vn_default_error },
226 	{ &vop_lookup_desc, vn_fifo_bypass },		/* lookup, ENOTDIR */
227 	{ &vop_create_desc, vn_fifo_bypass },		/* genfs_badop */
228 	{ &vop_mknod_desc, vn_fifo_bypass },		/* genfs_badop */
229 	{ &vop_open_desc, vn_fifo_bypass },		/* open */
230 	{ &vop_close_desc, vn_fifo_bypass },		/* close */
231 	{ &vop_access_desc, puffs_vnop_checkop },	/* access */
232 	{ &vop_accessx_desc, genfs_accessx },		/* accessx */
233 	{ &vop_getattr_desc, puffs_vnop_checkop },	/* getattr */
234 	{ &vop_setattr_desc, puffs_vnop_checkop },	/* setattr */
235 	{ &vop_read_desc, puffs_vnop_fifo_read },	/* read, update */
236 	{ &vop_write_desc, puffs_vnop_fifo_write },	/* write, update */
237 	{ &vop_fallocate_desc, vn_fifo_bypass },	/* fallocate */
238 	{ &vop_fdiscard_desc, vn_fifo_bypass },		/* fdiscard */
239 	{ &vop_ioctl_desc, vn_fifo_bypass },		/* ioctl */
240 	{ &vop_fcntl_desc, genfs_fcntl },		/* dummy */
241 	{ &vop_poll_desc, vn_fifo_bypass },		/* poll */
242 	{ &vop_kqfilter_desc, vn_fifo_bypass },		/* kqfilter */
243 	{ &vop_revoke_desc, vn_fifo_bypass },		/* genfs_revoke */
244 	{ &vop_mmap_desc, vn_fifo_bypass },		/* genfs_badop */
245 	{ &vop_fsync_desc, vn_fifo_bypass },		/* genfs_nullop*/
246 	{ &vop_seek_desc, vn_fifo_bypass },		/* genfs_badop */
247 	{ &vop_remove_desc, vn_fifo_bypass },		/* genfs_badop */
248 	{ &vop_link_desc, vn_fifo_bypass },		/* genfs_badop */
249 	{ &vop_rename_desc, vn_fifo_bypass },		/* genfs_badop */
250 	{ &vop_mkdir_desc, vn_fifo_bypass },		/* genfs_badop */
251 	{ &vop_rmdir_desc, vn_fifo_bypass },		/* genfs_badop */
252 	{ &vop_symlink_desc, vn_fifo_bypass },		/* genfs_badop */
253 	{ &vop_readdir_desc, vn_fifo_bypass },		/* genfs_badop */
254 	{ &vop_readlink_desc, vn_fifo_bypass },		/* genfs_badop */
255 	{ &vop_abortop_desc, vn_fifo_bypass },		/* genfs_badop */
256 	{ &vop_inactive_desc, puffs_vnop_inactive },	/* REAL inactive */
257 	{ &vop_reclaim_desc, puffs_vnop_reclaim },	/* REAL reclaim */
258 	{ &vop_lock_desc, puffs_vnop_lock },		/* REAL lock */
259 	{ &vop_unlock_desc, puffs_vnop_unlock },	/* REAL unlock */
260 	{ &vop_bmap_desc, vn_fifo_bypass },		/* dummy */
261 	{ &vop_strategy_desc, vn_fifo_bypass },		/* genfs_badop */
262 	{ &vop_print_desc, puffs_vnop_print },		/* REAL print */
263 	{ &vop_islocked_desc, puffs_vnop_islocked },	/* REAL islocked */
264 	{ &vop_pathconf_desc, vn_fifo_bypass },		/* pathconf */
265 	{ &vop_advlock_desc, vn_fifo_bypass },		/* genfs_einval */
266 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
267 	{ &vop_putpages_desc, vn_fifo_bypass }, 	/* genfs_null_putpages*/
268 #if 0
269 	{ &vop_openextattr_desc, _openextattr },	/* openextattr */
270 	{ &vop_closeextattr_desc, _closeextattr },	/* closeextattr */
271 #endif
272 	{ &vop_getextattr_desc, puffs_vnop_checkop },		/* getextattr */
273 	{ &vop_setextattr_desc, puffs_vnop_checkop },		/* setextattr */
274 	{ &vop_listextattr_desc, puffs_vnop_checkop },	/* listextattr */
275 	{ &vop_deleteextattr_desc, puffs_vnop_checkop },	/* deleteextattr */
276 	{ NULL, NULL }
277 };
278 const struct vnodeopv_desc puffs_fifoop_opv_desc =
279 	{ &puffs_fifoop_p, puffs_fifoop_entries };
280 
281 
282 /* "real" vnode operations */
283 int (**puffs_msgop_p)(void *);
284 const struct vnodeopv_entry_desc puffs_msgop_entries[] = {
285 	{ &vop_default_desc, vn_default_error },
286 	{ &vop_create_desc, puffs_vnop_create },	/* create */
287         { &vop_mknod_desc, puffs_vnop_mknod },		/* mknod */
288         { &vop_open_desc, puffs_vnop_open },		/* open */
289         { &vop_close_desc, puffs_vnop_close },		/* close */
290         { &vop_access_desc, puffs_vnop_access },	/* access */
291         { &vop_accessx_desc, genfs_accessx },		/* accessx */
292         { &vop_getattr_desc, puffs_vnop_getattr },	/* getattr */
293         { &vop_setattr_desc, puffs_vnop_setattr },	/* setattr */
294         { &vop_read_desc, puffs_vnop_read },		/* read */
295         { &vop_write_desc, puffs_vnop_write },		/* write */
296         { &vop_seek_desc, puffs_vnop_seek },		/* seek */
297         { &vop_remove_desc, puffs_vnop_remove },	/* remove */
298         { &vop_link_desc, puffs_vnop_link },		/* link */
299         { &vop_rename_desc, puffs_vnop_rename },	/* rename */
300         { &vop_mkdir_desc, puffs_vnop_mkdir },		/* mkdir */
301         { &vop_rmdir_desc, puffs_vnop_rmdir },		/* rmdir */
302         { &vop_symlink_desc, puffs_vnop_symlink },	/* symlink */
303         { &vop_readdir_desc, puffs_vnop_readdir },	/* readdir */
304         { &vop_readlink_desc, puffs_vnop_readlink },	/* readlink */
305         { &vop_print_desc, puffs_vnop_print },		/* print */
306         { &vop_islocked_desc, puffs_vnop_islocked },	/* islocked */
307         { &vop_pathconf_desc, puffs_vnop_pathconf },	/* pathconf */
308         { &vop_getpages_desc, puffs_vnop_getpages },	/* getpages */
309 	{ NULL, NULL }
310 };
311 const struct vnodeopv_desc puffs_msgop_opv_desc =
312 	{ &puffs_msgop_p, puffs_msgop_entries };
313 
314 /*
315  * for dosetattr / update_va
316  */
317 #define SETATTR_CHSIZE	0x01
318 #define SETATTR_ASYNC	0x02
319 
320 #define ERROUT(err)							\
321 do {									\
322 	error = err;							\
323 	goto out;							\
324 } while (/*CONSTCOND*/0)
325 
326 /*
327  * This is a generic vnode operation handler.  It checks if the necessary
328  * operations for the called vnode operation are implemented by userspace
329  * and either returns a dummy return value or proceeds to call the real
330  * vnode operation from puffs_msgop_v.
331  *
332  * XXX: this should described elsewhere and autogenerated, the complexity
333  * of the vnode operations vectors and their interrelationships is also
334  * getting a bit out of hand.  Another problem is that we need this same
335  * information in the fs server code, so keeping the two in sync manually
336  * is not a viable (long term) plan.
337  */
338 
339 /* not supported, handle locking protocol */
340 #define CHECKOP_NOTSUPP(op)						\
341 case VOP_##op##_DESCOFFSET:						\
342 	if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0)			\
343 		return genfs_eopnotsupp(v);				\
344 	break
345 
346 /* always succeed, no locking */
347 #define CHECKOP_SUCCESS(op)						\
348 case VOP_##op##_DESCOFFSET:						\
349 	if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0)			\
350 		return 0;						\
351 	break
352 
353 int
354 puffs_vnop_checkop(void *v)
355 {
356 	struct vop_generic_args /* {
357 		struct vnodeop_desc *a_desc;
358 		spooky mystery contents;
359 	} */ *ap = v;
360 	struct vnodeop_desc *desc = ap->a_desc;
361 	struct puffs_mount *pmp;
362 	struct vnode *vp;
363 	int offset, rv;
364 
365 	offset = ap->a_desc->vdesc_vp_offsets[0];
366 #ifdef DIAGNOSTIC
367 	if (offset == VDESC_NO_OFFSET)
368 		panic("puffs_checkop: no vnode, why did you call me?");
369 #endif
370 	vp = *VOPARG_OFFSETTO(struct vnode **, offset, ap);
371 	pmp = MPTOPUFFSMP(vp->v_mount);
372 
373 	DPRINTF_VERBOSE(("checkop call %s (%d), vp %p\n",
374 	    ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp));
375 
376 	if (!ALLOPS(pmp)) {
377 		switch (desc->vdesc_offset) {
378 			CHECKOP_NOTSUPP(CREATE);
379 			CHECKOP_NOTSUPP(MKNOD);
380 			CHECKOP_NOTSUPP(GETATTR);
381 			CHECKOP_NOTSUPP(SETATTR);
382 			CHECKOP_NOTSUPP(READ);
383 			CHECKOP_NOTSUPP(WRITE);
384 			CHECKOP_NOTSUPP(FCNTL);
385 			CHECKOP_NOTSUPP(IOCTL);
386 			CHECKOP_NOTSUPP(REMOVE);
387 			CHECKOP_NOTSUPP(LINK);
388 			CHECKOP_NOTSUPP(RENAME);
389 			CHECKOP_NOTSUPP(MKDIR);
390 			CHECKOP_NOTSUPP(RMDIR);
391 			CHECKOP_NOTSUPP(SYMLINK);
392 			CHECKOP_NOTSUPP(READDIR);
393 			CHECKOP_NOTSUPP(READLINK);
394 			CHECKOP_NOTSUPP(PRINT);
395 			CHECKOP_NOTSUPP(PATHCONF);
396 			CHECKOP_NOTSUPP(GETEXTATTR);
397 			CHECKOP_NOTSUPP(SETEXTATTR);
398 			CHECKOP_NOTSUPP(LISTEXTATTR);
399 			CHECKOP_NOTSUPP(DELETEEXTATTR);
400 
401 			CHECKOP_SUCCESS(ACCESS);
402 			CHECKOP_SUCCESS(CLOSE);
403 			CHECKOP_SUCCESS(SEEK);
404 
405 		case VOP_GETPAGES_DESCOFFSET:
406 			if (!EXISTSOP(pmp, READ))
407 				return genfs_eopnotsupp(v);
408 			break;
409 
410 		default:
411 			panic("puffs_checkop: unhandled vnop %d",
412 			    desc->vdesc_offset);
413 		}
414 	}
415 
416 	rv = VOCALL(puffs_msgop_p, ap->a_desc->vdesc_offset, v);
417 
418 	DPRINTF_VERBOSE(("checkop return %s (%d), vp %p: %d\n",
419 	    ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp, rv));
420 
421 	return rv;
422 }
423 
424 static int callremove(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
425 			    struct componentname *);
426 static int callrmdir(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
427 			   struct componentname *);
428 static void callinactive(struct puffs_mount *, puffs_cookie_t, int);
429 static void callreclaim(struct puffs_mount *, puffs_cookie_t, int);
430 static int  flushvncache(struct vnode *, off_t, off_t, bool);
431 static void update_va(struct vnode *, struct vattr *, struct vattr *,
432 		      struct timespec *, struct timespec *, int);
433 static void update_parent(struct vnode *, struct vnode *);
434 
435 
436 #define PUFFS_ABORT_LOOKUP	1
437 #define PUFFS_ABORT_CREATE	2
438 #define PUFFS_ABORT_MKNOD	3
439 #define PUFFS_ABORT_MKDIR	4
440 #define PUFFS_ABORT_SYMLINK	5
441 
442 /*
443  * Press the pani^Wabort button!  Kernel resource allocation failed.
444  */
445 static void
446 puffs_abortbutton(struct puffs_mount *pmp, int what,
447 	puffs_cookie_t dck, puffs_cookie_t ck, struct componentname *cnp)
448 {
449 
450 	switch (what) {
451 	case PUFFS_ABORT_CREATE:
452 	case PUFFS_ABORT_MKNOD:
453 	case PUFFS_ABORT_SYMLINK:
454 		callremove(pmp, dck, ck, cnp);
455 		break;
456 	case PUFFS_ABORT_MKDIR:
457 		callrmdir(pmp, dck, ck, cnp);
458 		break;
459 	}
460 
461 	callinactive(pmp, ck, 0);
462 	callreclaim(pmp, ck, 1);
463 }
464 
465 /*
466  * Begin vnode operations.
467  *
468  * A word from the keymaster about locks: generally we don't want
469  * to use the vnode locks at all: it creates an ugly dependency between
470  * the userlandia file server and the kernel.  But we'll play along with
471  * the kernel vnode locks for now.  However, even currently we attempt
472  * to release locks as early as possible.  This is possible for some
473  * operations which a) don't need a locked vnode after the userspace op
474  * and b) return with the vnode unlocked.  Theoretically we could
475  * unlock-do op-lock for others and order the graph in userspace, but I
476  * don't want to think of the consequences for the time being.
477  */
478 
479 #define TTL_TO_TIMEOUT(ts) \
480     (getticks() + (ts->tv_sec * hz) + (ts->tv_nsec * hz / 1000000000))
481 #define TTL_VALID(ts) \
482     ((ts != NULL) && !((ts->tv_sec == 0) && (ts->tv_nsec == 0)))
483 #define TIMED_OUT(expire) \
484     ((int)((unsigned int)getticks() - (unsigned int)expire) > 0)
485 int
486 puffs_vnop_lookup(void *v)
487 {
488         struct vop_lookup_v2_args /* {
489 		const struct vnodeop_desc *a_desc;
490 		struct vnode *a_dvp;
491 		struct vnode **a_vpp;
492 		struct componentname *a_cnp;
493         } */ *ap = v;
494 	PUFFS_MSG_VARS(vn, lookup);
495 	struct puffs_mount *pmp;
496 	struct componentname *cnp;
497 	struct vnode *vp, *dvp, *cvp;
498 	struct puffs_node *dpn, *cpn;
499 	int isdot;
500 	int error;
501 
502 	pmp = MPTOPUFFSMP(ap->a_dvp->v_mount);
503 	cnp = ap->a_cnp;
504 	dvp = ap->a_dvp;
505 	cvp = NULL;
506 	cpn = NULL;
507 	*ap->a_vpp = NULL;
508 
509 	/* r/o fs?  we check create later to handle EEXIST */
510 	if ((cnp->cn_flags & ISLASTCN)
511 	    && (dvp->v_mount->mnt_flag & MNT_RDONLY)
512 	    && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
513 		return EROFS;
514 
515 	isdot = cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.';
516 
517 	DPRINTF(("puffs_lookup: \"%s\", parent vnode %p, op: %x\n",
518 	    cnp->cn_nameptr, dvp, cnp->cn_nameiop));
519 
520 	/*
521 	 * If dotdot cache is enabled, add reference to .. and return.
522 	 */
523 	if (PUFFS_USE_DOTDOTCACHE(pmp) && (cnp->cn_flags & ISDOTDOT)) {
524 		vp = VPTOPP(ap->a_dvp)->pn_parent;
525 		vref(vp);
526 
527 		*ap->a_vpp = vp;
528 		return 0;
529 	}
530 
531 	/*
532 	 * Check if someone fed it into the cache
533 	 */
534 	if (!isdot && PUFFS_USE_NAMECACHE(pmp)) {
535 		int found, iswhiteout;
536 
537 		found = cache_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
538 				     cnp->cn_nameiop, cnp->cn_flags,
539 				     &iswhiteout, ap->a_vpp);
540 		if (iswhiteout) {
541 			cnp->cn_flags |= ISWHITEOUT;
542 		}
543 
544 		if (found && *ap->a_vpp != NULLVP && PUFFS_USE_FS_TTL(pmp)) {
545 			cvp = *ap->a_vpp;
546 			cpn = VPTOPP(cvp);
547 
548 			if (TIMED_OUT(cpn->pn_cn_timeout)) {
549 				cache_purge(cvp);
550 				/*
551 				 * cached vnode (cvp) is still referenced
552 				 * so that we can reuse it upon a new
553 				 * successful lookup.
554 				 */
555 				*ap->a_vpp = NULL;
556 				found = 0;
557 			}
558 		}
559 
560 		/*
561 		 * Do not use negative caching, since the filesystem
562 		 * provides no TTL for it.
