puffs_vnops.c revision 1.196 1 /* $NetBSD: puffs_vnops.c,v 1.196 2014/10/31 13:52:41 manu 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 1.196 2014/10/31 13:52:41 manu 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, hardclock_ticks */
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_getattr_desc, puffs_vnop_checkop }, /* getattr */
115 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */
116 { &vop_read_desc, puffs_vnop_checkop }, /* read */
117 { &vop_write_desc, puffs_vnop_checkop }, /* write */
118 { &vop_fallocate_desc, puffs_vnop_fallocate }, /* fallocate */
119 { &vop_fdiscard_desc, puffs_vnop_fdiscard }, /* fdiscard */
120 { &vop_fsync_desc, puffs_vnop_fsync }, /* REAL fsync */
121 { &vop_seek_desc, puffs_vnop_checkop }, /* seek */
122 { &vop_remove_desc, puffs_vnop_checkop }, /* remove */
123 { &vop_link_desc, puffs_vnop_checkop }, /* link */
124 { &vop_rename_desc, puffs_vnop_checkop }, /* rename */
125 { &vop_mkdir_desc, puffs_vnop_checkop }, /* mkdir */
126 { &vop_rmdir_desc, puffs_vnop_checkop }, /* rmdir */
127 { &vop_symlink_desc, puffs_vnop_checkop }, /* symlink */
128 { &vop_readdir_desc, puffs_vnop_checkop }, /* readdir */
129 { &vop_readlink_desc, puffs_vnop_checkop }, /* readlink */
130 { &vop_getpages_desc, puffs_vnop_checkop }, /* getpages */
131 { &vop_putpages_desc, genfs_putpages }, /* REAL putpages */
132 { &vop_pathconf_desc, puffs_vnop_checkop }, /* pathconf */
133 { &vop_advlock_desc, puffs_vnop_advlock }, /* advlock */
134 { &vop_strategy_desc, puffs_vnop_strategy }, /* REAL strategy */
135 { &vop_revoke_desc, genfs_revoke }, /* REAL revoke */
136 { &vop_abortop_desc, puffs_vnop_abortop }, /* REAL abortop */
137 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
138 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
139 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
140 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
141 { &vop_bmap_desc, puffs_vnop_bmap }, /* REAL bmap */
142 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
143 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
144 { &vop_bwrite_desc, genfs_nullop }, /* REAL bwrite */
145 { &vop_mmap_desc, puffs_vnop_mmap }, /* REAL mmap */
146 { &vop_poll_desc, puffs_vnop_poll }, /* REAL poll */
147 { &vop_getextattr_desc, puffs_vnop_getextattr }, /* getextattr */
148 { &vop_setextattr_desc, puffs_vnop_setextattr }, /* setextattr */
149 { &vop_listextattr_desc, puffs_vnop_listextattr }, /* listextattr */
150 { &vop_deleteextattr_desc, puffs_vnop_deleteextattr },/* deleteextattr */
151 #if 0
152 { &vop_openextattr_desc, puffs_vnop_checkop }, /* openextattr */
153 { &vop_closeextattr_desc, puffs_vnop_checkop }, /* closeextattr */
154 #endif
155 { &vop_kqfilter_desc, genfs_eopnotsupp }, /* kqfilter XXX */
156 { NULL, NULL }
157 };
158 const struct vnodeopv_desc puffs_vnodeop_opv_desc =
159 { &puffs_vnodeop_p, puffs_vnodeop_entries };
160
161
162 int (**puffs_specop_p)(void *);
163 const struct vnodeopv_entry_desc puffs_specop_entries[] = {
164 { &vop_default_desc, vn_default_error },
165 { &vop_lookup_desc, spec_lookup }, /* lookup, ENOTDIR */
166 { &vop_create_desc, spec_create }, /* genfs_badop */
167 { &vop_mknod_desc, spec_mknod }, /* genfs_badop */
168 { &vop_open_desc, spec_open }, /* spec_open */
169 { &vop_close_desc, spec_close }, /* spec_close */
170 { &vop_access_desc, puffs_vnop_checkop }, /* access */
171 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */
172 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */
173 { &vop_read_desc, puffs_vnop_spec_read }, /* update, read */
174 { &vop_write_desc, puffs_vnop_spec_write }, /* update, write */
175 { &vop_fallocate_desc, spec_fallocate }, /* fallocate */
176 { &vop_fdiscard_desc, spec_fdiscard }, /* fdiscard */
177 { &vop_ioctl_desc, spec_ioctl }, /* spec_ioctl */
178 { &vop_fcntl_desc, genfs_fcntl }, /* dummy */
179 { &vop_poll_desc, spec_poll }, /* spec_poll */
180 { &vop_kqfilter_desc, spec_kqfilter }, /* spec_kqfilter */
181 { &vop_revoke_desc, spec_revoke }, /* genfs_revoke */
182 { &vop_mmap_desc, spec_mmap }, /* spec_mmap */
183 { &vop_fsync_desc, spec_fsync }, /* vflushbuf */
184 { &vop_seek_desc, spec_seek }, /* genfs_nullop */
185 { &vop_remove_desc, spec_remove }, /* genfs_badop */
186 { &vop_link_desc, spec_link }, /* genfs_badop */
187 { &vop_rename_desc, spec_rename }, /* genfs_badop */
188 { &vop_mkdir_desc, spec_mkdir }, /* genfs_badop */
189 { &vop_rmdir_desc, spec_rmdir }, /* genfs_badop */
190 { &vop_symlink_desc, spec_symlink }, /* genfs_badop */
191 { &vop_readdir_desc, spec_readdir }, /* genfs_badop */
192 { &vop_readlink_desc, spec_readlink }, /* genfs_badop */
193 { &vop_abortop_desc, spec_abortop }, /* genfs_badop */
194 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
195 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
196 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
197 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
198 { &vop_bmap_desc, spec_bmap }, /* dummy */
199 { &vop_strategy_desc, spec_strategy }, /* dev strategy */
200 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
201 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
202 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
203 { &vop_advlock_desc, spec_advlock }, /* lf_advlock */
204 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
205 { &vop_getpages_desc, spec_getpages }, /* genfs_getpages */
206 { &vop_putpages_desc, spec_putpages }, /* genfs_putpages */
207 { &vop_getextattr_desc, puffs_vnop_checkop }, /* getextattr */
208 { &vop_setextattr_desc, puffs_vnop_checkop }, /* setextattr */
209 { &vop_listextattr_desc, puffs_vnop_checkop }, /* listextattr */
210 { &vop_deleteextattr_desc, puffs_vnop_checkop },/* deleteextattr */
211 #if 0
212 { &vop_openextattr_desc, _openextattr }, /* openextattr */
213 { &vop_closeextattr_desc, _closeextattr }, /* closeextattr */
214 #endif
215 { NULL, NULL }
216 };
217 const struct vnodeopv_desc puffs_specop_opv_desc =
218 { &puffs_specop_p, puffs_specop_entries };
219
220
221 int (**puffs_fifoop_p)(void *);
222 const struct vnodeopv_entry_desc puffs_fifoop_entries[] = {
223 { &vop_default_desc, vn_default_error },
224 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup, ENOTDIR */
225 { &vop_create_desc, vn_fifo_bypass }, /* genfs_badop */
226 { &vop_mknod_desc, vn_fifo_bypass }, /* genfs_badop */
227 { &vop_open_desc, vn_fifo_bypass }, /* open */
228 { &vop_close_desc, vn_fifo_bypass }, /* close */
229 { &vop_access_desc, puffs_vnop_checkop }, /* access */
230 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */
231 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */
232 { &vop_read_desc, puffs_vnop_fifo_read }, /* read, update */
233 { &vop_write_desc, puffs_vnop_fifo_write }, /* write, update */
234 { &vop_fallocate_desc, vn_fifo_bypass }, /* fallocate */
235 { &vop_fdiscard_desc, vn_fifo_bypass }, /* fdiscard */
236 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */
237 { &vop_fcntl_desc, genfs_fcntl }, /* dummy */
238 { &vop_poll_desc, vn_fifo_bypass }, /* poll */
239 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */
240 { &vop_revoke_desc, vn_fifo_bypass }, /* genfs_revoke */
241 { &vop_mmap_desc, vn_fifo_bypass }, /* genfs_badop */
242 { &vop_fsync_desc, vn_fifo_bypass }, /* genfs_nullop*/
243 { &vop_seek_desc, vn_fifo_bypass }, /* genfs_badop */
244 { &vop_remove_desc, vn_fifo_bypass }, /* genfs_badop */
245 { &vop_link_desc, vn_fifo_bypass }, /* genfs_badop */
246 { &vop_rename_desc, vn_fifo_bypass }, /* genfs_badop */
247 { &vop_mkdir_desc, vn_fifo_bypass }, /* genfs_badop */
248 { &vop_rmdir_desc, vn_fifo_bypass }, /* genfs_badop */
249 { &vop_symlink_desc, vn_fifo_bypass }, /* genfs_badop */
250 { &vop_readdir_desc, vn_fifo_bypass }, /* genfs_badop */
251 { &vop_readlink_desc, vn_fifo_bypass }, /* genfs_badop */
252 { &vop_abortop_desc, vn_fifo_bypass }, /* genfs_badop */
253 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
254 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
255 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
256 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
257 { &vop_bmap_desc, vn_fifo_bypass }, /* dummy */
258 { &vop_strategy_desc, vn_fifo_bypass }, /* genfs_badop */
259 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
260 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
261 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */
262 { &vop_advlock_desc, vn_fifo_bypass }, /* genfs_einval */
263 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
264 { &vop_putpages_desc, vn_fifo_bypass }, /* genfs_null_putpages*/
265 #if 0
266 { &vop_openextattr_desc, _openextattr }, /* openextattr */
267 { &vop_closeextattr_desc, _closeextattr }, /* closeextattr */
268 #endif
269 { &vop_getextattr_desc, puffs_vnop_checkop }, /* getextattr */
270 { &vop_setextattr_desc, puffs_vnop_checkop }, /* setextattr */
271 { &vop_listextattr_desc, puffs_vnop_checkop }, /* listextattr */
272 { &vop_deleteextattr_desc, puffs_vnop_checkop }, /* deleteextattr */
273 { NULL, NULL }
274 };
275 const struct vnodeopv_desc puffs_fifoop_opv_desc =
276 { &puffs_fifoop_p, puffs_fifoop_entries };
277
278
279 /* "real" vnode operations */
280 int (**puffs_msgop_p)(void *);
281 const struct vnodeopv_entry_desc puffs_msgop_entries[] = {
282 { &vop_default_desc, vn_default_error },
283 { &vop_create_desc, puffs_vnop_create }, /* create */
284 { &vop_mknod_desc, puffs_vnop_mknod }, /* mknod */
285 { &vop_open_desc, puffs_vnop_open }, /* open */
286 { &vop_close_desc, puffs_vnop_close }, /* close */
287 { &vop_access_desc, puffs_vnop_access }, /* access */
288 { &vop_getattr_desc, puffs_vnop_getattr }, /* getattr */
289 { &vop_setattr_desc, puffs_vnop_setattr }, /* setattr */
290 { &vop_read_desc, puffs_vnop_read }, /* read */
291 { &vop_write_desc, puffs_vnop_write }, /* write */
292 { &vop_seek_desc, puffs_vnop_seek }, /* seek */
293 { &vop_remove_desc, puffs_vnop_remove }, /* remove */
294 { &vop_link_desc, puffs_vnop_link }, /* link */
295 { &vop_rename_desc, puffs_vnop_rename }, /* rename */
296 { &vop_mkdir_desc, puffs_vnop_mkdir }, /* mkdir */
297 { &vop_rmdir_desc, puffs_vnop_rmdir }, /* rmdir */
298 { &vop_symlink_desc, puffs_vnop_symlink }, /* symlink */
299 { &vop_readdir_desc, puffs_vnop_readdir }, /* readdir */
300 { &vop_readlink_desc, puffs_vnop_readlink }, /* readlink */
301 { &vop_print_desc, puffs_vnop_print }, /* print */
302 { &vop_islocked_desc, puffs_vnop_islocked }, /* islocked */
303 { &vop_pathconf_desc, puffs_vnop_pathconf }, /* pathconf */
304 { &vop_getpages_desc, puffs_vnop_getpages }, /* getpages */
305 { NULL, NULL }
306 };
307 const struct vnodeopv_desc puffs_msgop_opv_desc =
308 { &puffs_msgop_p, puffs_msgop_entries };
309
310 /*
311 * for dosetattr / update_va
312 */
313 #define SETATTR_CHSIZE 0x01
314 #define SETATTR_ASYNC 0x02
315
316 #define ERROUT(err) \
317 do { \
318 error = err; \
319 goto out; \
320 } while (/*CONSTCOND*/0)
321
322 /*
323 * This is a generic vnode operation handler. It checks if the necessary
324 * operations for the called vnode operation are implemented by userspace
325 * and either returns a dummy return value or proceeds to call the real
326 * vnode operation from puffs_msgop_v.
327 *
328 * XXX: this should described elsewhere and autogenerated, the complexity
329 * of the vnode operations vectors and their interrelationships is also
330 * getting a bit out of hand. Another problem is that we need this same
331 * information in the fs server code, so keeping the two in sync manually
332 * is not a viable (long term) plan.
