xref: /netbsd-src/sys/fs/puffs/puffs_vnops.c (revision 6fc217346bb51c463d3a5a2a7883cb56515cd6d7)
1 /*	$NetBSD: puffs_vnops.c,v 1.142 2010/01/14 14:44:13 pooka 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 1.142 2010/01/14 14:44:13 pooka Exp $");
34 
35 #include <sys/param.h>
36 #include <sys/buf.h>
37 #include <sys/malloc.h>
38 #include <sys/mount.h>
39 #include <sys/namei.h>
40 #include <sys/vnode.h>
41 #include <sys/proc.h>
42 
43 #include <uvm/uvm.h>
44 
45 #include <fs/puffs/puffs_msgif.h>
46 #include <fs/puffs/puffs_sys.h>
47 
48 #include <miscfs/fifofs/fifo.h>
49 #include <miscfs/genfs/genfs.h>
50 #include <miscfs/specfs/specdev.h>
51 
52 int	puffs_vnop_lookup(void *);
53 int	puffs_vnop_create(void *);
54 int	puffs_vnop_access(void *);
55 int	puffs_vnop_mknod(void *);
56 int	puffs_vnop_open(void *);
57 int	puffs_vnop_close(void *);
58 int	puffs_vnop_getattr(void *);
59 int	puffs_vnop_setattr(void *);
60 int	puffs_vnop_reclaim(void *);
61 int	puffs_vnop_readdir(void *);
62 int	puffs_vnop_poll(void *);
63 int	puffs_vnop_fsync(void *);
64 int	puffs_vnop_seek(void *);
65 int	puffs_vnop_remove(void *);
66 int	puffs_vnop_mkdir(void *);
67 int	puffs_vnop_rmdir(void *);
68 int	puffs_vnop_link(void *);
69 int	puffs_vnop_readlink(void *);
70 int	puffs_vnop_symlink(void *);
71 int	puffs_vnop_rename(void *);
72 int	puffs_vnop_read(void *);
73 int	puffs_vnop_write(void *);
74 int	puffs_vnop_fcntl(void *);
75 int	puffs_vnop_ioctl(void *);
76 int	puffs_vnop_inactive(void *);
77 int	puffs_vnop_print(void *);
78 int	puffs_vnop_pathconf(void *);
79 int	puffs_vnop_advlock(void *);
80 int	puffs_vnop_strategy(void *);
81 int	puffs_vnop_bmap(void *);
82 int	puffs_vnop_mmap(void *);
83 int	puffs_vnop_getpages(void *);
84 int	puffs_vnop_abortop(void *);
85 
86 int	puffs_vnop_spec_read(void *);
87 int	puffs_vnop_spec_write(void *);
88 int	puffs_vnop_fifo_read(void *);
89 int	puffs_vnop_fifo_write(void *);
90 
91 int	puffs_vnop_checkop(void *);
92 
93 #define puffs_vnop_lock genfs_lock
94 #define puffs_vnop_unlock genfs_unlock
95 #define puffs_vnop_islocked genfs_islocked
96 
97 int (**puffs_vnodeop_p)(void *);
98 const struct vnodeopv_entry_desc puffs_vnodeop_entries[] = {
99 	{ &vop_default_desc, vn_default_error },
100 	{ &vop_lookup_desc, puffs_vnop_lookup },	/* REAL lookup */
101 	{ &vop_create_desc, puffs_vnop_checkop },	/* create */
102         { &vop_mknod_desc, puffs_vnop_checkop },	/* mknod */
103         { &vop_open_desc, puffs_vnop_open },		/* REAL open */
104         { &vop_close_desc, puffs_vnop_checkop },	/* close */
105         { &vop_access_desc, puffs_vnop_access },	/* REAL access */
106         { &vop_getattr_desc, puffs_vnop_checkop },	/* getattr */
107         { &vop_setattr_desc, puffs_vnop_checkop },	/* setattr */
108         { &vop_read_desc, puffs_vnop_checkop },		/* read */
109         { &vop_write_desc, puffs_vnop_checkop },	/* write */
110         { &vop_fsync_desc, puffs_vnop_fsync },		/* REAL fsync */
111         { &vop_seek_desc, puffs_vnop_checkop },		/* seek */
112         { &vop_remove_desc, puffs_vnop_checkop },	/* remove */
113         { &vop_link_desc, puffs_vnop_checkop },		/* link */
114         { &vop_rename_desc, puffs_vnop_checkop },	/* rename */
115         { &vop_mkdir_desc, puffs_vnop_checkop },	/* mkdir */
116         { &vop_rmdir_desc, puffs_vnop_checkop },	/* rmdir */
117         { &vop_symlink_desc, puffs_vnop_checkop },	/* symlink */
118         { &vop_readdir_desc, puffs_vnop_checkop },	/* readdir */
119         { &vop_readlink_desc, puffs_vnop_checkop },	/* readlink */
120         { &vop_getpages_desc, puffs_vnop_checkop },	/* getpages */
121         { &vop_putpages_desc, genfs_putpages },		/* REAL putpages */
122         { &vop_pathconf_desc, puffs_vnop_checkop },	/* pathconf */
123         { &vop_advlock_desc, puffs_vnop_checkop },	/* advlock */
124         { &vop_strategy_desc, puffs_vnop_strategy },	/* REAL strategy */
125         { &vop_revoke_desc, genfs_revoke },		/* REAL revoke */
126         { &vop_abortop_desc, puffs_vnop_abortop },	/* REAL abortop */
127         { &vop_inactive_desc, puffs_vnop_inactive },	/* REAL inactive */
128         { &vop_reclaim_desc, puffs_vnop_reclaim },	/* REAL reclaim */
129         { &vop_lock_desc, puffs_vnop_lock },		/* REAL lock */
130         { &vop_unlock_desc, puffs_vnop_unlock },	/* REAL unlock */
131         { &vop_bmap_desc, puffs_vnop_bmap },		/* REAL bmap */
132         { &vop_print_desc, puffs_vnop_print },		/* REAL print */
133         { &vop_islocked_desc, puffs_vnop_islocked },	/* REAL islocked */
134         { &vop_bwrite_desc, genfs_nullop },		/* REAL bwrite */
135         { &vop_mmap_desc, puffs_vnop_mmap },		/* REAL mmap */
136         { &vop_poll_desc, puffs_vnop_poll },		/* REAL poll */
137 
138         { &vop_kqfilter_desc, genfs_eopnotsupp },	/* kqfilter XXX */
139 	{ NULL, NULL }
140 };
141 const struct vnodeopv_desc puffs_vnodeop_opv_desc =
142 	{ &puffs_vnodeop_p, puffs_vnodeop_entries };
143 
144 
145 int (**puffs_specop_p)(void *);
146 const struct vnodeopv_entry_desc puffs_specop_entries[] = {
147 	{ &vop_default_desc, vn_default_error },
148 	{ &vop_lookup_desc, spec_lookup },		/* lookup, ENOTDIR */
149 	{ &vop_create_desc, spec_create },		/* genfs_badop */
150 	{ &vop_mknod_desc, spec_mknod },		/* genfs_badop */
151 	{ &vop_open_desc, spec_open },			/* spec_open */
152 	{ &vop_close_desc, spec_close },		/* spec_close */
153 	{ &vop_access_desc, puffs_vnop_checkop },	/* access */
154 	{ &vop_getattr_desc, puffs_vnop_checkop },	/* getattr */
155 	{ &vop_setattr_desc, puffs_vnop_checkop },	/* setattr */
156 	{ &vop_read_desc, puffs_vnop_spec_read },	/* update, read */
157 	{ &vop_write_desc, puffs_vnop_spec_write },	/* update, write */
158 	{ &vop_ioctl_desc, spec_ioctl },		/* spec_ioctl */
159 	{ &vop_fcntl_desc, genfs_fcntl },		/* dummy */
160 	{ &vop_poll_desc, spec_poll },			/* spec_poll */
161 	{ &vop_kqfilter_desc, spec_kqfilter },		/* spec_kqfilter */
162 	{ &vop_revoke_desc, spec_revoke },		/* genfs_revoke */
163 	{ &vop_mmap_desc, spec_mmap },			/* spec_mmap */
164 	{ &vop_fsync_desc, spec_fsync },		/* vflushbuf */
165 	{ &vop_seek_desc, spec_seek },			/* genfs_nullop */
166 	{ &vop_remove_desc, spec_remove },		/* genfs_badop */
167 	{ &vop_link_desc, spec_link },			/* genfs_badop */
168 	{ &vop_rename_desc, spec_rename },		/* genfs_badop */
169 	{ &vop_mkdir_desc, spec_mkdir },		/* genfs_badop */
170 	{ &vop_rmdir_desc, spec_rmdir },		/* genfs_badop */
171 	{ &vop_symlink_desc, spec_symlink },		/* genfs_badop */
172 	{ &vop_readdir_desc, spec_readdir },		/* genfs_badop */
173 	{ &vop_readlink_desc, spec_readlink },		/* genfs_badop */
174 	{ &vop_abortop_desc, spec_abortop },		/* genfs_badop */
175 	{ &vop_inactive_desc, puffs_vnop_inactive },	/* REAL inactive */
176 	{ &vop_reclaim_desc, puffs_vnop_reclaim },	/* REAL reclaim */
177 	{ &vop_lock_desc, puffs_vnop_lock },		/* REAL lock */
178 	{ &vop_unlock_desc, puffs_vnop_unlock },	/* REAL unlock */
179 	{ &vop_bmap_desc, spec_bmap },			/* dummy */
180 	{ &vop_strategy_desc, spec_strategy },		/* dev strategy */
181 	{ &vop_print_desc, puffs_vnop_print },		/* REAL print */
182 	{ &vop_islocked_desc, puffs_vnop_islocked },	/* REAL islocked */
183 	{ &vop_pathconf_desc, spec_pathconf },		/* pathconf */
184 	{ &vop_advlock_desc, spec_advlock },		/* lf_advlock */
185 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
186 	{ &vop_getpages_desc, spec_getpages },		/* genfs_getpages */
187 	{ &vop_putpages_desc, spec_putpages },		/* genfs_putpages */
188 #if 0
189 	{ &vop_openextattr_desc, _openextattr },	/* openextattr */
190 	{ &vop_closeextattr_desc, _closeextattr },	/* closeextattr */
191 	{ &vop_getextattr_desc, _getextattr },		/* getextattr */
192 	{ &vop_setextattr_desc, _setextattr },		/* setextattr */
193 	{ &vop_listextattr_desc, _listextattr },	/* listextattr */
194 	{ &vop_deleteextattr_desc, _deleteextattr },	/* deleteextattr */
195 #endif
196 	{ NULL, NULL }
197 };
198 const struct vnodeopv_desc puffs_specop_opv_desc =
199 	{ &puffs_specop_p, puffs_specop_entries };
200 
201 
202 int (**puffs_fifoop_p)(void *);
203 const struct vnodeopv_entry_desc puffs_fifoop_entries[] = {
204 	{ &vop_default_desc, vn_default_error },
205 	{ &vop_lookup_desc, fifo_lookup },		/* lookup, ENOTDIR */
206 	{ &vop_create_desc, fifo_create },		/* genfs_badop */
207 	{ &vop_mknod_desc, fifo_mknod },		/* genfs_badop */
208 	{ &vop_open_desc, fifo_open },			/* open */
209 	{ &vop_close_desc, fifo_close },		/* close */
210 	{ &vop_access_desc, puffs_vnop_checkop },	/* access */
211 	{ &vop_getattr_desc, puffs_vnop_checkop },	/* getattr */
212 	{ &vop_setattr_desc, puffs_vnop_checkop },	/* setattr */
213 	{ &vop_read_desc, puffs_vnop_fifo_read },	/* read, update */
214 	{ &vop_write_desc, puffs_vnop_fifo_write },	/* write, update */
215 	{ &vop_ioctl_desc, fifo_ioctl },		/* ioctl */
216 	{ &vop_fcntl_desc, genfs_fcntl },		/* dummy */
217 	{ &vop_poll_desc, fifo_poll },			/* poll */
218 	{ &vop_kqfilter_desc, fifo_kqfilter },		/* kqfilter */
219 	{ &vop_revoke_desc, fifo_revoke },		/* genfs_revoke */
220 	{ &vop_mmap_desc, fifo_mmap },			/* genfs_badop */
221 	{ &vop_fsync_desc, fifo_fsync },		/* genfs_nullop*/
222 	{ &vop_seek_desc, fifo_seek },			/* genfs_badop */
223 	{ &vop_remove_desc, fifo_remove },		/* genfs_badop */
224 	{ &vop_link_desc, fifo_link },			/* genfs_badop */
225 	{ &vop_rename_desc, fifo_rename },		/* genfs_badop */
226 	{ &vop_mkdir_desc, fifo_mkdir },		/* genfs_badop */
227 	{ &vop_rmdir_desc, fifo_rmdir },		/* genfs_badop */
228 	{ &vop_symlink_desc, fifo_symlink },		/* genfs_badop */
229 	{ &vop_readdir_desc, fifo_readdir },		/* genfs_badop */
230 	{ &vop_readlink_desc, fifo_readlink },		/* genfs_badop */
231 	{ &vop_abortop_desc, fifo_abortop },		/* genfs_badop */
232 	{ &vop_inactive_desc, puffs_vnop_inactive },	/* REAL inactive */
233 	{ &vop_reclaim_desc, puffs_vnop_reclaim },	/* REAL reclaim */
234 	{ &vop_lock_desc, puffs_vnop_lock },		/* REAL lock */
235 	{ &vop_unlock_desc, puffs_vnop_unlock },	/* REAL unlock */
236 	{ &vop_bmap_desc, fifo_bmap },			/* dummy */
237 	{ &vop_strategy_desc, fifo_strategy },		/* genfs_badop */
238 	{ &vop_print_desc, puffs_vnop_print },		/* REAL print */
239 	{ &vop_islocked_desc, puffs_vnop_islocked },	/* REAL islocked */
240 	{ &vop_pathconf_desc, fifo_pathconf },		/* pathconf */
241 	{ &vop_advlock_desc, fifo_advlock },		/* genfs_einval */
242 	{ &vop_bwrite_desc, vn_bwrite },		/* bwrite */
243 	{ &vop_putpages_desc, fifo_putpages }, 		/* genfs_null_putpages*/
244 #if 0
245 	{ &vop_openextattr_desc, _openextattr },	/* openextattr */
246 	{ &vop_closeextattr_desc, _closeextattr },	/* closeextattr */
247 	{ &vop_getextattr_desc, _getextattr },		/* getextattr */
248 	{ &vop_setextattr_desc, _setextattr },		/* setextattr */
249 	{ &vop_listextattr_desc, _listextattr },	/* listextattr */
250 	{ &vop_deleteextattr_desc, _deleteextattr },	/* deleteextattr */
251 #endif
252 	{ NULL, NULL }
253 };
254 const struct vnodeopv_desc puffs_fifoop_opv_desc =
255 	{ &puffs_fifoop_p, puffs_fifoop_entries };
256 
257 
258 /* "real" vnode operations */
259 int (**puffs_msgop_p)(void *);
260 const struct vnodeopv_entry_desc puffs_msgop_entries[] = {
261 	{ &vop_default_desc, vn_default_error },
262 	{ &vop_create_desc, puffs_vnop_create },	/* create */
263         { &vop_mknod_desc, puffs_vnop_mknod },		/* mknod */
264         { &vop_open_desc, puffs_vnop_open },		/* open */
265         { &vop_close_desc, puffs_vnop_close },		/* close */
266         { &vop_access_desc, puffs_vnop_access },	/* access */
267         { &vop_getattr_desc, puffs_vnop_getattr },	/* getattr */
268         { &vop_setattr_desc, puffs_vnop_setattr },	/* setattr */
269         { &vop_read_desc, puffs_vnop_read },		/* read */
270         { &vop_write_desc, puffs_vnop_write },		/* write */
271         { &vop_seek_desc, puffs_vnop_seek },		/* seek */
272         { &vop_remove_desc, puffs_vnop_remove },	/* remove */
273         { &vop_link_desc, puffs_vnop_link },		/* link */
274         { &vop_rename_desc, puffs_vnop_rename },	/* rename */
275         { &vop_mkdir_desc, puffs_vnop_mkdir },		/* mkdir */
276         { &vop_rmdir_desc, puffs_vnop_rmdir },		/* rmdir */
277         { &vop_symlink_desc, puffs_vnop_symlink },	/* symlink */
278         { &vop_readdir_desc, puffs_vnop_readdir },	/* readdir */
279         { &vop_readlink_desc, puffs_vnop_readlink },	/* readlink */
280         { &vop_print_desc, puffs_vnop_print },		/* print */
281         { &vop_islocked_desc, puffs_vnop_islocked },	/* islocked */
282         { &vop_pathconf_desc, puffs_vnop_pathconf },	/* pathconf */
283         { &vop_advlock_desc, puffs_vnop_advlock },	/* advlock */
284         { &vop_getpages_desc, puffs_vnop_getpages },	/* getpages */
285 	{ NULL, NULL }
286 };
287 const struct vnodeopv_desc puffs_msgop_opv_desc =
288 	{ &puffs_msgop_p, puffs_msgop_entries };
289 
290 
291 #define ERROUT(err)							\
292 do {									\
293 	error = err;							\
294 	goto out;							\
295 } while (/*CONSTCOND*/0)
296 
297 /*
298  * This is a generic vnode operation handler.  It checks if the necessary
299  * operations for the called vnode operation are implemented by userspace
300  * and either returns a dummy return value or proceeds to call the real
301  * vnode operation from puffs_msgop_v.