563 		 */
564 		if (found && *ap->a_vpp == NULLVP && PUFFS_USE_FS_TTL(pmp))
565 			found = 0;
566 
567 		if (found) {
568 			return *ap->a_vpp == NULLVP ? ENOENT : 0;
569 		}
570 
571 		/*
572 		 * This is what would have been left in ERROR before
573 		 * the rearrangement of cache_lookup(). What with all
574 		 * the macros, I am not sure if this is a dead value
575 		 * below or not.
576 		 */
577 		error = -1;
578 	}
579 
580 	if (isdot) {
581 		/* deal with rename lookup semantics */
582 		if (cnp->cn_nameiop == RENAME && (cnp->cn_flags & ISLASTCN))
583 			return EISDIR;
584 
585 		vp = ap->a_dvp;
586 		vref(vp);
587 		*ap->a_vpp = vp;
588 		return 0;
589 	}
590 
591 	if (cvp != NULL) {
592 		if (vn_lock(cvp, LK_EXCLUSIVE) != 0) {
593 			vrele(cvp);
594 			cvp = NULL;
595 		} else
596 			mutex_enter(&cpn->pn_sizemtx);
597 	}
598 
599 	PUFFS_MSG_ALLOC(vn, lookup);
600 	puffs_makecn(&lookup_msg->pvnr_cn, &lookup_msg->pvnr_cn_cred,
601 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
602 
603 	if (cnp->cn_flags & ISDOTDOT)
604 		VOP_UNLOCK(dvp);
605 
606 	puffs_msg_setinfo(park_lookup, PUFFSOP_VN,
607 	    PUFFS_VN_LOOKUP, VPTOPNC(dvp));
608 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_lookup, dvp->v_data, NULL, error);
609 	DPRINTF(("puffs_lookup: return of the userspace, part %d\n", error));
610 
611 	/*
612 	 * In case of error, there is no new vnode to play with, so be
613 	 * happy with the NULL value given to vpp in the beginning.
614 	 * Also, check if this really was an error or the target was not
615 	 * present.  Either treat it as a non-error for CREATE/RENAME or
616 	 * enter the component into the negative name cache (if desired).
617 	 */
618 	if (error) {
619 		error = checkerr(pmp, error, __func__);
620 		if (error == ENOENT) {
621 			/* don't allow to create files on r/o fs */
622 			if ((dvp->v_mount->mnt_flag & MNT_RDONLY)
623 			    && cnp->cn_nameiop == CREATE) {
624 				error = EROFS;
625 
626 			/* adjust values if we are creating */
627 			} else if ((cnp->cn_flags & ISLASTCN)
628 			    && (cnp->cn_nameiop == CREATE
629 			      || cnp->cn_nameiop == RENAME)) {
630 				error = EJUSTRETURN;
631 
632 			/* save negative cache entry */
633 			} else {
634 				if (PUFFS_USE_NAMECACHE(pmp) &&
635 				    !PUFFS_USE_FS_TTL(pmp))
636 					cache_enter(dvp, NULL, cnp->cn_nameptr,
637 						cnp->cn_namelen, cnp->cn_flags);
638 			}
639 		}
640 		goto out;
641 	}
642 
643 	/*
644 	 * Check that we don't get our parent node back, that would cause
645 	 * a pretty obvious deadlock.
646 	 */
647 	dpn = dvp->v_data;
648 	if (lookup_msg->pvnr_newnode == dpn->pn_cookie) {
649 		puffs_senderr(pmp, PUFFS_ERR_LOOKUP, EINVAL,
650 		    "lookup produced parent cookie", lookup_msg->pvnr_newnode);
651 		error = EPROTO;
652 		goto out;
653 	}
654 
655 	/*
656 	 * Check if we looked up the cached vnode
657 	 */
658 	vp = NULL;
659 	if (cvp && (VPTOPP(cvp)->pn_cookie == lookup_msg->pvnr_newnode)) {
660 		int grace;
661 
662 		/*
663 		 * Bump grace time of this node so that it does not get
664 		 * reclaimed too fast. We try to increase a bit more the
665 		 * lifetime of busiest * nodes - with some limits.
666 		 */
667 		grace = 10 * puffs_sopreq_expire_timeout;
668 		cpn->pn_cn_grace = getticks() + grace;
669 		vp = cvp;
670 	}
671 
672 	/*
673 	 * No cached vnode available, or the cached vnode does not
674 	 * match the userland cookie anymore: is the node known?
675 	 */
676 	if (vp == NULL) {
677 		error = puffs_getvnode(dvp->v_mount,
678 		    lookup_msg->pvnr_newnode, lookup_msg->pvnr_vtype,
679 		    lookup_msg->pvnr_size, lookup_msg->pvnr_rdev, &vp);
680 		if (error) {
681 			puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP,
682 			    VPTOPNC(dvp), lookup_msg->pvnr_newnode,
683 			    ap->a_cnp);
684 			goto out;
685 		}
686 
687 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
688 	}
689 
690 	/*
691 	 * Update cache and TTL
692 	 */
693 	if (PUFFS_USE_FS_TTL(pmp)) {
694 		struct timespec *va_ttl = &lookup_msg->pvnr_va_ttl;
695 		struct timespec *cn_ttl = &lookup_msg->pvnr_cn_ttl;
696 		update_va(vp, NULL, &lookup_msg->pvnr_va,
697 			  va_ttl, cn_ttl, SETATTR_CHSIZE);
698 	}
699 
700 	KASSERT(lookup_msg->pvnr_newnode == VPTOPP(vp)->pn_cookie);
701 	*ap->a_vpp = vp;
702 
703 	if (PUFFS_USE_NAMECACHE(pmp))
704 		cache_enter(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
705 			    cnp->cn_flags);
706 
707 	/* XXX */
708 	if ((lookup_msg->pvnr_cn.pkcn_flags & REQUIREDIR) == 0)
709 		cnp->cn_flags &= ~REQUIREDIR;
710 	if (lookup_msg->pvnr_cn.pkcn_consume)
711 		cnp->cn_consume = MIN(lookup_msg->pvnr_cn.pkcn_consume,
712 		    strlen(cnp->cn_nameptr) - cnp->cn_namelen);
713 
714 	VPTOPP(vp)->pn_nlookup++;
715 
716 	if (PUFFS_USE_DOTDOTCACHE(pmp) &&
717 	    (VPTOPP(vp)->pn_parent != dvp))
718 		update_parent(vp, dvp);
719 
720  out:
721 	if (cvp != NULL) {
722 		mutex_exit(&cpn->pn_sizemtx);
723 
724 		if (error || (cvp != vp))
725 			vput(cvp);
726 	}
727 	if (error == 0)
728 		VOP_UNLOCK(*ap->a_vpp);
729 
730 	if (cnp->cn_flags & ISDOTDOT)
731 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
732 
733 	DPRINTF(("puffs_lookup: returning %d %p\n", error, *ap->a_vpp));
734 	PUFFS_MSG_RELEASE(lookup);
735 	return error;
736 }
737 
738 #define REFPN_AND_UNLOCKVP(a, b)					\
739 do {									\
740 	mutex_enter(&b->pn_mtx);					\
741 	puffs_referencenode(b);						\
742 	mutex_exit(&b->pn_mtx);						\
743 	VOP_UNLOCK(a);						\
744 } while (/*CONSTCOND*/0)
745 
746 #define REFPN(b)							\
747 do {									\
748 	mutex_enter(&b->pn_mtx);					\
749 	puffs_referencenode(b);						\
750 	mutex_exit(&b->pn_mtx);						\
751 } while (/*CONSTCOND*/0)
752 
753 #define RELEPN_AND_VP(a, b)						\
754 do {									\
755 	puffs_releasenode(b);						\
756 	vrele(a);							\
757 } while (/*CONSTCOND*/0)
758 
759 int
760 puffs_vnop_create(void *v)
761 {
762 	struct vop_create_v3_args /* {
763 		const struct vnodeop_desc *a_desc;
764 		struct vnode *a_dvp;
765 		struct vnode **a_vpp;
766 		struct componentname *a_cnp;
767 		struct vattr *a_vap;
768 	} */ *ap = v;
769 	PUFFS_MSG_VARS(vn, create);
770 	struct vnode *dvp = ap->a_dvp;
771 	struct puffs_node *dpn = VPTOPP(dvp);
772 	struct componentname *cnp = ap->a_cnp;
773 	struct mount *mp = dvp->v_mount;
774 	struct puffs_mount *pmp = MPTOPUFFSMP(mp);
775 	int error;
776 
777 	DPRINTF(("puffs_create: dvp %p, cnp: %s\n",
778 	    dvp, ap->a_cnp->cn_nameptr));
779 
780 	PUFFS_MSG_ALLOC(vn, create);
781 	puffs_makecn(&create_msg->pvnr_cn, &create_msg->pvnr_cn_cred,
782 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
783 	create_msg->pvnr_va = *ap->a_vap;
784 	puffs_msg_setinfo(park_create, PUFFSOP_VN,
785 	    PUFFS_VN_CREATE, VPTOPNC(dvp));
786 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_create, dvp->v_data, NULL, error);
787 
788 	error = checkerr(pmp, error, __func__);
789 	if (error)
790 		goto out;
791 
792 	error = puffs_newnode(mp, dvp, ap->a_vpp,
793 	    create_msg->pvnr_newnode, cnp, ap->a_vap->va_type, 0);
794 	if (error) {
795 		puffs_abortbutton(pmp, PUFFS_ABORT_CREATE, dpn->pn_cookie,
796 		    create_msg->pvnr_newnode, cnp);
797 		goto out;
798 	}
799 
800 	if (PUFFS_USE_FS_TTL(pmp)) {
801 		struct timespec *va_ttl = &create_msg->pvnr_va_ttl;
802 		struct timespec *cn_ttl = &create_msg->pvnr_cn_ttl;
803 		struct vattr *rvap = &create_msg->pvnr_va;
804 
805 		update_va(*ap->a_vpp, NULL, rvap,
806 			  va_ttl, cn_ttl, SETATTR_CHSIZE);
807 	}
808 
809 	VPTOPP(*ap->a_vpp)->pn_nlookup++;
810 
811 	if (PUFFS_USE_DOTDOTCACHE(pmp) &&
812 	    (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
813 		update_parent(*ap->a_vpp, dvp);
814 
815  out:
816 	DPRINTF(("puffs_create: return %d\n", error));
817 	PUFFS_MSG_RELEASE(create);
818 	return error;
819 }
820 
821 int
822 puffs_vnop_mknod(void *v)
823 {
824 	struct vop_mknod_v3_args /* {
825 		const struct vnodeop_desc *a_desc;
826 		struct vnode *a_dvp;
827 		struct vnode **a_vpp;
828 		struct componentname *a_cnp;
829 		struct vattr *a_vap;
830 	} */ *ap = v;
831 	PUFFS_MSG_VARS(vn, mknod);
832 	struct vnode *dvp = ap->a_dvp;
833 	struct puffs_node *dpn = VPTOPP(dvp);
834 	struct componentname *cnp = ap->a_cnp;
835 	struct mount *mp = dvp->v_mount;
836 	struct puffs_mount *pmp = MPTOPUFFSMP(mp);
837 	int error;
838 
839 	PUFFS_MSG_ALLOC(vn, mknod);
840 	puffs_makecn(&mknod_msg->pvnr_cn, &mknod_msg->pvnr_cn_cred,
841 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
842 	mknod_msg->pvnr_va = *ap->a_vap;
843 	puffs_msg_setinfo(park_mknod, PUFFSOP_VN,
844 	    PUFFS_VN_MKNOD, VPTOPNC(dvp));
845 
846 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mknod, dvp->v_data, NULL, error);
847 
848 	error = checkerr(pmp, error, __func__);
849 	if (error)
850 		goto out;
851 
852 	error = puffs_newnode(mp, dvp, ap->a_vpp,
853 	    mknod_msg->pvnr_newnode, cnp, ap->a_vap->va_type,
854 	    ap->a_vap->va_rdev);
855 	if (error) {
856 		puffs_abortbutton(pmp, PUFFS_ABORT_MKNOD, dpn->pn_cookie,
857 		    mknod_msg->pvnr_newnode, cnp);
858 		goto out;
859 	}
860 
861 	if (PUFFS_USE_FS_TTL(pmp)) {
862 		struct timespec *va_ttl = &mknod_msg->pvnr_va_ttl;
863 		struct timespec *cn_ttl = &mknod_msg->pvnr_cn_ttl;
864 		struct vattr *rvap = &mknod_msg->pvnr_va;
865 
866 		update_va(*ap->a_vpp, NULL, rvap,
867 			   va_ttl, cn_ttl, SETATTR_CHSIZE);
868 	}
869 
870 	VPTOPP(*ap->a_vpp)->pn_nlookup++;
871 
872 	if (PUFFS_USE_DOTDOTCACHE(pmp) &&
873 	    (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
874 		update_parent(*ap->a_vpp, dvp);
875 
876  out:
877 	PUFFS_MSG_RELEASE(mknod);
878 	return error;
879 }
880 
881 int
882 puffs_vnop_open(void *v)
883 {
884 	struct vop_open_args /* {
885 		const struct vnodeop_desc *a_desc;
886 		struct vnode *a_vp;
887 		int a_mode;
888 		kauth_cred_t a_cred;
889 	} */ *ap = v;
890 	PUFFS_MSG_VARS(vn, open);
891 	struct vnode *vp = ap->a_vp;
892 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
893 	struct puffs_node *pn = VPTOPP(vp);
894 	int mode = ap->a_mode;
895 	int error;
896 
897 	DPRINTF(("puffs_open: vp %p, mode 0x%x\n", vp, mode));
898 
899 	if (vp->v_type == VREG && mode & FWRITE && !EXISTSOP(pmp, WRITE))
900 		ERROUT(EROFS);
901 
902 	if (!EXISTSOP(pmp, OPEN))
903 		ERROUT(0);
904 
905 	PUFFS_MSG_ALLOC(vn, open);
906 	open_msg->pvnr_mode = mode;
907 	puffs_credcvt(&open_msg->pvnr_cred, ap->a_cred);
908 	puffs_msg_setinfo(park_open, PUFFSOP_VN,
909 	    PUFFS_VN_OPEN, VPTOPNC(vp));
910 
911 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_open, vp->v_data, NULL, error);
912 	error = checkerr(pmp, error, __func__);
913 
914 	if (open_msg->pvnr_oflags & PUFFS_OPEN_IO_DIRECT) {
915 		/*
916 		 * Flush cache:
917 		 * - we do not want to discard cached write by direct write
918 		 * - read cache is now useless and should be freed
919 		 */
920 		mutex_enter(&pn->pn_sizemtx);
921 		flushvncache(vp, 0, 0, true);
922 		mutex_exit(&pn->pn_sizemtx);
923 
924 		if (mode & FREAD)
925 			pn->pn_stat |= PNODE_RDIRECT;
926 		if (mode & FWRITE)
927 			pn->pn_stat |= PNODE_WDIRECT;
928 	}
929  out:
930 	DPRINTF(("puffs_open: returning %d\n", error));
931 	PUFFS_MSG_RELEASE(open);
932 	return error;
933 }
934 
935 int
936 puffs_vnop_close(void *v)
937 {
938 	struct vop_close_args /* {
939 		const struct vnodeop_desc *a_desc;
940 		struct vnode *a_vp;
941 		int a_fflag;
942 		kauth_cred_t a_cred;
943 	} */ *ap = v;
944 	PUFFS_MSG_VARS(vn, close);
945 	struct vnode *vp = ap->a_vp;
946 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
947 
948 	PUFFS_MSG_ALLOC(vn, close);
949 	puffs_msg_setfaf(park_close);
950 	close_msg->pvnr_fflag = ap->a_fflag;
951 	puffs_credcvt(&close_msg->pvnr_cred, ap->a_cred);
952 	puffs_msg_setinfo(park_close, PUFFSOP_VN,
953 	    PUFFS_VN_CLOSE, VPTOPNC(vp));
954 
955 	puffs_msg_enqueue(pmp, park_close);
956 	PUFFS_MSG_RELEASE(close);
957 	return 0;
958 }
959 
960 int
961 puffs_vnop_access(void *v)
962 {
963 	struct vop_access_args /* {
964 		const struct vnodeop_desc *a_desc;
965 		struct vnode *a_vp;
966 		accmode_t a_accmode;
967 		kauth_cred_t a_cred;
968 	} */ *ap = v;
969 	PUFFS_MSG_VARS(vn, access);
970 	struct vnode *vp = ap->a_vp;
971 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
972 	accmode_t accmode = ap->a_accmode;
973 	int error;
974 
975 	if (accmode & VWRITE) {
976 		switch (vp->v_type) {
977 		case VDIR:
978 		case VLNK:
979 		case VREG:
980 			if ((vp->v_mount->mnt_flag & MNT_RDONLY)
981 			    || !EXISTSOP(pmp, WRITE))
982 				return EROFS;
983 			break;
984 		default:
985 			break;
986 		}
987 	}
988 
989 	if (!EXISTSOP(pmp, ACCESS))
990 		return 0;
991 
992 	PUFFS_MSG_ALLOC(vn, access);
993 	access_msg->pvnr_mode = ap->a_accmode;
994 	puffs_credcvt(&access_msg->pvnr_cred, ap->a_cred);
995 	puffs_msg_setinfo(park_access, PUFFSOP_VN,
996 	    PUFFS_VN_ACCESS, VPTOPNC(vp));
997 
998 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_access, vp->v_data, NULL, error);
999 	error = checkerr(pmp, error, __func__);
1000 	PUFFS_MSG_RELEASE(access);
1001 
1002 	return error;
1003 }
1004 
1005 static void
1006 update_va(struct vnode *vp, struct vattr *vap, struct vattr *rvap,
1007 	  struct timespec *va_ttl, struct timespec *cn_ttl, int flags)
1008 {
1009 	struct puffs_node *pn = VPTOPP(vp);
1010 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1011 	int use_metacache;
1012 
1013 	if (TTL_VALID(cn_ttl)) {
1014 		pn->pn_cn_timeout = TTL_TO_TIMEOUT(cn_ttl);
1015 		pn->pn_cn_grace = MAX(pn->pn_cn_timeout, pn->pn_cn_grace);
1016 	}
1017 
1018 	/*
1019 	 * Don't listen to the file server regarding special device
1020 	 * size info, the file server doesn't know anything about them.