333 */
334
335 /* not supported, handle locking protocol */
336 #define CHECKOP_NOTSUPP(op) \
337 case VOP_##op##_DESCOFFSET: \
338 if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0) \
339 return genfs_eopnotsupp(v); \
340 break
341
342 /* always succeed, no locking */
343 #define CHECKOP_SUCCESS(op) \
344 case VOP_##op##_DESCOFFSET: \
345 if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0) \
346 return 0; \
347 break
348
349 int
350 puffs_vnop_checkop(void *v)
351 {
352 struct vop_generic_args /* {
353 struct vnodeop_desc *a_desc;
354 spooky mystery contents;
355 } */ *ap = v;
356 struct vnodeop_desc *desc = ap->a_desc;
357 struct puffs_mount *pmp;
358 struct vnode *vp;
359 int offset, rv;
360
361 offset = ap->a_desc->vdesc_vp_offsets[0];
362 #ifdef DIAGNOSTIC
363 if (offset == VDESC_NO_OFFSET)
364 panic("puffs_checkop: no vnode, why did you call me?");
365 #endif
366 vp = *VOPARG_OFFSETTO(struct vnode **, offset, ap);
367 pmp = MPTOPUFFSMP(vp->v_mount);
368
369 DPRINTF_VERBOSE(("checkop call %s (%d), vp %p\n",
370 ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp));
371
372 if (!ALLOPS(pmp)) {
373 switch (desc->vdesc_offset) {
374 CHECKOP_NOTSUPP(CREATE);
375 CHECKOP_NOTSUPP(MKNOD);
376 CHECKOP_NOTSUPP(GETATTR);
377 CHECKOP_NOTSUPP(SETATTR);
378 CHECKOP_NOTSUPP(READ);
379 CHECKOP_NOTSUPP(WRITE);
380 CHECKOP_NOTSUPP(FCNTL);
381 CHECKOP_NOTSUPP(IOCTL);
382 CHECKOP_NOTSUPP(REMOVE);
383 CHECKOP_NOTSUPP(LINK);
384 CHECKOP_NOTSUPP(RENAME);
385 CHECKOP_NOTSUPP(MKDIR);
386 CHECKOP_NOTSUPP(RMDIR);
387 CHECKOP_NOTSUPP(SYMLINK);
388 CHECKOP_NOTSUPP(READDIR);
389 CHECKOP_NOTSUPP(READLINK);
390 CHECKOP_NOTSUPP(PRINT);
391 CHECKOP_NOTSUPP(PATHCONF);
392 CHECKOP_NOTSUPP(GETEXTATTR);
393 CHECKOP_NOTSUPP(SETEXTATTR);
394 CHECKOP_NOTSUPP(LISTEXTATTR);
395 CHECKOP_NOTSUPP(DELETEEXTATTR);
396
397 CHECKOP_SUCCESS(ACCESS);
398 CHECKOP_SUCCESS(CLOSE);
399 CHECKOP_SUCCESS(SEEK);
400
401 case VOP_GETPAGES_DESCOFFSET:
402 if (!EXISTSOP(pmp, READ))
403 return genfs_eopnotsupp(v);
404 break;
405
406 default:
407 panic("puffs_checkop: unhandled vnop %d",
408 desc->vdesc_offset);
409 }
410 }
411
412 rv = VOCALL(puffs_msgop_p, ap->a_desc->vdesc_offset, v);
413
414 DPRINTF_VERBOSE(("checkop return %s (%d), vp %p: %d\n",
415 ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp, rv));
416
417 return rv;
418 }
419
420 static int callremove(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
421 struct componentname *);
422 static int callrmdir(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
423 struct componentname *);
424 static void callinactive(struct puffs_mount *, puffs_cookie_t, int);
425 static void callreclaim(struct puffs_mount *, puffs_cookie_t, int);
426 static int flushvncache(struct vnode *, off_t, off_t, bool);
427 static void update_va(struct vnode *, struct vattr *, struct vattr *,
428 struct timespec *, struct timespec *, int);
429 static void update_parent(struct vnode *, struct vnode *);
430
431
432 #define PUFFS_ABORT_LOOKUP 1
433 #define PUFFS_ABORT_CREATE 2
434 #define PUFFS_ABORT_MKNOD 3
435 #define PUFFS_ABORT_MKDIR 4
436 #define PUFFS_ABORT_SYMLINK 5
437
438 /*
439 * Press the pani^Wabort button! Kernel resource allocation failed.
440 */
441 static void
442 puffs_abortbutton(struct puffs_mount *pmp, int what,
443 puffs_cookie_t dck, puffs_cookie_t ck, struct componentname *cnp)
444 {
445
446 switch (what) {
447 case PUFFS_ABORT_CREATE:
448 case PUFFS_ABORT_MKNOD:
449 case PUFFS_ABORT_SYMLINK:
450 callremove(pmp, dck, ck, cnp);
451 break;
452 case PUFFS_ABORT_MKDIR:
453 callrmdir(pmp, dck, ck, cnp);
454 break;
455 }
456
457 callinactive(pmp, ck, 0);
458 callreclaim(pmp, ck, 1);
459 }
460
461 /*
462 * Begin vnode operations.
463 *
464 * A word from the keymaster about locks: generally we don't want
465 * to use the vnode locks at all: it creates an ugly dependency between
466 * the userlandia file server and the kernel. But we'll play along with
467 * the kernel vnode locks for now. However, even currently we attempt
468 * to release locks as early as possible. This is possible for some
469 * operations which a) don't need a locked vnode after the userspace op
470 * and b) return with the vnode unlocked. Theoretically we could
471 * unlock-do op-lock for others and order the graph in userspace, but I
472 * don't want to think of the consequences for the time being.
473 */
474
475 #define TTL_TO_TIMEOUT(ts) \
476 (hardclock_ticks + (ts->tv_sec * hz) + (ts->tv_nsec * hz / 1000000000))
477 #define TTL_VALID(ts) \
478 ((ts != NULL) && !((ts->tv_sec == 0) && (ts->tv_nsec == 0)))
479 #define TIMED_OUT(expire) \
480 ((int)((unsigned int)hardclock_ticks - (unsigned int)expire) > 0)
481 int
482 puffs_vnop_lookup(void *v)
483 {
484 struct vop_lookup_v2_args /* {
485 const struct vnodeop_desc *a_desc;
486 struct vnode *a_dvp;
487 struct vnode **a_vpp;
488 struct componentname *a_cnp;
489 } */ *ap = v;
490 PUFFS_MSG_VARS(vn, lookup);
491 struct puffs_mount *pmp;
492 struct componentname *cnp;
493 struct vnode *vp, *dvp, *cvp;
494 struct puffs_node *dpn, *cpn;
495 int isdot;
496 int error;
497
498 pmp = MPTOPUFFSMP(ap->a_dvp->v_mount);
499 cnp = ap->a_cnp;
500 dvp = ap->a_dvp;
501 cvp = NULL;
502 cpn = NULL;
503 *ap->a_vpp = NULL;
504
505 /* r/o fs? we check create later to handle EEXIST */
506 if ((cnp->cn_flags & ISLASTCN)
507 && (dvp->v_mount->mnt_flag & MNT_RDONLY)
508 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
509 return EROFS;
510
511 isdot = cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.';
512
513 DPRINTF(("puffs_lookup: \"%s\", parent vnode %p, op: %x\n",
514 cnp->cn_nameptr, dvp, cnp->cn_nameiop));
515
516 /*
517 * If dotdot cache is enabled, add reference to .. and return.
518 */
519 if (PUFFS_USE_DOTDOTCACHE(pmp) && (cnp->cn_flags & ISDOTDOT)) {
520 vp = VPTOPP(ap->a_dvp)->pn_parent;
521 vref(vp);
522
523 *ap->a_vpp = vp;
524 return 0;
525 }
526
527 /*
528 * Check if someone fed it into the cache
529 */
530 if (!isdot && PUFFS_USE_NAMECACHE(pmp)) {
531 int found, iswhiteout;
532
533 found = cache_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
534 cnp->cn_nameiop, cnp->cn_flags,
535 &iswhiteout, ap->a_vpp);
536 if (iswhiteout) {
537 cnp->cn_flags |= ISWHITEOUT;
538 }
539
540 if (found && *ap->a_vpp != NULLVP && PUFFS_USE_FS_TTL(pmp)) {
541 cvp = *ap->a_vpp;
542 cpn = VPTOPP(cvp);
543
544 if (TIMED_OUT(cpn->pn_cn_timeout)) {
545 cache_purge(cvp);
546 /*
547 * cached vnode (cvp) is still referenced
548 * so that we can reuse it upon a new
549 * successful lookup.
550 */
551 *ap->a_vpp = NULL;
552 found = 0;
553 }
554 }
555
556 /*
557 * Do not use negative caching, since the filesystem
558 * provides no TTL for it.
559 */
560 if (found && *ap->a_vpp == NULLVP && PUFFS_USE_FS_TTL(pmp))
561 found = 0;
562
563 if (found) {
564 return *ap->a_vpp == NULLVP ? ENOENT : 0;
565 }
566
567 /*
568 * This is what would have been left in ERROR before
569 * the rearrangement of cache_lookup(). What with all
570 * the macros, I am not sure if this is a dead value
571 * below or not.
572 */
573 error = -1;
574 }
575
576 if (isdot) {
577 /* deal with rename lookup semantics */
578 if (cnp->cn_nameiop == RENAME && (cnp->cn_flags & ISLASTCN))
579 return EISDIR;
580
581 vp = ap->a_dvp;
582 vref(vp);
583 *ap->a_vpp = vp;
584 return 0;
585 }
586
587 if (cvp != NULL) {
588 if (vn_lock(cvp, LK_EXCLUSIVE) != 0) {
589 vrele(cvp);
590 cvp = NULL;
591 } else
592 mutex_enter(&cpn->pn_sizemtx);
593 }
594
595 PUFFS_MSG_ALLOC(vn, lookup);
596 puffs_makecn(&lookup_msg->pvnr_cn, &lookup_msg->pvnr_cn_cred,
597 cnp, PUFFS_USE_FULLPNBUF(pmp));
598
599 if (cnp->cn_flags & ISDOTDOT)
600 VOP_UNLOCK(dvp);
601
602 puffs_msg_setinfo(park_lookup, PUFFSOP_VN,
603 PUFFS_VN_LOOKUP, VPTOPNC(dvp));
604 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_lookup, dvp->v_data, NULL, error);
605 DPRINTF(("puffs_lookup: return of the userspace, part %d\n", error));
606
607 /*
608 * In case of error, there is no new vnode to play with, so be
609 * happy with the NULL value given to vpp in the beginning.
610 * Also, check if this really was an error or the target was not
611 * present. Either treat it as a non-error for CREATE/RENAME or
612 * enter the component into the negative name cache (if desired).
613 */
614 if (error) {
615 error = checkerr(pmp, error, __func__);
616 if (error == ENOENT) {
617 /* don't allow to create files on r/o fs */
618 if ((dvp->v_mount->mnt_flag & MNT_RDONLY)
619 && cnp->cn_nameiop == CREATE) {
620 error = EROFS;
621
622 /* adjust values if we are creating */
623 } else if ((cnp->cn_flags & ISLASTCN)
624 && (cnp->cn_nameiop == CREATE
625 || cnp->cn_nameiop == RENAME)) {
626 error = EJUSTRETURN;
627
628 /* save negative cache entry */
629 } else {
630 if (PUFFS_USE_NAMECACHE(pmp) &&
631 !PUFFS_USE_FS_TTL(pmp))
632 cache_enter(dvp, NULL, cnp->cn_nameptr,
633 cnp->cn_namelen, cnp->cn_flags);
634 }
635 }
636 goto out;
637 }
638
639 /*
640 * Check that we don't get our parent node back, that would cause
641 * a pretty obvious deadlock.
642 */
643 dpn = dvp->v_data;
644 if (lookup_msg->pvnr_newnode == dpn->pn_cookie) {
645 puffs_senderr(pmp, PUFFS_ERR_LOOKUP, EINVAL,
646 "lookup produced parent cookie", lookup_msg->pvnr_newnode);
647 error = EPROTO;
648 goto out;
649 }
650
651 /*
652 * Check if we looked up the cached vnode
653 */
654 vp = NULL;
655 if (cvp && (VPTOPP(cvp)->pn_cookie == lookup_msg->pvnr_newnode)) {
656 int grace;
657
658 /*
659 * Bump grace time of this node so that it does not get
660 * reclaimed too fast. We try to increase a bit more the
661 * lifetime of busiest * nodes - with some limits.
662 */
663 grace = 10 * puffs_sopreq_expire_timeout;
664 cpn->pn_cn_grace = hardclock_ticks + grace;
665 vp = cvp;
666 }
667
668 /*
669 * No cached vnode available, or the cached vnode does not
670 * match the userland cookie anymore: is the node known?