302  *
303  * XXX: this should described elsewhere and autogenerated, the complexity
304  * of the vnode operations vectors and their interrelationships is also
305  * getting a bit out of hand.  Another problem is that we need this same
306  * information in the fs server code, so keeping the two in sync manually
307  * is not a viable (long term) plan.
308  */
309 
310 /* not supported, handle locking protocol */
311 #define CHECKOP_NOTSUPP(op)						\
312 case VOP_##op##_DESCOFFSET:						\
313 	if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0)			\
314 		return genfs_eopnotsupp(v);				\
315 	break
316 
317 /* always succeed, no locking */
318 #define CHECKOP_SUCCESS(op)						\
319 case VOP_##op##_DESCOFFSET:						\
320 	if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0)			\
321 		return 0;						\
322 	break
323 
324 int
325 puffs_vnop_checkop(void *v)
326 {
327 	struct vop_generic_args /* {
328 		struct vnodeop_desc *a_desc;
329 		spooky mystery contents;
330 	} */ *ap = v;
331 	struct vnodeop_desc *desc = ap->a_desc;
332 	struct puffs_mount *pmp;
333 	struct vnode *vp;
334 	int offset, rv;
335 
336 	offset = ap->a_desc->vdesc_vp_offsets[0];
337 #ifdef DIAGNOSTIC
338 	if (offset == VDESC_NO_OFFSET)
339 		panic("puffs_checkop: no vnode, why did you call me?");
340 #endif
341 	vp = *VOPARG_OFFSETTO(struct vnode **, offset, ap);
342 	pmp = MPTOPUFFSMP(vp->v_mount);
343 
344 	DPRINTF_VERBOSE(("checkop call %s (%d), vp %p\n",
345 	    ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp));
346 
347 	if (!ALLOPS(pmp)) {
348 		switch (desc->vdesc_offset) {
349 			CHECKOP_NOTSUPP(CREATE);
350 			CHECKOP_NOTSUPP(MKNOD);
351 			CHECKOP_NOTSUPP(GETATTR);
352 			CHECKOP_NOTSUPP(SETATTR);
353 			CHECKOP_NOTSUPP(READ);
354 			CHECKOP_NOTSUPP(WRITE);
355 			CHECKOP_NOTSUPP(FCNTL);
356 			CHECKOP_NOTSUPP(IOCTL);
357 			CHECKOP_NOTSUPP(REMOVE);
358 			CHECKOP_NOTSUPP(LINK);
359 			CHECKOP_NOTSUPP(RENAME);
360 			CHECKOP_NOTSUPP(MKDIR);
361 			CHECKOP_NOTSUPP(RMDIR);
362 			CHECKOP_NOTSUPP(SYMLINK);
363 			CHECKOP_NOTSUPP(READDIR);
364 			CHECKOP_NOTSUPP(READLINK);
365 			CHECKOP_NOTSUPP(PRINT);
366 			CHECKOP_NOTSUPP(PATHCONF);
367 			CHECKOP_NOTSUPP(ADVLOCK);
368 
369 			CHECKOP_SUCCESS(ACCESS);
370 			CHECKOP_SUCCESS(CLOSE);
371 			CHECKOP_SUCCESS(SEEK);
372 
373 		case VOP_GETPAGES_DESCOFFSET:
374 			if (!EXISTSOP(pmp, READ))
375 				return genfs_eopnotsupp(v);
376 			break;
377 
378 		default:
379 			panic("puffs_checkop: unhandled vnop %d",
380 			    desc->vdesc_offset);
381 		}
382 	}
383 
384 	rv = VOCALL(puffs_msgop_p, ap->a_desc->vdesc_offset, v);
385 
386 	DPRINTF_VERBOSE(("checkop return %s (%d), vp %p: %d\n",
387 	    ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp, rv));
388 
389 	return rv;
390 }
391 
392 static int callremove(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
393 			    struct componentname *);
394 static int callrmdir(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
395 			   struct componentname *);
396 static void callinactive(struct puffs_mount *, puffs_cookie_t, int);
397 static void callreclaim(struct puffs_mount *, puffs_cookie_t);
398 static int  flushvncache(struct vnode *, off_t, off_t, bool);
399 
400 
401 #define PUFFS_ABORT_LOOKUP	1
402 #define PUFFS_ABORT_CREATE	2
403 #define PUFFS_ABORT_MKNOD	3
404 #define PUFFS_ABORT_MKDIR	4
405 #define PUFFS_ABORT_SYMLINK	5
406 
407 /*
408  * Press the pani^Wabort button!  Kernel resource allocation failed.
409  */
410 static void
411 puffs_abortbutton(struct puffs_mount *pmp, int what,
412 	puffs_cookie_t dck, puffs_cookie_t ck, struct componentname *cnp)
413 {
414 
415 	switch (what) {
416 	case PUFFS_ABORT_CREATE:
417 	case PUFFS_ABORT_MKNOD:
418 	case PUFFS_ABORT_SYMLINK:
419 		callremove(pmp, dck, ck, cnp);
420 		break;
421 	case PUFFS_ABORT_MKDIR:
422 		callrmdir(pmp, dck, ck, cnp);
423 		break;
424 	}
425 
426 	callinactive(pmp, ck, 0);
427 	callreclaim(pmp, ck);
428 }
429 
430 /*
431  * Begin vnode operations.
432  *
433  * A word from the keymaster about locks: generally we don't want
434  * to use the vnode locks at all: it creates an ugly dependency between
435  * the userlandia file server and the kernel.  But we'll play along with
436  * the kernel vnode locks for now.  However, even currently we attempt
437  * to release locks as early as possible.  This is possible for some
438  * operations which a) don't need a locked vnode after the userspace op
439  * and b) return with the vnode unlocked.  Theoretically we could
440  * unlock-do op-lock for others and order the graph in userspace, but I
441  * don't want to think of the consequences for the time being.
442  */
443 
444 int
445 puffs_vnop_lookup(void *v)
446 {
447         struct vop_lookup_args /* {
448 		const struct vnodeop_desc *a_desc;
449 		struct vnode *a_dvp;
450 		struct vnode **a_vpp;
451 		struct componentname *a_cnp;
452         } */ *ap = v;
453 	PUFFS_MSG_VARS(vn, lookup);
454 	struct puffs_mount *pmp;
455 	struct componentname *cnp;
456 	struct vnode *vp, *dvp;
457 	struct puffs_node *dpn;
458 	int isdot;
459 	int error;
460 
461 	pmp = MPTOPUFFSMP(ap->a_dvp->v_mount);
462 	cnp = ap->a_cnp;
463 	dvp = ap->a_dvp;
464 	*ap->a_vpp = NULL;
465 
466 	/* r/o fs?  we check create later to handle EEXIST */
467 	if ((cnp->cn_flags & ISLASTCN)
468 	    && (dvp->v_mount->mnt_flag & MNT_RDONLY)
469 	    && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
470 		return EROFS;
471 
472 	isdot = cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.';
473 
474 	DPRINTF(("puffs_lookup: \"%s\", parent vnode %p, op: %x\n",
475 	    cnp->cn_nameptr, dvp, cnp->cn_nameiop));
476 
477 	/*
478 	 * Check if someone fed it into the cache
479 	 */
480 	if (PUFFS_USE_NAMECACHE(pmp)) {
481 		error = cache_lookup(dvp, ap->a_vpp, cnp);
482 
483 		if (error >= 0)
484 			return error;
485 	}
486 
487 	if (isdot) {
488 		vp = ap->a_dvp;
489 		vref(vp);
490 		*ap->a_vpp = vp;
491 		return 0;
492 	}
493 
494 	PUFFS_MSG_ALLOC(vn, lookup);
495 	puffs_makecn(&lookup_msg->pvnr_cn, &lookup_msg->pvnr_cn_cred,
496 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
497 
498 	if (cnp->cn_flags & ISDOTDOT)
499 		VOP_UNLOCK(dvp, 0);
500 
501 	puffs_msg_setinfo(park_lookup, PUFFSOP_VN,
502 	    PUFFS_VN_LOOKUP, VPTOPNC(dvp));
503 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_lookup, dvp->v_data, NULL, error);
504 	DPRINTF(("puffs_lookup: return of the userspace, part %d\n", error));
505 
506 	/*
507 	 * In case of error, there is no new vnode to play with, so be
508 	 * happy with the NULL value given to vpp in the beginning.
509 	 * Also, check if this really was an error or the target was not
510 	 * present.  Either treat it as a non-error for CREATE/RENAME or
511 	 * enter the component into the negative name cache (if desired).
512 	 */
513 	if (error) {
514 		error = checkerr(pmp, error, __func__);
515 		if (error == ENOENT) {
516 			/* don't allow to create files on r/o fs */
517 			if ((dvp->v_mount->mnt_flag & MNT_RDONLY)
518 			    && cnp->cn_nameiop == CREATE) {
519 				error = EROFS;
520 
521 			/* adjust values if we are creating */
522 			} else if ((cnp->cn_flags & ISLASTCN)
523 			    && (cnp->cn_nameiop == CREATE
524 			      || cnp->cn_nameiop == RENAME)) {
525 				cnp->cn_flags |= SAVENAME;
526 				error = EJUSTRETURN;
527 
528 			/* save negative cache entry */
529 			} else {
530 				if ((cnp->cn_flags & MAKEENTRY)
531 				    && PUFFS_USE_NAMECACHE(pmp))
532 					cache_enter(dvp, NULL, cnp);
533 			}
534 		}
535 		goto out;
536 	}
537 
538 	/*
539 	 * Check that we don't get our parent node back, that would cause
540 	 * a pretty obvious deadlock.