1021 	 */
1022 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1023 		rvap->va_size = vp->v_size;
1024 
1025 	/* Ditto for blocksize (ufs comment: this doesn't belong here) */
1026 	if (vp->v_type == VBLK)
1027 		rvap->va_blocksize = BLKDEV_IOSIZE;
1028 	else if (vp->v_type == VCHR)
1029 		rvap->va_blocksize = MAXBSIZE;
1030 
1031 	if (vap != NULL) {
1032 		(void) memcpy(vap, rvap, sizeof(struct vattr));
1033 		vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1034 
1035 		if (PUFFS_USE_METAFLUSH(pmp)) {
1036 			if (pn->pn_stat & PNODE_METACACHE_ATIME)
1037 				vap->va_atime = pn->pn_mc_atime;
1038 			if (pn->pn_stat & PNODE_METACACHE_CTIME)
1039 				vap->va_ctime = pn->pn_mc_ctime;
1040 			if (pn->pn_stat & PNODE_METACACHE_MTIME)
1041 				vap->va_mtime = pn->pn_mc_mtime;
1042 			if (pn->pn_stat & PNODE_METACACHE_SIZE)
1043 				vap->va_size = pn->pn_mc_size;
1044 		}
1045 	}
1046 
1047 	use_metacache = PUFFS_USE_METAFLUSH(pmp) &&
1048 			(pn->pn_stat & PNODE_METACACHE_SIZE);
1049 	if (!use_metacache && (flags & SETATTR_CHSIZE)) {
1050 		if (rvap->va_size != VNOVAL
1051 		    && vp->v_type != VBLK && vp->v_type != VCHR) {
1052 			uvm_vnp_setsize(vp, rvap->va_size);
1053 			pn->pn_serversize = rvap->va_size;
1054 		}
1055 	}
1056 
1057 	if ((va_ttl != NULL) && TTL_VALID(va_ttl)) {
1058 		if (pn->pn_va_cache == NULL)
1059 			pn->pn_va_cache = pool_get(&puffs_vapool, PR_WAITOK);
1060 
1061 		(void)memcpy(pn->pn_va_cache, rvap, sizeof(*rvap));
1062 
1063 		pn->pn_va_timeout = TTL_TO_TIMEOUT(va_ttl);
1064 	}
1065 }
1066 
1067 static void
1068 update_parent(struct vnode *vp, struct vnode *dvp)
1069 {
1070 	struct puffs_node *pn = VPTOPP(vp);
1071 
1072 	if (pn->pn_parent != NULL) {
1073 		KASSERT(pn->pn_parent != dvp);
1074 		vrele(pn->pn_parent);
1075 	}
1076 
1077 	vref(dvp);
1078 	pn->pn_parent = dvp;
1079 }
1080 
1081 int
1082 puffs_vnop_getattr(void *v)
1083 {
1084 	struct vop_getattr_args /* {
1085 		const struct vnodeop_desc *a_desc;
1086 		struct vnode *a_vp;
1087 		struct vattr *a_vap;
1088 		kauth_cred_t a_cred;
1089 	} */ *ap = v;
1090 	PUFFS_MSG_VARS(vn, getattr);
1091 	struct vnode *vp = ap->a_vp;
1092 	struct mount *mp = vp->v_mount;
1093 	struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1094 	struct vattr *vap, *rvap;
1095 	struct puffs_node *pn = VPTOPP(vp);
1096 	struct timespec *va_ttl = NULL;
1097 	int error = 0;
1098 
1099 	/*
1100 	 * A lock is required so that we do not race with
1101 	 * setattr, write and fsync when changing vp->v_size.
1102 	 * This is critical, since setting a stall smaler value
1103 	 * triggers a file truncate in uvm_vnp_setsize(), which
1104 	 * most of the time means data corruption (a chunk of
1105 	 * data is replaced by zeroes). This can be removed if
1106 	 * we decide one day that VOP_GETATTR must operate on
1107 	 * a locked vnode.
1108 	 *
1109 	 * XXX Should be useless now that VOP_GETATTR has been
1110 	 *     fixed to always require a shared lock at least.
1111 	 */
1112 	mutex_enter(&pn->pn_sizemtx);
1113 
1114 	REFPN(pn);
1115 	vap = ap->a_vap;
1116 
1117 	if (PUFFS_USE_FS_TTL(pmp)) {
1118 		if (!TIMED_OUT(pn->pn_va_timeout)) {
1119 			update_va(vp, vap, pn->pn_va_cache,
1120 				  NULL, NULL, SETATTR_CHSIZE);
1121 			goto out2;
1122 		}
1123 	}
1124 
1125 	PUFFS_MSG_ALLOC(vn, getattr);
1126 	vattr_null(&getattr_msg->pvnr_va);
1127 	puffs_credcvt(&getattr_msg->pvnr_cred, ap->a_cred);
1128 	puffs_msg_setinfo(park_getattr, PUFFSOP_VN,
1129 	    PUFFS_VN_GETATTR, VPTOPNC(vp));
1130 
1131 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getattr, vp->v_data, NULL, error);
1132 	error = checkerr(pmp, error, __func__);
1133 	if (error)
1134 		goto out;
1135 
1136 	rvap = &getattr_msg->pvnr_va;
1137 
1138 	if (PUFFS_USE_FS_TTL(pmp))
1139 		va_ttl = &getattr_msg->pvnr_va_ttl;
1140 
1141 	update_va(vp, vap, rvap, va_ttl, NULL, SETATTR_CHSIZE);
1142 
1143  out:
1144 	PUFFS_MSG_RELEASE(getattr);
1145 
1146  out2:
1147 	puffs_releasenode(pn);
1148 
1149 	mutex_exit(&pn->pn_sizemtx);
1150 
1151 	return error;
1152 }
1153 
1154 static void
1155 zerofill_lastpage(struct vnode *vp, voff_t off)
1156 {
1157 
1158 	if (trunc_page(off) == off)
1159 		return;
1160 
1161 	if (vp->v_writecount == 0)
1162 		return;
1163 
1164 	vsize_t len = round_page(off) - off;
1165 	ubc_zerorange(&vp->v_uobj, off, len, UBC_WRITE|UBC_VNODE_FLAGS(vp));
1166 }
1167 
1168 static int
1169 dosetattr(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, int flags)
1170 {
1171 	PUFFS_MSG_VARS(vn, setattr);
1172 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1173 	struct puffs_node *pn = vp->v_data;
1174 	vsize_t oldsize = vp->v_size;
1175 	int error = 0;
1176 
1177 	KASSERT(!(flags & SETATTR_CHSIZE) || mutex_owned(&pn->pn_sizemtx));
1178 
1179 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) &&
1180 	    (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL
1181 	    || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL
1182 	    || vap->va_mode != (mode_t)VNOVAL))
1183 		return EROFS;
1184 
1185 	if ((vp->v_mount->mnt_flag & MNT_RDONLY)
1186 	    && vp->v_type == VREG && vap->va_size != VNOVAL)
1187 		return EROFS;
1188 
1189 	/*
1190 	 * Flush metacache first.  If we are called with some explicit
1191 	 * parameters, treat them as information overriding metacache
1192 	 * information.
1193 	 */
1194 	if (PUFFS_USE_METAFLUSH(pmp) && pn->pn_stat & PNODE_METACACHE_MASK) {
1195 		if ((pn->pn_stat & PNODE_METACACHE_ATIME)
1196 		    && vap->va_atime.tv_sec == VNOVAL)
1197 			vap->va_atime = pn->pn_mc_atime;
1198 		if ((pn->pn_stat & PNODE_METACACHE_CTIME)
1199 		    && vap->va_ctime.tv_sec == VNOVAL)
1200 			vap->va_ctime = pn->pn_mc_ctime;
1201 		if ((pn->pn_stat & PNODE_METACACHE_MTIME)
1202 		    && vap->va_mtime.tv_sec == VNOVAL)
1203 			vap->va_mtime = pn->pn_mc_mtime;
1204 		if ((pn->pn_stat & PNODE_METACACHE_SIZE)
1205 		    && vap->va_size == VNOVAL)
1206 			vap->va_size = pn->pn_mc_size;
1207 
1208 		pn->pn_stat &= ~PNODE_METACACHE_MASK;
1209 	}
1210 
1211 	/*
1212 	 * Flush attribute cache so that another thread do
1213 	 * not get a stale value during the operation.
1214 	 */
1215 	if (PUFFS_USE_FS_TTL(pmp))
1216 		pn->pn_va_timeout = 0;
1217 
1218 	PUFFS_MSG_ALLOC(vn, setattr);
1219 	(void)memcpy(&setattr_msg->pvnr_va, vap, sizeof(struct vattr));
1220 	puffs_credcvt(&setattr_msg->pvnr_cred, cred);
1221 	puffs_msg_setinfo(park_setattr, PUFFSOP_VN,
1222 	    PUFFS_VN_SETATTR, VPTOPNC(vp));
1223 	if (flags & SETATTR_ASYNC)
1224 		puffs_msg_setfaf(park_setattr);
1225 
1226 	puffs_msg_enqueue(pmp, park_setattr);
1227 	if ((flags & SETATTR_ASYNC) == 0) {
1228 		error = puffs_msg_wait2(pmp, park_setattr, vp->v_data, NULL);
1229 
1230 		if ((error == 0) && PUFFS_USE_FS_TTL(pmp)) {
1231 			struct timespec *va_ttl = &setattr_msg->pvnr_va_ttl;
1232 			struct vattr *rvap = &setattr_msg->pvnr_va;
1233 
1234 			update_va(vp, NULL, rvap, va_ttl, NULL, flags);
1235 		}
1236 	}
1237 
1238 	PUFFS_MSG_RELEASE(setattr);
1239 	if ((flags & SETATTR_ASYNC) == 0) {
1240 		error = checkerr(pmp, error, __func__);
1241 		if (error)
1242 			return error;
1243 	} else {
1244 		error = 0;
1245 	}
1246 
1247 	if (vap->va_size != VNOVAL) {
1248 		/*
1249 		 * If we truncated the file, make sure the data beyond
1250 		 * EOF in last page does not remain in cache, otherwise
1251 		 * if the file is later truncated to a larger size (creating
1252 		 * a hole), that area will not return zeroes as it
1253 		 * should.
1254 		 */
1255 		if ((flags & SETATTR_CHSIZE) && PUFFS_USE_PAGECACHE(pmp) &&
1256 		    (vap->va_size < oldsize))
1257 			zerofill_lastpage(vp, vap->va_size);
1258 
1259 		pn->pn_serversize = vap->va_size;
1260 		if (flags & SETATTR_CHSIZE)
1261 			uvm_vnp_setsize(vp, vap->va_size);
1262 		puffs_updatenode(pn, PUFFS_UPDATECTIME | PUFFS_UPDATEMTIME, 0);
1263 	}
1264 
1265 	return 0;
1266 }
1267 
1268 int
1269 puffs_vnop_setattr(void *v)
1270 {
1271 	struct vop_getattr_args /* {
1272 		const struct vnodeop_desc *a_desc;
1273 		struct vnode *a_vp;
1274 		struct vattr *a_vap;
1275 		kauth_cred_t a_cred;
1276 	} */ *ap = v;
1277 	struct puffs_node *pn = ap->a_vp->v_data;
1278 	int error;
1279 
1280 	mutex_enter(&pn->pn_sizemtx);
1281 	error = dosetattr(ap->a_vp, ap->a_vap, ap->a_cred, SETATTR_CHSIZE);
1282 	mutex_exit(&pn->pn_sizemtx);
1283 
1284 	return error;
1285 }
1286 
1287 static __inline int
1288 doinact(struct puffs_mount *pmp, int iaflag)
1289 {
1290 
1291 	if (EXISTSOP(pmp, INACTIVE))
1292 		if (pmp->pmp_flags & PUFFS_KFLAG_IAONDEMAND)
1293 			if (iaflag || ALLOPS(pmp))
1294 				return 1;
1295 			else
1296 				return 0;
1297 		else
1298 			return 1;
1299 	else
1300 		return 0;
1301 }
1302 
1303 static void
1304 callinactive(struct puffs_mount *pmp, puffs_cookie_t ck, int iaflag)
1305 {
1306 	PUFFS_MSG_VARS(vn, inactive);
1307 
1308 	if (doinact(pmp, iaflag)) {
1309 		PUFFS_MSG_ALLOC(vn, inactive);
1310 		puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1311 		    PUFFS_VN_INACTIVE, ck);
1312 		PUFFS_MSG_ENQUEUEWAIT_NOERROR(pmp, park_inactive);
1313 		PUFFS_MSG_RELEASE(inactive);
1314 	}
1315 }
1316 
1317 /* XXX: callinactive can't setback */
1318 int
1319 puffs_vnop_inactive(void *v)
1320 {
1321 	struct vop_inactive_v2_args /* {
1322 		const struct vnodeop_desc *a_desc;
1323 		struct vnode *a_vp;
1324 	} */ *ap = v;
1325 	PUFFS_MSG_VARS(vn, inactive);
1326 	struct vnode *vp = ap->a_vp;
1327 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1328 	struct puffs_node *pnode;
1329 	bool recycle = false;
1330 
1331 	/*
1332 	 * When puffs_cookie2vnode() misses an entry, vcache_get()
1333 	 * creates a new node (puffs_vfsop_loadvnode being called to
1334 	 * initialize the PUFFS part), then it discovers it is VNON,
1335 	 * and tries to vrele() it. This leads us there, while the
1336 	 * cookie was stall and the node likely already reclaimed.
1337 	 */
1338 	if (vp->v_type == VNON) {
1339 		return 0;
1340 	}
1341 
1342 	pnode = vp->v_data;
1343 	mutex_enter(&pnode->pn_sizemtx);
1344 
1345 	if (doinact(pmp, pnode->pn_stat & PNODE_DOINACT)) {
1346 		flushvncache(vp, 0, 0, false);
1347 		PUFFS_MSG_ALLOC(vn, inactive);
1348 		puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1349 		    PUFFS_VN_INACTIVE, VPTOPNC(vp));
1350 		PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_inactive, vp->v_data,
1351 		    NULL);
1352 		PUFFS_MSG_RELEASE(inactive);
1353 	}
1354 	pnode->pn_stat &= ~PNODE_DOINACT;
1355 
1356 	/*
1357 	 * file server thinks it's gone?  then don't be afraid care,
1358 	 * node's life was already all it would ever be
1359 	 */
1360 	if (pnode->pn_stat & PNODE_NOREFS) {
1361 		pnode->pn_stat |= PNODE_DYING;
1362 		recycle = true;
1363 	}
1364 
1365 	/*
1366 	 * Handle node TTL.
1367 	 * If grace has already timed out, make it reclaimed.
1368 	 * Otherwise, we queue its expiration by sop thread, so
1369 	 * that it does not remain for ages in the freelist,
1370 	 * holding memory in userspace, while we will have
1371 	 * to look it up again anyway.
1372 	 */
1373 	if (PUFFS_USE_FS_TTL(pmp) && !(vp->v_vflag & VV_ROOT) && !recycle) {
1374 		bool incache = !TIMED_OUT(pnode->pn_cn_timeout);
1375 		bool ingrace = !TIMED_OUT(pnode->pn_cn_grace);
1376 		bool reclaimqueued = pnode->pn_stat & PNODE_SOPEXP;
1377 
1378 		if (!incache && !ingrace && !reclaimqueued) {
1379 			pnode->pn_stat |= PNODE_DYING;
1380 			recycle = true;
1381 		}
1382 
1383 		if (!recycle && !reclaimqueued) {
1384 			struct puffs_sopreq *psopr;
1385 			int at = MAX(pnode->pn_cn_grace, pnode->pn_cn_timeout);
1386 
1387 			KASSERT(curlwp != uvm.pagedaemon_lwp);
1388 			psopr = kmem_alloc(sizeof(*psopr), KM_SLEEP);
1389 			psopr->psopr_ck = VPTOPNC(pnode->pn_vp);
1390 			psopr->psopr_sopreq = PUFFS_SOPREQ_EXPIRE;
1391 			psopr->psopr_at = at;
1392 
1393 			mutex_enter(&pmp->pmp_sopmtx);
1394 
1395 			/*
1396 			 * If thread has disapeared, just give up. The
1397 			 * fs is being unmounted and the node will be
1398 			 * be reclaimed anyway.