671 */
672 if (vp == NULL) {
673 error = puffs_getvnode(dvp->v_mount,
674 lookup_msg->pvnr_newnode, lookup_msg->pvnr_vtype,
675 lookup_msg->pvnr_size, lookup_msg->pvnr_rdev, &vp);
676 if (error) {
677 puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP,
678 VPTOPNC(dvp), lookup_msg->pvnr_newnode,
679 ap->a_cnp);
680 goto out;
681 }
682
683 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
684 }
685
686 /*
687 * Update cache and TTL
688 */
689 if (PUFFS_USE_FS_TTL(pmp)) {
690 struct timespec *va_ttl = &lookup_msg->pvnr_va_ttl;
691 struct timespec *cn_ttl = &lookup_msg->pvnr_cn_ttl;
692 update_va(vp, NULL, &lookup_msg->pvnr_va,
693 va_ttl, cn_ttl, SETATTR_CHSIZE);
694 }
695
696 KASSERT(lookup_msg->pvnr_newnode == VPTOPP(vp)->pn_cookie);
697 *ap->a_vpp = vp;
698
699 if (PUFFS_USE_NAMECACHE(pmp))
700 cache_enter(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
701 cnp->cn_flags);
702
703 /* XXX */
704 if ((lookup_msg->pvnr_cn.pkcn_flags & REQUIREDIR) == 0)
705 cnp->cn_flags &= ~REQUIREDIR;
706 if (lookup_msg->pvnr_cn.pkcn_consume)
707 cnp->cn_consume = MIN(lookup_msg->pvnr_cn.pkcn_consume,
708 strlen(cnp->cn_nameptr) - cnp->cn_namelen);
709
710 VPTOPP(vp)->pn_nlookup++;
711
712 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
713 (VPTOPP(vp)->pn_parent != dvp))
714 update_parent(vp, dvp);
715
716 out:
717 if (cvp != NULL) {
718 mutex_exit(&cpn->pn_sizemtx);
719
720 if (error || (cvp != vp))
721 vput(cvp);
722 }
723 if (error == 0)
724 VOP_UNLOCK(*ap->a_vpp);
725
726 if (cnp->cn_flags & ISDOTDOT)
727 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
728
729 DPRINTF(("puffs_lookup: returning %d %p\n", error, *ap->a_vpp));
730 PUFFS_MSG_RELEASE(lookup);
731 return error;
732 }
733
734 #define REFPN_AND_UNLOCKVP(a, b) \
735 do { \
736 mutex_enter(&b->pn_mtx); \
737 puffs_referencenode(b); \
738 mutex_exit(&b->pn_mtx); \
739 VOP_UNLOCK(a); \
740 } while (/*CONSTCOND*/0)
741
742 #define REFPN(b) \
743 do { \
744 mutex_enter(&b->pn_mtx); \
745 puffs_referencenode(b); \
746 mutex_exit(&b->pn_mtx); \
747 } while (/*CONSTCOND*/0)
748
749 #define RELEPN_AND_VP(a, b) \
750 do { \
751 puffs_releasenode(b); \
752 vrele(a); \
753 } while (/*CONSTCOND*/0)
754
755 int
756 puffs_vnop_create(void *v)
757 {
758 struct vop_create_v3_args /* {
759 const struct vnodeop_desc *a_desc;
760 struct vnode *a_dvp;
761 struct vnode **a_vpp;
762 struct componentname *a_cnp;
763 struct vattr *a_vap;
764 } */ *ap = v;
765 PUFFS_MSG_VARS(vn, create);
766 struct vnode *dvp = ap->a_dvp;
767 struct puffs_node *dpn = VPTOPP(dvp);
768 struct componentname *cnp = ap->a_cnp;
769 struct mount *mp = dvp->v_mount;
770 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
771 int error;
772
773 DPRINTF(("puffs_create: dvp %p, cnp: %s\n",
774 dvp, ap->a_cnp->cn_nameptr));
775
776 PUFFS_MSG_ALLOC(vn, create);
777 puffs_makecn(&create_msg->pvnr_cn, &create_msg->pvnr_cn_cred,
778 cnp, PUFFS_USE_FULLPNBUF(pmp));
779 create_msg->pvnr_va = *ap->a_vap;
780 puffs_msg_setinfo(park_create, PUFFSOP_VN,
781 PUFFS_VN_CREATE, VPTOPNC(dvp));
782 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_create, dvp->v_data, NULL, error);
783
784 error = checkerr(pmp, error, __func__);
785 if (error)
786 goto out;
787
788 error = puffs_newnode(mp, dvp, ap->a_vpp,
789 create_msg->pvnr_newnode, cnp, ap->a_vap->va_type, 0);
790 if (error) {
791 puffs_abortbutton(pmp, PUFFS_ABORT_CREATE, dpn->pn_cookie,
792 create_msg->pvnr_newnode, cnp);
793 goto out;
794 }
795
796 if (PUFFS_USE_FS_TTL(pmp)) {
797 struct timespec *va_ttl = &create_msg->pvnr_va_ttl;
798 struct timespec *cn_ttl = &create_msg->pvnr_cn_ttl;
799 struct vattr *rvap = &create_msg->pvnr_va;
800
801 update_va(*ap->a_vpp, NULL, rvap,
802 va_ttl, cn_ttl, SETATTR_CHSIZE);
803 }
804
805 VPTOPP(*ap->a_vpp)->pn_nlookup++;
806
807 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
808 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
809 update_parent(*ap->a_vpp, dvp);
810
811 out:
812 DPRINTF(("puffs_create: return %d\n", error));
813 PUFFS_MSG_RELEASE(create);
814 return error;
815 }
816
817 int
818 puffs_vnop_mknod(void *v)
819 {
820 struct vop_mknod_v3_args /* {
821 const struct vnodeop_desc *a_desc;
822 struct vnode *a_dvp;
823 struct vnode **a_vpp;
824 struct componentname *a_cnp;
825 struct vattr *a_vap;
826 } */ *ap = v;
827 PUFFS_MSG_VARS(vn, mknod);
828 struct vnode *dvp = ap->a_dvp;
829 struct puffs_node *dpn = VPTOPP(dvp);
830 struct componentname *cnp = ap->a_cnp;
831 struct mount *mp = dvp->v_mount;
832 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
833 int error;
834
835 PUFFS_MSG_ALLOC(vn, mknod);
836 puffs_makecn(&mknod_msg->pvnr_cn, &mknod_msg->pvnr_cn_cred,
837 cnp, PUFFS_USE_FULLPNBUF(pmp));
838 mknod_msg->pvnr_va = *ap->a_vap;
839 puffs_msg_setinfo(park_mknod, PUFFSOP_VN,
840 PUFFS_VN_MKNOD, VPTOPNC(dvp));
841
842 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mknod, dvp->v_data, NULL, error);
843
844 error = checkerr(pmp, error, __func__);
845 if (error)
846 goto out;
847
848 error = puffs_newnode(mp, dvp, ap->a_vpp,
849 mknod_msg->pvnr_newnode, cnp, ap->a_vap->va_type,
850 ap->a_vap->va_rdev);
851 if (error) {
852 puffs_abortbutton(pmp, PUFFS_ABORT_MKNOD, dpn->pn_cookie,
853 mknod_msg->pvnr_newnode, cnp);
854 goto out;
855 }
856
857 if (PUFFS_USE_FS_TTL(pmp)) {
858 struct timespec *va_ttl = &mknod_msg->pvnr_va_ttl;
859 struct timespec *cn_ttl = &mknod_msg->pvnr_cn_ttl;
860 struct vattr *rvap = &mknod_msg->pvnr_va;
861
862 update_va(*ap->a_vpp, NULL, rvap,
863 va_ttl, cn_ttl, SETATTR_CHSIZE);
864 }
865
866 VPTOPP(*ap->a_vpp)->pn_nlookup++;
867
868 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
869 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
870 update_parent(*ap->a_vpp, dvp);
871
872 out:
873 PUFFS_MSG_RELEASE(mknod);
874 return error;
875 }
876
877 int
878 puffs_vnop_open(void *v)
879 {
880 struct vop_open_args /* {
881 const struct vnodeop_desc *a_desc;
882 struct vnode *a_vp;
883 int a_mode;
884 kauth_cred_t a_cred;
885 } */ *ap = v;
886 PUFFS_MSG_VARS(vn, open);
887 struct vnode *vp = ap->a_vp;
888 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
889 struct puffs_node *pn = VPTOPP(vp);
890 int mode = ap->a_mode;
891 int error;
892
893 DPRINTF(("puffs_open: vp %p, mode 0x%x\n", vp, mode));
894
895 if (vp->v_type == VREG && mode & FWRITE && !EXISTSOP(pmp, WRITE))
896 ERROUT(EROFS);
897
898 if (!EXISTSOP(pmp, OPEN))
899 ERROUT(0);
900
901 PUFFS_MSG_ALLOC(vn, open);
902 open_msg->pvnr_mode = mode;
903 puffs_credcvt(&open_msg->pvnr_cred, ap->a_cred);
904 puffs_msg_setinfo(park_open, PUFFSOP_VN,
905 PUFFS_VN_OPEN, VPTOPNC(vp));
906
907 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_open, vp->v_data, NULL, error);
908 error = checkerr(pmp, error, __func__);
909
910 if (open_msg->pvnr_oflags & PUFFS_OPEN_IO_DIRECT) {
911 if (mode & FREAD)
912 pn->pn_stat |= PNODE_RDIRECT;
913 if (mode & FWRITE)
914 pn->pn_stat |= PNODE_WDIRECT;
915 }
916 out:
917 DPRINTF(("puffs_open: returning %d\n", error));
918 PUFFS_MSG_RELEASE(open);
919 return error;
920 }
921
922 int
923 puffs_vnop_close(void *v)
924 {
925 struct vop_close_args /* {
926 const struct vnodeop_desc *a_desc;
927 struct vnode *a_vp;
928 int a_fflag;
929 kauth_cred_t a_cred;
930 } */ *ap = v;
931 PUFFS_MSG_VARS(vn, close);
932 struct vnode *vp = ap->a_vp;
933 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
934
935 PUFFS_MSG_ALLOC(vn, close);
936 puffs_msg_setfaf(park_close);
937 close_msg->pvnr_fflag = ap->a_fflag;
938 puffs_credcvt(&close_msg->pvnr_cred, ap->a_cred);
939 puffs_msg_setinfo(park_close, PUFFSOP_VN,
940 PUFFS_VN_CLOSE, VPTOPNC(vp));
941
942 puffs_msg_enqueue(pmp, park_close);
943 PUFFS_MSG_RELEASE(close);
944 return 0;
945 }
946
947 int
948 puffs_vnop_access(void *v)
949 {
950 struct vop_access_args /* {
951 const struct vnodeop_desc *a_desc;
952 struct vnode *a_vp;
953 int a_mode;
954 kauth_cred_t a_cred;
955 } */ *ap = v;
956 PUFFS_MSG_VARS(vn, access);
957 struct vnode *vp = ap->a_vp;
958 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
959 int mode = ap->a_mode;
960 int error;
961
962 if (mode & VWRITE) {
963 switch (vp->v_type) {
964 case VDIR:
965 case VLNK:
966 case VREG:
967 if ((vp->v_mount->mnt_flag & MNT_RDONLY)
968 || !EXISTSOP(pmp, WRITE))
969 return EROFS;
970 break;
971 default:
972 break;
973 }
974 }
975
976 if (!EXISTSOP(pmp, ACCESS))
977 return 0;
978
979 PUFFS_MSG_ALLOC(vn, access);
980 access_msg->pvnr_mode = ap->a_mode;
981 puffs_credcvt(&access_msg->pvnr_cred, ap->a_cred);
982 puffs_msg_setinfo(park_access, PUFFSOP_VN,
983 PUFFS_VN_ACCESS, VPTOPNC(vp));
984
985 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_access, vp->v_data, NULL, error);
986 error = checkerr(pmp, error, __func__);
987 PUFFS_MSG_RELEASE(access);
988
989 return error;
990 }
991
992 static void
993 update_va(struct vnode *vp, struct vattr *vap, struct vattr *rvap,
994 struct timespec *va_ttl, struct timespec *cn_ttl, int flags)
995 {
996 struct puffs_node *pn = VPTOPP(vp);
997
998 if (TTL_VALID(cn_ttl)) {
999 pn->pn_cn_timeout = TTL_TO_TIMEOUT(cn_ttl);
1000 pn->pn_cn_grace = MAX(pn->pn_cn_timeout, pn->pn_cn_grace);
1001 }
1002
1003 /*
1004 * Don't listen to the file server regarding special device
1005 * size info, the file server doesn't know anything about them.