541 	 */
542 	dpn = dvp->v_data;
543 	if (lookup_msg->pvnr_newnode == dpn->pn_cookie) {
544 		puffs_senderr(pmp, PUFFS_ERR_LOOKUP, EINVAL,
545 		    "lookup produced parent cookie", lookup_msg->pvnr_newnode);
546 		error = EPROTO;
547 		goto out;
548 	}
549 
550 	error = puffs_cookie2vnode(pmp, lookup_msg->pvnr_newnode, 1, 1, &vp);
551 	if (error == PUFFS_NOSUCHCOOKIE) {
552 		error = puffs_getvnode(dvp->v_mount,
553 		    lookup_msg->pvnr_newnode, lookup_msg->pvnr_vtype,
554 		    lookup_msg->pvnr_size, lookup_msg->pvnr_rdev, &vp);
555 		if (error) {
556 			puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP, VPTOPNC(dvp),
557 			    lookup_msg->pvnr_newnode, ap->a_cnp);
558 			goto out;
559 		}
560 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
561 	} else if (error) {
562 		puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP, VPTOPNC(dvp),
563 		    lookup_msg->pvnr_newnode, ap->a_cnp);
564 		goto out;
565 	}
566 
567 	*ap->a_vpp = vp;
568 
569 	if ((cnp->cn_flags & MAKEENTRY) != 0 && PUFFS_USE_NAMECACHE(pmp))
570 		cache_enter(dvp, vp, cnp);
571 
572 	/* XXX */
573 	if ((lookup_msg->pvnr_cn.pkcn_flags & REQUIREDIR) == 0)
574 		cnp->cn_flags &= ~REQUIREDIR;
575 	if (lookup_msg->pvnr_cn.pkcn_consume)
576 		cnp->cn_consume = MIN(lookup_msg->pvnr_cn.pkcn_consume,
577 		    strlen(cnp->cn_nameptr) - cnp->cn_namelen);
578 
579 	/*
580 	 * We need the name in remove and rmdir (well, rename too, but
581 	 * SAVESTART takes care of that)
582 	 */
583 	if (cnp->cn_nameiop == DELETE)
584 		cnp->cn_flags |= SAVENAME;
585 
586  out:
587 	if (cnp->cn_flags & ISDOTDOT)
588 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
589 
590 	DPRINTF(("puffs_lookup: returning %d %p\n", error, *ap->a_vpp));
591 	PUFFS_MSG_RELEASE(lookup);
592 	return error;
593 }
594 
595 #define REFPN_AND_UNLOCKVP(a, b)					\
596 do {									\
597 	mutex_enter(&b->pn_mtx);					\
598 	puffs_referencenode(b);						\
599 	mutex_exit(&b->pn_mtx);						\
600 	VOP_UNLOCK(a, 0);						\
601 } while (/*CONSTCOND*/0)
602 
603 #define REFPN(b)							\
604 do {									\
605 	mutex_enter(&b->pn_mtx);					\
606 	puffs_referencenode(b);						\
607 	mutex_exit(&b->pn_mtx);						\
608 } while (/*CONSTCOND*/0)
609 
610 #define RELEPN_AND_VP(a, b)						\
611 do {									\
612 	puffs_releasenode(b);						\
613 	vrele(a);							\
614 } while (/*CONSTCOND*/0)
615 
616 int
617 puffs_vnop_create(void *v)
618 {
619 	struct vop_create_args /* {
620 		const struct vnodeop_desc *a_desc;
621 		struct vnode *a_dvp;
622 		struct vnode **a_vpp;
623 		struct componentname *a_cnp;
624 		struct vattr *a_vap;
625 	} */ *ap = v;
626 	PUFFS_MSG_VARS(vn, create);
627 	struct vnode *dvp = ap->a_dvp;
628 	struct puffs_node *dpn = VPTOPP(dvp);
629 	struct componentname *cnp = ap->a_cnp;
630 	struct mount *mp = dvp->v_mount;
631 	struct puffs_mount *pmp = MPTOPUFFSMP(mp);
632 	int error;
633 
634 	DPRINTF(("puffs_create: dvp %p, cnp: %s\n",
635 	    dvp, ap->a_cnp->cn_nameptr));
636 
637 	PUFFS_MSG_ALLOC(vn, create);
638 	puffs_makecn(&create_msg->pvnr_cn, &create_msg->pvnr_cn_cred,
639 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
640 	create_msg->pvnr_va = *ap->a_vap;
641 	puffs_msg_setinfo(park_create, PUFFSOP_VN,
642 	    PUFFS_VN_CREATE, VPTOPNC(dvp));
643 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_create, dvp->v_data, NULL, error);
644 
645 	error = checkerr(pmp, error, __func__);
646 	if (error)
647 		goto out;
648 
649 	error = puffs_newnode(mp, dvp, ap->a_vpp,
650 	    create_msg->pvnr_newnode, cnp, ap->a_vap->va_type, 0);
651 	if (error)
652 		puffs_abortbutton(pmp, PUFFS_ABORT_CREATE, dpn->pn_cookie,
653 		    create_msg->pvnr_newnode, cnp);
654 
655  out:
656 	vput(dvp);
657 	if (error || (cnp->cn_flags & SAVESTART) == 0)
658 		PNBUF_PUT(cnp->cn_pnbuf);
659 
660 	DPRINTF(("puffs_create: return %d\n", error));
661 	PUFFS_MSG_RELEASE(create);
662 	return error;
663 }
664 
665 int
666 puffs_vnop_mknod(void *v)
667 {
668 	struct vop_mknod_args /* {
669 		const struct vnodeop_desc *a_desc;
670 		struct vnode *a_dvp;
671 		struct vnode **a_vpp;
672 		struct componentname *a_cnp;
673 		struct vattr *a_vap;
674 	} */ *ap = v;
675 	PUFFS_MSG_VARS(vn, mknod);
676 	struct vnode *dvp = ap->a_dvp;
677 	struct puffs_node *dpn = VPTOPP(dvp);
678 	struct componentname *cnp = ap->a_cnp;
679 	struct mount *mp = dvp->v_mount;
680 	struct puffs_mount *pmp = MPTOPUFFSMP(mp);
681 	int error;
682 
683 	PUFFS_MSG_ALLOC(vn, mknod);
684 	puffs_makecn(&mknod_msg->pvnr_cn, &mknod_msg->pvnr_cn_cred,
685 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
686 	mknod_msg->pvnr_va = *ap->a_vap;
687 	puffs_msg_setinfo(park_mknod, PUFFSOP_VN,
688 	    PUFFS_VN_MKNOD, VPTOPNC(dvp));
689 
690 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mknod, dvp->v_data, NULL, error);
691 
692 	error = checkerr(pmp, error, __func__);
693 	if (error)
694 		goto out;
695 
696 	error = puffs_newnode(mp, dvp, ap->a_vpp,
697 	    mknod_msg->pvnr_newnode, cnp, ap->a_vap->va_type,
698 	    ap->a_vap->va_rdev);
699 	if (error)
700 		puffs_abortbutton(pmp, PUFFS_ABORT_MKNOD, dpn->pn_cookie,
701 		    mknod_msg->pvnr_newnode, cnp);
702 
703  out:
704 	vput(dvp);
705 	PUFFS_MSG_RELEASE(mknod);
706 	if (error || (cnp->cn_flags & SAVESTART) == 0)
707 		PNBUF_PUT(cnp->cn_pnbuf);
708 	return error;
709 }
710 
711 int
712 puffs_vnop_open(void *v)
713 {
714 	struct vop_open_args /* {
715 		const struct vnodeop_desc *a_desc;
716 		struct vnode *a_vp;
717 		int a_mode;
718 		kauth_cred_t a_cred;
719 	} */ *ap = v;
720 	PUFFS_MSG_VARS(vn, open);
721 	struct vnode *vp = ap->a_vp;
722 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
723 	int mode = ap->a_mode;
724 	int error;
725 
726 	DPRINTF(("puffs_open: vp %p, mode 0x%x\n", vp, mode));
727 
728 	if (vp->v_type == VREG && mode & FWRITE && !EXISTSOP(pmp, WRITE))
729 		ERROUT(EROFS);
730 
731 	if (!EXISTSOP(pmp, OPEN))
732 		ERROUT(0);
733 
734 	PUFFS_MSG_ALLOC(vn, open);
735 	open_msg->pvnr_mode = mode;
736 	puffs_credcvt(&open_msg->pvnr_cred, ap->a_cred);
737 	puffs_msg_setinfo(park_open, PUFFSOP_VN,
738 	    PUFFS_VN_OPEN, VPTOPNC(vp));
739 
740 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_open, vp->v_data, NULL, error);
741 	error = checkerr(pmp, error, __func__);
742 
743  out:
744 	DPRINTF(("puffs_open: returning %d\n", error));
745 	PUFFS_MSG_RELEASE(open);
746 	return error;
747 }
748 
749 int
750 puffs_vnop_close(void *v)
751 {
752 	struct vop_close_args /* {
753 		const struct vnodeop_desc *a_desc;
754 		struct vnode *a_vp;
755 		int a_fflag;
756 		kauth_cred_t a_cred;
757 	} */ *ap = v;
758 	PUFFS_MSG_VARS(vn, close);
759 	struct vnode *vp = ap->a_vp;
760 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
761 
762 	PUFFS_MSG_ALLOC(vn, close);
763 	puffs_msg_setfaf(park_close);
764 	close_msg->pvnr_fflag = ap->a_fflag;
765 	puffs_credcvt(&close_msg->pvnr_cred, ap->a_cred);
766 	puffs_msg_setinfo(park_close, PUFFSOP_VN,
767 	    PUFFS_VN_CLOSE, VPTOPNC(vp));
768 
769 	puffs_msg_enqueue(pmp, park_close);
770 	PUFFS_MSG_RELEASE(close);
771 	return 0;
772 }
773 
774 int
775 puffs_vnop_access(void *v)
776 {
777 	struct vop_access_args /* {
778 		const struct vnodeop_desc *a_desc;
779 		struct vnode *a_vp;
780 		int a_mode;
781 		kauth_cred_t a_cred;
782 	} */ *ap = v;
783 	PUFFS_MSG_VARS(vn, access);
784 	struct vnode *vp = ap->a_vp;
785 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
786 	int mode = ap->a_mode;
787 	int error;
788 
789 	if (mode & VWRITE) {
790 		switch (vp->v_type) {
791 		case VDIR:
792 		case VLNK:
793 		case VREG:
794 			if ((vp->v_mount->mnt_flag & MNT_RDONLY)
795 			    || !EXISTSOP(pmp, WRITE))
796 				return EROFS;
797 			break;
798 		default:
799 			break;
800 		}
801 	}
802 
803 	if (!EXISTSOP(pmp, ACCESS))
804 		return 0;
805 
806 	PUFFS_MSG_ALLOC(vn, access);
807 	access_msg->pvnr_mode = ap->a_mode;
808 	puffs_credcvt(&access_msg->pvnr_cred, ap->a_cred);
809 	puffs_msg_setinfo(park_access, PUFFSOP_VN,
810 	    PUFFS_VN_ACCESS, VPTOPNC(vp));
811 
812 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_access, vp->v_data, NULL, error);
813 	error = checkerr(pmp, error, __func__);
814 	PUFFS_MSG_RELEASE(access);
815 
816 	return error;
817 }
818 
819 int
820 puffs_vnop_getattr(void *v)
821 {
822 	struct vop_getattr_args /* {
823 		const struct vnodeop_desc *a_desc;
824 		struct vnode *a_vp;
825 		struct vattr *a_vap;
826 		kauth_cred_t a_cred;
827 	} */ *ap = v;
828 	PUFFS_MSG_VARS(vn, getattr);
829 	struct vnode *vp = ap->a_vp;
830 	struct mount *mp = vp->v_mount;
831 	struct puffs_mount *pmp = MPTOPUFFSMP(mp);
832 	struct vattr *vap, *rvap;
833 	struct puffs_node *pn = VPTOPP(vp);
834 	int error = 0;
835 
836 	REFPN(pn);
837 	vap = ap->a_vap;
838 
839 	PUFFS_MSG_ALLOC(vn, getattr);
840 	vattr_null(&getattr_msg->pvnr_va);
841 	puffs_credcvt(&getattr_msg->pvnr_cred, ap->a_cred);
842 	puffs_msg_setinfo(park_getattr, PUFFSOP_VN,
843 	    PUFFS_VN_GETATTR, VPTOPNC(vp));
844 
845 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getattr, vp->v_data, NULL, error);
846 	error = checkerr(pmp, error, __func__);
847 	if (error)
848 		goto out;
849 
850 	rvap = &getattr_msg->pvnr_va;
851 	/*
852 	 * Don't listen to the file server regarding special device
853 	 * size info, the file server doesn't know anything about them.
854 	 */
855 	if (vp->v_type == VBLK || vp->v_type == VCHR)
856 		rvap->va_size = vp->v_size;
857 
858 	/* Ditto for blocksize (ufs comment: this doesn't belong here) */
859 	if (vp->v_type == VBLK)
860 		rvap->va_blocksize = BLKDEV_IOSIZE;
861 	else if (vp->v_type == VCHR)
862 		rvap->va_blocksize = MAXBSIZE;
863 
864 	(void) memcpy(vap, rvap, sizeof(struct vattr));
865 	vap->va_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
866 
867 	if (pn->pn_stat & PNODE_METACACHE_ATIME)
868 		vap->va_atime = pn->pn_mc_atime;
869 	if (pn->pn_stat & PNODE_METACACHE_CTIME)
870 		vap->va_ctime = pn->pn_mc_ctime;
871 	if (pn->pn_stat & PNODE_METACACHE_MTIME)
872 		vap->va_mtime = pn->pn_mc_mtime;
873 	if (pn->pn_stat & PNODE_METACACHE_SIZE) {
874 		vap->va_size = pn->pn_mc_size;
875 	} else {
876 		if (rvap->va_size != VNOVAL
877 		    && vp->v_type != VBLK && vp->v_type != VCHR) {
878 			uvm_vnp_setsize(vp, rvap->va_size);
879 			pn->pn_serversize = rvap->va_size;
880 		}
881 	}
882 
883  out:
884 	puffs_releasenode(pn);
885 	PUFFS_MSG_RELEASE(getattr);
886 	return error;
887 }
888 
889 #define SETATTR_CHSIZE	0x01
890 #define SETATTR_ASYNC	0x02
891 static int
892 dosetattr(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, int flags)
893 {
894 	PUFFS_MSG_VARS(vn, setattr);
895 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
896 	struct puffs_node *pn = vp->v_data;
897 	int error = 0;
898 
899 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) &&
900 	    (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL
901 	    || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL
902 	    || vap->va_mode != (mode_t)VNOVAL))
903 		return EROFS;
904 
905 	if ((vp->v_mount->mnt_flag & MNT_RDONLY)
906 	    && vp->v_type == VREG && vap->va_size != VNOVAL)
907 		return EROFS;
908 
909 	/*
910 	 * Flush metacache first.  If we are called with some explicit
911 	 * parameters, treat them as information overriding metacache
912 	 * information.