1399 			 *
1400 			 * Otherwise, we queue the request but do not
1401 			 * immediatly signal the thread, as the node
1402 			 * has not been expired yet.
1403 			 */
1404 			if (pmp->pmp_sopthrcount == 0) {
1405 				kmem_free(psopr, sizeof(*psopr));
1406 			} else {
1407 				TAILQ_INSERT_TAIL(&pmp->pmp_sopnodereqs,
1408 				    psopr, psopr_entries);
1409 				pnode->pn_stat |= PNODE_SOPEXP;
1410 			}
1411 
1412 			mutex_exit(&pmp->pmp_sopmtx);
1413 		}
1414 	}
1415 
1416 	/*
1417 	 * Wipe direct I/O flags
1418 	 */
1419 	pnode->pn_stat &= ~(PNODE_RDIRECT|PNODE_WDIRECT);
1420 
1421 	*ap->a_recycle = recycle;
1422 
1423 	mutex_exit(&pnode->pn_sizemtx);
1424 
1425 	return 0;
1426 }
1427 
1428 static void
1429 callreclaim(struct puffs_mount *pmp, puffs_cookie_t ck, int nlookup)
1430 {
1431 	PUFFS_MSG_VARS(vn, reclaim);
1432 
1433 	if (!EXISTSOP(pmp, RECLAIM))
1434 		return;
1435 
1436 	PUFFS_MSG_ALLOC(vn, reclaim);
1437 	reclaim_msg->pvnr_nlookup = nlookup;
1438 	puffs_msg_setfaf(park_reclaim);
1439 	puffs_msg_setinfo(park_reclaim, PUFFSOP_VN, PUFFS_VN_RECLAIM, ck);
1440 
1441 	puffs_msg_enqueue(pmp, park_reclaim);
1442 	PUFFS_MSG_RELEASE(reclaim);
1443 	return;
1444 }
1445 
1446 /*
1447  * always FAF, we don't really care if the server wants to fail to
1448  * reclaim the node or not
1449  */
1450 int
1451 puffs_vnop_reclaim(void *v)
1452 {
1453 	struct vop_reclaim_v2_args /* {
1454 		const struct vnodeop_desc *a_desc;
1455 		struct vnode *a_vp;
1456 	} */ *ap = v;
1457 	struct vnode *vp = ap->a_vp;
1458 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1459 	bool notifyserver = true;
1460 
1461 	VOP_UNLOCK(vp);
1462 
1463 	/*
1464 	 * first things first: check if someone is trying to reclaim the
1465 	 * root vnode.  do not allow that to travel to userspace.
1466 	 * Note that we don't need to take the lock similarly to
1467 	 * puffs_root(), since there is only one of us.
1468 	 */
1469 	if (vp->v_vflag & VV_ROOT) {
1470 		mutex_enter(&pmp->pmp_lock);
1471 		KASSERT(pmp->pmp_root != NULL);
1472 		pmp->pmp_root = NULL;
1473 		mutex_exit(&pmp->pmp_lock);
1474 		notifyserver = false;
1475 	}
1476 
1477 	/* See the comment on top of puffs_vnop_inactive(). */
1478 	if (vp->v_type == VNON)
1479 		notifyserver = false;
1480 
1481 	/*
1482 	 * purge info from kernel before issueing FAF, since we
1483 	 * don't really know when we'll get around to it after
1484 	 * that and someone might race us into node creation
1485 	 */
1486 	mutex_enter(&pmp->pmp_lock);
1487 	if (PUFFS_USE_NAMECACHE(pmp))
1488 		cache_purge(vp);
1489 	mutex_exit(&pmp->pmp_lock);
1490 
1491 	if (notifyserver) {
1492 		int nlookup = VPTOPP(vp)->pn_nlookup;
1493 
1494 		callreclaim(MPTOPUFFSMP(vp->v_mount), VPTOPNC(vp), nlookup);
1495 	}
1496 
1497 	if (PUFFS_USE_DOTDOTCACHE(pmp)) {
1498 		if (__predict_true(VPTOPP(vp)->pn_parent != NULL))
1499 			vrele(VPTOPP(vp)->pn_parent);
1500 		else
1501 			KASSERT(vp->v_type == VNON || (vp->v_vflag & VV_ROOT));
1502 	}
1503 
1504 	puffs_putvnode(vp);
1505 
1506 	return 0;
1507 }
1508 
1509 #define CSIZE sizeof(**ap->a_cookies)
1510 int
1511 puffs_vnop_readdir(void *v)
1512 {
1513 	struct vop_readdir_args /* {
1514 		const struct vnodeop_desc *a_desc;
1515 		struct vnode *a_vp;
1516 		struct uio *a_uio;
1517 		kauth_cred_t a_cred;
1518 		int *a_eofflag;
1519 		off_t **a_cookies;
1520 		int *a_ncookies;
1521 	} */ *ap = v;
1522 	PUFFS_MSG_VARS(vn, readdir);
1523 	struct vnode *vp = ap->a_vp;
1524 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1525 	size_t argsize, tomove, cookiemem, cookiesmax;
1526 	struct uio *uio = ap->a_uio;
1527 	size_t howmuch, resid;
1528 	int error;
1529 
1530 	/*
1531 	 * ok, so we need: resid + cookiemem = maxreq
1532 	 * => resid + cookiesize * (resid/minsize) = maxreq
1533 	 * => resid + cookiesize/minsize * resid = maxreq
1534 	 * => (cookiesize/minsize + 1) * resid = maxreq
1535 	 * => resid = maxreq / (cookiesize/minsize + 1)
1536 	 *
1537 	 * Since cookiesize <= minsize and we're not very big on floats,
1538 	 * we approximate that to be 1.  Therefore:
1539 	 *
1540 	 * resid = maxreq / 2;
1541 	 *
1542 	 * Well, at least we didn't have to use differential equations
1543 	 * or the Gram-Schmidt process.
1544 	 *
1545 	 * (yes, I'm very afraid of this)
1546 	 */
1547 	KASSERT(CSIZE <= _DIRENT_MINSIZE((struct dirent *)0));
1548 
1549 	if (ap->a_cookies) {
1550 		KASSERT(ap->a_ncookies != NULL);
1551 		if (pmp->pmp_args.pa_fhsize == 0)
1552 			return EOPNOTSUPP;
1553 		resid = PUFFS_TOMOVE(uio->uio_resid, pmp) / 2;
1554 		cookiesmax = resid/_DIRENT_MINSIZE((struct dirent *)0);
1555 		cookiemem = ALIGN(cookiesmax*CSIZE); /* play safe */
1556 	} else {
1557 		resid = PUFFS_TOMOVE(uio->uio_resid, pmp);
1558 		cookiesmax = 0;
1559 		cookiemem = 0;
1560 	}
1561 
1562 	argsize = sizeof(struct puffs_vnmsg_readdir);
1563 	tomove = resid + cookiemem;
1564 	puffs_msgmem_alloc(argsize + tomove, &park_readdir,
1565 	    (void *)&readdir_msg, 1);
1566 
1567 	puffs_credcvt(&readdir_msg->pvnr_cred, ap->a_cred);
1568 	readdir_msg->pvnr_offset = uio->uio_offset;
1569 	readdir_msg->pvnr_resid = resid;
1570 	readdir_msg->pvnr_ncookies = cookiesmax;
1571 	readdir_msg->pvnr_eofflag = 0;
1572 	readdir_msg->pvnr_dentoff = cookiemem;
1573 	puffs_msg_setinfo(park_readdir, PUFFSOP_VN,
1574 	    PUFFS_VN_READDIR, VPTOPNC(vp));
1575 	puffs_msg_setdelta(park_readdir, tomove);
1576 
1577 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readdir, vp->v_data, NULL, error);
1578 	error = checkerr(pmp, error, __func__);
1579 	if (error)
1580 		goto out;
1581 
1582 	/* userspace is cheating? */
1583 	if (readdir_msg->pvnr_resid > resid) {
1584 		puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1585 		    "resid grew", VPTOPNC(vp));
1586 		ERROUT(EPROTO);
1587 	}
1588 	if (readdir_msg->pvnr_ncookies > cookiesmax) {
1589 		puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1590 		    "too many cookies", VPTOPNC(vp));
1591 		ERROUT(EPROTO);
1592 	}
1593 
1594 	/* check eof */
1595 	if (readdir_msg->pvnr_eofflag)
1596 		*ap->a_eofflag = 1;
1597 
1598 	/* bouncy-wouncy with the directory data */
1599 	howmuch = resid - readdir_msg->pvnr_resid;
1600 
1601 	/* force eof if no data was returned (getcwd() needs this) */
1602 	if (howmuch == 0) {
1603 		*ap->a_eofflag = 1;
1604 		goto out;
1605 	}
1606 
1607 	error = uiomove(readdir_msg->pvnr_data + cookiemem, howmuch, uio);
1608 	if (error)
1609 		goto out;
1610 
1611 	/* provide cookies to caller if so desired */
1612 	if (ap->a_cookies) {
1613 		KASSERT(curlwp != uvm.pagedaemon_lwp);
1614 		*ap->a_cookies = malloc(readdir_msg->pvnr_ncookies*CSIZE,
1615 		    M_TEMP, M_WAITOK);
1616 		*ap->a_ncookies = readdir_msg->pvnr_ncookies;
1617 		memcpy(*ap->a_cookies, readdir_msg->pvnr_data,
1618 		    *ap->a_ncookies*CSIZE);
1619 	}
1620 
1621 	/* next readdir starts here */
1622 	uio->uio_offset = readdir_msg->pvnr_offset;
1623 
1624  out:
1625 	puffs_msgmem_release(park_readdir);
1626 	return error;
1627 }
1628 #undef CSIZE
1629 
1630 /*
1631  * poll works by consuming the bitmask in pn_revents.  If there are
1632  * events available, poll returns immediately.  If not, it issues a
1633  * poll to userspace, selrecords itself and returns with no available
1634  * events.  When the file server returns, it executes puffs_parkdone_poll(),
1635  * where available events are added to the bitmask.  selnotify() is
1636  * then also executed by that function causing us to enter here again
1637  * and hopefully find the missing bits (unless someone got them first,
1638  * in which case it starts all over again).
1639  */
1640 int
1641 puffs_vnop_poll(void *v)
1642 {
1643 	struct vop_poll_args /* {
1644 		const struct vnodeop_desc *a_desc;
1645 		struct vnode *a_vp;
1646 		int a_events;
1647 	} */ *ap = v;
1648 	PUFFS_MSG_VARS(vn, poll);
1649 	struct vnode *vp = ap->a_vp;
1650 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1651 	struct puffs_node *pn = vp->v_data;
1652 	int events;
1653 
1654 	if (EXISTSOP(pmp, POLL)) {
1655 		mutex_enter(&pn->pn_mtx);
1656 		events = pn->pn_revents & ap->a_events;
1657 		if (events & ap->a_events) {
1658 			pn->pn_revents &= ~ap->a_events;
1659 			mutex_exit(&pn->pn_mtx);
1660 
1661 			return events;
1662 		} else {
1663 			puffs_referencenode(pn);
1664 			mutex_exit(&pn->pn_mtx);
1665 
1666 			PUFFS_MSG_ALLOC(vn, poll);
1667 			poll_msg->pvnr_events = ap->a_events;
1668 			puffs_msg_setinfo(park_poll, PUFFSOP_VN,
1669 			    PUFFS_VN_POLL, VPTOPNC(vp));
1670 			puffs_msg_setcall(park_poll, puffs_parkdone_poll, pn);
1671 			selrecord(curlwp, &pn->pn_sel);
1672 
1673 			PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_poll,
1674 			    vp->v_data, NULL);
1675 			PUFFS_MSG_RELEASE(poll);
1676 
1677 			return 0;
1678 		}
1679 	} else {
1680 		return genfs_poll(v);
1681 	}
1682 }
1683 
1684 static int
1685 flushvncache(struct vnode *vp, off_t offlo, off_t offhi, bool wait)
1686 {
1687 	struct puffs_node *pn = VPTOPP(vp);
1688 	struct vattr va;
1689 	int pflags, error;
1690 
1691 	/* flush out information from our metacache, see vop_setattr */
1692 	if (pn->pn_stat & PNODE_METACACHE_MASK
1693 	    && (pn->pn_stat & PNODE_DYING) == 0) {
1694 		vattr_null(&va);
1695 		error = dosetattr(vp, &va, FSCRED,
1696 		    SETATTR_CHSIZE | (wait ? 0 : SETATTR_ASYNC));
1697 		if (error)
1698 			return error;
1699 	}
1700 
1701 	/*
1702 	 * flush pages to avoid being overly dirty
1703 	 */
1704 	pflags = PGO_CLEANIT;
1705 	if (wait)
1706 		pflags |= PGO_SYNCIO;
1707 
1708 	rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
1709 	return VOP_PUTPAGES(vp, trunc_page(offlo), round_page(offhi), pflags);
1710 }
1711 
1712 int
1713 puffs_vnop_fsync(void *v)
1714 {
1715 	struct vop_fsync_args /* {
1716 		const struct vnodeop_desc *a_desc;
1717 		struct vnode *a_vp;
1718 		kauth_cred_t a_cred;
1719 		int a_flags;
1720 		off_t a_offlo;
1721 		off_t a_offhi;
1722 	} */ *ap = v;
1723 	PUFFS_MSG_VARS(vn, fsync);
1724 	struct vnode *vp;
1725 	struct puffs_node *pn;
1726 	struct puffs_mount *pmp;
1727 	int error, dofaf;
1728 
1729 	vp = ap->a_vp;
1730 	KASSERT(vp != NULL);
1731 	pn = VPTOPP(vp);
1732 	KASSERT(pn != NULL);
1733 	pmp = MPTOPUFFSMP(vp->v_mount);
1734 
1735 	/* See the comment on top of puffs_vnop_inactive(). */
1736 	if (vp->v_type == VNON)
1737 		return 0;
1738 
1739 	if (ap->a_flags & FSYNC_WAIT) {
1740 		mutex_enter(&pn->pn_sizemtx);
1741 	} else {
1742 		if (mutex_tryenter(&pn->pn_sizemtx) == 0)
1743 			return EDEADLK;
1744 	}
1745 
1746 	error = flushvncache(vp, ap->a_offlo, ap->a_offhi,
1747 	    (ap->a_flags & FSYNC_WAIT) == FSYNC_WAIT);
1748 	if (error)
1749 		goto out;
1750 
1751 	/*
1752 	 * HELLO!  We exit already here if the user server does not
1753 	 * support fsync OR if we should call fsync for a node which
1754 	 * has references neither in the kernel or the fs server.
1755 	 * Otherwise we continue to issue fsync() forward.
1756 	 */
1757 	error = 0;
1758 	if (!EXISTSOP(pmp, FSYNC) || (pn->pn_stat & PNODE_DYING))
1759 		goto out;
1760 
1761 	dofaf = (ap->a_flags & FSYNC_WAIT) == 0 || ap->a_flags == FSYNC_LAZY;
1762 	/*
1763 	 * We abuse VXLOCK to mean "vnode is going to die", so we issue
1764 	 * only FAFs for those.  Otherwise there's a danger of deadlock,
1765 	 * since the execution context here might be the user server
1766 	 * doing some operation on another fs, which in turn caused a
1767 	 * vnode to be reclaimed from the freelist for this fs.