1006 */
1007 if (vp->v_type == VBLK || vp->v_type == VCHR)
1008 rvap->va_size = vp->v_size;
1009
1010 /* Ditto for blocksize (ufs comment: this doesn't belong here) */
1011 if (vp->v_type == VBLK)
1012 rvap->va_blocksize = BLKDEV_IOSIZE;
1013 else if (vp->v_type == VCHR)
1014 rvap->va_blocksize = MAXBSIZE;
1015
1016 if (vap != NULL) {
1017 (void) memcpy(vap, rvap, sizeof(struct vattr));
1018 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1019
1020 if (pn->pn_stat & PNODE_METACACHE_ATIME)
1021 vap->va_atime = pn->pn_mc_atime;
1022 if (pn->pn_stat & PNODE_METACACHE_CTIME)
1023 vap->va_ctime = pn->pn_mc_ctime;
1024 if (pn->pn_stat & PNODE_METACACHE_MTIME)
1025 vap->va_mtime = pn->pn_mc_mtime;
1026 if (pn->pn_stat & PNODE_METACACHE_SIZE)
1027 vap->va_size = pn->pn_mc_size;
1028 }
1029
1030 if (!(pn->pn_stat & PNODE_METACACHE_SIZE) && (flags & SETATTR_CHSIZE)) {
1031 if (rvap->va_size != VNOVAL
1032 && vp->v_type != VBLK && vp->v_type != VCHR) {
1033 uvm_vnp_setsize(vp, rvap->va_size);
1034 pn->pn_serversize = rvap->va_size;
1035 }
1036 }
1037
1038 if ((va_ttl != NULL) && TTL_VALID(va_ttl)) {
1039 if (pn->pn_va_cache == NULL)
1040 pn->pn_va_cache = pool_get(&puffs_vapool, PR_WAITOK);
1041
1042 (void)memcpy(pn->pn_va_cache, rvap, sizeof(*rvap));
1043
1044 pn->pn_va_timeout = TTL_TO_TIMEOUT(va_ttl);
1045 }
1046 }
1047
1048 static void
1049 update_parent(struct vnode *vp, struct vnode *dvp)
1050 {
1051 struct puffs_node *pn = VPTOPP(vp);
1052
1053 if (pn->pn_parent != NULL) {
1054 KASSERT(pn->pn_parent != dvp);
1055 vrele(pn->pn_parent);
1056 }
1057
1058 vref(dvp);
1059 pn->pn_parent = dvp;
1060 }
1061
1062 int
1063 puffs_vnop_getattr(void *v)
1064 {
1065 struct vop_getattr_args /* {
1066 const struct vnodeop_desc *a_desc;
1067 struct vnode *a_vp;
1068 struct vattr *a_vap;
1069 kauth_cred_t a_cred;
1070 } */ *ap = v;
1071 PUFFS_MSG_VARS(vn, getattr);
1072 struct vnode *vp = ap->a_vp;
1073 struct mount *mp = vp->v_mount;
1074 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1075 struct vattr *vap, *rvap;
1076 struct puffs_node *pn = VPTOPP(vp);
1077 struct timespec *va_ttl = NULL;
1078 int error = 0;
1079
1080 /*
1081 * A lock is required so that we do not race with
1082 * setattr, write and fsync when changing vp->v_size.
1083 * This is critical, since setting a stall smaler value
1084 * triggers a file truncate in uvm_vnp_setsize(), which
1085 * most of the time means data corruption (a chunk of
1086 * data is replaced by zeroes). This can be removed if
1087 * we decide one day that VOP_GETATTR must operate on
1088 * a locked vnode.
1089 *
1090 * XXX Should be useless now that VOP_GETATTR has been
1091 * fixed to always require a shared lock at least.
1092 */
1093 mutex_enter(&pn->pn_sizemtx);
1094
1095 REFPN(pn);
1096 vap = ap->a_vap;
1097
1098 if (PUFFS_USE_FS_TTL(pmp)) {
1099 if (!TIMED_OUT(pn->pn_va_timeout)) {
1100 update_va(vp, vap, pn->pn_va_cache,
1101 NULL, NULL, SETATTR_CHSIZE);
1102 goto out2;
1103 }
1104 }
1105
1106 PUFFS_MSG_ALLOC(vn, getattr);
1107 vattr_null(&getattr_msg->pvnr_va);
1108 puffs_credcvt(&getattr_msg->pvnr_cred, ap->a_cred);
1109 puffs_msg_setinfo(park_getattr, PUFFSOP_VN,
1110 PUFFS_VN_GETATTR, VPTOPNC(vp));
1111
1112 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getattr, vp->v_data, NULL, error);
1113 error = checkerr(pmp, error, __func__);
1114 if (error)
1115 goto out;
1116
1117 rvap = &getattr_msg->pvnr_va;
1118
1119 if (PUFFS_USE_FS_TTL(pmp))
1120 va_ttl = &getattr_msg->pvnr_va_ttl;
1121
1122 update_va(vp, vap, rvap, va_ttl, NULL, SETATTR_CHSIZE);
1123
1124 out:
1125 PUFFS_MSG_RELEASE(getattr);
1126
1127 out2:
1128 puffs_releasenode(pn);
1129
1130 mutex_exit(&pn->pn_sizemtx);
1131
1132 return error;
1133 }
1134
1135 static void
1136 zerofill_lastpage(struct vnode *vp, voff_t off)
1137 {
1138 char zbuf[PAGE_SIZE];
1139 struct iovec iov;
1140 struct uio uio;
1141 vsize_t len;
1142 int error;
1143
1144 if (trunc_page(off) == off)
1145 return;
1146
1147 if (vp->v_writecount == 0)
1148 return;
1149
1150 len = round_page(off) - off;
1151 memset(zbuf, 0, len);
1152
1153 iov.iov_base = zbuf;
1154 iov.iov_len = len;
1155 UIO_SETUP_SYSSPACE(&uio);
1156 uio.uio_iov = &iov;
1157 uio.uio_iovcnt = 1;
1158 uio.uio_offset = off;
1159 uio.uio_resid = len;
1160 uio.uio_rw = UIO_WRITE;
1161
1162 error = ubc_uiomove(&vp->v_uobj, &uio, len,
1163 UVM_ADV_SEQUENTIAL, UBC_WRITE|UBC_UNMAP_FLAG(vp));
1164 if (error) {
1165 DPRINTF(("zero-fill 0x%" PRIxVSIZE "@0x%" PRIx64
1166 " failed: error = %d\n", len, off, error));
1167 }
1168
1169 return;
1170 }
1171
1172 static int
1173 dosetattr(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, int flags)
1174 {
1175 PUFFS_MSG_VARS(vn, setattr);
1176 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1177 struct puffs_node *pn = vp->v_data;
1178 vsize_t oldsize = vp->v_size;
1179 int error = 0;
1180
1181 KASSERT(!(flags & SETATTR_CHSIZE) || mutex_owned(&pn->pn_sizemtx));
1182
1183 if ((vp->v_mount->mnt_flag & MNT_RDONLY) &&
1184 (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL
1185 || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL
1186 || vap->va_mode != (mode_t)VNOVAL))
1187 return EROFS;
1188
1189 if ((vp->v_mount->mnt_flag & MNT_RDONLY)
1190 && vp->v_type == VREG && vap->va_size != VNOVAL)
1191 return EROFS;
1192
1193 /*
1194 * Flush metacache first. If we are called with some explicit
1195 * parameters, treat them as information overriding metacache
1196 * information.
1197 */
1198 if (pn->pn_stat & PNODE_METACACHE_MASK) {
1199 if ((pn->pn_stat & PNODE_METACACHE_ATIME)
1200 && vap->va_atime.tv_sec == VNOVAL)
1201 vap->va_atime = pn->pn_mc_atime;
1202 if ((pn->pn_stat & PNODE_METACACHE_CTIME)
1203 && vap->va_ctime.tv_sec == VNOVAL)
1204 vap->va_ctime = pn->pn_mc_ctime;
1205 if ((pn->pn_stat & PNODE_METACACHE_MTIME)
1206 && vap->va_mtime.tv_sec == VNOVAL)
1207 vap->va_mtime = pn->pn_mc_mtime;
1208 if ((pn->pn_stat & PNODE_METACACHE_SIZE)
1209 && vap->va_size == VNOVAL)
1210 vap->va_size = pn->pn_mc_size;
1211
1212 pn->pn_stat &= ~PNODE_METACACHE_MASK;
1213 }
1214
1215 /*
1216 * Flush attribute cache so that another thread do
1217 * not get a stale value during the operation.
1218 */
1219 if (PUFFS_USE_FS_TTL(pmp))
1220 pn->pn_va_timeout = 0;
1221
1222 PUFFS_MSG_ALLOC(vn, setattr);
1223 (void)memcpy(&setattr_msg->pvnr_va, vap, sizeof(struct vattr));
1224 puffs_credcvt(&setattr_msg->pvnr_cred, cred);
1225 puffs_msg_setinfo(park_setattr, PUFFSOP_VN,
1226 PUFFS_VN_SETATTR, VPTOPNC(vp));
1227 if (flags & SETATTR_ASYNC)
1228 puffs_msg_setfaf(park_setattr);
1229
1230 puffs_msg_enqueue(pmp, park_setattr);
1231 if ((flags & SETATTR_ASYNC) == 0)
1232 error = puffs_msg_wait2(pmp, park_setattr, vp->v_data, NULL);
1233
1234 if ((error == 0) && PUFFS_USE_FS_TTL(pmp)) {
1235 struct timespec *va_ttl = &setattr_msg->pvnr_va_ttl;
1236 struct vattr *rvap = &setattr_msg->pvnr_va;
1237
1238 update_va(vp, NULL, rvap, va_ttl, NULL, flags);
1239 }
1240
1241 PUFFS_MSG_RELEASE(setattr);
1242 if ((flags & SETATTR_ASYNC) == 0) {
1243 error = checkerr(pmp, error, __func__);
1244 if (error)
1245 return error;
1246 } else {
1247 error = 0;
1248 }
1249
1250 if (vap->va_size != VNOVAL) {
1251 /*
1252 * If we truncated the file, make sure the data beyond
1253 * EOF in last page does not remain in cache, otherwise
1254 * if the file is later truncated to a larger size (creating
1255 * a hole), that area will not return zeroes as it
1256 * should.
1257 */
1258 if ((flags & SETATTR_CHSIZE) && PUFFS_USE_PAGECACHE(pmp) &&
1259 (vap->va_size < oldsize))
1260 zerofill_lastpage(vp, vap->va_size);
1261
1262 pn->pn_serversize = vap->va_size;
1263 if (flags & SETATTR_CHSIZE)
1264 uvm_vnp_setsize(vp, vap->va_size);
1265 }
1266
1267 return 0;
1268 }
1269
1270 int
1271 puffs_vnop_setattr(void *v)
1272 {
1273 struct vop_getattr_args /* {
1274 const struct vnodeop_desc *a_desc;
1275 struct vnode *a_vp;
1276 struct vattr *a_vap;
1277 kauth_cred_t a_cred;
1278 } */ *ap = v;
1279 struct puffs_node *pn = ap->a_vp->v_data;
1280 int error;
1281
1282 mutex_enter(&pn->pn_sizemtx);
1283 error = dosetattr(ap->a_vp, ap->a_vap, ap->a_cred, SETATTR_CHSIZE);
1284 mutex_exit(&pn->pn_sizemtx);
1285
1286 return error;
1287 }
1288
1289 static __inline int
1290 doinact(struct puffs_mount *pmp, int iaflag)
1291 {
1292
1293 if (EXISTSOP(pmp, INACTIVE))
1294 if (pmp->pmp_flags & PUFFS_KFLAG_IAONDEMAND)
1295 if (iaflag || ALLOPS(pmp))
1296 return 1;
1297 else
1298 return 0;
1299 else
1300 return 1;
1301 else
1302 return 0;
1303 }
1304
1305 static void
1306 callinactive(struct puffs_mount *pmp, puffs_cookie_t ck, int iaflag)
1307 {
1308 PUFFS_MSG_VARS(vn, inactive);
1309
1310 if (doinact(pmp, iaflag)) {
1311 PUFFS_MSG_ALLOC(vn, inactive);
1312 puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1313 PUFFS_VN_INACTIVE, ck);
1314 PUFFS_MSG_ENQUEUEWAIT_NOERROR(pmp, park_inactive);
1315 PUFFS_MSG_RELEASE(inactive);
1316 }
1317 }
1318
1319 /* XXX: callinactive can't setback */
1320 int
1321 puffs_vnop_inactive(void *v)
1322 {
1323 struct vop_inactive_args /* {
1324 const struct vnodeop_desc *a_desc;
1325 struct vnode *a_vp;
1326 } */ *ap = v;
1327 PUFFS_MSG_VARS(vn, inactive);
1328 struct vnode *vp = ap->a_vp;
1329 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1330 struct puffs_node *pnode;
1331 bool recycle = false;
1332
1333 pnode = vp->v_data;
1334 mutex_enter(&pnode->pn_sizemtx);
1335
1336 if (doinact(pmp, pnode->pn_stat & PNODE_DOINACT)) {
1337 flushvncache(vp, 0, 0, false);
1338 PUFFS_MSG_ALLOC(vn, inactive);
1339 puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1340 PUFFS_VN_INACTIVE, VPTOPNC(vp));
1341 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_inactive, vp->v_data,
1342 NULL);
1343 PUFFS_MSG_RELEASE(inactive);
1344 }
1345 pnode->pn_stat &= ~PNODE_DOINACT;
1346
1347 /*
1348 * file server thinks it's gone? then don't be afraid care,
1349 * node's life was already all it would ever be
1350 */
1351 if (pnode->pn_stat & PNODE_NOREFS) {
1352 pnode->pn_stat |= PNODE_DYING;
1353 recycle = true;
1354 }
1355
1356 /*
1357 * Handle node TTL.
1358 * If grace has already timed out, make it reclaimed.
1359 * Otherwise, we queue its expiration by sop thread, so
1360 * that it does not remain for ages in the freelist,
1361 * holding memory in userspace, while we will have
1362 * to look it up again anyway.
1363 */
1364 if (PUFFS_USE_FS_TTL(pmp) && !(vp->v_vflag & VV_ROOT) && !recycle) {
1365 bool incache = !TIMED_OUT(pnode->pn_cn_timeout);
1366 bool ingrace = !TIMED_OUT(pnode->pn_cn_grace);
1367 bool reclaimqueued = pnode->pn_stat & PNODE_SOPEXP;
1368
1369 if (!incache && !ingrace && !reclaimqueued) {
1370 pnode->pn_stat |= PNODE_DYING;
1371 recycle = true;
1372 }
1373
1374 if (!recycle && !reclaimqueued) {
1375 struct puffs_sopreq *psopr;
1376 int at = MAX(pnode->pn_cn_grace, pnode->pn_cn_timeout);
1377
1378 KASSERT(curlwp != uvm.pagedaemon_lwp);
1379 psopr = kmem_alloc(sizeof(*psopr), KM_SLEEP);
1380 psopr->psopr_ck = VPTOPNC(pnode->pn_vp);
1381 psopr->psopr_sopreq = PUFFS_SOPREQ_EXPIRE;
1382 psopr->psopr_at = at;
1383
1384 mutex_enter(&pmp->pmp_sopmtx);
1385
1386 /*
1387 * If thread has disapeared, just give up. The
1388 * fs is being unmounted and the node will be
1389 * be reclaimed anyway.