913 	 */
914 	if (pn->pn_stat & PNODE_METACACHE_MASK) {
915 		if ((pn->pn_stat & PNODE_METACACHE_ATIME)
916 		    && vap->va_atime.tv_sec == VNOVAL)
917 			vap->va_atime = pn->pn_mc_atime;
918 		if ((pn->pn_stat & PNODE_METACACHE_CTIME)
919 		    && vap->va_ctime.tv_sec == VNOVAL)
920 			vap->va_ctime = pn->pn_mc_ctime;
921 		if ((pn->pn_stat & PNODE_METACACHE_MTIME)
922 		    && vap->va_mtime.tv_sec == VNOVAL)
923 			vap->va_mtime = pn->pn_mc_mtime;
924 		if ((pn->pn_stat & PNODE_METACACHE_SIZE)
925 		    && vap->va_size == VNOVAL)
926 			vap->va_size = pn->pn_mc_size;
927 
928 		pn->pn_stat &= ~PNODE_METACACHE_MASK;
929 	}
930 
931 	PUFFS_MSG_ALLOC(vn, setattr);
932 	(void)memcpy(&setattr_msg->pvnr_va, vap, sizeof(struct vattr));
933 	puffs_credcvt(&setattr_msg->pvnr_cred, cred);
934 	puffs_msg_setinfo(park_setattr, PUFFSOP_VN,
935 	    PUFFS_VN_SETATTR, VPTOPNC(vp));
936 	if (flags & SETATTR_ASYNC)
937 		puffs_msg_setfaf(park_setattr);
938 
939 	puffs_msg_enqueue(pmp, park_setattr);
940 	if ((flags & SETATTR_ASYNC) == 0)
941 		error = puffs_msg_wait2(pmp, park_setattr, vp->v_data, NULL);
942 	PUFFS_MSG_RELEASE(setattr);
943 	if ((flags & SETATTR_ASYNC) == 0) {
944 		error = checkerr(pmp, error, __func__);
945 		if (error)
946 			return error;
947 	} else {
948 		error = 0;
949 	}
950 
951 	if (vap->va_size != VNOVAL) {
952 		pn->pn_serversize = vap->va_size;
953 		if (flags & SETATTR_CHSIZE)
954 			uvm_vnp_setsize(vp, vap->va_size);
955 	}
956 
957 	return 0;
958 }
959 
960 int
961 puffs_vnop_setattr(void *v)
962 {
963 	struct vop_getattr_args /* {
964 		const struct vnodeop_desc *a_desc;
965 		struct vnode *a_vp;
966 		struct vattr *a_vap;
967 		kauth_cred_t a_cred;
968 	} */ *ap = v;
969 
970 	return dosetattr(ap->a_vp, ap->a_vap, ap->a_cred, SETATTR_CHSIZE);
971 }
972 
973 static __inline int
974 doinact(struct puffs_mount *pmp, int iaflag)
975 {
976 
977 	if (EXISTSOP(pmp, INACTIVE))
978 		if (pmp->pmp_flags & PUFFS_KFLAG_IAONDEMAND)
979 			if (iaflag || ALLOPS(pmp))
980 				return 1;
981 			else
982 				return 0;
983 		else
984 			return 1;
985 	else
986 		return 0;
987 }
988 
989 static void
990 callinactive(struct puffs_mount *pmp, puffs_cookie_t ck, int iaflag)
991 {
992 	int error;
993 	PUFFS_MSG_VARS(vn, inactive);
994 
995 	if (doinact(pmp, iaflag)) {
996 		PUFFS_MSG_ALLOC(vn, inactive);
997 		puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
998 		    PUFFS_VN_INACTIVE, ck);
999 
1000 		PUFFS_MSG_ENQUEUEWAIT(pmp, park_inactive, error);
1001 		PUFFS_MSG_RELEASE(inactive);
1002 	}
1003 }
1004 
1005 /* XXX: callinactive can't setback */
1006 int
1007 puffs_vnop_inactive(void *v)
1008 {
1009 	struct vop_inactive_args /* {
1010 		const struct vnodeop_desc *a_desc;
1011 		struct vnode *a_vp;
1012 	} */ *ap = v;
1013 	PUFFS_MSG_VARS(vn, inactive);
1014 	struct vnode *vp = ap->a_vp;
1015 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1016 	struct puffs_node *pnode;
1017 	int error;
1018 
1019 	pnode = vp->v_data;
1020 
1021 	if (doinact(pmp, pnode->pn_stat & PNODE_DOINACT)) {
1022 		flushvncache(vp, 0, 0, false);
1023 		PUFFS_MSG_ALLOC(vn, inactive);
1024 		puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1025 		    PUFFS_VN_INACTIVE, VPTOPNC(vp));
1026 
1027 		PUFFS_MSG_ENQUEUEWAIT2(pmp, park_inactive, vp->v_data,
1028 		    NULL, error);
1029 		PUFFS_MSG_RELEASE(inactive);
1030 	}
1031 	pnode->pn_stat &= ~PNODE_DOINACT;
1032 
1033 	/*
1034 	 * file server thinks it's gone?  then don't be afraid care,
1035 	 * node's life was already all it would ever be
1036 	 */
1037 	if (pnode->pn_stat & PNODE_NOREFS) {
1038 		pnode->pn_stat |= PNODE_DYING;
1039 		*ap->a_recycle = true;
1040 	}
1041 
1042 	VOP_UNLOCK(vp, 0);
1043 
1044 	return 0;
1045 }
1046 
1047 static void
1048 callreclaim(struct puffs_mount *pmp, puffs_cookie_t ck)
1049 {
1050 	PUFFS_MSG_VARS(vn, reclaim);
1051 
1052 	if (!EXISTSOP(pmp, RECLAIM))
1053 		return;
1054 
1055 	PUFFS_MSG_ALLOC(vn, reclaim);
1056 	puffs_msg_setfaf(park_reclaim);
1057 	puffs_msg_setinfo(park_reclaim, PUFFSOP_VN, PUFFS_VN_RECLAIM, ck);
1058 
1059 	puffs_msg_enqueue(pmp, park_reclaim);
1060 	PUFFS_MSG_RELEASE(reclaim);
1061 }
1062 
1063 /*
1064  * always FAF, we don't really care if the server wants to fail to
1065  * reclaim the node or not
1066  */
1067 int
1068 puffs_vnop_reclaim(void *v)
1069 {
1070 	struct vop_reclaim_args /* {
1071 		const struct vnodeop_desc *a_desc;
1072 		struct vnode *a_vp;
1073 	} */ *ap = v;
1074 	struct vnode *vp = ap->a_vp;
1075 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1076 	struct puffs_node *pnode = vp->v_data;
1077 	bool notifyserver = true;
1078 
1079 	/*
1080 	 * first things first: check if someone is trying to reclaim the
1081 	 * root vnode.  do not allow that to travel to userspace.
1082 	 * Note that we don't need to take the lock similarly to
1083 	 * puffs_root(), since there is only one of us.
1084 	 */
1085 	if (vp->v_vflag & VV_ROOT) {
1086 		mutex_enter(&pmp->pmp_lock);
1087 		KASSERT(pmp->pmp_root != NULL);
1088 		pmp->pmp_root = NULL;
1089 		mutex_exit(&pmp->pmp_lock);
1090 		notifyserver = false;
1091 	}
1092 
1093 	/*
1094 	 * purge info from kernel before issueing FAF, since we
1095 	 * don't really know when we'll get around to it after
1096 	 * that and someone might race us into node creation
1097 	 */
1098 	mutex_enter(&pmp->pmp_lock);
1099 	LIST_REMOVE(pnode, pn_hashent);
1100 	mutex_exit(&pmp->pmp_lock);
1101 	if (PUFFS_USE_NAMECACHE(pmp))
1102 		cache_purge(vp);
1103 
1104 	if (notifyserver)
1105 		callreclaim(MPTOPUFFSMP(vp->v_mount), VPTOPNC(vp));
1106 
1107 	puffs_putvnode(vp);
1108 
1109 	return 0;
1110 }
1111 
1112 #define CSIZE sizeof(**ap->a_cookies)
1113 int
1114 puffs_vnop_readdir(void *v)
1115 {
1116 	struct vop_readdir_args /* {
1117 		const struct vnodeop_desc *a_desc;
1118 		struct vnode *a_vp;
1119 		struct uio *a_uio;
1120 		kauth_cred_t a_cred;
1121 		int *a_eofflag;
1122 		off_t **a_cookies;
1123 		int *a_ncookies;
1124 	} */ *ap = v;
1125 	PUFFS_MSG_VARS(vn, readdir);
1126 	struct vnode *vp = ap->a_vp;
1127 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1128 	size_t argsize, tomove, cookiemem, cookiesmax;
1129 	struct uio *uio = ap->a_uio;
1130 	size_t howmuch, resid;
1131 	int error;
1132 
1133 	/*
1134 	 * ok, so we need: resid + cookiemem = maxreq
1135 	 * => resid + cookiesize * (resid/minsize) = maxreq
1136 	 * => resid + cookiesize/minsize * resid = maxreq
1137 	 * => (cookiesize/minsize + 1) * resid = maxreq
1138 	 * => resid = maxreq / (cookiesize/minsize + 1)
1139 	 *
1140 	 * Since cookiesize <= minsize and we're not very big on floats,
1141 	 * we approximate that to be 1.  Therefore:
1142 	 *
1143 	 * resid = maxreq / 2;
1144 	 *
1145 	 * Well, at least we didn't have to use differential equations
1146 	 * or the Gram-Schmidt process.
1147 	 *
1148 	 * (yes, I'm very afraid of this)
1149 	 */
1150 	KASSERT(CSIZE <= _DIRENT_MINSIZE((struct dirent *)0));
1151 
1152 	if (ap->a_cookies) {
1153 		KASSERT(ap->a_ncookies != NULL);
1154 		if (pmp->pmp_args.pa_fhsize == 0)
1155 			return EOPNOTSUPP;
1156 		resid = PUFFS_TOMOVE(uio->uio_resid, pmp) / 2;
1157 		cookiesmax = resid/_DIRENT_MINSIZE((struct dirent *)0);
1158 		cookiemem = ALIGN(cookiesmax*CSIZE); /* play safe */
1159 	} else {
1160 		resid = PUFFS_TOMOVE(uio->uio_resid, pmp);
1161 		cookiesmax = 0;
1162 		cookiemem = 0;
1163 	}
1164 
1165 	argsize = sizeof(struct puffs_vnmsg_readdir);
1166 	tomove = resid + cookiemem;
1167 	puffs_msgmem_alloc(argsize + tomove, &park_readdir,
1168 	    (void *)&readdir_msg, 1);
1169 
1170 	puffs_credcvt(&readdir_msg->pvnr_cred, ap->a_cred);
1171 	readdir_msg->pvnr_offset = uio->uio_offset;
1172 	readdir_msg->pvnr_resid = resid;
1173 	readdir_msg->pvnr_ncookies = cookiesmax;
1174 	readdir_msg->pvnr_eofflag = 0;
1175 	readdir_msg->pvnr_dentoff = cookiemem;
1176 	puffs_msg_setinfo(park_readdir, PUFFSOP_VN,
1177 	    PUFFS_VN_READDIR, VPTOPNC(vp));
1178 	puffs_msg_setdelta(park_readdir, tomove);
1179 
1180 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readdir, vp->v_data, NULL, error);
1181 	error = checkerr(pmp, error, __func__);
1182 	if (error)
1183 		goto out;
1184 
1185 	/* userspace is cheating? */
1186 	if (readdir_msg->pvnr_resid > resid) {
1187 		puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1188 		    "resid grew", VPTOPNC(vp));
1189 		ERROUT(EPROTO);
1190 	}
1191 	if (readdir_msg->pvnr_ncookies > cookiesmax) {
1192 		puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1193 		    "too many cookies", VPTOPNC(vp));
1194 		ERROUT(EPROTO);
1195 	}
1196 
1197 	/* check eof */
1198 	if (readdir_msg->pvnr_eofflag)
1199 		*ap->a_eofflag = 1;
1200 
1201 	/* bouncy-wouncy with the directory data */
1202 	howmuch = resid - readdir_msg->pvnr_resid;
1203 
1204 	/* force eof if no data was returned (getcwd() needs this) */
1205 	if (howmuch == 0) {
1206 		*ap->a_eofflag = 1;
1207 		goto out;
1208 	}
1209 
1210 	error = uiomove(readdir_msg->pvnr_data + cookiemem, howmuch, uio);
1211 	if (error)
1212 		goto out;
1213 
1214 	/* provide cookies to caller if so desired */
1215 	if (ap->a_cookies) {
1216 		*ap->a_cookies = malloc(readdir_msg->pvnr_ncookies*CSIZE,
1217 		    M_TEMP, M_WAITOK);
1218 		*ap->a_ncookies = readdir_msg->pvnr_ncookies;
1219 		memcpy(*ap->a_cookies, readdir_msg->pvnr_data,
1220 		    *ap->a_ncookies*CSIZE);
1221 	}
1222 
1223 	/* next readdir starts here */
1224 	uio->uio_offset = readdir_msg->pvnr_offset;
1225 
1226  out:
1227 	puffs_msgmem_release(park_readdir);
1228 	return error;
1229 }
1230 #undef CSIZE
1231 
1232 /*
1233  * poll works by consuming the bitmask in pn_revents.  If there are
1234  * events available, poll returns immediately.  If not, it issues a
1235  * poll to userspace, selrecords itself and returns with no available
1236  * events.  When the file server returns, it executes puffs_parkdone_poll(),
1237  * where available events are added to the bitmask.  selnotify() is
1238  * then also executed by that function causing us to enter here again
1239  * and hopefully find the missing bits (unless someone got them first,
1240  * in which case it starts all over again).