1768 	 */
1769 	if (dofaf == 0) {
1770 		mutex_enter(vp->v_interlock);
1771 		if (vdead_check(vp, VDEAD_NOWAIT) != 0)
1772 			dofaf = 1;
1773 		mutex_exit(vp->v_interlock);
1774 	}
1775 
1776 	PUFFS_MSG_ALLOC(vn, fsync);
1777 	if (dofaf)
1778 		puffs_msg_setfaf(park_fsync);
1779 
1780 	puffs_credcvt(&fsync_msg->pvnr_cred, ap->a_cred);
1781 	fsync_msg->pvnr_flags = ap->a_flags;
1782 	fsync_msg->pvnr_offlo = ap->a_offlo;
1783 	fsync_msg->pvnr_offhi = ap->a_offhi;
1784 	puffs_msg_setinfo(park_fsync, PUFFSOP_VN,
1785 	    PUFFS_VN_FSYNC, VPTOPNC(vp));
1786 
1787 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fsync, vp->v_data, NULL, error);
1788 	PUFFS_MSG_RELEASE(fsync);
1789 
1790 	error = checkerr(pmp, error, __func__);
1791 
1792 out:
1793 	mutex_exit(&pn->pn_sizemtx);
1794 	return error;
1795 }
1796 
1797 int
1798 puffs_vnop_seek(void *v)
1799 {
1800 	struct vop_seek_args /* {
1801 		const struct vnodeop_desc *a_desc;
1802 		struct vnode *a_vp;
1803 		off_t a_oldoff;
1804 		off_t a_newoff;
1805 		kauth_cred_t a_cred;
1806 	} */ *ap = v;
1807 	PUFFS_MSG_VARS(vn, seek);
1808 	struct vnode *vp = ap->a_vp;
1809 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1810 	int error;
1811 
1812 	PUFFS_MSG_ALLOC(vn, seek);
1813 	seek_msg->pvnr_oldoff = ap->a_oldoff;
1814 	seek_msg->pvnr_newoff = ap->a_newoff;
1815 	puffs_credcvt(&seek_msg->pvnr_cred, ap->a_cred);
1816 	puffs_msg_setinfo(park_seek, PUFFSOP_VN,
1817 	    PUFFS_VN_SEEK, VPTOPNC(vp));
1818 
1819 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_seek, vp->v_data, NULL, error);
1820 	PUFFS_MSG_RELEASE(seek);
1821 	return checkerr(pmp, error, __func__);
1822 }
1823 
1824 static int
1825 callremove(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1826 	struct componentname *cnp)
1827 {
1828 	PUFFS_MSG_VARS(vn, remove);
1829 	int error;
1830 
1831 	PUFFS_MSG_ALLOC(vn, remove);
1832 	remove_msg->pvnr_cookie_targ = ck;
1833 	puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1834 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1835 	puffs_msg_setinfo(park_remove, PUFFSOP_VN, PUFFS_VN_REMOVE, dck);
1836 
1837 	PUFFS_MSG_ENQUEUEWAIT(pmp, park_remove, error);
1838 	PUFFS_MSG_RELEASE(remove);
1839 
1840 	return checkerr(pmp, error, __func__);
1841 }
1842 
1843 /*
1844  * XXX: can't use callremove now because can't catch setbacks with
1845  * it due to lack of a pnode argument.
1846  */
1847 int
1848 puffs_vnop_remove(void *v)
1849 {
1850 	struct vop_remove_v2_args /* {
1851 		const struct vnodeop_desc *a_desc;
1852 		struct vnode *a_dvp;
1853 		struct vnode *a_vp;
1854 		struct componentname *a_cnp;
1855 	} */ *ap = v;
1856 	PUFFS_MSG_VARS(vn, remove);
1857 	struct vnode *dvp = ap->a_dvp;
1858 	struct vnode *vp = ap->a_vp;
1859 	struct puffs_node *dpn = VPTOPP(dvp);
1860 	struct puffs_node *pn = VPTOPP(vp);
1861 	struct componentname *cnp = ap->a_cnp;
1862 	struct mount *mp = dvp->v_mount;
1863 	struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1864 	int error;
1865 
1866 	PUFFS_MSG_ALLOC(vn, remove);
1867 	remove_msg->pvnr_cookie_targ = VPTOPNC(vp);
1868 	puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1869 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1870 	puffs_msg_setinfo(park_remove, PUFFSOP_VN,
1871 	    PUFFS_VN_REMOVE, VPTOPNC(dvp));
1872 
1873 	puffs_msg_enqueue(pmp, park_remove);
1874 	vref(dvp);		/* hang onto caller's reference at end */
1875 	REFPN(dpn);
1876 	if (dvp == vp)
1877 		REFPN(pn);
1878 	else
1879 		REFPN_AND_UNLOCKVP(vp, pn);
1880 	error = puffs_msg_wait2(pmp, park_remove, dpn, pn);
1881 
1882 	PUFFS_MSG_RELEASE(remove);
1883 
1884 	puffs_updatenode(VPTOPP(dvp), PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0);
1885 
1886 	RELEPN_AND_VP(dvp, dpn);
1887 	RELEPN_AND_VP(vp, pn);
1888 
1889 	error = checkerr(pmp, error, __func__);
1890 	return error;
1891 }
1892 
1893 int
1894 puffs_vnop_mkdir(void *v)
1895 {
1896 	struct vop_mkdir_v3_args /* {
1897 		const struct vnodeop_desc *a_desc;
1898 		struct vnode *a_dvp;
1899 		struct vnode **a_vpp;
1900 		struct componentname *a_cnp;
1901 		struct vattr *a_vap;
1902 	} */ *ap = v;
1903 	PUFFS_MSG_VARS(vn, mkdir);
1904 	struct vnode *dvp = ap->a_dvp;
1905 	struct puffs_node *dpn = VPTOPP(dvp);
1906 	struct componentname *cnp = ap->a_cnp;
1907 	struct mount *mp = dvp->v_mount;
1908 	struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1909 	int error;
1910 
1911 	PUFFS_MSG_ALLOC(vn, mkdir);
1912 	puffs_makecn(&mkdir_msg->pvnr_cn, &mkdir_msg->pvnr_cn_cred,
1913 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1914 	mkdir_msg->pvnr_va = *ap->a_vap;
1915 	puffs_msg_setinfo(park_mkdir, PUFFSOP_VN,
1916 	    PUFFS_VN_MKDIR, VPTOPNC(dvp));
1917 
1918 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mkdir, dvp->v_data, NULL, error);
1919 
1920 	error = checkerr(pmp, error, __func__);
1921 	if (error)
1922 		goto out;
1923 
1924 	error = puffs_newnode(mp, dvp, ap->a_vpp,
1925 	    mkdir_msg->pvnr_newnode, cnp, VDIR, 0);
1926 	if (error) {
1927 		puffs_abortbutton(pmp, PUFFS_ABORT_MKDIR, dpn->pn_cookie,
1928 		    mkdir_msg->pvnr_newnode, cnp);
1929 		goto out;
1930 	}
1931 
1932 	if (PUFFS_USE_FS_TTL(pmp)) {
1933 		struct timespec *va_ttl = &mkdir_msg->pvnr_va_ttl;
1934 		struct timespec *cn_ttl = &mkdir_msg->pvnr_cn_ttl;
1935 		struct vattr *rvap = &mkdir_msg->pvnr_va;
1936 
1937 		update_va(*ap->a_vpp, NULL, rvap,
1938 			  va_ttl, cn_ttl, SETATTR_CHSIZE);
1939 	}
1940 
1941 	VPTOPP(*ap->a_vpp)->pn_nlookup++;
1942 
1943 	if (PUFFS_USE_DOTDOTCACHE(pmp) &&
1944 	    (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
1945 		update_parent(*ap->a_vpp, dvp);
1946 
1947  out:
1948 	PUFFS_MSG_RELEASE(mkdir);
1949 	return error;
1950 }
1951 
1952 static int
1953 callrmdir(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1954 	struct componentname *cnp)
1955 {
1956 	PUFFS_MSG_VARS(vn, rmdir);
1957 	int error;
1958 
1959 	PUFFS_MSG_ALLOC(vn, rmdir);
1960 	rmdir_msg->pvnr_cookie_targ = ck;
1961 	puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1962 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1963 	puffs_msg_setinfo(park_rmdir, PUFFSOP_VN, PUFFS_VN_RMDIR, dck);
1964 
1965 	PUFFS_MSG_ENQUEUEWAIT(pmp, park_rmdir, error);
1966 	PUFFS_MSG_RELEASE(rmdir);
1967 
1968 	return checkerr(pmp, error, __func__);
1969 }
1970 
1971 int
1972 puffs_vnop_rmdir(void *v)
1973 {
1974 	struct vop_rmdir_v2_args /* {
1975 		const struct vnodeop_desc *a_desc;
1976 		struct vnode *a_dvp;
1977 		struct vnode *a_vp;
1978 		struct componentname *a_cnp;
1979 	} */ *ap = v;
1980 	PUFFS_MSG_VARS(vn, rmdir);
1981 	struct vnode *dvp = ap->a_dvp;
1982 	struct vnode *vp = ap->a_vp;
1983 	struct puffs_node *dpn = VPTOPP(dvp);
1984 	struct puffs_node *pn = VPTOPP(vp);
1985 	struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1986 	struct componentname *cnp = ap->a_cnp;
1987 	int error;
1988 
1989 	PUFFS_MSG_ALLOC(vn, rmdir);
1990 	rmdir_msg->pvnr_cookie_targ = VPTOPNC(vp);
1991 	puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1992 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1993 	puffs_msg_setinfo(park_rmdir, PUFFSOP_VN,
1994 	    PUFFS_VN_RMDIR, VPTOPNC(dvp));
1995 
1996 	puffs_msg_enqueue(pmp, park_rmdir);
1997 	vref(dvp);		/* hang onto caller's reference at end */
1998 	KASSERTMSG((dvp != vp), "rmdir .");
1999 	REFPN(dpn);
2000 	REFPN_AND_UNLOCKVP(vp, pn);
2001 	error = puffs_msg_wait2(pmp, park_rmdir, dpn, pn);
2002 
2003 	PUFFS_MSG_RELEASE(rmdir);
2004 
2005 	puffs_updatenode(VPTOPP(dvp), PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0);
2006 
2007 	/* XXX: some call cache_purge() *for both vnodes* here, investigate */
2008 	RELEPN_AND_VP(dvp, dpn);
2009 	RELEPN_AND_VP(vp, pn);
2010 
2011 	return error;
2012 }
2013 
2014 int
2015 puffs_vnop_link(void *v)
2016 {
2017 	struct vop_link_v2_args /* {
2018 		const struct vnodeop_desc *a_desc;
2019 		struct vnode *a_dvp;
2020 		struct vnode *a_vp;
2021 		struct componentname *a_cnp;
2022 	} */ *ap = v;
2023 	PUFFS_MSG_VARS(vn, link);
2024 	struct vnode *dvp = ap->a_dvp;
2025 	struct vnode *vp = ap->a_vp;
2026 	struct puffs_node *dpn = VPTOPP(dvp);
2027 	struct puffs_node *pn = VPTOPP(vp);
2028 	struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2029 	struct componentname *cnp = ap->a_cnp;
2030 	int error;
2031 
2032 	PUFFS_MSG_ALLOC(vn, link);
2033 	link_msg->pvnr_cookie_targ = VPTOPNC(vp);
2034 	puffs_makecn(&link_msg->pvnr_cn, &link_msg->pvnr_cn_cred,
2035 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
2036 	puffs_msg_setinfo(park_link, PUFFSOP_VN,
2037 	    PUFFS_VN_LINK, VPTOPNC(dvp));
2038 
2039 	puffs_msg_enqueue(pmp, park_link);
2040 	error = puffs_msg_wait2(pmp, park_link, dpn, pn);
2041 
2042 	PUFFS_MSG_RELEASE(link);
2043 
2044 	error = checkerr(pmp, error, __func__);
2045 
2046 	/*
2047 	 * XXX: stay in touch with the cache.  I don't like this, but
2048 	 * don't have a better solution either.  See also puffs_rename().
2049 	 */
2050 	if (error == 0) {
2051 		puffs_updatenode(pn, PUFFS_UPDATECTIME, 0);
2052 		puffs_updatenode(VPTOPP(dvp),
2053 				 PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0);
2054 	}
2055 
2056 	return error;
2057 }
2058 
2059 int
2060 puffs_vnop_symlink(void *v)
2061 {
2062 	struct vop_symlink_v3_args /* {
2063 		const struct vnodeop_desc *a_desc;
2064 		struct vnode *a_dvp;
2065 		struct vnode **a_vpp;
2066 		struct componentname *a_cnp;
2067 		struct vattr *a_vap;
2068 		char *a_target;
2069 	} */ *ap = v;
2070 	PUFFS_MSG_VARS(vn, symlink);
2071 	struct vnode *dvp = ap->a_dvp;
2072 	struct puffs_node *dpn = VPTOPP(dvp);
2073 	struct mount *mp = dvp->v_mount;
2074 	struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2075 	struct componentname *cnp = ap->a_cnp;
2076 	int error;
2077 
2078 	*ap->a_vpp = NULL;
2079 
2080 	PUFFS_MSG_ALLOC(vn, symlink);
2081 	puffs_makecn(&symlink_msg->pvnr_cn, &symlink_msg->pvnr_cn_cred,
2082 		cnp, PUFFS_USE_FULLPNBUF(pmp));
2083 	symlink_msg->pvnr_va = *ap->a_vap;
2084 	(void)strlcpy(symlink_msg->pvnr_link, ap->a_target,
2085 	    sizeof(symlink_msg->pvnr_link));
2086 	puffs_msg_setinfo(park_symlink, PUFFSOP_VN,
2087 	    PUFFS_VN_SYMLINK, VPTOPNC(dvp));
2088 
2089 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_symlink, dvp->v_data, NULL, error);
2090 
2091 	error = checkerr(pmp, error, __func__);
2092 	if (error)
2093 		goto out;
2094 
2095 	error = puffs_newnode(mp, dvp, ap->a_vpp,
2096 	    symlink_msg->pvnr_newnode, cnp, VLNK, 0);
2097 	if (error) {
2098 		puffs_abortbutton(pmp, PUFFS_ABORT_SYMLINK, dpn->pn_cookie,
2099 		    symlink_msg->pvnr_newnode, cnp);
2100 		goto out;
2101 	}
2102 
2103 	if (PUFFS_USE_FS_TTL(pmp)) {
2104 		struct timespec *va_ttl = &symlink_msg->pvnr_va_ttl;
2105 		struct timespec *cn_ttl = &symlink_msg->pvnr_cn_ttl;
2106 		struct vattr *rvap = &symlink_msg->pvnr_va;
2107 
2108 		update_va(*ap->a_vpp, NULL, rvap,
2109 			  va_ttl, cn_ttl, SETATTR_CHSIZE);
2110 	}
2111 
2112 	VPTOPP(*ap->a_vpp)->pn_nlookup++;
2113 
2114 	if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2115 	    (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
2116 		update_parent(*ap->a_vpp, dvp);
2117 
2118  out:
2119 	PUFFS_MSG_RELEASE(symlink);
2120 
2121 	return error;
2122 }
2123 
2124 int
2125 puffs_vnop_readlink(void *v)
2126 {
2127 	struct vop_readlink_args /* {
2128 		const struct vnodeop_desc *a_desc;
2129 		struct vnode *a_vp;
2130 		struct uio *a_uio;
2131 		kauth_cred_t a_cred;
2132 	} */ *ap = v;
2133 	PUFFS_MSG_VARS(vn, readlink);
2134 	struct vnode *vp = ap->a_vp;
2135 	struct puffs_mount *pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2136 	size_t linklen;
2137 	int error;
2138 
2139 	PUFFS_MSG_ALLOC(vn, readlink);
2140 	puffs_credcvt(&readlink_msg->pvnr_cred, ap->a_cred);
2141 	linklen = sizeof(readlink_msg->pvnr_link);
2142 	readlink_msg->pvnr_linklen = linklen;
2143 	puffs_msg_setinfo(park_readlink, PUFFSOP_VN,
2144 	    PUFFS_VN_READLINK, VPTOPNC(vp));
2145 
2146 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readlink, vp->v_data, NULL, error);
2147 	error = checkerr(pmp, error, __func__);
2148 	if (error)
2149 		goto out;
2150 
2151 	/* bad bad user file server */
2152 	if (readlink_msg->pvnr_linklen > linklen) {
2153 		puffs_senderr(pmp, PUFFS_ERR_READLINK, E2BIG,
2154 		    "linklen too big", VPTOPNC(ap->a_vp));
2155 		error = EPROTO;
2156 		goto out;
2157 	}
2158 
2159 	error = uiomove(&readlink_msg->pvnr_link, readlink_msg->pvnr_linklen,
2160 	    ap->a_uio);
2161  out:
2162 	PUFFS_MSG_RELEASE(readlink);
2163 	return error;
2164 }
2165 
2166 int
2167 puffs_vnop_rename(void *v)
2168 {
2169 	struct vop_rename_args /* {
2170 		const struct vnodeop_desc *a_desc;
2171 		struct vnode *a_fdvp;
2172 		struct vnode *a_fvp;
2173 		struct componentname *a_fcnp;
2174 		struct vnode *a_tdvp;
2175 		struct vnode *a_tvp;
2176 		struct componentname *a_tcnp;
2177 	} */ *ap = v;
2178 	PUFFS_MSG_VARS(vn, rename);
2179 	struct vnode *fdvp = ap->a_fdvp, *fvp = ap->a_fvp;
2180 	struct vnode *tdvp = ap->a_tdvp, *tvp = ap->a_tvp;
2181 	struct puffs_node *fpn = ap->a_fvp->v_data;
2182 	struct puffs_mount *pmp = MPTOPUFFSMP(fdvp->v_mount);
2183 	int error;
2184 	bool doabort = true;
2185 
2186 	if ((fvp->v_mount != tdvp->v_mount) ||
2187 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
2188 		ERROUT(EXDEV);
2189 	}
2190 
2191 	PUFFS_MSG_ALLOC(vn, rename);
2192 	rename_msg->pvnr_cookie_src = VPTOPNC(fvp);
2193 	rename_msg->pvnr_cookie_targdir = VPTOPNC(tdvp);
2194 	if (tvp)
2195 		rename_msg->pvnr_cookie_targ = VPTOPNC(tvp);
2196 	else
2197 		rename_msg->pvnr_cookie_targ = NULL;
2198 	puffs_makecn(&rename_msg->pvnr_cn_src, &rename_msg->pvnr_cn_src_cred,
2199 	    ap->a_fcnp, PUFFS_USE_FULLPNBUF(pmp));
2200 	puffs_makecn(&rename_msg->pvnr_cn_targ, &rename_msg->pvnr_cn_targ_cred,
2201 	    ap->a_tcnp, PUFFS_USE_FULLPNBUF(pmp));
2202 	puffs_msg_setinfo(park_rename, PUFFSOP_VN,
2203 	    PUFFS_VN_RENAME, VPTOPNC(fdvp));
2204 
2205 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rename, fdvp->v_data, NULL, error);
2206 	doabort = false;
2207 	PUFFS_MSG_RELEASE(rename);
2208 	error = checkerr(pmp, error, __func__);
2209 
2210 	/*
2211 	 * XXX: stay in touch with the cache.  I don't like this, but
2212 	 * don't have a better solution either.  See also puffs_link().