1390 *
1391 * Otherwise, we queue the request but do not
1392 * immediatly signal the thread, as the node
1393 * has not been expired yet.
1394 */
1395 if (pmp->pmp_sopthrcount == 0) {
1396 kmem_free(psopr, sizeof(*psopr));
1397 } else {
1398 TAILQ_INSERT_TAIL(&pmp->pmp_sopnodereqs,
1399 psopr, psopr_entries);
1400 pnode->pn_stat |= PNODE_SOPEXP;
1401 }
1402
1403 mutex_exit(&pmp->pmp_sopmtx);
1404 }
1405 }
1406
1407 *ap->a_recycle = recycle;
1408
1409 mutex_exit(&pnode->pn_sizemtx);
1410 VOP_UNLOCK(vp);
1411
1412 return 0;
1413 }
1414
1415 static void
1416 callreclaim(struct puffs_mount *pmp, puffs_cookie_t ck, int nlookup)
1417 {
1418 PUFFS_MSG_VARS(vn, reclaim);
1419
1420 if (!EXISTSOP(pmp, RECLAIM))
1421 return;
1422
1423 PUFFS_MSG_ALLOC(vn, reclaim);
1424 reclaim_msg->pvnr_nlookup = nlookup;
1425 puffs_msg_setfaf(park_reclaim);
1426 puffs_msg_setinfo(park_reclaim, PUFFSOP_VN, PUFFS_VN_RECLAIM, ck);
1427
1428 puffs_msg_enqueue(pmp, park_reclaim);
1429 PUFFS_MSG_RELEASE(reclaim);
1430 return;
1431 }
1432
1433 /*
1434 * always FAF, we don't really care if the server wants to fail to
1435 * reclaim the node or not
1436 */
1437 int
1438 puffs_vnop_reclaim(void *v)
1439 {
1440 struct vop_reclaim_args /* {
1441 const struct vnodeop_desc *a_desc;
1442 struct vnode *a_vp;
1443 } */ *ap = v;
1444 struct vnode *vp = ap->a_vp;
1445 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1446 bool notifyserver = true;
1447
1448 /*
1449 * first things first: check if someone is trying to reclaim the
1450 * root vnode. do not allow that to travel to userspace.
1451 * Note that we don't need to take the lock similarly to
1452 * puffs_root(), since there is only one of us.
1453 */
1454 if (vp->v_vflag & VV_ROOT) {
1455 mutex_enter(&pmp->pmp_lock);
1456 KASSERT(pmp->pmp_root != NULL);
1457 pmp->pmp_root = NULL;
1458 mutex_exit(&pmp->pmp_lock);
1459 notifyserver = false;
1460 }
1461
1462 /*
1463 * purge info from kernel before issueing FAF, since we
1464 * don't really know when we'll get around to it after
1465 * that and someone might race us into node creation
1466 */
1467 mutex_enter(&pmp->pmp_lock);
1468 if (PUFFS_USE_NAMECACHE(pmp))
1469 cache_purge(vp);
1470 mutex_exit(&pmp->pmp_lock);
1471
1472 if (notifyserver) {
1473 int nlookup = VPTOPP(vp)->pn_nlookup;
1474
1475 callreclaim(MPTOPUFFSMP(vp->v_mount), VPTOPNC(vp), nlookup);
1476 }
1477
1478 if (PUFFS_USE_DOTDOTCACHE(pmp)) {
1479 if (__predict_true(VPTOPP(vp)->pn_parent != NULL))
1480 vrele(VPTOPP(vp)->pn_parent);
1481 else
1482 KASSERT(vp->v_type == VNON || (vp->v_vflag & VV_ROOT));
1483 }
1484
1485 puffs_putvnode(vp);
1486
1487 return 0;
1488 }
1489
1490 #define CSIZE sizeof(**ap->a_cookies)
1491 int
1492 puffs_vnop_readdir(void *v)
1493 {
1494 struct vop_readdir_args /* {
1495 const struct vnodeop_desc *a_desc;
1496 struct vnode *a_vp;
1497 struct uio *a_uio;
1498 kauth_cred_t a_cred;
1499 int *a_eofflag;
1500 off_t **a_cookies;
1501 int *a_ncookies;
1502 } */ *ap = v;
1503 PUFFS_MSG_VARS(vn, readdir);
1504 struct vnode *vp = ap->a_vp;
1505 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1506 size_t argsize, tomove, cookiemem, cookiesmax;
1507 struct uio *uio = ap->a_uio;
1508 size_t howmuch, resid;
1509 int error;
1510
1511 /*
1512 * ok, so we need: resid + cookiemem = maxreq
1513 * => resid + cookiesize * (resid/minsize) = maxreq
1514 * => resid + cookiesize/minsize * resid = maxreq
1515 * => (cookiesize/minsize + 1) * resid = maxreq
1516 * => resid = maxreq / (cookiesize/minsize + 1)
1517 *
1518 * Since cookiesize <= minsize and we're not very big on floats,
1519 * we approximate that to be 1. Therefore:
1520 *
1521 * resid = maxreq / 2;
1522 *
1523 * Well, at least we didn't have to use differential equations
1524 * or the Gram-Schmidt process.
1525 *
1526 * (yes, I'm very afraid of this)
1527 */
1528 KASSERT(CSIZE <= _DIRENT_MINSIZE((struct dirent *)0));
1529
1530 if (ap->a_cookies) {
1531 KASSERT(ap->a_ncookies != NULL);
1532 if (pmp->pmp_args.pa_fhsize == 0)
1533 return EOPNOTSUPP;
1534 resid = PUFFS_TOMOVE(uio->uio_resid, pmp) / 2;
1535 cookiesmax = resid/_DIRENT_MINSIZE((struct dirent *)0);
1536 cookiemem = ALIGN(cookiesmax*CSIZE); /* play safe */
1537 } else {
1538 resid = PUFFS_TOMOVE(uio->uio_resid, pmp);
1539 cookiesmax = 0;
1540 cookiemem = 0;
1541 }
1542
1543 argsize = sizeof(struct puffs_vnmsg_readdir);
1544 tomove = resid + cookiemem;
1545 puffs_msgmem_alloc(argsize + tomove, &park_readdir,
1546 (void *)&readdir_msg, 1);
1547
1548 puffs_credcvt(&readdir_msg->pvnr_cred, ap->a_cred);
1549 readdir_msg->pvnr_offset = uio->uio_offset;
1550 readdir_msg->pvnr_resid = resid;
1551 readdir_msg->pvnr_ncookies = cookiesmax;
1552 readdir_msg->pvnr_eofflag = 0;
1553 readdir_msg->pvnr_dentoff = cookiemem;
1554 puffs_msg_setinfo(park_readdir, PUFFSOP_VN,
1555 PUFFS_VN_READDIR, VPTOPNC(vp));
1556 puffs_msg_setdelta(park_readdir, tomove);
1557
1558 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readdir, vp->v_data, NULL, error);
1559 error = checkerr(pmp, error, __func__);
1560 if (error)
1561 goto out;
1562
1563 /* userspace is cheating? */
1564 if (readdir_msg->pvnr_resid > resid) {
1565 puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1566 "resid grew", VPTOPNC(vp));
1567 ERROUT(EPROTO);
1568 }
1569 if (readdir_msg->pvnr_ncookies > cookiesmax) {
1570 puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1571 "too many cookies", VPTOPNC(vp));
1572 ERROUT(EPROTO);
1573 }
1574
1575 /* check eof */
1576 if (readdir_msg->pvnr_eofflag)
1577 *ap->a_eofflag = 1;
1578
1579 /* bouncy-wouncy with the directory data */
1580 howmuch = resid - readdir_msg->pvnr_resid;
1581
1582 /* force eof if no data was returned (getcwd() needs this) */
1583 if (howmuch == 0) {
1584 *ap->a_eofflag = 1;
1585 goto out;
1586 }
1587
1588 error = uiomove(readdir_msg->pvnr_data + cookiemem, howmuch, uio);
1589 if (error)
1590 goto out;
1591
1592 /* provide cookies to caller if so desired */
1593 if (ap->a_cookies) {
1594 KASSERT(curlwp != uvm.pagedaemon_lwp);
1595 *ap->a_cookies = malloc(readdir_msg->pvnr_ncookies*CSIZE,
1596 M_TEMP, M_WAITOK);
1597 *ap->a_ncookies = readdir_msg->pvnr_ncookies;
1598 memcpy(*ap->a_cookies, readdir_msg->pvnr_data,
1599 *ap->a_ncookies*CSIZE);
1600 }
1601
1602 /* next readdir starts here */
1603 uio->uio_offset = readdir_msg->pvnr_offset;
1604
1605 out:
1606 puffs_msgmem_release(park_readdir);
1607 return error;
1608 }
1609 #undef CSIZE
1610
1611 /*
1612 * poll works by consuming the bitmask in pn_revents. If there are
1613 * events available, poll returns immediately. If not, it issues a
1614 * poll to userspace, selrecords itself and returns with no available
1615 * events. When the file server returns, it executes puffs_parkdone_poll(),
1616 * where available events are added to the bitmask. selnotify() is
1617 * then also executed by that function causing us to enter here again
1618 * and hopefully find the missing bits (unless someone got them first,
1619 * in which case it starts all over again).
1620 */
1621 int
1622 puffs_vnop_poll(void *v)
1623 {
1624 struct vop_poll_args /* {
1625 const struct vnodeop_desc *a_desc;
1626 struct vnode *a_vp;
1627 int a_events;
1628 } */ *ap = v;
1629 PUFFS_MSG_VARS(vn, poll);
1630 struct vnode *vp = ap->a_vp;
1631 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1632 struct puffs_node *pn = vp->v_data;
1633 int events;
1634
1635 if (EXISTSOP(pmp, POLL)) {
1636 mutex_enter(&pn->pn_mtx);
1637 events = pn->pn_revents & ap->a_events;
1638 if (events & ap->a_events) {
1639 pn->pn_revents &= ~ap->a_events;
1640 mutex_exit(&pn->pn_mtx);
1641
1642 return events;
1643 } else {
1644 puffs_referencenode(pn);
1645 mutex_exit(&pn->pn_mtx);
1646
1647 PUFFS_MSG_ALLOC(vn, poll);
1648 poll_msg->pvnr_events = ap->a_events;
1649 puffs_msg_setinfo(park_poll, PUFFSOP_VN,
1650 PUFFS_VN_POLL, VPTOPNC(vp));
1651 puffs_msg_setcall(park_poll, puffs_parkdone_poll, pn);
1652 selrecord(curlwp, &pn->pn_sel);
1653
1654 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_poll,
1655 vp->v_data, NULL);
1656 PUFFS_MSG_RELEASE(poll);
1657
1658 return 0;
1659 }
1660 } else {
1661 return genfs_poll(v);
1662 }
1663 }
1664
1665 static int
1666 flushvncache(struct vnode *vp, off_t offlo, off_t offhi, bool wait)
1667 {
1668 struct puffs_node *pn = VPTOPP(vp);
1669 struct vattr va;
1670 int pflags, error;
1671
1672 /* flush out information from our metacache, see vop_setattr */
1673 if (pn->pn_stat & PNODE_METACACHE_MASK
1674 && (pn->pn_stat & PNODE_DYING) == 0) {
1675 vattr_null(&va);
1676 error = dosetattr(vp, &va, FSCRED,
1677 SETATTR_CHSIZE | (wait ? 0 : SETATTR_ASYNC));
1678 if (error)
1679 return error;
1680 }
1681
1682 /*
1683 * flush pages to avoid being overly dirty
1684 */
1685 pflags = PGO_CLEANIT;
1686 if (wait)
1687 pflags |= PGO_SYNCIO;
1688
1689 mutex_enter(vp->v_interlock);
1690 return VOP_PUTPAGES(vp, trunc_page(offlo), round_page(offhi), pflags);
1691 }
1692
1693 int
1694 puffs_vnop_fsync(void *v)
1695 {
1696 struct vop_fsync_args /* {
1697 const struct vnodeop_desc *a_desc;
1698 struct vnode *a_vp;
1699 kauth_cred_t a_cred;
1700 int a_flags;
1701 off_t a_offlo;
1702 off_t a_offhi;
1703 } */ *ap = v;
1704 PUFFS_MSG_VARS(vn, fsync);
1705 struct vnode *vp;
1706 struct puffs_node *pn;
1707 struct puffs_mount *pmp;
1708 int error, dofaf;
1709
1710 vp = ap->a_vp;
1711 KASSERT(vp != NULL);
1712 pn = VPTOPP(vp);
1713 KASSERT(pn != NULL);
1714 pmp = MPTOPUFFSMP(vp->v_mount);
1715 if (ap->a_flags & FSYNC_WAIT) {
1716 mutex_enter(&pn->pn_sizemtx);
1717 } else {
1718 if (mutex_tryenter(&pn->pn_sizemtx) == 0)
1719 return EDEADLK;
1720 }
1721
1722 error = flushvncache(vp, ap->a_offlo, ap->a_offhi,
1723 (ap->a_flags & FSYNC_WAIT) == FSYNC_WAIT);
1724 if (error)
1725 goto out;
1726
1727 /*
1728 * HELLO! We exit already here if the user server does not
1729 * support fsync OR if we should call fsync for a node which
1730 * has references neither in the kernel or the fs server.