1241  */
1242 int
1243 puffs_vnop_poll(void *v)
1244 {
1245 	struct vop_poll_args /* {
1246 		const struct vnodeop_desc *a_desc;
1247 		struct vnode *a_vp;
1248 		int a_events;
1249 	} */ *ap = v;
1250 	PUFFS_MSG_VARS(vn, poll);
1251 	struct vnode *vp = ap->a_vp;
1252 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1253 	struct puffs_node *pn = vp->v_data;
1254 	int events, error;
1255 
1256 	if (EXISTSOP(pmp, POLL)) {
1257 		mutex_enter(&pn->pn_mtx);
1258 		events = pn->pn_revents & ap->a_events;
1259 		if (events & ap->a_events) {
1260 			pn->pn_revents &= ~ap->a_events;
1261 			mutex_exit(&pn->pn_mtx);
1262 
1263 			return events;
1264 		} else {
1265 			puffs_referencenode(pn);
1266 			mutex_exit(&pn->pn_mtx);
1267 
1268 			PUFFS_MSG_ALLOC(vn, poll);
1269 			poll_msg->pvnr_events = ap->a_events;
1270 			puffs_msg_setinfo(park_poll, PUFFSOP_VN,
1271 			    PUFFS_VN_POLL, VPTOPNC(vp));
1272 			puffs_msg_setcall(park_poll, puffs_parkdone_poll, pn);
1273 			selrecord(curlwp, &pn->pn_sel);
1274 
1275 			PUFFS_MSG_ENQUEUEWAIT2(pmp, park_poll, vp->v_data,
1276 			    NULL, error);
1277 			PUFFS_MSG_RELEASE(poll);
1278 
1279 			return 0;
1280 		}
1281 	} else {
1282 		return genfs_poll(v);
1283 	}
1284 }
1285 
1286 static int
1287 flushvncache(struct vnode *vp, off_t offlo, off_t offhi, bool wait)
1288 {
1289 	struct puffs_node *pn = VPTOPP(vp);
1290 	struct vattr va;
1291 	int pflags, error;
1292 
1293 	/* flush out information from our metacache, see vop_setattr */
1294 	if (pn->pn_stat & PNODE_METACACHE_MASK
1295 	    && (pn->pn_stat & PNODE_DYING) == 0) {
1296 		vattr_null(&va);
1297 		error = dosetattr(vp, &va, FSCRED,
1298 		    SETATTR_CHSIZE | (wait ? 0 : SETATTR_ASYNC));
1299 		if (error)
1300 			return error;
1301 	}
1302 
1303 	/*
1304 	 * flush pages to avoid being overly dirty
1305 	 */
1306 	pflags = PGO_CLEANIT;
1307 	if (wait)
1308 		pflags |= PGO_SYNCIO;
1309 	mutex_enter(&vp->v_interlock);
1310 	return VOP_PUTPAGES(vp, trunc_page(offlo), round_page(offhi), pflags);
1311 }
1312 
1313 int
1314 puffs_vnop_fsync(void *v)
1315 {
1316 	struct vop_fsync_args /* {
1317 		const struct vnodeop_desc *a_desc;
1318 		struct vnode *a_vp;
1319 		kauth_cred_t a_cred;
1320 		int a_flags;
1321 		off_t a_offlo;
1322 		off_t a_offhi;
1323 	} */ *ap = v;
1324 	PUFFS_MSG_VARS(vn, fsync);
1325 	struct vnode *vp = ap->a_vp;
1326 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1327 	struct puffs_node *pn = VPTOPP(vp);
1328 	int error, dofaf;
1329 
1330 	error = flushvncache(vp, ap->a_offlo, ap->a_offhi,
1331 	    (ap->a_flags & FSYNC_WAIT) == FSYNC_WAIT);
1332 	if (error)
1333 		return error;
1334 
1335 	/*
1336 	 * HELLO!  We exit already here if the user server does not
1337 	 * support fsync OR if we should call fsync for a node which
1338 	 * has references neither in the kernel or the fs server.
1339 	 * Otherwise we continue to issue fsync() forward.
1340 	 */
1341 	if (!EXISTSOP(pmp, FSYNC) || (pn->pn_stat & PNODE_DYING))
1342 		return 0;
1343 
1344 	dofaf = (ap->a_flags & FSYNC_WAIT) == 0 || ap->a_flags == FSYNC_LAZY;
1345 	/*
1346 	 * We abuse VXLOCK to mean "vnode is going to die", so we issue
1347 	 * only FAFs for those.  Otherwise there's a danger of deadlock,
1348 	 * since the execution context here might be the user server
1349 	 * doing some operation on another fs, which in turn caused a
1350 	 * vnode to be reclaimed from the freelist for this fs.
1351 	 */
1352 	if (dofaf == 0) {
1353 		mutex_enter(&vp->v_interlock);
1354 		if (vp->v_iflag & VI_XLOCK)
1355 			dofaf = 1;
1356 		mutex_exit(&vp->v_interlock);
1357 	}
1358 
1359 	PUFFS_MSG_ALLOC(vn, fsync);
1360 	if (dofaf)
1361 		puffs_msg_setfaf(park_fsync);
1362 
1363 	puffs_credcvt(&fsync_msg->pvnr_cred, ap->a_cred);
1364 	fsync_msg->pvnr_flags = ap->a_flags;
1365 	fsync_msg->pvnr_offlo = ap->a_offlo;
1366 	fsync_msg->pvnr_offhi = ap->a_offhi;
1367 	puffs_msg_setinfo(park_fsync, PUFFSOP_VN,
1368 	    PUFFS_VN_FSYNC, VPTOPNC(vp));
1369 
1370 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fsync, vp->v_data, NULL, error);
1371 	PUFFS_MSG_RELEASE(fsync);
1372 
1373 	error = checkerr(pmp, error, __func__);
1374 
1375 	return error;
1376 }
1377 
1378 int
1379 puffs_vnop_seek(void *v)
1380 {
1381 	struct vop_seek_args /* {
1382 		const struct vnodeop_desc *a_desc;
1383 		struct vnode *a_vp;
1384 		off_t a_oldoff;
1385 		off_t a_newoff;
1386 		kauth_cred_t a_cred;
1387 	} */ *ap = v;
1388 	PUFFS_MSG_VARS(vn, seek);
1389 	struct vnode *vp = ap->a_vp;
1390 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1391 	int error;
1392 
1393 	PUFFS_MSG_ALLOC(vn, seek);
1394 	seek_msg->pvnr_oldoff = ap->a_oldoff;
1395 	seek_msg->pvnr_newoff = ap->a_newoff;
1396 	puffs_credcvt(&seek_msg->pvnr_cred, ap->a_cred);
1397 	puffs_msg_setinfo(park_seek, PUFFSOP_VN,
1398 	    PUFFS_VN_SEEK, VPTOPNC(vp));
1399 
1400 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_seek, vp->v_data, NULL, error);
1401 	PUFFS_MSG_RELEASE(seek);
1402 	return checkerr(pmp, error, __func__);
1403 }
1404 
1405 static int
1406 callremove(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1407 	struct componentname *cnp)
1408 {
1409 	PUFFS_MSG_VARS(vn, remove);
1410 	int error;
1411 
1412 	PUFFS_MSG_ALLOC(vn, remove);
1413 	remove_msg->pvnr_cookie_targ = ck;
1414 	puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1415 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1416 	puffs_msg_setinfo(park_remove, PUFFSOP_VN, PUFFS_VN_REMOVE, dck);
1417 
1418 	PUFFS_MSG_ENQUEUEWAIT(pmp, park_remove, error);
1419 	PUFFS_MSG_RELEASE(remove);
1420 
1421 	return checkerr(pmp, error, __func__);
1422 }
1423 
1424 /*
1425  * XXX: can't use callremove now because can't catch setbacks with
1426  * it due to lack of a pnode argument.
1427  */
1428 int
1429 puffs_vnop_remove(void *v)
1430 {
1431 	struct vop_remove_args /* {
1432 		const struct vnodeop_desc *a_desc;
1433 		struct vnode *a_dvp;
1434 		struct vnode *a_vp;
1435 		struct componentname *a_cnp;
1436 	} */ *ap = v;
1437 	PUFFS_MSG_VARS(vn, remove);
1438 	struct vnode *dvp = ap->a_dvp;
1439 	struct vnode *vp = ap->a_vp;
1440 	struct puffs_node *dpn = VPTOPP(dvp);
1441 	struct puffs_node *pn = VPTOPP(vp);
1442 	struct componentname *cnp = ap->a_cnp;
1443 	struct mount *mp = dvp->v_mount;
1444 	struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1445 	int error;
1446 
1447 	PUFFS_MSG_ALLOC(vn, remove);
1448 	remove_msg->pvnr_cookie_targ = VPTOPNC(vp);
1449 	puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1450 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1451 	puffs_msg_setinfo(park_remove, PUFFSOP_VN,
1452 	    PUFFS_VN_REMOVE, VPTOPNC(dvp));
1453 
1454 	puffs_msg_enqueue(pmp, park_remove);
1455 	REFPN_AND_UNLOCKVP(dvp, dpn);
1456 	if (dvp == vp)
1457 		REFPN(pn);
1458 	else
1459 		REFPN_AND_UNLOCKVP(vp, pn);
1460 	error = puffs_msg_wait2(pmp, park_remove, dpn, pn);
1461 
1462 	PUFFS_MSG_RELEASE(remove);
1463 
1464 	RELEPN_AND_VP(dvp, dpn);
1465 	RELEPN_AND_VP(vp, pn);
1466 
1467 	error = checkerr(pmp, error, __func__);
1468 	if (error || (cnp->cn_flags & SAVESTART) == 0)
1469 		PNBUF_PUT(cnp->cn_pnbuf);
1470 	return error;
1471 }
1472 
1473 int
1474 puffs_vnop_mkdir(void *v)
1475 {
1476 	struct vop_mkdir_args /* {
1477 		const struct vnodeop_desc *a_desc;
1478 		struct vnode *a_dvp;
1479 		struct vnode **a_vpp;
1480 		struct componentname *a_cnp;
1481 		struct vattr *a_vap;
1482 	} */ *ap = v;
1483 	PUFFS_MSG_VARS(vn, mkdir);
1484 	struct vnode *dvp = ap->a_dvp;
1485 	struct puffs_node *dpn = VPTOPP(dvp);
1486 	struct componentname *cnp = ap->a_cnp;
1487 	struct mount *mp = dvp->v_mount;
1488 	struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1489 	int error;
1490 
1491 	PUFFS_MSG_ALLOC(vn, mkdir);
1492 	puffs_makecn(&mkdir_msg->pvnr_cn, &mkdir_msg->pvnr_cn_cred,
1493 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1494 	mkdir_msg->pvnr_va = *ap->a_vap;
1495 	puffs_msg_setinfo(park_mkdir, PUFFSOP_VN,
1496 	    PUFFS_VN_MKDIR, VPTOPNC(dvp));
1497 
1498 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mkdir, dvp->v_data, NULL, error);
1499 
1500 	error = checkerr(pmp, error, __func__);
1501 	if (error)
1502 		goto out;
1503 
1504 	error = puffs_newnode(mp, dvp, ap->a_vpp,
1505 	    mkdir_msg->pvnr_newnode, cnp, VDIR, 0);
1506 	if (error)
1507 		puffs_abortbutton(pmp, PUFFS_ABORT_MKDIR, dpn->pn_cookie,
1508 		    mkdir_msg->pvnr_newnode, cnp);
1509 
1510  out:
1511 	vput(dvp);
1512 	PUFFS_MSG_RELEASE(mkdir);
1513 	if (error || (cnp->cn_flags & SAVESTART) == 0)
1514 		PNBUF_PUT(cnp->cn_pnbuf);
1515 	return error;
1516 }
1517 
1518 static int
1519 callrmdir(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1520 	struct componentname *cnp)
1521 {
1522 	PUFFS_MSG_VARS(vn, rmdir);
1523 	int error;
1524 
1525 	PUFFS_MSG_ALLOC(vn, rmdir);
1526 	rmdir_msg->pvnr_cookie_targ = ck;
1527 	puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1528 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1529 	puffs_msg_setinfo(park_rmdir, PUFFSOP_VN, PUFFS_VN_RMDIR, dck);
1530 
1531 	PUFFS_MSG_ENQUEUEWAIT(pmp, park_rmdir, error);
1532 	PUFFS_MSG_RELEASE(rmdir);
1533 
1534 	return checkerr(pmp, error, __func__);
1535 }
1536 
1537 int
1538 puffs_vnop_rmdir(void *v)
1539 {
1540 	struct vop_rmdir_args /* {
1541 		const struct vnodeop_desc *a_desc;
1542 		struct vnode *a_dvp;
1543 		struct vnode *a_vp;
1544 		struct componentname *a_cnp;
1545 	} */ *ap = v;
1546 	PUFFS_MSG_VARS(vn, rmdir);
1547 	struct vnode *dvp = ap->a_dvp;
1548 	struct vnode *vp = ap->a_vp;
1549 	struct puffs_node *dpn = VPTOPP(dvp);
1550 	struct puffs_node *pn = VPTOPP(vp);
1551 	struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1552 	struct componentname *cnp = ap->a_cnp;
1553 	int error;
1554 
1555 	PUFFS_MSG_ALLOC(vn, rmdir);
1556 	rmdir_msg->pvnr_cookie_targ = VPTOPNC(vp);
1557 	puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1558 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1559 	puffs_msg_setinfo(park_rmdir, PUFFSOP_VN,
1560 	    PUFFS_VN_RMDIR, VPTOPNC(dvp));
1561 
1562 	puffs_msg_enqueue(pmp, park_rmdir);
1563 	REFPN_AND_UNLOCKVP(dvp, dpn);
1564 	REFPN_AND_UNLOCKVP(vp, pn);
1565 	error = puffs_msg_wait2(pmp, park_rmdir, dpn, pn);
1566 
1567 	PUFFS_MSG_RELEASE(rmdir);
1568 
1569 	/* XXX: some call cache_purge() *for both vnodes* here, investigate */
1570 	RELEPN_AND_VP(dvp, dpn);
1571 	RELEPN_AND_VP(vp, pn);
1572 
1573 	if (error || (cnp->cn_flags & SAVESTART) == 0)
1574 		PNBUF_PUT(cnp->cn_pnbuf);
1575 
1576 	return error;
1577 }
1578 
1579 int
1580 puffs_vnop_link(void *v)
1581 {
1582 	struct vop_link_args /* {
1583 		const struct vnodeop_desc *a_desc;
1584 		struct vnode *a_dvp;
1585 		struct vnode *a_vp;
1586 		struct componentname *a_cnp;
1587 	} */ *ap = v;
1588 	PUFFS_MSG_VARS(vn, link);
1589 	struct vnode *dvp = ap->a_dvp;
1590 	struct vnode *vp = ap->a_vp;
1591 	struct puffs_node *dpn = VPTOPP(dvp);
1592 	struct puffs_node *pn = VPTOPP(vp);
1593 	struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1594 	struct componentname *cnp = ap->a_cnp;
1595 	int error;
1596 
1597 	PUFFS_MSG_ALLOC(vn, link);
1598 	link_msg->pvnr_cookie_targ = VPTOPNC(vp);
1599 	puffs_makecn(&link_msg->pvnr_cn, &link_msg->pvnr_cn_cred,
1600 	    cnp, PUFFS_USE_FULLPNBUF(pmp));
1601 	puffs_msg_setinfo(park_link, PUFFSOP_VN,
1602 	    PUFFS_VN_LINK, VPTOPNC(dvp));
1603 
1604 	puffs_msg_enqueue(pmp, park_link);
1605 	REFPN_AND_UNLOCKVP(dvp, dpn);
1606 	REFPN(pn);
1607 	error = puffs_msg_wait2(pmp, park_link, dpn, pn);
1608 
1609 	PUFFS_MSG_RELEASE(link);
1610 
1611 	error = checkerr(pmp, error, __func__);
1612 
1613 	/*
1614 	 * XXX: stay in touch with the cache.  I don't like this, but
1615 	 * don't have a better solution either.  See also puffs_rename().