2213 	 */
2214 	if (error == 0) {
2215 		puffs_updatenode(fpn, PUFFS_UPDATECTIME, 0);
2216 		puffs_updatenode(VPTOPP(fdvp),
2217 				 PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0);
2218 		if (fdvp != tdvp)
2219 			puffs_updatenode(VPTOPP(tdvp),
2220 					 PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME,
2221 					 0);
2222 
2223 		if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2224 		    (VPTOPP(fvp)->pn_parent != tdvp))
2225 			update_parent(fvp, tdvp);
2226 	}
2227 
2228 
2229  out:
2230 	if (doabort)
2231 		VOP_ABORTOP(tdvp, ap->a_tcnp);
2232 	if (tvp != NULL)
2233 		vput(tvp);
2234 	if (tdvp == tvp)
2235 		vrele(tdvp);
2236 	else
2237 		vput(tdvp);
2238 
2239 	if (doabort)
2240 		VOP_ABORTOP(fdvp, ap->a_fcnp);
2241 	vrele(fdvp);
2242 	vrele(fvp);
2243 
2244 	return error;
2245 }
2246 
2247 #define RWARGS(cont, iofl, move, offset, creds)				\
2248 	(cont)->pvnr_ioflag = (iofl);					\
2249 	(cont)->pvnr_resid = (move);					\
2250 	(cont)->pvnr_offset = (offset);					\
2251 	puffs_credcvt(&(cont)->pvnr_cred, creds)
2252 
2253 int
2254 puffs_vnop_read(void *v)
2255 {
2256 	struct vop_read_args /* {
2257 		const struct vnodeop_desc *a_desc;
2258 		struct vnode *a_vp;
2259 		struct uio *a_uio;
2260 		int a_ioflag;
2261 		kauth_cred_t a_cred;
2262 	} */ *ap = v;
2263 	PUFFS_MSG_VARS(vn, read);
2264 	struct vnode *vp = ap->a_vp;
2265 	struct puffs_node *pn = VPTOPP(vp);
2266 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2267 	struct uio *uio = ap->a_uio;
2268 	size_t tomove, argsize;
2269 	vsize_t bytelen;
2270 	int error;
2271 
2272 	read_msg = NULL;
2273 	error = 0;
2274 
2275 	/* std sanity */
2276 	if (uio->uio_resid == 0)
2277 		return 0;
2278 	if (uio->uio_offset < 0)
2279 		return EFBIG;
2280 
2281 	/*
2282 	 * On the case of reading empty files and (vp->v_size != 0) below:
2283 	 * some filesystems (hint: FUSE and distributed filesystems) still
2284 	 * expect to get the READ in order to update atime. Reading through
2285 	 * the case filters empty files, therefore we prefer to bypass the
2286 	 * cache here.
2287 	 */
2288 	if (vp->v_type == VREG &&
2289 	    PUFFS_USE_PAGECACHE(pmp) &&
2290 	    !(pn->pn_stat & PNODE_RDIRECT) &&
2291 	    (vp->v_size != 0)) {
2292 		const int advice = IO_ADV_DECODE(ap->a_ioflag);
2293 
2294 		while (uio->uio_resid > 0) {
2295 			if (vp->v_size <= uio->uio_offset) {
2296 				break;
2297 			}
2298 			bytelen = MIN(uio->uio_resid,
2299 			    vp->v_size - uio->uio_offset);
2300 			if (bytelen == 0)
2301 				break;
2302 
2303 			error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
2304 			    UBC_READ | UBC_PARTIALOK | UBC_VNODE_FLAGS(vp));
2305 			if (error)
2306 				break;
2307 		}
2308 
2309 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
2310 			puffs_updatenode(VPTOPP(vp), PUFFS_UPDATEATIME, 0);
2311 	} else {
2312 		/*
2313 		 * in case it's not a regular file or we're operating
2314 		 * uncached, do read in the old-fashioned style,
2315 		 * i.e. explicit read operations
2316 		 */
2317 
2318 		tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2319 		argsize = sizeof(struct puffs_vnmsg_read);
2320 		puffs_msgmem_alloc(argsize + tomove, &park_read,
2321 		    (void *)&read_msg, 1);
2322 
2323 		error = 0;
2324 		while (uio->uio_resid > 0) {
2325 			tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2326 			memset(read_msg, 0, argsize); /* XXX: touser KASSERT */
2327 			RWARGS(read_msg, ap->a_ioflag, tomove,
2328 			    uio->uio_offset, ap->a_cred);
2329 			puffs_msg_setinfo(park_read, PUFFSOP_VN,
2330 			    PUFFS_VN_READ, VPTOPNC(vp));
2331 			puffs_msg_setdelta(park_read, tomove);
2332 
2333 			PUFFS_MSG_ENQUEUEWAIT2(pmp, park_read, vp->v_data,
2334 			    NULL, error);
2335 			error = checkerr(pmp, error, __func__);
2336 			if (error)
2337 				break;
2338 
2339 			if (read_msg->pvnr_resid > tomove) {
2340 				puffs_senderr(pmp, PUFFS_ERR_READ,
2341 				    E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2342 				error = EPROTO;
2343 				break;
2344 			}
2345 
2346 			error = uiomove(read_msg->pvnr_data,
2347 			    tomove - read_msg->pvnr_resid, uio);
2348 
2349 			/*
2350 			 * in case the file is out of juice, resid from
2351 			 * userspace is != 0.  and the error-case is
2352 			 * quite obvious
2353 			 */
2354 			if (error || read_msg->pvnr_resid)
2355 				break;
2356 		}
2357 
2358 		puffs_msgmem_release(park_read);
2359 	}
2360 
2361 	return error;
2362 }
2363 
2364 /*
2365  * XXX: in case of a failure, this leaves uio in a bad state.
2366  * We could theoretically copy the uio and iovecs and "replay"
2367  * them the right amount after the userspace trip, but don't
2368  * bother for now.
2369  */
2370 int
2371 puffs_vnop_write(void *v)
2372 {
2373 	struct vop_write_args /* {
2374 		const struct vnodeop_desc *a_desc;
2375 		struct vnode *a_vp;
2376 		struct uio *a_uio;
2377 		int a_ioflag;
2378 		kauth_cred_t a_cred;
2379 	} */ *ap = v;
2380 	PUFFS_MSG_VARS(vn, write);
2381 	struct vnode *vp = ap->a_vp;
2382 	struct puffs_node *pn = VPTOPP(vp);
2383 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2384 	struct uio *uio = ap->a_uio;
2385 	size_t tomove, argsize;
2386 	off_t oldoff, newoff, origoff;
2387 	vsize_t bytelen;
2388 	int error, uflags;
2389 	int ubcflags;
2390 
2391 	error = uflags = 0;
2392 	write_msg = NULL;
2393 
2394 	/* std sanity */
2395 	if (uio->uio_resid == 0)
2396 		return 0;
2397 	if (uio->uio_offset < 0)
2398 		return EFBIG;
2399 
2400 	mutex_enter(&pn->pn_sizemtx);
2401 
2402 	/*
2403 	 * userspace *should* be allowed to control this,
2404 	 * but with UBC it's a bit unclear how to handle it
2405 	 */
2406 	if (ap->a_ioflag & IO_APPEND)
2407 		uio->uio_offset = vp->v_size;
2408 
2409 	origoff = uio->uio_offset;
2410 
2411 	if (vp->v_type == VREG &&
2412 	    PUFFS_USE_PAGECACHE(pmp) &&
2413 	    !(pn->pn_stat & PNODE_WDIRECT)) {
2414 		ubcflags = UBC_WRITE | UBC_PARTIALOK | UBC_VNODE_FLAGS(vp);
2415 
2416 		while (uio->uio_resid > 0) {
2417 			oldoff = uio->uio_offset;
2418 			bytelen = uio->uio_resid;
2419 
2420 			newoff = oldoff + bytelen;
2421 			if (vp->v_size < newoff) {
2422 				uvm_vnp_setwritesize(vp, newoff);
2423 			}
2424 			error = ubc_uiomove(&vp->v_uobj, uio, bytelen,
2425 			    UVM_ADV_RANDOM, ubcflags);
2426 
2427 			/*
2428 			 * In case of a ubc_uiomove() error,
2429 			 * opt to not extend the file at all and
2430 			 * return an error.  Otherwise, if we attempt
2431 			 * to clear the memory we couldn't fault to,
2432 			 * we might generate a kernel page fault.
2433 			 */
2434 			if (vp->v_size < newoff) {
2435 				if (error == 0) {
2436 					uflags |= PUFFS_UPDATESIZE;
2437 					uvm_vnp_setsize(vp, newoff);
2438 				} else {
2439 					uvm_vnp_setwritesize(vp, vp->v_size);
2440 				}
2441 			}
2442 			if (error)
2443 				break;
2444 
2445 			/*
2446 			 * If we're writing large files, flush to file server
2447 			 * every 64k.  Otherwise we can very easily exhaust
2448 			 * kernel and user memory, as the file server cannot
2449 			 * really keep up with our writing speed.
2450 			 *
2451 			 * Note: this does *NOT* honor MNT_ASYNC, because
2452 			 * that gives userland too much say in the kernel.
2453 			 */
2454 			if (oldoff >> 16 != uio->uio_offset >> 16) {
2455 				rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
2456 				error = VOP_PUTPAGES(vp, oldoff & ~0xffff,
2457 				    uio->uio_offset & ~0xffff,
2458 				    PGO_CLEANIT | PGO_SYNCIO);
2459 				if (error)
2460 					break;
2461 			}
2462 		}
2463 
2464 		/* synchronous I/O? */
2465 		if (error == 0 && ap->a_ioflag & IO_SYNC) {
2466 			rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
2467 			error = VOP_PUTPAGES(vp, trunc_page(origoff),
2468 			    round_page(uio->uio_offset),
2469 			    PGO_CLEANIT | PGO_SYNCIO);
2470 
2471 		/* write through page cache? */
2472 		} else if (error == 0 && pmp->pmp_flags & PUFFS_KFLAG_WTCACHE) {
2473 			rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
2474 			error = VOP_PUTPAGES(vp, trunc_page(origoff),
2475 			    round_page(uio->uio_offset), PGO_CLEANIT);
2476 		}
2477 	} else {
2478 		/* tomove is non-increasing */
2479 		tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2480 		argsize = sizeof(struct puffs_vnmsg_write) + tomove;
2481 		puffs_msgmem_alloc(argsize, &park_write, (void *)&write_msg,1);
2482 
2483 		while (uio->uio_resid > 0) {
2484 			/* move data to buffer */
2485 			tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2486 			memset(write_msg, 0, argsize); /* XXX: touser KASSERT */
2487 			RWARGS(write_msg, ap->a_ioflag, tomove,
2488 			    uio->uio_offset, ap->a_cred);
2489 			error = uiomove(write_msg->pvnr_data, tomove, uio);
2490 			if (error)
2491 				break;
2492 
2493 			/* move buffer to userspace */
2494 			puffs_msg_setinfo(park_write, PUFFSOP_VN,
2495 			    PUFFS_VN_WRITE, VPTOPNC(vp));
2496 			PUFFS_MSG_ENQUEUEWAIT2(pmp, park_write, vp->v_data,
2497 			    NULL, error);
2498 			error = checkerr(pmp, error, __func__);
2499 			if (error)
2500 				break;
2501 
2502 			if (write_msg->pvnr_resid > tomove) {
2503 				puffs_senderr(pmp, PUFFS_ERR_WRITE,
2504 				    E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2505 				error = EPROTO;
2506 				break;
2507 			}
2508 
2509 			/* adjust file size */
2510 			if (vp->v_size < uio->uio_offset) {
2511 				uflags |= PUFFS_UPDATESIZE;
2512 				uvm_vnp_setsize(vp, uio->uio_offset);
2513 			}
2514 
2515 			/* didn't move everything?  bad userspace.  bail */
2516 			if (write_msg->pvnr_resid != 0) {
2517 				error = EIO;
2518 				break;
2519 			}
2520 		}
2521 		puffs_msgmem_release(park_write);
2522 
2523 		/*
2524 		 * Direct I/O on write but not on read: we must
2525 		 * invlidate the written pages so that we read
2526 		 * the written data and not the stalled cache.