1731 * Otherwise we continue to issue fsync() forward.
1732 */
1733 error = 0;
1734 if (!EXISTSOP(pmp, FSYNC) || (pn->pn_stat & PNODE_DYING))
1735 goto out;
1736
1737 dofaf = (ap->a_flags & FSYNC_WAIT) == 0 || ap->a_flags == FSYNC_LAZY;
1738 /*
1739 * We abuse VXLOCK to mean "vnode is going to die", so we issue
1740 * only FAFs for those. Otherwise there's a danger of deadlock,
1741 * since the execution context here might be the user server
1742 * doing some operation on another fs, which in turn caused a
1743 * vnode to be reclaimed from the freelist for this fs.
1744 */
1745 if (dofaf == 0) {
1746 mutex_enter(vp->v_interlock);
1747 if (vdead_check(vp, VDEAD_NOWAIT) != 0)
1748 dofaf = 1;
1749 mutex_exit(vp->v_interlock);
1750 }
1751
1752 PUFFS_MSG_ALLOC(vn, fsync);
1753 if (dofaf)
1754 puffs_msg_setfaf(park_fsync);
1755
1756 puffs_credcvt(&fsync_msg->pvnr_cred, ap->a_cred);
1757 fsync_msg->pvnr_flags = ap->a_flags;
1758 fsync_msg->pvnr_offlo = ap->a_offlo;
1759 fsync_msg->pvnr_offhi = ap->a_offhi;
1760 puffs_msg_setinfo(park_fsync, PUFFSOP_VN,
1761 PUFFS_VN_FSYNC, VPTOPNC(vp));
1762
1763 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fsync, vp->v_data, NULL, error);
1764 PUFFS_MSG_RELEASE(fsync);
1765
1766 error = checkerr(pmp, error, __func__);
1767
1768 out:
1769 mutex_exit(&pn->pn_sizemtx);
1770 return error;
1771 }
1772
1773 int
1774 puffs_vnop_seek(void *v)
1775 {
1776 struct vop_seek_args /* {
1777 const struct vnodeop_desc *a_desc;
1778 struct vnode *a_vp;
1779 off_t a_oldoff;
1780 off_t a_newoff;
1781 kauth_cred_t a_cred;
1782 } */ *ap = v;
1783 PUFFS_MSG_VARS(vn, seek);
1784 struct vnode *vp = ap->a_vp;
1785 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1786 int error;
1787
1788 PUFFS_MSG_ALLOC(vn, seek);
1789 seek_msg->pvnr_oldoff = ap->a_oldoff;
1790 seek_msg->pvnr_newoff = ap->a_newoff;
1791 puffs_credcvt(&seek_msg->pvnr_cred, ap->a_cred);
1792 puffs_msg_setinfo(park_seek, PUFFSOP_VN,
1793 PUFFS_VN_SEEK, VPTOPNC(vp));
1794
1795 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_seek, vp->v_data, NULL, error);
1796 PUFFS_MSG_RELEASE(seek);
1797 return checkerr(pmp, error, __func__);
1798 }
1799
1800 static int
1801 callremove(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1802 struct componentname *cnp)
1803 {
1804 PUFFS_MSG_VARS(vn, remove);
1805 int error;
1806
1807 PUFFS_MSG_ALLOC(vn, remove);
1808 remove_msg->pvnr_cookie_targ = ck;
1809 puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1810 cnp, PUFFS_USE_FULLPNBUF(pmp));
1811 puffs_msg_setinfo(park_remove, PUFFSOP_VN, PUFFS_VN_REMOVE, dck);
1812
1813 PUFFS_MSG_ENQUEUEWAIT(pmp, park_remove, error);
1814 PUFFS_MSG_RELEASE(remove);
1815
1816 return checkerr(pmp, error, __func__);
1817 }
1818
1819 /*
1820 * XXX: can't use callremove now because can't catch setbacks with
1821 * it due to lack of a pnode argument.
1822 */
1823 int
1824 puffs_vnop_remove(void *v)
1825 {
1826 struct vop_remove_args /* {
1827 const struct vnodeop_desc *a_desc;
1828 struct vnode *a_dvp;
1829 struct vnode *a_vp;
1830 struct componentname *a_cnp;
1831 } */ *ap = v;
1832 PUFFS_MSG_VARS(vn, remove);
1833 struct vnode *dvp = ap->a_dvp;
1834 struct vnode *vp = ap->a_vp;
1835 struct puffs_node *dpn = VPTOPP(dvp);
1836 struct puffs_node *pn = VPTOPP(vp);
1837 struct componentname *cnp = ap->a_cnp;
1838 struct mount *mp = dvp->v_mount;
1839 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1840 int error;
1841
1842 PUFFS_MSG_ALLOC(vn, remove);
1843 remove_msg->pvnr_cookie_targ = VPTOPNC(vp);
1844 puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1845 cnp, PUFFS_USE_FULLPNBUF(pmp));
1846 puffs_msg_setinfo(park_remove, PUFFSOP_VN,
1847 PUFFS_VN_REMOVE, VPTOPNC(dvp));
1848
1849 puffs_msg_enqueue(pmp, park_remove);
1850 REFPN_AND_UNLOCKVP(dvp, dpn);
1851 if (dvp == vp)
1852 REFPN(pn);
1853 else
1854 REFPN_AND_UNLOCKVP(vp, pn);
1855 error = puffs_msg_wait2(pmp, park_remove, dpn, pn);
1856
1857 PUFFS_MSG_RELEASE(remove);
1858
1859 puffs_updatenode(VPTOPP(dvp), PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0);
1860
1861 RELEPN_AND_VP(dvp, dpn);
1862 RELEPN_AND_VP(vp, pn);
1863
1864 error = checkerr(pmp, error, __func__);
1865 return error;
1866 }
1867
1868 int
1869 puffs_vnop_mkdir(void *v)
1870 {
1871 struct vop_mkdir_v3_args /* {
1872 const struct vnodeop_desc *a_desc;
1873 struct vnode *a_dvp;
1874 struct vnode **a_vpp;
1875 struct componentname *a_cnp;
1876 struct vattr *a_vap;
1877 } */ *ap = v;
1878 PUFFS_MSG_VARS(vn, mkdir);
1879 struct vnode *dvp = ap->a_dvp;
1880 struct puffs_node *dpn = VPTOPP(dvp);
1881 struct componentname *cnp = ap->a_cnp;
1882 struct mount *mp = dvp->v_mount;
1883 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1884 int error;
1885
1886 PUFFS_MSG_ALLOC(vn, mkdir);
1887 puffs_makecn(&mkdir_msg->pvnr_cn, &mkdir_msg->pvnr_cn_cred,
1888 cnp, PUFFS_USE_FULLPNBUF(pmp));
1889 mkdir_msg->pvnr_va = *ap->a_vap;
1890 puffs_msg_setinfo(park_mkdir, PUFFSOP_VN,
1891 PUFFS_VN_MKDIR, VPTOPNC(dvp));
1892
1893 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mkdir, dvp->v_data, NULL, error);
1894
1895 error = checkerr(pmp, error, __func__);
1896 if (error)
1897 goto out;
1898
1899 error = puffs_newnode(mp, dvp, ap->a_vpp,
1900 mkdir_msg->pvnr_newnode, cnp, VDIR, 0);
1901 if (error) {
1902 puffs_abortbutton(pmp, PUFFS_ABORT_MKDIR, dpn->pn_cookie,
1903 mkdir_msg->pvnr_newnode, cnp);
1904 goto out;
1905 }
1906
1907 if (PUFFS_USE_FS_TTL(pmp)) {
1908 struct timespec *va_ttl = &mkdir_msg->pvnr_va_ttl;
1909 struct timespec *cn_ttl = &mkdir_msg->pvnr_cn_ttl;
1910 struct vattr *rvap = &mkdir_msg->pvnr_va;
1911
1912 update_va(*ap->a_vpp, NULL, rvap,
1913 va_ttl, cn_ttl, SETATTR_CHSIZE);
1914 }
1915
1916 VPTOPP(*ap->a_vpp)->pn_nlookup++;
1917
1918 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
1919 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
1920 update_parent(*ap->a_vpp, dvp);
1921
1922 out:
1923 PUFFS_MSG_RELEASE(mkdir);
1924 return error;
1925 }
1926
1927 static int
1928 callrmdir(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1929 struct componentname *cnp)
1930 {
1931 PUFFS_MSG_VARS(vn, rmdir);
1932 int error;
1933
1934 PUFFS_MSG_ALLOC(vn, rmdir);
1935 rmdir_msg->pvnr_cookie_targ = ck;
1936 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1937 cnp, PUFFS_USE_FULLPNBUF(pmp));
1938 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN, PUFFS_VN_RMDIR, dck);
1939
1940 PUFFS_MSG_ENQUEUEWAIT(pmp, park_rmdir, error);
1941 PUFFS_MSG_RELEASE(rmdir);
1942
1943 return checkerr(pmp, error, __func__);
1944 }
1945
1946 int
1947 puffs_vnop_rmdir(void *v)
1948 {
1949 struct vop_rmdir_args /* {
1950 const struct vnodeop_desc *a_desc;
1951 struct vnode *a_dvp;
1952 struct vnode *a_vp;
1953 struct componentname *a_cnp;
1954 } */ *ap = v;
1955 PUFFS_MSG_VARS(vn, rmdir);
1956 struct vnode *dvp = ap->a_dvp;
1957 struct vnode *vp = ap->a_vp;
1958 struct puffs_node *dpn = VPTOPP(dvp);
1959 struct puffs_node *pn = VPTOPP(vp);
1960 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1961 struct componentname *cnp = ap->a_cnp;
1962 int error;
1963
1964 PUFFS_MSG_ALLOC(vn, rmdir);
1965 rmdir_msg->pvnr_cookie_targ = VPTOPNC(vp);
1966 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1967 cnp, PUFFS_USE_FULLPNBUF(pmp));
1968 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN,
1969 PUFFS_VN_RMDIR, VPTOPNC(dvp));
1970
1971 puffs_msg_enqueue(pmp, park_rmdir);
1972 REFPN_AND_UNLOCKVP(dvp, dpn);
1973 REFPN_AND_UNLOCKVP(vp, pn);
1974 error = puffs_msg_wait2(pmp, park_rmdir, dpn, pn);
1975
1976 PUFFS_MSG_RELEASE(rmdir);
1977
1978 puffs_updatenode(VPTOPP(dvp), PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0);
1979
1980 /* XXX: some call cache_purge() *for both vnodes* here, investigate */
1981 RELEPN_AND_VP(dvp, dpn);
1982 RELEPN_AND_VP(vp, pn);
1983
1984 return error;
1985 }
1986
1987 int
1988 puffs_vnop_link(void *v)
1989 {
1990 struct vop_link_args /* {
1991 const struct vnodeop_desc *a_desc;
1992 struct vnode *a_dvp;
1993 struct vnode *a_vp;
1994 struct componentname *a_cnp;
1995 } */ *ap = v;
1996 PUFFS_MSG_VARS(vn, link);
1997 struct vnode *dvp = ap->a_dvp;
1998 struct vnode *vp = ap->a_vp;
1999 struct puffs_node *dpn = VPTOPP(dvp);
2000 struct puffs_node *pn = VPTOPP(vp);
2001 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2002 struct componentname *cnp = ap->a_cnp;
2003 int error;
2004
2005 PUFFS_MSG_ALLOC(vn, link);
2006 link_msg->pvnr_cookie_targ = VPTOPNC(vp);
2007 puffs_makecn(&link_msg->pvnr_cn, &link_msg->pvnr_cn_cred,
2008 cnp, PUFFS_USE_FULLPNBUF(pmp));
2009 puffs_msg_setinfo(park_link, PUFFSOP_VN,
2010 PUFFS_VN_LINK, VPTOPNC(dvp));
2011
2012 puffs_msg_enqueue(pmp, park_link);
2013 REFPN_AND_UNLOCKVP(dvp, dpn);
2014 REFPN(pn);
2015 error = puffs_msg_wait2(pmp, park_link, dpn, pn);
2016
2017 PUFFS_MSG_RELEASE(link);
2018
2019 error = checkerr(pmp, error, __func__);
2020
2021 /*
2022 * XXX: stay in touch with the cache. I don't like this, but
2023 * don't have a better solution either. See also puffs_rename().