1616 	 */
1617 	if (error == 0)
1618 		puffs_updatenode(pn, PUFFS_UPDATECTIME, 0);
1619 
1620 	PNBUF_PUT(cnp->cn_pnbuf);
1621 	RELEPN_AND_VP(dvp, dpn);
1622 	puffs_releasenode(pn);
1623 
1624 	return error;
1625 }
1626 
1627 int
1628 puffs_vnop_symlink(void *v)
1629 {
1630 	struct vop_symlink_args /* {
1631 		const struct vnodeop_desc *a_desc;
1632 		struct vnode *a_dvp;
1633 		struct vnode **a_vpp;
1634 		struct componentname *a_cnp;
1635 		struct vattr *a_vap;
1636 		char *a_target;
1637 	} */ *ap = v;
1638 	PUFFS_MSG_VARS(vn, symlink);
1639 	struct vnode *dvp = ap->a_dvp;
1640 	struct puffs_node *dpn = VPTOPP(dvp);
1641 	struct mount *mp = dvp->v_mount;
1642 	struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1643 	struct componentname *cnp = ap->a_cnp;
1644 	int error;
1645 
1646 	*ap->a_vpp = NULL;
1647 
1648 	PUFFS_MSG_ALLOC(vn, symlink);
1649 	puffs_makecn(&symlink_msg->pvnr_cn, &symlink_msg->pvnr_cn_cred,
1650 		cnp, PUFFS_USE_FULLPNBUF(pmp));
1651 	symlink_msg->pvnr_va = *ap->a_vap;
1652 	(void)strlcpy(symlink_msg->pvnr_link, ap->a_target,
1653 	    sizeof(symlink_msg->pvnr_link));
1654 	puffs_msg_setinfo(park_symlink, PUFFSOP_VN,
1655 	    PUFFS_VN_SYMLINK, VPTOPNC(dvp));
1656 
1657 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_symlink, dvp->v_data, NULL, error);
1658 
1659 	error = checkerr(pmp, error, __func__);
1660 	if (error)
1661 		goto out;
1662 
1663 	error = puffs_newnode(mp, dvp, ap->a_vpp,
1664 	    symlink_msg->pvnr_newnode, cnp, VLNK, 0);
1665 	if (error)
1666 		puffs_abortbutton(pmp, PUFFS_ABORT_SYMLINK, dpn->pn_cookie,
1667 		    symlink_msg->pvnr_newnode, cnp);
1668 
1669  out:
1670 	vput(dvp);
1671 	PUFFS_MSG_RELEASE(symlink);
1672 	if (error || (cnp->cn_flags & SAVESTART) == 0)
1673 		PNBUF_PUT(cnp->cn_pnbuf);
1674 
1675 	return error;
1676 }
1677 
1678 int
1679 puffs_vnop_readlink(void *v)
1680 {
1681 	struct vop_readlink_args /* {
1682 		const struct vnodeop_desc *a_desc;
1683 		struct vnode *a_vp;
1684 		struct uio *a_uio;
1685 		kauth_cred_t a_cred;
1686 	} */ *ap = v;
1687 	PUFFS_MSG_VARS(vn, readlink);
1688 	struct vnode *vp = ap->a_vp;
1689 	struct puffs_mount *pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
1690 	size_t linklen;
1691 	int error;
1692 
1693 	PUFFS_MSG_ALLOC(vn, readlink);
1694 	puffs_credcvt(&readlink_msg->pvnr_cred, ap->a_cred);
1695 	linklen = sizeof(readlink_msg->pvnr_link);
1696 	readlink_msg->pvnr_linklen = linklen;
1697 	puffs_msg_setinfo(park_readlink, PUFFSOP_VN,
1698 	    PUFFS_VN_READLINK, VPTOPNC(vp));
1699 
1700 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readlink, vp->v_data, NULL, error);
1701 	error = checkerr(pmp, error, __func__);
1702 	if (error)
1703 		goto out;
1704 
1705 	/* bad bad user file server */
1706 	if (readlink_msg->pvnr_linklen > linklen) {
1707 		puffs_senderr(pmp, PUFFS_ERR_READLINK, E2BIG,
1708 		    "linklen too big", VPTOPNC(ap->a_vp));
1709 		error = EPROTO;
1710 		goto out;
1711 	}
1712 
1713 	error = uiomove(&readlink_msg->pvnr_link, readlink_msg->pvnr_linklen,
1714 	    ap->a_uio);
1715  out:
1716 	PUFFS_MSG_RELEASE(readlink);
1717 	return error;
1718 }
1719 
1720 int
1721 puffs_vnop_rename(void *v)
1722 {
1723 	struct vop_rename_args /* {
1724 		const struct vnodeop_desc *a_desc;
1725 		struct vnode *a_fdvp;
1726 		struct vnode *a_fvp;
1727 		struct componentname *a_fcnp;
1728 		struct vnode *a_tdvp;
1729 		struct vnode *a_tvp;
1730 		struct componentname *a_tcnp;
1731 	} */ *ap = v;
1732 	PUFFS_MSG_VARS(vn, rename);
1733 	struct vnode *fdvp = ap->a_fdvp, *fvp = ap->a_fvp;
1734 	struct vnode *tdvp = ap->a_tdvp, *tvp = ap->a_tvp;
1735 	struct puffs_node *fpn = ap->a_fvp->v_data;
1736 	struct puffs_mount *pmp = MPTOPUFFSMP(fdvp->v_mount);
1737 	int error;
1738 	bool doabort = true;
1739 
1740 	if ((fvp->v_mount != tdvp->v_mount) ||
1741 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1742 		ERROUT(EXDEV);
1743 	}
1744 
1745 	PUFFS_MSG_ALLOC(vn, rename);
1746 	rename_msg->pvnr_cookie_src = VPTOPNC(fvp);
1747 	rename_msg->pvnr_cookie_targdir = VPTOPNC(tdvp);
1748 	if (tvp)
1749 		rename_msg->pvnr_cookie_targ = VPTOPNC(tvp);
1750 	else
1751 		rename_msg->pvnr_cookie_targ = NULL;
1752 	puffs_makecn(&rename_msg->pvnr_cn_src, &rename_msg->pvnr_cn_src_cred,
1753 	    ap->a_fcnp, PUFFS_USE_FULLPNBUF(pmp));
1754 	puffs_makecn(&rename_msg->pvnr_cn_targ, &rename_msg->pvnr_cn_targ_cred,
1755 	    ap->a_tcnp, PUFFS_USE_FULLPNBUF(pmp));
1756 	puffs_msg_setinfo(park_rename, PUFFSOP_VN,
1757 	    PUFFS_VN_RENAME, VPTOPNC(fdvp));
1758 
1759 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rename, fdvp->v_data, NULL, error);
1760 	doabort = false;
1761 	PUFFS_MSG_RELEASE(rename);
1762 	error = checkerr(pmp, error, __func__);
1763 
1764 	/*
1765 	 * XXX: stay in touch with the cache.  I don't like this, but
1766 	 * don't have a better solution either.  See also puffs_link().
1767 	 */
1768 	if (error == 0)
1769 		puffs_updatenode(fpn, PUFFS_UPDATECTIME, 0);
1770 
1771  out:
1772 	if (doabort)
1773 		VOP_ABORTOP(tdvp, ap->a_tcnp);
1774 	if (tvp != NULL)
1775 		vput(tvp);
1776 	if (tdvp == tvp)
1777 		vrele(tdvp);
1778 	else
1779 		vput(tdvp);
1780 
1781 	if (doabort)
1782 		VOP_ABORTOP(fdvp, ap->a_fcnp);
1783 	vrele(fdvp);
1784 	vrele(fvp);
1785 
1786 	return error;
1787 }
1788 
1789 #define RWARGS(cont, iofl, move, offset, creds)				\
1790 	(cont)->pvnr_ioflag = (iofl);					\
1791 	(cont)->pvnr_resid = (move);					\
1792 	(cont)->pvnr_offset = (offset);					\
1793 	puffs_credcvt(&(cont)->pvnr_cred, creds)
1794 
1795 int
1796 puffs_vnop_read(void *v)
1797 {
1798 	struct vop_read_args /* {
1799 		const struct vnodeop_desc *a_desc;
1800 		struct vnode *a_vp;
1801 		struct uio *a_uio;
1802 		int a_ioflag;
1803 		kauth_cred_t a_cred;
1804 	} */ *ap = v;
1805 	PUFFS_MSG_VARS(vn, read);
1806 	struct vnode *vp = ap->a_vp;
1807 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1808 	struct uio *uio = ap->a_uio;
1809 	size_t tomove, argsize;
1810 	vsize_t bytelen;
1811 	int error;
1812 
1813 	read_msg = NULL;
1814 	error = 0;
1815 
1816 	/* std sanity */
1817 	if (uio->uio_resid == 0)
1818 		return 0;
1819 	if (uio->uio_offset < 0)
1820 		return EINVAL;
1821 
1822 	if (vp->v_type == VREG && PUFFS_USE_PAGECACHE(pmp)) {
1823 		const int advice = IO_ADV_DECODE(ap->a_ioflag);
1824 
1825 		while (uio->uio_resid > 0) {
1826 			bytelen = MIN(uio->uio_resid,
1827 			    vp->v_size - uio->uio_offset);
1828 			if (bytelen == 0)
1829 				break;
1830 
1831 			error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
1832 			    UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
1833 			if (error)
1834 				break;
1835 		}
1836 
1837 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1838 			puffs_updatenode(VPTOPP(vp), PUFFS_UPDATEATIME, 0);
1839 	} else {
1840 		/*
1841 		 * in case it's not a regular file or we're operating
1842 		 * uncached, do read in the old-fashioned style,
1843 		 * i.e. explicit read operations
1844 		 */
1845 
1846 		tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
1847 		argsize = sizeof(struct puffs_vnmsg_read);
1848 		puffs_msgmem_alloc(argsize + tomove, &park_read,
1849 		    (void *)&read_msg, 1);
1850 
1851 		error = 0;
1852 		while (uio->uio_resid > 0) {
1853 			tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
1854 			memset(read_msg, 0, argsize); /* XXX: touser KASSERT */
1855 			RWARGS(read_msg, ap->a_ioflag, tomove,
1856 			    uio->uio_offset, ap->a_cred);
1857 			puffs_msg_setinfo(park_read, PUFFSOP_VN,
1858 			    PUFFS_VN_READ, VPTOPNC(vp));
1859 			puffs_msg_setdelta(park_read, tomove);
1860 
1861 			PUFFS_MSG_ENQUEUEWAIT2(pmp, park_read, vp->v_data,
1862 			    NULL, error);
1863 			error = checkerr(pmp, error, __func__);
1864 			if (error)
1865 				break;
1866 
1867 			if (read_msg->pvnr_resid > tomove) {
1868 				puffs_senderr(pmp, PUFFS_ERR_READ,
1869 				    E2BIG, "resid grew", VPTOPNC(ap->a_vp));
1870 				error = EPROTO;
1871 				break;
1872 			}
1873 
1874 			error = uiomove(read_msg->pvnr_data,
1875 			    tomove - read_msg->pvnr_resid, uio);
1876 
1877 			/*
1878 			 * in case the file is out of juice, resid from
1879 			 * userspace is != 0.  and the error-case is
1880 			 * quite obvious
1881 			 */
1882 			if (error || read_msg->pvnr_resid)
1883 				break;
1884 		}
1885 
1886 		puffs_msgmem_release(park_read);
1887 	}
1888 
1889 	return error;
1890 }
1891 
1892 /*
1893  * XXX: in case of a failure, this leaves uio in a bad state.
1894  * We could theoretically copy the uio and iovecs and "replay"
1895  * them the right amount after the userspace trip, but don't
1896  * bother for now.
1897  */
1898 int
1899 puffs_vnop_write(void *v)
1900 {
1901 	struct vop_write_args /* {
1902 		const struct vnodeop_desc *a_desc;
1903 		struct vnode *a_vp;
1904 		struct uio *a_uio;
1905 		int a_ioflag;
1906 		kauth_cred_t a_cred;
1907 	} */ *ap = v;
1908 	PUFFS_MSG_VARS(vn, write);
1909 	struct vnode *vp = ap->a_vp;
1910 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1911 	struct uio *uio = ap->a_uio;
1912 	size_t tomove, argsize;
1913 	off_t oldoff, newoff, origoff;
1914 	vsize_t bytelen;
1915 	int error, uflags;
1916 	int ubcflags;
1917 
1918 	error = uflags = 0;
1919 	write_msg = NULL;
1920 
1921 	if (vp->v_type == VREG && PUFFS_USE_PAGECACHE(pmp)) {
1922 		ubcflags = UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp);
1923 
1924 		/*
1925 		 * userspace *should* be allowed to control this,
1926 		 * but with UBC it's a bit unclear how to handle it
1927 		 */
1928 		if (ap->a_ioflag & IO_APPEND)
1929 			uio->uio_offset = vp->v_size;
1930 
1931 		origoff = uio->uio_offset;
1932 		while (uio->uio_resid > 0) {
1933 			uflags |= PUFFS_UPDATECTIME;
1934 			uflags |= PUFFS_UPDATEMTIME;
1935 			oldoff = uio->uio_offset;
1936 			bytelen = uio->uio_resid;
1937 
1938 			newoff = oldoff + bytelen;
1939 			if (vp->v_size < newoff) {
1940 				uvm_vnp_setwritesize(vp, newoff);
1941 			}
1942 			error = ubc_uiomove(&vp->v_uobj, uio, bytelen,
1943 			    UVM_ADV_RANDOM, ubcflags);
1944 
1945 			/*
1946 			 * In case of a ubc_uiomove() error,
1947 			 * opt to not extend the file at all and
1948 			 * return an error.  Otherwise, if we attempt
1949 			 * to clear the memory we couldn't fault to,
1950 			 * we might generate a kernel page fault.