2527 		 */
2528 		if ((error == 0) &&
2529 		    (vp->v_type == VREG) && PUFFS_USE_PAGECACHE(pmp) &&
2530 		    (pn->pn_stat & PNODE_WDIRECT) &&
2531 		    !(pn->pn_stat & PNODE_RDIRECT)) {
2532 			voff_t off_lo = trunc_page(origoff);
2533 			voff_t off_hi = round_page(uio->uio_offset);
2534 
2535 			rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
2536 			error = VOP_PUTPAGES(vp, off_lo, off_hi, PGO_FREE);
2537 		}
2538 	}
2539 
2540 	if (vp->v_mount->mnt_flag & MNT_RELATIME)
2541 		uflags |= PUFFS_UPDATEATIME;
2542 	uflags |= PUFFS_UPDATECTIME;
2543 	uflags |= PUFFS_UPDATEMTIME;
2544 	puffs_updatenode(VPTOPP(vp), uflags, vp->v_size);
2545 
2546 	/*
2547 	 * If we do not use meta flush, we need to update the
2548 	 * filesystem now, otherwise we will get a stale value
2549 	 * on the next GETATTR
2550 	 */
2551 	if (!PUFFS_USE_METAFLUSH(pmp) && (uflags & PUFFS_UPDATESIZE)) {
2552 		struct vattr va;
2553 		int ret;
2554 
2555 		vattr_null(&va);
2556 		va.va_size = vp->v_size;
2557 		ret = dosetattr(vp, &va, FSCRED, 0);
2558 		if (ret) {
2559 			DPRINTF(("dosetattr set size to %jd failed: %d\n",
2560 			    (intmax_t)vp->v_size, ret));
2561 		}
2562 	}
2563 	mutex_exit(&pn->pn_sizemtx);
2564 	return error;
2565 }
2566 
2567 int
2568 puffs_vnop_fallocate(void *v)
2569 {
2570 	struct vop_fallocate_args /* {
2571 		const struct vnodeop_desc *a_desc;
2572 		struct vnode *a_vp;
2573 		off_t a_pos;
2574 		off_t a_len;
2575 	} */ *ap = v;
2576 	struct vnode *vp = ap->a_vp;
2577 	struct puffs_node *pn = VPTOPP(vp);
2578 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2579 	PUFFS_MSG_VARS(vn, fallocate);
2580 	int error;
2581 
2582 	mutex_enter(&pn->pn_sizemtx);
2583 
2584 	PUFFS_MSG_ALLOC(vn, fallocate);
2585 	fallocate_msg->pvnr_off = ap->a_pos;
2586 	fallocate_msg->pvnr_len = ap->a_len;
2587 	puffs_msg_setinfo(park_fallocate, PUFFSOP_VN,
2588 	    PUFFS_VN_FALLOCATE, VPTOPNC(vp));
2589 
2590 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fallocate, vp->v_data, NULL, error);
2591 	error = checkerr(pmp, error, __func__);
2592 	PUFFS_MSG_RELEASE(fallocate);
2593 
2594 	switch (error) {
2595 	case 0:
2596 		break;
2597 	case EAGAIN:
2598 		error = EIO;
2599 		/* FALLTHROUGH */
2600 	default:
2601 		goto out;
2602 	}
2603 
2604 	if (ap->a_pos + ap->a_len > vp->v_size) {
2605 		uvm_vnp_setsize(vp, ap->a_pos + ap->a_len);
2606 		puffs_updatenode(pn, PUFFS_UPDATESIZE, vp->v_size);
2607 	}
2608 out:
2609  	mutex_exit(&pn->pn_sizemtx);
2610 
2611  	return error;
2612 }
2613 
2614 int
2615 puffs_vnop_fdiscard(void *v)
2616 {
2617 	struct vop_fdiscard_args /* {
2618 		const struct vnodeop_desc *a_desc;
2619 		struct vnode *a_vp;
2620 		off_t a_pos;
2621 		off_t a_len;
2622 	} */ *ap = v;
2623 	struct vnode *vp = ap->a_vp;
2624 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2625 	PUFFS_MSG_VARS(vn, fdiscard);
2626 	int error;
2627 
2628 	PUFFS_MSG_ALLOC(vn, fdiscard);
2629 	fdiscard_msg->pvnr_off = ap->a_pos;
2630 	fdiscard_msg->pvnr_len = ap->a_len;
2631 	puffs_msg_setinfo(park_fdiscard, PUFFSOP_VN,
2632 	    PUFFS_VN_FALLOCATE, VPTOPNC(vp));
2633 
2634 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fdiscard, vp->v_data, NULL, error);
2635 	error = checkerr(pmp, error, __func__);
2636 	PUFFS_MSG_RELEASE(fdiscard);
2637 
2638  	return error;
2639 }
2640 
2641 int
2642 puffs_vnop_print(void *v)
2643 {
2644 	struct vop_print_args /* {
2645 		struct vnode *a_vp;
2646 	} */ *ap = v;
2647 	PUFFS_MSG_VARS(vn, print);
2648 	struct vnode *vp = ap->a_vp;
2649 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2650 	struct puffs_node *pn = vp->v_data;
2651 
2652 	/* kernel portion */
2653 	printf("tag VT_PUFFS, vnode %p, puffs node: %p,\n"
2654 	    "\tuserspace cookie: %p", vp, pn, pn->pn_cookie);
2655 	if (vp->v_type == VFIFO)
2656 		VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
2657 	printf("\n");
2658 
2659 	/* userspace portion */
2660 	if (EXISTSOP(pmp, PRINT)) {
2661 		PUFFS_MSG_ALLOC(vn, print);
2662 		puffs_msg_setinfo(park_print, PUFFSOP_VN,
2663 		    PUFFS_VN_PRINT, VPTOPNC(vp));
2664 		PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_print, vp->v_data,
2665 		    NULL);
2666 		PUFFS_MSG_RELEASE(print);
2667 	}
2668 
2669 	return 0;
2670 }
2671 
2672 int
2673 puffs_vnop_pathconf(void *v)
2674 {
2675 	struct vop_pathconf_args /* {
2676 		const struct vnodeop_desc *a_desc;
2677 		struct vnode *a_vp;
2678 		int a_name;
2679 		register_t *a_retval;
2680 	} */ *ap = v;
2681 	PUFFS_MSG_VARS(vn, pathconf);
2682 	struct vnode *vp = ap->a_vp;
2683 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2684 	int error;
2685 
2686 	PUFFS_MSG_ALLOC(vn, pathconf);
2687 	pathconf_msg->pvnr_name = ap->a_name;
2688 	puffs_msg_setinfo(park_pathconf, PUFFSOP_VN,
2689 	    PUFFS_VN_PATHCONF, VPTOPNC(vp));
2690 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_pathconf, vp->v_data, NULL, error);
2691 	error = checkerr(pmp, error, __func__);
2692 	if (!error)
2693 		*ap->a_retval = pathconf_msg->pvnr_retval;
2694 	PUFFS_MSG_RELEASE(pathconf);
2695 
2696 	return error;
2697 }
2698 
2699 int
2700 puffs_vnop_advlock(void *v)
2701 {
2702 	struct vop_advlock_args /* {
2703 		const struct vnodeop_desc *a_desc;
2704 		struct vnode *a_vp;
2705 		void *a_id;
2706 		int a_op;
2707 		struct flock *a_fl;
2708 		int a_flags;
2709 	} */ *ap = v;
2710 	PUFFS_MSG_VARS(vn, advlock);
2711 	struct vnode *vp = ap->a_vp;
2712 	struct puffs_node *pn = VPTOPP(vp);
2713 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2714 	int error;
2715 
2716 	if (!EXISTSOP(pmp, ADVLOCK))
2717 		return lf_advlock(ap, &pn->pn_lockf, vp->v_size);
2718 
2719 	PUFFS_MSG_ALLOC(vn, advlock);
2720 	(void)memcpy(&advlock_msg->pvnr_fl, ap->a_fl,
2721 		     sizeof(advlock_msg->pvnr_fl));
2722 	advlock_msg->pvnr_id = ap->a_id;
2723 	advlock_msg->pvnr_op = ap->a_op;
2724 	advlock_msg->pvnr_flags = ap->a_flags;
2725 	puffs_msg_setinfo(park_advlock, PUFFSOP_VN,
2726 	    PUFFS_VN_ADVLOCK, VPTOPNC(vp));
2727 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_advlock, vp->v_data, NULL, error);
2728 	error = checkerr(pmp, error, __func__);
2729 	PUFFS_MSG_RELEASE(advlock);
2730 
2731 	return error;
2732 }
2733 
2734 int
2735 puffs_vnop_abortop(void *v)
2736 {
2737 	struct vop_abortop_args /* {
2738 		struct vnode *a_dvp;
2739 		struct componentname *a_cnp;
2740 	}; */ *ap = v;
2741 	PUFFS_MSG_VARS(vn, abortop);
2742 	struct vnode *dvp = ap->a_dvp;
2743 	struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2744 	struct componentname *cnp = ap->a_cnp;
2745 
2746 	if (EXISTSOP(pmp, ABORTOP)) {
2747 		PUFFS_MSG_ALLOC(vn, abortop);
2748 		puffs_makecn(&abortop_msg->pvnr_cn, &abortop_msg->pvnr_cn_cred,
2749 		    cnp, PUFFS_USE_FULLPNBUF(pmp));
2750 		puffs_msg_setfaf(park_abortop);
2751 		puffs_msg_setinfo(park_abortop, PUFFSOP_VN,
2752 		    PUFFS_VN_ABORTOP, VPTOPNC(dvp));
2753 
2754 		puffs_msg_enqueue(pmp, park_abortop);
2755 		PUFFS_MSG_RELEASE(abortop);
2756 	}
2757 
2758 	return genfs_abortop(v);
2759 }
2760 
2761 #define BIOASYNC(bp) (bp->b_flags & B_ASYNC)
2762 
2763 /*
2764  * This maps itself to PUFFS_VN_READ/WRITE for data transfer.
2765  */
2766 int
2767 puffs_vnop_strategy(void *v)
2768 {
2769 	struct vop_strategy_args /* {
2770 		const struct vnodeop_desc *a_desc;
2771 		struct vnode *a_vp;
2772 		struct buf *a_bp;
2773 	} */ *ap = v;
2774 	PUFFS_MSG_VARS(vn, rw);
2775 	struct vnode *vp = ap->a_vp;
2776 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2777 	struct puffs_node *pn;
2778 	struct buf *bp;
2779 	size_t argsize;
2780 	size_t tomove, moved;
2781 	int error, dofaf, cansleep, dobiodone;
2782 
2783 	pmp = MPTOPUFFSMP(vp->v_mount);
2784 	bp = ap->a_bp;
2785 	error = 0;
2786 	dofaf = 0;
2787 	cansleep = 0;
2788 	pn = VPTOPP(vp);
2789 	park_rw = NULL; /* explicit */
2790 	dobiodone = 1;
2791 
2792 	if ((BUF_ISREAD(bp) && !EXISTSOP(pmp, READ))
2793 	    || (BUF_ISWRITE(bp) && !EXISTSOP(pmp, WRITE)))
2794 		ERROUT(EOPNOTSUPP);
2795 
2796 	/*
2797 	 * Short-circuit optimization: don't flush buffer in between
2798 	 * VOP_INACTIVE and VOP_RECLAIM in case the node has no references.
2799 	 */
2800 	if (pn->pn_stat & PNODE_DYING) {
2801 		KASSERT(BUF_ISWRITE(bp));
2802 		bp->b_resid = 0;
2803 		goto out;
2804 	}
2805 
2806 #ifdef DIAGNOSTIC
2807 	if (bp->b_bcount > pmp->pmp_msg_maxsize - PUFFS_MSGSTRUCT_MAX)
2808 		panic("puffs_strategy: wildly inappropriate buf bcount %d",
2809 		    bp->b_bcount);
2810 #endif
2811 
2812 	/*
2813 	 * See explanation for the necessity of a FAF in puffs_fsync.
2814 	 *
2815 	 * Also, do FAF in case we're suspending.
2816 	 * See puffs_vfsops.c:pageflush()
2817 	 */
2818 	if (BUF_ISWRITE(bp)) {
2819 		mutex_enter(vp->v_interlock);
2820 		if (vdead_check(vp, VDEAD_NOWAIT) != 0)
2821 			dofaf = 1;
2822 		if (pn->pn_stat & PNODE_FAF)
2823 			dofaf = 1;
2824 		mutex_exit(vp->v_interlock);
2825 	}
2826 
2827 	cansleep = (curlwp == uvm.pagedaemon_lwp || dofaf) ? 0 : 1;
2828 
2829 	KASSERT(curlwp != uvm.pagedaemon_lwp || dofaf || BIOASYNC(bp));
2830 
2831 	/* allocate transport structure */
2832 	tomove = PUFFS_TOMOVE(bp->b_bcount, pmp);
2833 	argsize = sizeof(struct puffs_vnmsg_rw);
2834 	error = puffs_msgmem_alloc(argsize + tomove, &park_rw,
2835 	    (void *)&rw_msg, cansleep);
2836 	if (error)
2837 		goto out;
2838 	RWARGS(rw_msg, 0, tomove, bp->b_blkno << DEV_BSHIFT, FSCRED);
2839 
2840 	/* 2x2 cases: read/write, faf/nofaf */
2841 	if (BUF_ISREAD(bp)) {
2842 		puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2843 		    PUFFS_VN_READ, VPTOPNC(vp));
2844 		puffs_msg_setdelta(park_rw, tomove);
2845 		if (BIOASYNC(bp)) {
2846 			puffs_msg_setcall(park_rw,
2847 			    puffs_parkdone_asyncbioread, bp);
2848 			puffs_msg_enqueue(pmp, park_rw);
2849 			dobiodone = 0;
2850 		} else {
2851 			PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data,
2852 			    NULL, error);
2853 			error = checkerr(pmp, error, __func__);
2854 			if (error)
2855 				goto out;
2856 
2857 			if (rw_msg->pvnr_resid > tomove) {
2858 				puffs_senderr(pmp, PUFFS_ERR_READ,
2859 				    E2BIG, "resid grew", VPTOPNC(vp));
2860 				ERROUT(EPROTO);
2861 			}
2862 
2863 			moved = tomove - rw_msg->pvnr_resid;
2864 
2865 			(void)memcpy(bp->b_data, rw_msg->pvnr_data, moved);
2866 			bp->b_resid = bp->b_bcount - moved;
2867 		}
2868 	} else {
2869 		puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2870 		    PUFFS_VN_WRITE, VPTOPNC(vp));
2871 		/*
2872 		 * make pages read-only before we write them if we want
2873 		 * write caching info
2874 		 */
2875 		if (PUFFS_WCACHEINFO(pmp)) {
2876 			struct uvm_object *uobj = &vp->v_uobj;
2877 			int npages = (bp->b_bcount + PAGE_SIZE-1) >> PAGE_SHIFT;
2878 			struct vm_page *vmp;
2879 			int i;
2880 
2881 			for (i = 0; i < npages; i++) {
2882 				vmp= uvm_pageratop((vaddr_t)bp->b_data
2883 				    + (i << PAGE_SHIFT));
2884 				DPRINTF(("puffs_strategy: write-protecting "
2885 				    "vp %p page %p, offset %" PRId64"\n",
2886 				    vp, vmp, vmp->offset));
2887 				rw_enter(uobj->vmobjlock, RW_WRITER);
2888 				vmp->flags |= PG_RDONLY;
2889 				pmap_page_protect(vmp, VM_PROT_READ);
2890 				rw_exit(uobj->vmobjlock);
2891 			}
2892 		}
2893 
2894 		(void)memcpy(&rw_msg->pvnr_data, bp->b_data, tomove);
2895 		if (dofaf) {
2896 			puffs_msg_setfaf(park_rw);
2897 		} else if (BIOASYNC(bp)) {
2898 			puffs_msg_setcall(park_rw,
2899 			    puffs_parkdone_asyncbiowrite, bp);
2900 			dobiodone = 0;
2901 		}
2902 
2903 		PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data, NULL, error);
2904 
2905 		if (dobiodone == 0)
2906 			goto out;
2907 
2908 		error = checkerr(pmp, error, __func__);
2909 		if (error)
2910 			goto out;
2911 
2912 		if (rw_msg->pvnr_resid > tomove) {
2913 			puffs_senderr(pmp, PUFFS_ERR_WRITE,
2914 			    E2BIG, "resid grew", VPTOPNC(vp));
2915 			ERROUT(EPROTO);
2916 		}
2917 
2918 		/*
2919 		 * FAF moved everything.  Frankly, we don't
2920 		 * really have a choice.
2921 		 */
2922 		if (dofaf && error == 0)
2923 			moved = tomove;
2924 		else
2925 			moved = tomove - rw_msg->pvnr_resid;
2926 
2927 		bp->b_resid = bp->b_bcount - moved;
2928 		if (bp->b_resid != 0) {
2929 			ERROUT(EIO);
2930 		}
2931 	}
2932 
2933  out:
2934 	if (park_rw)
2935 		puffs_msgmem_release(park_rw);
2936 
2937 	if (error)
2938 		bp->b_error = error;
2939 
2940 	if (error || dobiodone)
2941 		biodone(bp);
2942 
2943 	return error;
2944 }
2945 
2946 int
2947 puffs_vnop_mmap(void *v)
2948 {
2949 	struct vop_mmap_args /* {
2950 		const struct vnodeop_desc *a_desc;
2951 		struct vnode *a_vp;
2952 		vm_prot_t a_prot;
2953 		kauth_cred_t a_cred;
2954 	} */ *ap = v;
2955 	PUFFS_MSG_VARS(vn, mmap);
2956 	struct vnode *vp = ap->a_vp;
2957 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2958 	int error;
2959 
2960 	if (!PUFFS_USE_PAGECACHE(pmp))
2961 		return genfs_eopnotsupp(v);
2962 
2963 	if (EXISTSOP(pmp, MMAP)) {
2964 		PUFFS_MSG_ALLOC(vn, mmap);
2965 		mmap_msg->pvnr_prot = ap->a_prot;
2966 		puffs_credcvt(&mmap_msg->pvnr_cred, ap->a_cred);
2967 		puffs_msg_setinfo(park_mmap, PUFFSOP_VN,
2968 		    PUFFS_VN_MMAP, VPTOPNC(vp));
2969 
2970 		PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mmap, vp->v_data, NULL, error);
2971 		error = checkerr(pmp, error, __func__);
2972 		PUFFS_MSG_RELEASE(mmap);
2973 	} else {
2974 		error = genfs_mmap(v);
2975 	}
2976 
2977 	return error;
2978 }
2979 
2980 
2981 /*
2982  * The rest don't get a free trip to userspace and back, they
2983  * have to stay within the kernel.
2984  */
2985 
2986 /*
2987  * bmap doesn't really make any sense for puffs, so just 1:1 map it.
2988  * well, maybe somehow, somewhere, some day ....
2989  */
2990 int
2991 puffs_vnop_bmap(void *v)
2992 {
2993 	struct vop_bmap_args /* {
2994 		const struct vnodeop_desc *a_desc;
2995 		struct vnode *a_vp;
2996 		daddr_t a_bn;
2997 		struct vnode **a_vpp;
2998 		daddr_t *a_bnp;
2999 		int *a_runp;
3000 	} */ *ap = v;
3001 	struct puffs_mount *pmp;
3002 
3003 	pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
3004 
3005 	if (ap->a_vpp)
3006 		*ap->a_vpp = ap->a_vp;
3007 	if (ap->a_bnp)
3008 		*ap->a_bnp = ap->a_bn;
3009 	if (ap->a_runp)
3010 		*ap->a_runp
3011 		    = (PUFFS_TOMOVE(pmp->pmp_msg_maxsize, pmp)>>DEV_BSHIFT) - 1;
3012 
3013 	return 0;
3014 }
3015 
3016 /*
3017  * Handle getpages faults in puffs.  We let genfs_getpages() do most
3018  * of the dirty work, but we come in this route to do accounting tasks.
3019  * If the user server has specified functions for cache notifications
3020  * about reads and/or writes, we record which type of operation we got,
3021  * for which page range, and proceed to issue a FAF notification to the
3022  * server about it.