2024 */
2025 if (error == 0) {
2026 puffs_updatenode(pn, PUFFS_UPDATECTIME, 0);
2027 puffs_updatenode(VPTOPP(dvp),
2028 PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0);
2029 }
2030
2031 RELEPN_AND_VP(dvp, dpn);
2032 puffs_releasenode(pn);
2033
2034 return error;
2035 }
2036
2037 int
2038 puffs_vnop_symlink(void *v)
2039 {
2040 struct vop_symlink_v3_args /* {
2041 const struct vnodeop_desc *a_desc;
2042 struct vnode *a_dvp;
2043 struct vnode **a_vpp;
2044 struct componentname *a_cnp;
2045 struct vattr *a_vap;
2046 char *a_target;
2047 } */ *ap = v;
2048 PUFFS_MSG_VARS(vn, symlink);
2049 struct vnode *dvp = ap->a_dvp;
2050 struct puffs_node *dpn = VPTOPP(dvp);
2051 struct mount *mp = dvp->v_mount;
2052 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2053 struct componentname *cnp = ap->a_cnp;
2054 int error;
2055
2056 *ap->a_vpp = NULL;
2057
2058 PUFFS_MSG_ALLOC(vn, symlink);
2059 puffs_makecn(&symlink_msg->pvnr_cn, &symlink_msg->pvnr_cn_cred,
2060 cnp, PUFFS_USE_FULLPNBUF(pmp));
2061 symlink_msg->pvnr_va = *ap->a_vap;
2062 (void)strlcpy(symlink_msg->pvnr_link, ap->a_target,
2063 sizeof(symlink_msg->pvnr_link));
2064 puffs_msg_setinfo(park_symlink, PUFFSOP_VN,
2065 PUFFS_VN_SYMLINK, VPTOPNC(dvp));
2066
2067 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_symlink, dvp->v_data, NULL, error);
2068
2069 error = checkerr(pmp, error, __func__);
2070 if (error)
2071 goto out;
2072
2073 error = puffs_newnode(mp, dvp, ap->a_vpp,
2074 symlink_msg->pvnr_newnode, cnp, VLNK, 0);
2075 if (error) {
2076 puffs_abortbutton(pmp, PUFFS_ABORT_SYMLINK, dpn->pn_cookie,
2077 symlink_msg->pvnr_newnode, cnp);
2078 goto out;
2079 }
2080
2081 if (PUFFS_USE_FS_TTL(pmp)) {
2082 struct timespec *va_ttl = &symlink_msg->pvnr_va_ttl;
2083 struct timespec *cn_ttl = &symlink_msg->pvnr_cn_ttl;
2084 struct vattr *rvap = &symlink_msg->pvnr_va;
2085
2086 update_va(*ap->a_vpp, NULL, rvap,
2087 va_ttl, cn_ttl, SETATTR_CHSIZE);
2088 }
2089
2090 VPTOPP(*ap->a_vpp)->pn_nlookup++;
2091
2092 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2093 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
2094 update_parent(*ap->a_vpp, dvp);
2095
2096 out:
2097 PUFFS_MSG_RELEASE(symlink);
2098
2099 return error;
2100 }
2101
2102 int
2103 puffs_vnop_readlink(void *v)
2104 {
2105 struct vop_readlink_args /* {
2106 const struct vnodeop_desc *a_desc;
2107 struct vnode *a_vp;
2108 struct uio *a_uio;
2109 kauth_cred_t a_cred;
2110 } */ *ap = v;
2111 PUFFS_MSG_VARS(vn, readlink);
2112 struct vnode *vp = ap->a_vp;
2113 struct puffs_mount *pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2114 size_t linklen;
2115 int error;
2116
2117 PUFFS_MSG_ALLOC(vn, readlink);
2118 puffs_credcvt(&readlink_msg->pvnr_cred, ap->a_cred);
2119 linklen = sizeof(readlink_msg->pvnr_link);
2120 readlink_msg->pvnr_linklen = linklen;
2121 puffs_msg_setinfo(park_readlink, PUFFSOP_VN,
2122 PUFFS_VN_READLINK, VPTOPNC(vp));
2123
2124 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readlink, vp->v_data, NULL, error);
2125 error = checkerr(pmp, error, __func__);
2126 if (error)
2127 goto out;
2128
2129 /* bad bad user file server */
2130 if (readlink_msg->pvnr_linklen > linklen) {
2131 puffs_senderr(pmp, PUFFS_ERR_READLINK, E2BIG,
2132 "linklen too big", VPTOPNC(ap->a_vp));
2133 error = EPROTO;
2134 goto out;
2135 }
2136
2137 error = uiomove(&readlink_msg->pvnr_link, readlink_msg->pvnr_linklen,
2138 ap->a_uio);
2139 out:
2140 PUFFS_MSG_RELEASE(readlink);
2141 return error;
2142 }
2143
2144 int
2145 puffs_vnop_rename(void *v)
2146 {
2147 struct vop_rename_args /* {
2148 const struct vnodeop_desc *a_desc;
2149 struct vnode *a_fdvp;
2150 struct vnode *a_fvp;
2151 struct componentname *a_fcnp;
2152 struct vnode *a_tdvp;
2153 struct vnode *a_tvp;
2154 struct componentname *a_tcnp;
2155 } */ *ap = v;
2156 PUFFS_MSG_VARS(vn, rename);
2157 struct vnode *fdvp = ap->a_fdvp, *fvp = ap->a_fvp;
2158 struct vnode *tdvp = ap->a_tdvp, *tvp = ap->a_tvp;
2159 struct puffs_node *fpn = ap->a_fvp->v_data;
2160 struct puffs_mount *pmp = MPTOPUFFSMP(fdvp->v_mount);
2161 int error;
2162 bool doabort = true;
2163
2164 if ((fvp->v_mount != tdvp->v_mount) ||
2165 (tvp && (fvp->v_mount != tvp->v_mount))) {
2166 ERROUT(EXDEV);
2167 }
2168
2169 PUFFS_MSG_ALLOC(vn, rename);
2170 rename_msg->pvnr_cookie_src = VPTOPNC(fvp);
2171 rename_msg->pvnr_cookie_targdir = VPTOPNC(tdvp);
2172 if (tvp)
2173 rename_msg->pvnr_cookie_targ = VPTOPNC(tvp);
2174 else
2175 rename_msg->pvnr_cookie_targ = NULL;
2176 puffs_makecn(&rename_msg->pvnr_cn_src, &rename_msg->pvnr_cn_src_cred,
2177 ap->a_fcnp, PUFFS_USE_FULLPNBUF(pmp));
2178 puffs_makecn(&rename_msg->pvnr_cn_targ, &rename_msg->pvnr_cn_targ_cred,
2179 ap->a_tcnp, PUFFS_USE_FULLPNBUF(pmp));
2180 puffs_msg_setinfo(park_rename, PUFFSOP_VN,
2181 PUFFS_VN_RENAME, VPTOPNC(fdvp));
2182
2183 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rename, fdvp->v_data, NULL, error);
2184 doabort = false;
2185 PUFFS_MSG_RELEASE(rename);
2186 error = checkerr(pmp, error, __func__);
2187
2188 /*
2189 * XXX: stay in touch with the cache. I don't like this, but
2190 * don't have a better solution either. See also puffs_link().
2191 */
2192 if (error == 0) {
2193 puffs_updatenode(fpn, PUFFS_UPDATECTIME, 0);
2194 puffs_updatenode(VPTOPP(fdvp),
2195 PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0);
2196 if (fdvp != tdvp)
2197 puffs_updatenode(VPTOPP(tdvp),
2198 PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME,
2199 0);
2200
2201 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2202 (VPTOPP(fvp)->pn_parent != tdvp))
2203 update_parent(fvp, tdvp);
2204 }
2205
2206
2207 out:
2208 if (doabort)
2209 VOP_ABORTOP(tdvp, ap->a_tcnp);
2210 if (tvp != NULL)
2211 vput(tvp);
2212 if (tdvp == tvp)
2213 vrele(tdvp);
2214 else
2215 vput(tdvp);
2216
2217 if (doabort)
2218 VOP_ABORTOP(fdvp, ap->a_fcnp);
2219 vrele(fdvp);
2220 vrele(fvp);
2221
2222 return error;
2223 }
2224
2225 #define RWARGS(cont, iofl, move, offset, creds) \
2226 (cont)->pvnr_ioflag = (iofl); \
2227 (cont)->pvnr_resid = (move); \
2228 (cont)->pvnr_offset = (offset); \
2229 puffs_credcvt(&(cont)->pvnr_cred, creds)
2230
2231 int
2232 puffs_vnop_read(void *v)
2233 {
2234 struct vop_read_args /* {
2235 const struct vnodeop_desc *a_desc;
2236 struct vnode *a_vp;
2237 struct uio *a_uio;
2238 int a_ioflag;
2239 kauth_cred_t a_cred;
2240 } */ *ap = v;
2241 PUFFS_MSG_VARS(vn, read);
2242 struct vnode *vp = ap->a_vp;
2243 struct puffs_node *pn = VPTOPP(vp);
2244 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2245 struct uio *uio = ap->a_uio;
2246 size_t tomove, argsize;
2247 vsize_t bytelen;
2248 int error;
2249
2250 read_msg = NULL;
2251 error = 0;
2252
2253 /* std sanity */
2254 if (uio->uio_resid == 0)
2255 return 0;
2256 if (uio->uio_offset < 0)
2257 return EFBIG;
2258
2259 if (vp->v_type == VREG &&
2260 PUFFS_USE_PAGECACHE(pmp) &&
2261 !(pn->pn_stat & PNODE_RDIRECT)) {
2262 const int advice = IO_ADV_DECODE(ap->a_ioflag);
2263
2264 while (uio->uio_resid > 0) {
2265 if (vp->v_size <= uio->uio_offset) {
2266 break;
2267 }
2268 bytelen = MIN(uio->uio_resid,
2269 vp->v_size - uio->uio_offset);
2270 if (bytelen == 0)
2271 break;
2272
2273 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
2274 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
2275 if (error)
2276 break;
2277 }
2278
2279 if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
2280 puffs_updatenode(VPTOPP(vp), PUFFS_UPDATEATIME, 0);
2281 } else {
2282 /*
2283 * in case it's not a regular file or we're operating
2284 * uncached, do read in the old-fashioned style,
2285 * i.e. explicit read operations
2286 */
2287
2288 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2289 argsize = sizeof(struct puffs_vnmsg_read);
2290 puffs_msgmem_alloc(argsize + tomove, &park_read,
2291 (void *)&read_msg, 1);
2292
2293 error = 0;
2294 while (uio->uio_resid > 0) {
2295 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2296 memset(read_msg, 0, argsize); /* XXX: touser KASSERT */
2297 RWARGS(read_msg, ap->a_ioflag, tomove,
2298 uio->uio_offset, ap->a_cred);
2299 puffs_msg_setinfo(park_read, PUFFSOP_VN,
2300 PUFFS_VN_READ, VPTOPNC(vp));
2301 puffs_msg_setdelta(park_read, tomove);
2302
2303 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_read, vp->v_data,
2304 NULL, error);
2305 error = checkerr(pmp, error, __func__);
2306 if (error)
2307 break;
2308
2309 if (read_msg->pvnr_resid > tomove) {
2310 puffs_senderr(pmp, PUFFS_ERR_READ,
2311 E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2312 error = EPROTO;
2313 break;
2314 }
2315
2316 error = uiomove(read_msg->pvnr_data,
2317 tomove - read_msg->pvnr_resid, uio);
2318
2319 /*
2320 * in case the file is out of juice, resid from
2321 * userspace is != 0. and the error-case is
2322 * quite obvious
2323 */
2324 if (error || read_msg->pvnr_resid)
2325 break;
2326 }
2327
2328 puffs_msgmem_release(park_read);
2329 }
2330
2331 return error;
2332 }
2333
2334 /*
2335 * XXX: in case of a failure, this leaves uio in a bad state.
2336 * We could theoretically copy the uio and iovecs and "replay"
2337 * them the right amount after the userspace trip, but don't
2338 * bother for now.
2339 */
2340 int
2341 puffs_vnop_write(void *v)
2342 {
2343 struct vop_write_args /* {
2344 const struct vnodeop_desc *a_desc;
2345 struct vnode *a_vp;
2346 struct uio *a_uio;
2347 int a_ioflag;
2348 kauth_cred_t a_cred;
2349 } */ *ap = v;
2350 PUFFS_MSG_VARS(vn, write);
2351 struct vnode *vp = ap->a_vp;
2352 struct puffs_node *pn = VPTOPP(vp);
2353 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2354 struct uio *uio = ap->a_uio;
2355 size_t tomove, argsize;
2356 off_t oldoff, newoff, origoff;
2357 vsize_t bytelen;
2358 int error, uflags;
2359 int ubcflags;
2360
2361 error = uflags = 0;
2362 write_msg = NULL;
2363
2364 /* std sanity */
2365 if (uio->uio_resid == 0)
2366 return 0;
2367 if (uio->uio_offset < 0)
2368 return EFBIG;
2369
2370 mutex_enter(&pn->pn_sizemtx);
2371
2372 if (vp->v_type == VREG &&
2373 PUFFS_USE_PAGECACHE(pmp) &&
2374 !(pn->pn_stat & PNODE_WDIRECT)) {
2375 ubcflags = UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp);
2376
2377 /*
2378 * userspace *should* be allowed to control this,
2379 * but with UBC it's a bit unclear how to handle it
2380 */
2381 if (ap->a_ioflag & IO_APPEND)
2382 uio->uio_offset = vp->v_size;
2383
2384 origoff = uio->uio_offset;
2385 while (uio->uio_resid > 0) {
2386 oldoff = uio->uio_offset;
2387 bytelen = uio->uio_resid;
2388
2389 newoff = oldoff + bytelen;
2390 if (vp->v_size < newoff) {
2391 uvm_vnp_setwritesize(vp, newoff);
2392 }
2393 error = ubc_uiomove(&vp->v_uobj, uio, bytelen,
2394 UVM_ADV_RANDOM, ubcflags);
2395
2396 /*
2397 * In case of a ubc_uiomove() error,
2398 * opt to not extend the file at all and
2399 * return an error. Otherwise, if we attempt
2400 * to clear the memory we couldn't fault to,
2401 * we might generate a kernel page fault.