1951 			 */
1952 			if (vp->v_size < newoff) {
1953 				if (error == 0) {
1954 					uflags |= PUFFS_UPDATESIZE;
1955 					uvm_vnp_setsize(vp, newoff);
1956 				} else {
1957 					uvm_vnp_setwritesize(vp, vp->v_size);
1958 				}
1959 			}
1960 			if (error)
1961 				break;
1962 
1963 			/*
1964 			 * If we're writing large files, flush to file server
1965 			 * every 64k.  Otherwise we can very easily exhaust
1966 			 * kernel and user memory, as the file server cannot
1967 			 * really keep up with our writing speed.
1968 			 *
1969 			 * Note: this does *NOT* honor MNT_ASYNC, because
1970 			 * that gives userland too much say in the kernel.
1971 			 */
1972 			if (oldoff >> 16 != uio->uio_offset >> 16) {
1973 				mutex_enter(&vp->v_interlock);
1974 				error = VOP_PUTPAGES(vp, oldoff & ~0xffff,
1975 				    uio->uio_offset & ~0xffff,
1976 				    PGO_CLEANIT | PGO_SYNCIO);
1977 				if (error)
1978 					break;
1979 			}
1980 		}
1981 
1982 		/* synchronous I/O? */
1983 		if (error == 0 && ap->a_ioflag & IO_SYNC) {
1984 			mutex_enter(&vp->v_interlock);
1985 			error = VOP_PUTPAGES(vp, trunc_page(origoff),
1986 			    round_page(uio->uio_offset),
1987 			    PGO_CLEANIT | PGO_SYNCIO);
1988 
1989 		/* write through page cache? */
1990 		} else if (error == 0 && pmp->pmp_flags & PUFFS_KFLAG_WTCACHE) {
1991 			mutex_enter(&vp->v_interlock);
1992 			error = VOP_PUTPAGES(vp, trunc_page(origoff),
1993 			    round_page(uio->uio_offset), PGO_CLEANIT);
1994 		}
1995 
1996 		puffs_updatenode(VPTOPP(vp), uflags, vp->v_size);
1997 	} else {
1998 		/* tomove is non-increasing */
1999 		tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2000 		argsize = sizeof(struct puffs_vnmsg_write) + tomove;
2001 		puffs_msgmem_alloc(argsize, &park_write, (void *)&write_msg,1);
2002 
2003 		while (uio->uio_resid > 0) {
2004 			/* move data to buffer */
2005 			tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2006 			memset(write_msg, 0, argsize); /* XXX: touser KASSERT */
2007 			RWARGS(write_msg, ap->a_ioflag, tomove,
2008 			    uio->uio_offset, ap->a_cred);
2009 			error = uiomove(write_msg->pvnr_data, tomove, uio);
2010 			if (error)
2011 				break;
2012 
2013 			/* move buffer to userspace */
2014 			puffs_msg_setinfo(park_write, PUFFSOP_VN,
2015 			    PUFFS_VN_WRITE, VPTOPNC(vp));
2016 			PUFFS_MSG_ENQUEUEWAIT2(pmp, park_write, vp->v_data,
2017 			    NULL, error);
2018 			error = checkerr(pmp, error, __func__);
2019 			if (error)
2020 				break;
2021 
2022 			if (write_msg->pvnr_resid > tomove) {
2023 				puffs_senderr(pmp, PUFFS_ERR_WRITE,
2024 				    E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2025 				error = EPROTO;
2026 				break;
2027 			}
2028 
2029 			/* adjust file size */
2030 			if (vp->v_size < uio->uio_offset)
2031 				uvm_vnp_setsize(vp, uio->uio_offset);
2032 
2033 			/* didn't move everything?  bad userspace.  bail */
2034 			if (write_msg->pvnr_resid != 0) {
2035 				error = EIO;
2036 				break;
2037 			}
2038 		}
2039 		puffs_msgmem_release(park_write);
2040 	}
2041 
2042 	return error;
2043 }
2044 
2045 int
2046 puffs_vnop_print(void *v)
2047 {
2048 	struct vop_print_args /* {
2049 		struct vnode *a_vp;
2050 	} */ *ap = v;
2051 	PUFFS_MSG_VARS(vn, print);
2052 	struct vnode *vp = ap->a_vp;
2053 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2054 	struct puffs_node *pn = vp->v_data;
2055 	int error;
2056 
2057 	/* kernel portion */
2058 	printf("tag VT_PUFFS, vnode %p, puffs node: %p,\n"
2059 	    "    userspace cookie: %p\n", vp, pn, pn->pn_cookie);
2060 	if (vp->v_type == VFIFO)
2061 		fifo_printinfo(vp);
2062 
2063 	/* userspace portion */
2064 	if (EXISTSOP(pmp, PRINT)) {
2065 		PUFFS_MSG_ALLOC(vn, print);
2066 		puffs_msg_setinfo(park_print, PUFFSOP_VN,
2067 		    PUFFS_VN_PRINT, VPTOPNC(vp));
2068 		PUFFS_MSG_ENQUEUEWAIT2(pmp, park_print, vp->v_data,
2069 		    NULL, error);
2070 		PUFFS_MSG_RELEASE(print);
2071 	}
2072 
2073 	return 0;
2074 }
2075 
2076 int
2077 puffs_vnop_pathconf(void *v)
2078 {
2079 	struct vop_pathconf_args /* {
2080 		const struct vnodeop_desc *a_desc;
2081 		struct vnode *a_vp;
2082 		int a_name;
2083 		register_t *a_retval;
2084 	} */ *ap = v;
2085 	PUFFS_MSG_VARS(vn, pathconf);
2086 	struct vnode *vp = ap->a_vp;
2087 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2088 	int error;
2089 
2090 	PUFFS_MSG_ALLOC(vn, pathconf);
2091 	pathconf_msg->pvnr_name = ap->a_name;
2092 	puffs_msg_setinfo(park_pathconf, PUFFSOP_VN,
2093 	    PUFFS_VN_PATHCONF, VPTOPNC(vp));
2094 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_pathconf, vp->v_data, NULL, error);
2095 	error = checkerr(pmp, error, __func__);
2096 	if (!error)
2097 		*ap->a_retval = pathconf_msg->pvnr_retval;
2098 	PUFFS_MSG_RELEASE(pathconf);
2099 
2100 	return error;
2101 }
2102 
2103 int
2104 puffs_vnop_advlock(void *v)
2105 {
2106 	struct vop_advlock_args /* {
2107 		const struct vnodeop_desc *a_desc;
2108 		struct vnode *a_vp;
2109 		void *a_id;
2110 		int a_op;
2111 		struct flock *a_fl;
2112 		int a_flags;
2113 	} */ *ap = v;
2114 	PUFFS_MSG_VARS(vn, advlock);
2115 	struct vnode *vp = ap->a_vp;
2116 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2117 	int error;
2118 
2119 	PUFFS_MSG_ALLOC(vn, advlock);
2120 	error = copyin(ap->a_fl, &advlock_msg->pvnr_fl, sizeof(struct flock));
2121 	if (error)
2122 		goto out;
2123 	advlock_msg->pvnr_id = ap->a_id;
2124 	advlock_msg->pvnr_op = ap->a_op;
2125 	advlock_msg->pvnr_flags = ap->a_flags;
2126 	puffs_msg_setinfo(park_advlock, PUFFSOP_VN,
2127 	    PUFFS_VN_ADVLOCK, VPTOPNC(vp));
2128 
2129 	PUFFS_MSG_ENQUEUEWAIT2(pmp, park_advlock, vp->v_data, NULL, error);
2130 	error = checkerr(pmp, error, __func__);
2131 
2132  out:
2133 	PUFFS_MSG_RELEASE(advlock);
2134 	return error;
2135 }
2136 
2137 int
2138 puffs_vnop_abortop(void *v)
2139 {
2140 	struct vop_abortop_args /* {
2141 		struct vnode *a_dvp;
2142 		struct componentname *a_cnp;
2143 	}; */ *ap = v;
2144 	PUFFS_MSG_VARS(vn, abortop);
2145 	struct vnode *dvp = ap->a_dvp;
2146 	struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2147 	struct componentname *cnp = ap->a_cnp;
2148 
2149 	if (EXISTSOP(pmp, ABORTOP)) {
2150 		PUFFS_MSG_ALLOC(vn, abortop);
2151 		puffs_makecn(&abortop_msg->pvnr_cn, &abortop_msg->pvnr_cn_cred,
2152 		    cnp, PUFFS_USE_FULLPNBUF(pmp));
2153 		puffs_msg_setfaf(park_abortop);
2154 		puffs_msg_setinfo(park_abortop, PUFFSOP_VN,
2155 		    PUFFS_VN_ABORTOP, VPTOPNC(dvp));
2156 
2157 		puffs_msg_enqueue(pmp, park_abortop);
2158 		PUFFS_MSG_RELEASE(abortop);
2159 	}
2160 
2161 	return genfs_abortop(v);
2162 }
2163 
2164 #define BIOASYNC(bp) (bp->b_flags & B_ASYNC)
2165 
2166 /*
2167  * This maps itself to PUFFS_VN_READ/WRITE for data transfer.
2168  */
2169 int
2170 puffs_vnop_strategy(void *v)
2171 {
2172 	struct vop_strategy_args /* {
2173 		const struct vnodeop_desc *a_desc;
2174 		struct vnode *a_vp;
2175 		struct buf *a_bp;
2176 	} */ *ap = v;
2177 	PUFFS_MSG_VARS(vn, rw);
2178 	struct vnode *vp = ap->a_vp;
2179 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2180 	struct puffs_node *pn;
2181 	struct buf *bp;
2182 	size_t argsize;
2183 	size_t tomove, moved;
2184 	int error, dofaf, dobiodone;
2185 
2186 	pmp = MPTOPUFFSMP(vp->v_mount);
2187 	bp = ap->a_bp;
2188 	error = 0;
2189 	dofaf = 0;
2190 	pn = VPTOPP(vp);
2191 	park_rw = NULL; /* explicit */
2192 	dobiodone = 1;
2193 
2194 	if ((BUF_ISREAD(bp) && !EXISTSOP(pmp, READ))
2195 	    || (BUF_ISWRITE(bp) && !EXISTSOP(pmp, WRITE)))
2196 		ERROUT(EOPNOTSUPP);
2197 
2198 	/*
2199 	 * Short-circuit optimization: don't flush buffer in between
2200 	 * VOP_INACTIVE and VOP_RECLAIM in case the node has no references.
2201 	 */
2202 	if (pn->pn_stat & PNODE_DYING) {
2203 		KASSERT(BUF_ISWRITE(bp));
2204 		bp->b_resid = 0;
2205 		goto out;
2206 	}
2207 
2208 #ifdef DIAGNOSTIC
2209 	if (bp->b_bcount > pmp->pmp_msg_maxsize - PUFFS_MSGSTRUCT_MAX)
2210 		panic("puffs_strategy: wildly inappropriate buf bcount %d",
2211 		    bp->b_bcount);
2212 #endif
2213 
2214 	/*
2215 	 * See explanation for the necessity of a FAF in puffs_fsync.
2216 	 *
2217 	 * Also, do FAF in case we're suspending.