3023  */
3024 int
3025 puffs_vnop_getpages(void *v)
3026 {
3027 	struct vop_getpages_args /* {
3028 		const struct vnodeop_desc *a_desc;
3029 		struct vnode *a_vp;
3030 		voff_t a_offset;
3031 		struct vm_page **a_m;
3032 		int *a_count;
3033 		int a_centeridx;
3034 		vm_prot_t a_access_type;
3035 		int a_advice;
3036 		int a_flags;
3037 	} */ *ap = v;
3038 	struct puffs_mount *pmp;
3039 	struct puffs_node *pn;
3040 	struct vnode *vp;
3041 	struct vm_page **pgs;
3042 	struct puffs_cacheinfo *pcinfo = NULL;
3043 	struct puffs_cacherun *pcrun;
3044 	void *parkmem = NULL;
3045 	size_t runsizes;
3046 	int i, npages, si, streakon;
3047 	int error, locked, write;
3048 
3049 	pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
3050 	npages = *ap->a_count;
3051 	pgs = ap->a_m;
3052 	vp = ap->a_vp;
3053 	pn = vp->v_data;
3054 	locked = (ap->a_flags & PGO_LOCKED) != 0;
3055 	write = (ap->a_access_type & VM_PROT_WRITE) != 0;
3056 
3057 	/* ccg xnaht - gets Wuninitialized wrong */
3058 	pcrun = NULL;
3059 	runsizes = 0;
3060 
3061 	/*
3062 	 * Check that we aren't trying to fault in pages which our file
3063 	 * server doesn't know about.  This happens if we extend a file by
3064 	 * skipping some pages and later try to fault in pages which
3065 	 * are between pn_serversize and vp_size.  This check optimizes
3066 	 * away the common case where a file is being extended.
3067 	 */
3068 	if (ap->a_offset >= pn->pn_serversize && ap->a_offset < vp->v_size) {
3069 		struct vattr va;
3070 
3071 		/* try again later when we can block */
3072 		if (locked)
3073 			ERROUT(EBUSY);
3074 
3075 		rw_exit(vp->v_uobj.vmobjlock);
3076 		vattr_null(&va);
3077 		va.va_size = vp->v_size;
3078 		error = dosetattr(vp, &va, FSCRED, 0);
3079 		if (error)
3080 			ERROUT(error);
3081 		rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
3082 	}
3083 
3084 	if (write && PUFFS_WCACHEINFO(pmp)) {
3085 #ifdef notnowjohn
3086 		/* allocate worst-case memory */
3087 		runsizes = ((npages / 2) + 1) * sizeof(struct puffs_cacherun);
3088 		KASSERT(curlwp != uvm.pagedaemon_lwp || locked);
3089 		pcinfo = kmem_zalloc(sizeof(struct puffs_cacheinfo) + runsize,
3090 		    locked ? KM_NOSLEEP : KM_SLEEP);
3091 
3092 		/*
3093 		 * can't block if we're locked and can't mess up caching
3094 		 * information for fs server.  so come back later, please
3095 		 */
3096 		if (pcinfo == NULL)
3097 			ERROUT(ENOMEM);
3098 
3099 		parkmem = puffs_park_alloc(locked == 0);
3100 		if (parkmem == NULL)
3101 			ERROUT(ENOMEM);
3102 
3103 		pcrun = pcinfo->pcache_runs;
3104 #else
3105 		(void)parkmem;
3106 #endif
3107 	}
3108 
3109 	error = genfs_getpages(v);
3110 	if (error)
3111 		goto out;
3112 
3113 	if (PUFFS_WCACHEINFO(pmp) == 0)
3114 		goto out;
3115 
3116 	/*
3117 	 * Let's see whose fault it was and inform the user server of
3118 	 * possibly read/written pages.  Map pages from read faults
3119 	 * strictly read-only, since otherwise we might miss info on
3120 	 * when the page is actually write-faulted to.
3121 	 */
3122 	if (!locked)
3123 		rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
3124 	for (i = 0, si = 0, streakon = 0; i < npages; i++) {
3125 		if (pgs[i] == NULL || pgs[i] == PGO_DONTCARE) {
3126 			if (streakon && write) {
3127 				streakon = 0;
3128 				pcrun[si].pcache_runend
3129 				    = trunc_page(pgs[i]->offset) + PAGE_MASK;
3130 				si++;
3131 			}
3132 			continue;
3133 		}
3134 		if (streakon == 0 && write) {
3135 			streakon = 1;
3136 			pcrun[si].pcache_runstart = pgs[i]->offset;
3137 		}
3138 
3139 		if (!write)
3140 			pgs[i]->flags |= PG_RDONLY;
3141 	}
3142 	/* was the last page part of our streak? */
3143 	if (streakon) {
3144 		pcrun[si].pcache_runend
3145 		    = trunc_page(pgs[i-1]->offset) + PAGE_MASK;
3146 		si++;
3147 	}
3148 	if (!locked)
3149 		rw_exit(vp->v_uobj.vmobjlock);
3150 
3151 	KASSERT(si <= (npages / 2) + 1);
3152 
3153 #ifdef notnowjohn
3154 	/* send results to userspace */
3155 	if (write)
3156 		puffs_cacheop(pmp, parkmem, pcinfo,
3157 		    sizeof(struct puffs_cacheinfo) + runsizes, VPTOPNC(vp));
3158 #endif
3159 
3160  out:
3161 	if (error) {
3162 		if (pcinfo != NULL)
3163 			kmem_free(pcinfo,
3164 			    sizeof(struct puffs_cacheinfo) + runsizes);
3165 #ifdef notnowjohn
3166 		if (parkmem != NULL)
3167 			puffs_park_release(parkmem, 1);
3168 #endif
3169 	}
3170 
3171 	return error;
3172 }
3173 
3174 /*
3175  * Extended attribute support.
3176  */
3177 
3178 int
3179 puffs_vnop_getextattr(void *v)
3180 {
3181 	struct vop_getextattr_args /*
3182 		struct vnode *a_vp;
3183 		int a_attrnamespace;
3184 		const char *a_name;
3185 		struct uio *a_uio;
3186 		size_t *a_size;
3187 		kauth_cred_t a_cred;
3188 	}; */ *ap = v;
3189 	PUFFS_MSG_VARS(vn, getextattr);
3190 	struct vnode *vp = ap->a_vp;
3191 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3192 	int attrnamespace = ap->a_attrnamespace;
3193 	const char *name = ap->a_name;
3194 	struct uio *uio = ap->a_uio;
3195 	size_t *sizep = ap->a_size;
3196 	size_t tomove, resid;
3197 	int error;
3198 
3199 	if (uio)
3200 		resid = uio->uio_resid;
3201 	else
3202 		resid = 0;
3203 
3204 	tomove = PUFFS_TOMOVE(resid, pmp);
3205 	if (tomove != resid) {
3206 		error = E2BIG;
3207 		goto out;
3208 	}
3209 
3210 	puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_getextattr) + tomove,
3211 	    &park_getextattr, (void *)&getextattr_msg, 1);
3212 
3213 	getextattr_msg->pvnr_attrnamespace = attrnamespace;
3214 	strlcpy(getextattr_msg->pvnr_attrname, name,
3215 	    sizeof(getextattr_msg->pvnr_attrname));
3216 	puffs_credcvt(&getextattr_msg->pvnr_cred, ap->a_cred);
3217 	if (sizep)
3218 		getextattr_msg->pvnr_datasize = 1;
3219 	getextattr_msg->pvnr_resid = tomove;
3220 
3221 	puffs_msg_setinfo(park_getextattr,
3222 	    PUFFSOP_VN, PUFFS_VN_GETEXTATTR, VPTOPNC(vp));
3223 	puffs_msg_setdelta(park_getextattr, tomove);
3224 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getextattr, vp->v_data, NULL, error);
3225 
3226 	error = checkerr(pmp, error, __func__);
3227 	if (error)
3228 		goto out;
3229 
3230 	resid = getextattr_msg->pvnr_resid;
3231 	if (resid > tomove) {
3232 		puffs_senderr(pmp, PUFFS_ERR_GETEXTATTR, E2BIG,
3233 		    "resid grew", VPTOPNC(vp));
3234 		error = EPROTO;
3235 		goto out;
3236 	}
3237 
3238 	if (sizep)
3239 		*sizep = getextattr_msg->pvnr_datasize;
3240 	if (uio)
3241 		error = uiomove(getextattr_msg->pvnr_data, tomove - resid, uio);
3242 
3243  out:
3244 	PUFFS_MSG_RELEASE(getextattr);
3245 	return error;
3246 }
3247 
3248 int
3249 puffs_vnop_setextattr(void *v)
3250 {
3251 	struct vop_setextattr_args /* {
3252 		struct vnode *a_vp;
3253 		int a_attrnamespace;
3254 		const char *a_name;
3255 		struct uio *a_uio;
3256 		kauth_cred_t a_cred;
3257 	}; */ *ap = v;
3258 	PUFFS_MSG_VARS(vn, setextattr);
3259 	struct vnode *vp = ap->a_vp;
3260 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3261 	int attrnamespace = ap->a_attrnamespace;
3262 	const char *name = ap->a_name;
3263 	struct uio *uio = ap->a_uio;
3264 	size_t tomove, resid;
3265 	int error;
3266 
3267 	if (uio)
3268 		resid = uio->uio_resid;
3269 	else
3270 		resid = 0;
3271 
3272 	tomove = PUFFS_TOMOVE(resid, pmp);
3273 	if (tomove != resid) {
3274 		error = E2BIG;
3275 		goto out;
3276 	}
3277 
3278 	puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_setextattr) + tomove,
3279 	    &park_setextattr, (void *)&setextattr_msg, 1);
3280 
3281 	setextattr_msg->pvnr_attrnamespace = attrnamespace;
3282 	strlcpy(setextattr_msg->pvnr_attrname, name,
3283 	    sizeof(setextattr_msg->pvnr_attrname));
3284 	puffs_credcvt(&setextattr_msg->pvnr_cred, ap->a_cred);
3285 	setextattr_msg->pvnr_resid = tomove;
3286 
3287 	if (uio) {
3288 		error = uiomove(setextattr_msg->pvnr_data, tomove, uio);
3289 		if (error)
3290 			goto out;
3291 	}
3292 
3293 	puffs_msg_setinfo(park_setextattr,
3294 	    PUFFSOP_VN, PUFFS_VN_SETEXTATTR, VPTOPNC(vp));
3295 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_setextattr, vp->v_data, NULL, error);
3296 
3297 	error = checkerr(pmp, error, __func__);
3298 	if (error)
3299 		goto out;
3300 
3301 	if (setextattr_msg->pvnr_resid != 0)
3302 		error = EIO;
3303 
3304  out:
3305 	PUFFS_MSG_RELEASE(setextattr);
3306 
3307 	return error;
3308 }
3309 
3310 int
3311 puffs_vnop_listextattr(void *v)
3312 {
3313 	struct vop_listextattr_args /* {
3314 		struct vnode *a_vp;
3315 		int a_attrnamespace;
3316 		struct uio *a_uio;
3317 		size_t *a_size;
3318 		int a_flag,
3319 		kauth_cred_t a_cred;
3320 	}; */ *ap = v;
3321 	PUFFS_MSG_VARS(vn, listextattr);
3322 	struct vnode *vp = ap->a_vp;
3323 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3324 	int attrnamespace = ap->a_attrnamespace;
3325 	struct uio *uio = ap->a_uio;
3326 	size_t *sizep = ap->a_size;
3327 	int flag = ap->a_flag;
3328 	size_t tomove, resid;
3329 	int error;
3330 
3331 	if (uio)
3332 		resid = uio->uio_resid;
3333 	else
3334 		resid = 0;
3335 
3336 	tomove = PUFFS_TOMOVE(resid, pmp);
3337 	if (tomove != resid) {
3338 		error = E2BIG;
3339 		goto out;
3340 	}
3341 
3342 	puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_listextattr) + tomove,
3343 	    &park_listextattr, (void *)&listextattr_msg, 1);
3344 
3345 	listextattr_msg->pvnr_attrnamespace = attrnamespace;
3346 	listextattr_msg->pvnr_flag = flag;
3347 	puffs_credcvt(&listextattr_msg->pvnr_cred, ap->a_cred);
3348 	listextattr_msg->pvnr_resid = tomove;
3349 	if (sizep)
3350 		listextattr_msg->pvnr_datasize = 1;
3351 
3352 	puffs_msg_setinfo(park_listextattr,
3353 	    PUFFSOP_VN, PUFFS_VN_LISTEXTATTR, VPTOPNC(vp));
3354 	puffs_msg_setdelta(park_listextattr, tomove);
3355 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_listextattr, vp->v_data, NULL, error);
3356 
3357 	error = checkerr(pmp, error, __func__);
3358 	if (error)
3359 		goto out;
3360 
3361 	resid = listextattr_msg->pvnr_resid;
3362 	if (resid > tomove) {
3363 		puffs_senderr(pmp, PUFFS_ERR_LISTEXTATTR, E2BIG,
3364 		    "resid grew", VPTOPNC(vp));
3365 		error = EPROTO;
3366 		goto out;
3367 	}
3368 
3369 	if (sizep)
3370 		*sizep = listextattr_msg->pvnr_datasize;
3371 	if (uio)
3372 		error = uiomove(listextattr_msg->pvnr_data, tomove-resid, uio);
3373 
3374  out:
3375 	PUFFS_MSG_RELEASE(listextattr);
3376 	return error;
3377 }
3378 
3379 int
3380 puffs_vnop_deleteextattr(void *v)
3381 {
3382 	struct vop_deleteextattr_args /* {
3383 		struct vnode *a_vp;
3384 		int a_attrnamespace;
3385 		const char *a_name;
3386 		kauth_cred_t a_cred;
3387 	}; */ *ap = v;
3388 	PUFFS_MSG_VARS(vn, deleteextattr);
3389 	struct vnode *vp = ap->a_vp;
3390 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3391 	int attrnamespace = ap->a_attrnamespace;
3392 	const char *name = ap->a_name;
3393 	int error;
3394 
3395 	PUFFS_MSG_ALLOC(vn, deleteextattr);
3396 	deleteextattr_msg->pvnr_attrnamespace = attrnamespace;
3397 	strlcpy(deleteextattr_msg->pvnr_attrname, name,
3398 	    sizeof(deleteextattr_msg->pvnr_attrname));
3399 	puffs_credcvt(&deleteextattr_msg->pvnr_cred, ap->a_cred);
3400 
3401 	puffs_msg_setinfo(park_deleteextattr,
3402 	    PUFFSOP_VN, PUFFS_VN_DELETEEXTATTR, VPTOPNC(vp));
3403 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_deleteextattr,
3404 	    vp->v_data, NULL, error);
3405 
3406 	error = checkerr(pmp, error, __func__);
3407 
3408 	PUFFS_MSG_RELEASE(deleteextattr);
3409 	return error;
3410 }
3411 
3412 /*
3413  * spec & fifo.  These call the miscfs spec and fifo vectors, but issue
3414  * FAF update information for the puffs node first.
3415  */
3416 int
3417 puffs_vnop_spec_read(void *v)
3418 {
3419 	struct vop_read_args /* {
3420 		const struct vnodeop_desc *a_desc;
3421 		struct vnode *a_vp;
3422 		struct uio *a_uio;
3423 		int a_ioflag;
3424 		kauth_cred_t a_cred;
3425 	} */ *ap = v;
3426 
3427 	puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3428 	return VOCALL(spec_vnodeop_p, VOFFSET(vop_read), v);
3429 }
3430 
3431 int
3432 puffs_vnop_spec_write(void *v)
3433 {
3434 	struct vop_write_args /* {
3435 		const struct vnodeop_desc *a_desc;
3436 		struct vnode *a_vp;
3437 		struct uio *a_uio;
3438 		int a_ioflag;
3439 		kauth_cred_t a_cred;
3440 	} */ *ap = v;
3441 
3442 	puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3443 	return VOCALL(spec_vnodeop_p, VOFFSET(vop_write), v);
3444 }
3445 
3446 int
3447 puffs_vnop_fifo_read(void *v)
3448 {
3449 	struct vop_read_args /* {
3450 		const struct vnodeop_desc *a_desc;
3451 		struct vnode *a_vp;
3452 		struct uio *a_uio;
3453 		int a_ioflag;
3454 		kauth_cred_t a_cred;
3455 	} */ *ap = v;
3456 
3457 	puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3458 	return VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), v);
3459 }
3460 
3461 int
3462 puffs_vnop_fifo_write(void *v)
3463 {
3464 	struct vop_write_args /* {
3465 		const struct vnodeop_desc *a_desc;
3466 		struct vnode *a_vp;
3467 		struct uio *a_uio;
3468 		int a_ioflag;
3469 		kauth_cred_t a_cred;
3470 	} */ *ap = v;
3471 
3472 	puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3473 	return VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), v);
3474 }
3475