2218 	 * See puffs_vfsops.c:pageflush()
2219 	 */
2220 	if (BUF_ISWRITE(bp)) {
2221 		mutex_enter(&vp->v_interlock);
2222 		if (vp->v_iflag & VI_XLOCK)
2223 			dofaf = 1;
2224 		if (pn->pn_stat & PNODE_FAF)
2225 			dofaf = 1;
2226 		mutex_exit(&vp->v_interlock);
2227 	}
2228 
2229 #ifdef DIAGNOSTIC
2230 		if (curlwp == uvm.pagedaemon_lwp)
2231 			KASSERT(dofaf || BIOASYNC(bp));
2232 #endif
2233 
2234 	/* allocate transport structure */
2235 	tomove = PUFFS_TOMOVE(bp->b_bcount, pmp);
2236 	argsize = sizeof(struct puffs_vnmsg_rw);
2237 	error = puffs_msgmem_alloc(argsize + tomove, &park_rw,
2238 	    (void *)&rw_msg, dofaf ? 0 : 1);
2239 	if (error)
2240 		goto out;
2241 	RWARGS(rw_msg, 0, tomove, bp->b_blkno << DEV_BSHIFT, FSCRED);
2242 
2243 	/* 2x2 cases: read/write, faf/nofaf */
2244 	if (BUF_ISREAD(bp)) {
2245 		puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2246 		    PUFFS_VN_READ, VPTOPNC(vp));
2247 		puffs_msg_setdelta(park_rw, tomove);
2248 		if (BIOASYNC(bp)) {
2249 			puffs_msg_setcall(park_rw,
2250 			    puffs_parkdone_asyncbioread, bp);
2251 			puffs_msg_enqueue(pmp, park_rw);
2252 			dobiodone = 0;
2253 		} else {
2254 			PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data,
2255 			    NULL, error);
2256 			error = checkerr(pmp, error, __func__);
2257 			if (error)
2258 				goto out;
2259 
2260 			if (rw_msg->pvnr_resid > tomove) {
2261 				puffs_senderr(pmp, PUFFS_ERR_READ,
2262 				    E2BIG, "resid grew", VPTOPNC(vp));
2263 				ERROUT(EPROTO);
2264 			}
2265 
2266 			moved = tomove - rw_msg->pvnr_resid;
2267 
2268 			(void)memcpy(bp->b_data, rw_msg->pvnr_data, moved);
2269 			bp->b_resid = bp->b_bcount - moved;
2270 		}
2271 	} else {
2272 		puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2273 		    PUFFS_VN_WRITE, VPTOPNC(vp));
2274 		/*
2275 		 * make pages read-only before we write them if we want
2276 		 * write caching info
2277 		 */
2278 		if (PUFFS_WCACHEINFO(pmp)) {
2279 			struct uvm_object *uobj = &vp->v_uobj;
2280 			int npages = (bp->b_bcount + PAGE_SIZE-1) >> PAGE_SHIFT;
2281 			struct vm_page *vmp;
2282 			int i;
2283 
2284 			for (i = 0; i < npages; i++) {
2285 				vmp= uvm_pageratop((vaddr_t)bp->b_data
2286 				    + (i << PAGE_SHIFT));
2287 				DPRINTF(("puffs_strategy: write-protecting "
2288 				    "vp %p page %p, offset %" PRId64"\n",
2289 				    vp, vmp, vmp->offset));
2290 				mutex_enter(&uobj->vmobjlock);
2291 				vmp->flags |= PG_RDONLY;
2292 				pmap_page_protect(vmp, VM_PROT_READ);
2293 				mutex_exit(&uobj->vmobjlock);
2294 			}
2295 		}
2296 
2297 		(void)memcpy(&rw_msg->pvnr_data, bp->b_data, tomove);
2298 		if (dofaf) {
2299 			puffs_msg_setfaf(park_rw);
2300 		} else if (BIOASYNC(bp)) {
2301 			puffs_msg_setcall(park_rw,
2302 			    puffs_parkdone_asyncbiowrite, bp);
2303 			dobiodone = 0;
2304 		}
2305 
2306 		PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data, NULL, error);
2307 
2308 		if (dobiodone == 0)
2309 			goto out;
2310 
2311 		/*
2312 		 * XXXXXXXX: wrong, but kernel can't survive strategy
2313 		 * failure currently.  Here, have one more X: X.
2314 		 */
2315 		if (error != ENOMEM)
2316 			error = 0;
2317 
2318 		error = checkerr(pmp, error, __func__);
2319 		if (error)
2320 			goto out;
2321 
2322 		if (rw_msg->pvnr_resid > tomove) {
2323 			puffs_senderr(pmp, PUFFS_ERR_WRITE,
2324 			    E2BIG, "resid grew", VPTOPNC(vp));
2325 			ERROUT(EPROTO);
2326 		}
2327 
2328 		/*
2329 		 * FAF moved everything.  Frankly, we don't
2330 		 * really have a choice.
2331 		 */
2332 		if (dofaf && error == 0)
2333 			moved = tomove;
2334 		else
2335 			moved = tomove - rw_msg->pvnr_resid;
2336 
2337 		bp->b_resid = bp->b_bcount - moved;
2338 		if (bp->b_resid != 0) {
2339 			ERROUT(EIO);
2340 		}
2341 	}
2342 
2343  out:
2344 	if (park_rw)
2345 		puffs_msgmem_release(park_rw);
2346 
2347 	if (error)
2348 		bp->b_error = error;
2349 
2350 	if (error || dobiodone)
2351 		biodone(bp);
2352 
2353 	return error;
2354 }
2355 
2356 int
2357 puffs_vnop_mmap(void *v)
2358 {
2359 	struct vop_mmap_args /* {
2360 		const struct vnodeop_desc *a_desc;
2361 		struct vnode *a_vp;
2362 		vm_prot_t a_prot;
2363 		kauth_cred_t a_cred;
2364 	} */ *ap = v;
2365 	PUFFS_MSG_VARS(vn, mmap);
2366 	struct vnode *vp = ap->a_vp;
2367 	struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2368 	int error;
2369 
2370 	if (!PUFFS_USE_PAGECACHE(pmp))
2371 		return genfs_eopnotsupp(v);
2372 
2373 	if (EXISTSOP(pmp, MMAP)) {
2374 		PUFFS_MSG_ALLOC(vn, mmap);
2375 		mmap_msg->pvnr_prot = ap->a_prot;
2376 		puffs_credcvt(&mmap_msg->pvnr_cred, ap->a_cred);
2377 		puffs_msg_setinfo(park_mmap, PUFFSOP_VN,
2378 		    PUFFS_VN_MMAP, VPTOPNC(vp));
2379 
2380 		PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mmap, vp->v_data, NULL, error);
2381 		error = checkerr(pmp, error, __func__);
2382 		PUFFS_MSG_RELEASE(mmap);
2383 	} else {
2384 		error = genfs_mmap(v);
2385 	}
2386 
2387 	return error;
2388 }
2389 
2390 
2391 /*
2392  * The rest don't get a free trip to userspace and back, they
2393  * have to stay within the kernel.
2394  */
2395 
2396 /*
2397  * bmap doesn't really make any sense for puffs, so just 1:1 map it.
2398  * well, maybe somehow, somewhere, some day ....
2399  */
2400 int
2401 puffs_vnop_bmap(void *v)
2402 {
2403 	struct vop_bmap_args /* {
2404 		const struct vnodeop_desc *a_desc;
2405 		struct vnode *a_vp;
2406 		daddr_t a_bn;
2407 		struct vnode **a_vpp;
2408 		daddr_t *a_bnp;
2409 		int *a_runp;
2410 	} */ *ap = v;
2411 	struct puffs_mount *pmp;
2412 
2413 	pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2414 
2415 	if (ap->a_vpp)
2416 		*ap->a_vpp = ap->a_vp;
2417 	if (ap->a_bnp)
2418 		*ap->a_bnp = ap->a_bn;
2419 	if (ap->a_runp)
2420 		*ap->a_runp
2421 		    = (PUFFS_TOMOVE(pmp->pmp_msg_maxsize, pmp)>>DEV_BSHIFT) - 1;
2422 
2423 	return 0;
2424 }
2425 
2426 /*
2427  * Handle getpages faults in puffs.  We let genfs_getpages() do most
2428  * of the dirty work, but we come in this route to do accounting tasks.
2429  * If the user server has specified functions for cache notifications
2430  * about reads and/or writes, we record which type of operation we got,
2431  * for which page range, and proceed to issue a FAF notification to the
2432  * server about it.
2433  */
2434 int
2435 puffs_vnop_getpages(void *v)
2436 {
2437 	struct vop_getpages_args /* {
2438 		const struct vnodeop_desc *a_desc;
2439 		struct vnode *a_vp;
2440 		voff_t a_offset;
2441 		struct vm_page **a_m;
2442 		int *a_count;
2443 		int a_centeridx;
2444 		vm_prot_t a_access_type;
2445 		int a_advice;
2446 		int a_flags;
2447 	} */ *ap = v;
2448 	struct puffs_mount *pmp;
2449 	struct puffs_node *pn;
2450 	struct vnode *vp;
2451 	struct vm_page **pgs;
2452 	struct puffs_cacheinfo *pcinfo = NULL;
2453 	struct puffs_cacherun *pcrun;
2454 	void *parkmem = NULL;
2455 	size_t runsizes;
2456 	int i, npages, si, streakon;
2457 	int error, locked, write;
2458 
2459 	pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2460 	npages = *ap->a_count;
2461 	pgs = ap->a_m;
2462 	vp = ap->a_vp;
2463 	pn = vp->v_data;
2464 	locked = (ap->a_flags & PGO_LOCKED) != 0;
2465 	write = (ap->a_access_type & VM_PROT_WRITE) != 0;
2466 
2467 	/* ccg xnaht - gets Wuninitialized wrong */
2468 	pcrun = NULL;
2469 	runsizes = 0;
2470 
2471 	/*
2472 	 * Check that we aren't trying to fault in pages which our file
2473 	 * server doesn't know about.  This happens if we extend a file by
2474 	 * skipping some pages and later try to fault in pages which
2475 	 * are between pn_serversize and vp_size.  This check optimizes
2476 	 * away the common case where a file is being extended.
2477 	 */
2478 	if (ap->a_offset >= pn->pn_serversize && ap->a_offset < vp->v_size) {
2479 		struct vattr va;
2480 
2481 		/* try again later when we can block */
2482 		if (locked)
2483 			ERROUT(EBUSY);
2484 
2485 		mutex_exit(&vp->v_interlock);
2486 		vattr_null(&va);
2487 		va.va_size = vp->v_size;
2488 		error = dosetattr(vp, &va, FSCRED, 0);
2489 		if (error)
2490 			ERROUT(error);
2491 		mutex_enter(&vp->v_interlock);
2492 	}
2493 
2494 	if (write && PUFFS_WCACHEINFO(pmp)) {
2495 #ifdef notnowjohn
2496 		/* allocate worst-case memory */
2497 		runsizes = ((npages / 2) + 1) * sizeof(struct puffs_cacherun);
2498 		pcinfo = kmem_zalloc(sizeof(struct puffs_cacheinfo) + runsize,
2499 		    locked ? KM_NOSLEEP : KM_SLEEP);
2500 
2501 		/*
2502 		 * can't block if we're locked and can't mess up caching
2503 		 * information for fs server.  so come back later, please
2504 		 */
2505 		if (pcinfo == NULL)
2506 			ERROUT(ENOMEM);
2507 
2508 		parkmem = puffs_park_alloc(locked == 0);
2509 		if (parkmem == NULL)
2510 			ERROUT(ENOMEM);
2511 
2512 		pcrun = pcinfo->pcache_runs;
2513 #else
2514 		(void)parkmem;
2515 #endif
2516 	}
2517 
2518 	error = genfs_getpages(v);
2519 	if (error)
2520 		goto out;
2521 
2522 	if (PUFFS_WCACHEINFO(pmp) == 0)
2523 		goto out;
2524 
2525 	/*
2526 	 * Let's see whose fault it was and inform the user server of
2527 	 * possibly read/written pages.  Map pages from read faults
2528 	 * strictly read-only, since otherwise we might miss info on
2529 	 * when the page is actually write-faulted to.
2530 	 */
2531 	if (!locked)
2532 		mutex_enter(&vp->v_uobj.vmobjlock);
2533 	for (i = 0, si = 0, streakon = 0; i < npages; i++) {
2534 		if (pgs[i] == NULL || pgs[i] == PGO_DONTCARE) {
2535 			if (streakon && write) {
2536 				streakon = 0;
2537 				pcrun[si].pcache_runend
2538 				    = trunc_page(pgs[i]->offset) + PAGE_MASK;
2539 				si++;
2540 			}
2541 			continue;
2542 		}
2543 		if (streakon == 0 && write) {
2544 			streakon = 1;
2545 			pcrun[si].pcache_runstart = pgs[i]->offset;
2546 		}
2547 
2548 		if (!write)
2549 			pgs[i]->flags |= PG_RDONLY;
2550 	}
2551 	/* was the last page part of our streak? */
2552 	if (streakon) {
2553 		pcrun[si].pcache_runend
2554 		    = trunc_page(pgs[i-1]->offset) + PAGE_MASK;
2555 		si++;
2556 	}
2557 	if (!locked)
2558 		mutex_exit(&vp->v_uobj.vmobjlock);
2559 
2560 	KASSERT(si <= (npages / 2) + 1);
2561 
2562 #ifdef notnowjohn
2563 	/* send results to userspace */
2564 	if (write)
2565 		puffs_cacheop(pmp, parkmem, pcinfo,
2566 		    sizeof(struct puffs_cacheinfo) + runsizes, VPTOPNC(vp));
2567 #endif
2568 
2569  out:
2570 	if (error) {
2571 		if (pcinfo != NULL)
2572 			kmem_free(pcinfo,
2573 			    sizeof(struct puffs_cacheinfo) + runsizes);
2574 #ifdef notnowjohn
2575 		if (parkmem != NULL)
2576 			puffs_park_release(parkmem, 1);
2577 #endif
2578 	}
2579 
2580 	return error;
2581 }
2582 
2583 /*
2584  * spec & fifo.  These call the miscfs spec and fifo vectors, but issue
2585  * FAF update information for the puffs node first.
2586  */
2587 int
2588 puffs_vnop_spec_read(void *v)
2589 {
2590 	struct vop_read_args /* {
2591 		const struct vnodeop_desc *a_desc;
2592 		struct vnode *a_vp;
2593 		struct uio *a_uio;
2594 		int a_ioflag;
2595 		kauth_cred_t a_cred;
2596 	} */ *ap = v;
2597 
2598 	puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
2599 	return VOCALL(spec_vnodeop_p, VOFFSET(vop_read), v);
2600 }
2601 
2602 int
2603 puffs_vnop_spec_write(void *v)
2604 {
2605 	struct vop_write_args /* {
2606 		const struct vnodeop_desc *a_desc;
2607 		struct vnode *a_vp;
2608 		struct uio *a_uio;
2609 		int a_ioflag;
2610 		kauth_cred_t a_cred;
2611 	} */ *ap = v;
2612 
2613 	puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
2614 	return VOCALL(spec_vnodeop_p, VOFFSET(vop_write), v);
2615 }
2616 
2617 int
2618 puffs_vnop_fifo_read(void *v)
2619 {
2620 	struct vop_read_args /* {
2621 		const struct vnodeop_desc *a_desc;
2622 		struct vnode *a_vp;
2623 		struct uio *a_uio;
2624 		int a_ioflag;
2625 		kauth_cred_t a_cred;
2626 	} */ *ap = v;
2627 
2628 	puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
2629 	return VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), v);
2630 }
2631 
2632 int
2633 puffs_vnop_fifo_write(void *v)
2634 {
2635 	struct vop_write_args /* {
2636 		const struct vnodeop_desc *a_desc;
2637 		struct vnode *a_vp;
2638 		struct uio *a_uio;
2639 		int a_ioflag;
2640 		kauth_cred_t a_cred;
2641 	} */ *ap = v;
2642 
2643 	puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
2644 	return VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), v);
2645 }
2646