2402 */
2403 if (vp->v_size < newoff) {
2404 if (error == 0) {
2405 uflags |= PUFFS_UPDATESIZE;
2406 uvm_vnp_setsize(vp, newoff);
2407 } else {
2408 uvm_vnp_setwritesize(vp, vp->v_size);
2409 }
2410 }
2411 if (error)
2412 break;
2413
2414 /*
2415 * If we're writing large files, flush to file server
2416 * every 64k. Otherwise we can very easily exhaust
2417 * kernel and user memory, as the file server cannot
2418 * really keep up with our writing speed.
2419 *
2420 * Note: this does *NOT* honor MNT_ASYNC, because
2421 * that gives userland too much say in the kernel.
2422 */
2423 if (oldoff >> 16 != uio->uio_offset >> 16) {
2424 mutex_enter(vp->v_interlock);
2425 error = VOP_PUTPAGES(vp, oldoff & ~0xffff,
2426 uio->uio_offset & ~0xffff,
2427 PGO_CLEANIT | PGO_SYNCIO);
2428 if (error)
2429 break;
2430 }
2431 }
2432
2433 /* synchronous I/O? */
2434 if (error == 0 && ap->a_ioflag & IO_SYNC) {
2435 mutex_enter(vp->v_interlock);
2436 error = VOP_PUTPAGES(vp, trunc_page(origoff),
2437 round_page(uio->uio_offset),
2438 PGO_CLEANIT | PGO_SYNCIO);
2439
2440 /* write through page cache? */
2441 } else if (error == 0 && pmp->pmp_flags & PUFFS_KFLAG_WTCACHE) {
2442 mutex_enter(vp->v_interlock);
2443 error = VOP_PUTPAGES(vp, trunc_page(origoff),
2444 round_page(uio->uio_offset), PGO_CLEANIT);
2445 }
2446 } else {
2447 /* tomove is non-increasing */
2448 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2449 argsize = sizeof(struct puffs_vnmsg_write) + tomove;
2450 puffs_msgmem_alloc(argsize, &park_write, (void *)&write_msg,1);
2451
2452 while (uio->uio_resid > 0) {
2453 /* move data to buffer */
2454 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2455 memset(write_msg, 0, argsize); /* XXX: touser KASSERT */
2456 RWARGS(write_msg, ap->a_ioflag, tomove,
2457 uio->uio_offset, ap->a_cred);
2458 error = uiomove(write_msg->pvnr_data, tomove, uio);
2459 if (error)
2460 break;
2461
2462 /* move buffer to userspace */
2463 puffs_msg_setinfo(park_write, PUFFSOP_VN,
2464 PUFFS_VN_WRITE, VPTOPNC(vp));
2465 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_write, vp->v_data,
2466 NULL, error);
2467 error = checkerr(pmp, error, __func__);
2468 if (error)
2469 break;
2470
2471 if (write_msg->pvnr_resid > tomove) {
2472 puffs_senderr(pmp, PUFFS_ERR_WRITE,
2473 E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2474 error = EPROTO;
2475 break;
2476 }
2477
2478 /* adjust file size */
2479 if (vp->v_size < uio->uio_offset) {
2480 uflags |= PUFFS_UPDATESIZE;
2481 uvm_vnp_setsize(vp, uio->uio_offset);
2482 }
2483
2484 /* didn't move everything? bad userspace. bail */
2485 if (write_msg->pvnr_resid != 0) {
2486 error = EIO;
2487 break;
2488 }
2489 }
2490 puffs_msgmem_release(park_write);
2491 }
2492
2493 if (vp->v_mount->mnt_flag & MNT_RELATIME)
2494 uflags |= PUFFS_UPDATEATIME;
2495 uflags |= PUFFS_UPDATECTIME;
2496 uflags |= PUFFS_UPDATEMTIME;
2497 puffs_updatenode(VPTOPP(vp), uflags, vp->v_size);
2498
2499 mutex_exit(&pn->pn_sizemtx);
2500 return error;
2501 }
2502
2503 int
2504 puffs_vnop_fallocate(void *v)
2505 {
2506 struct vop_fallocate_args /* {
2507 const struct vnodeop_desc *a_desc;
2508 struct vnode *a_vp;
2509 off_t a_pos;
2510 off_t a_len;
2511 } */ *ap = v;
2512 struct vnode *vp = ap->a_vp;
2513 struct puffs_node *pn = VPTOPP(vp);
2514 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2515 PUFFS_MSG_VARS(vn, fallocate);
2516 int error;
2517
2518 mutex_enter(&pn->pn_sizemtx);
2519
2520 PUFFS_MSG_ALLOC(vn, fallocate);
2521 fallocate_msg->pvnr_off = ap->a_pos;
2522 fallocate_msg->pvnr_len = ap->a_len;
2523 puffs_msg_setinfo(park_fallocate, PUFFSOP_VN,
2524 PUFFS_VN_FALLOCATE, VPTOPNC(vp));
2525
2526 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fallocate, vp->v_data, NULL, error);
2527 error = checkerr(pmp, error, __func__);
2528 PUFFS_MSG_RELEASE(fallocate);
2529
2530 switch (error) {
2531 case 0:
2532 break;
2533 case EAGAIN:
2534 error = EIO;
2535 /* FALLTHROUGH */
2536 default:
2537 goto out;
2538 }
2539
2540 if (ap->a_pos + ap->a_len > vp->v_size) {
2541 uvm_vnp_setsize(vp, ap->a_pos + ap->a_len);
2542 puffs_updatenode(pn, PUFFS_UPDATESIZE, vp->v_size);
2543 }
2544 out:
2545 mutex_exit(&pn->pn_sizemtx);
2546
2547 return error;
2548 }
2549
2550 int
2551 puffs_vnop_fdiscard(void *v)
2552 {
2553 struct vop_fdiscard_args /* {
2554 const struct vnodeop_desc *a_desc;
2555 struct vnode *a_vp;
2556 off_t a_pos;
2557 off_t a_len;
2558 } */ *ap = v;
2559 struct vnode *vp = ap->a_vp;
2560 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2561 PUFFS_MSG_VARS(vn, fdiscard);
2562 int error;
2563
2564 PUFFS_MSG_ALLOC(vn, fdiscard);
2565 fdiscard_msg->pvnr_off = ap->a_pos;
2566 fdiscard_msg->pvnr_len = ap->a_len;
2567 puffs_msg_setinfo(park_fdiscard, PUFFSOP_VN,
2568 PUFFS_VN_FALLOCATE, VPTOPNC(vp));
2569
2570 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fdiscard, vp->v_data, NULL, error);
2571 error = checkerr(pmp, error, __func__);
2572 PUFFS_MSG_RELEASE(fdiscard);
2573
2574 return error;
2575 }
2576
2577 int
2578 puffs_vnop_print(void *v)
2579 {
2580 struct vop_print_args /* {
2581 struct vnode *a_vp;
2582 } */ *ap = v;
2583 PUFFS_MSG_VARS(vn, print);
2584 struct vnode *vp = ap->a_vp;
2585 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2586 struct puffs_node *pn = vp->v_data;
2587
2588 /* kernel portion */
2589 printf("tag VT_PUFFS, vnode %p, puffs node: %p,\n"
2590 "\tuserspace cookie: %p", vp, pn, pn->pn_cookie);
2591 if (vp->v_type == VFIFO)
2592 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
2593 printf("\n");
2594
2595 /* userspace portion */
2596 if (EXISTSOP(pmp, PRINT)) {
2597 PUFFS_MSG_ALLOC(vn, print);
2598 puffs_msg_setinfo(park_print, PUFFSOP_VN,
2599 PUFFS_VN_PRINT, VPTOPNC(vp));
2600 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_print, vp->v_data,
2601 NULL);
2602 PUFFS_MSG_RELEASE(print);
2603 }
2604
2605 return 0;
2606 }
2607
2608 int
2609 puffs_vnop_pathconf(void *v)
2610 {
2611 struct vop_pathconf_args /* {
2612 const struct vnodeop_desc *a_desc;
2613 struct vnode *a_vp;
2614 int a_name;
2615 register_t *a_retval;
2616 } */ *ap = v;
2617 PUFFS_MSG_VARS(vn, pathconf);
2618 struct vnode *vp = ap->a_vp;
2619 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2620 int error;
2621
2622 PUFFS_MSG_ALLOC(vn, pathconf);
2623 pathconf_msg->pvnr_name = ap->a_name;
2624 puffs_msg_setinfo(park_pathconf, PUFFSOP_VN,
2625 PUFFS_VN_PATHCONF, VPTOPNC(vp));
2626 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_pathconf, vp->v_data, NULL, error);
2627 error = checkerr(pmp, error, __func__);
2628 if (!error)
2629 *ap->a_retval = pathconf_msg->pvnr_retval;
2630 PUFFS_MSG_RELEASE(pathconf);
2631
2632 return error;
2633 }
2634
2635 int
2636 puffs_vnop_advlock(void *v)
2637 {
2638 struct vop_advlock_args /* {
2639 const struct vnodeop_desc *a_desc;
2640 struct vnode *a_vp;
2641 void *a_id;
2642 int a_op;
2643 struct flock *a_fl;
2644 int a_flags;
2645 } */ *ap = v;
2646 PUFFS_MSG_VARS(vn, advlock);
2647 struct vnode *vp = ap->a_vp;
2648 struct puffs_node *pn = VPTOPP(vp);
2649 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2650 int error;
2651
2652 if (!EXISTSOP(pmp, ADVLOCK))
2653 return lf_advlock(ap, &pn->pn_lockf, vp->v_size);
2654
2655 PUFFS_MSG_ALLOC(vn, advlock);
2656 (void)memcpy(&advlock_msg->pvnr_fl, ap->a_fl,
2657 sizeof(advlock_msg->pvnr_fl));
2658 advlock_msg->pvnr_id = ap->a_id;
2659 advlock_msg->pvnr_op = ap->a_op;
2660 advlock_msg->pvnr_flags = ap->a_flags;
2661 puffs_msg_setinfo(park_advlock, PUFFSOP_VN,
2662 PUFFS_VN_ADVLOCK, VPTOPNC(vp));
2663 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_advlock, vp->v_data, NULL, error);
2664 error = checkerr(pmp, error, __func__);
2665 PUFFS_MSG_RELEASE(advlock);
2666
2667 return error;
2668 }
2669
2670 int
2671 puffs_vnop_abortop(void *v)
2672 {
2673 struct vop_abortop_args /* {
2674 struct vnode *a_dvp;
2675 struct componentname *a_cnp;
2676 }; */ *ap = v;
2677 PUFFS_MSG_VARS(vn, abortop);
2678 struct vnode *dvp = ap->a_dvp;
2679 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2680 struct componentname *cnp = ap->a_cnp;
2681
2682 if (EXISTSOP(pmp, ABORTOP)) {
2683 PUFFS_MSG_ALLOC(vn, abortop);
2684 puffs_makecn(&abortop_msg->pvnr_cn, &abortop_msg->pvnr_cn_cred,
2685 cnp, PUFFS_USE_FULLPNBUF(pmp));
2686 puffs_msg_setfaf(park_abortop);
2687 puffs_msg_setinfo(park_abortop, PUFFSOP_VN,
2688 PUFFS_VN_ABORTOP, VPTOPNC(dvp));
2689
2690 puffs_msg_enqueue(pmp, park_abortop);
2691 PUFFS_MSG_RELEASE(abortop);
2692 }
2693
2694 return genfs_abortop(v);
2695 }
2696
2697 #define BIOASYNC(bp) (bp->b_flags & B_ASYNC)
2698
2699 /*
2700 * This maps itself to PUFFS_VN_READ/WRITE for data transfer.
2701 */
2702 int
2703 puffs_vnop_strategy(void *v)
2704 {
2705 struct vop_strategy_args /* {
2706 const struct vnodeop_desc *a_desc;
2707 struct vnode *a_vp;
2708 struct buf *a_bp;
2709 } */ *ap = v;
2710 PUFFS_MSG_VARS(vn, rw);
2711 struct vnode *vp = ap->a_vp;
2712 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2713 struct puffs_node *pn;
2714 struct buf *bp;
2715 size_t argsize;
2716 size_t tomove, moved;
2717 int error, dofaf, cansleep, dobiodone;
2718
2719 pmp = MPTOPUFFSMP(vp->v_mount);
2720 bp = ap->a_bp;
2721 error = 0;
2722 dofaf = 0;
2723 cansleep = 0;
2724 pn = VPTOPP(vp);
2725 park_rw = NULL; /* explicit */
2726 dobiodone = 1;
2727
2728 if ((BUF_ISREAD(bp) && !EXISTSOP(pmp, READ))
2729 || (BUF_ISWRITE(bp) && !EXISTSOP(pmp, WRITE)))
2730 ERROUT(EOPNOTSUPP);
2731
2732 /*
2733 * Short-circuit optimization: don't flush buffer in between
2734 * VOP_INACTIVE and VOP_RECLAIM in case the node has no references.
2735 */
2736 if (pn->pn_stat & PNODE_DYING) {
2737 KASSERT(BUF_ISWRITE(bp));
2738 bp->b_resid = 0;
2739 goto out;
2740 }
2741
2742 #ifdef DIAGNOSTIC
2743 if (bp->b_bcount > pmp->pmp_msg_maxsize - PUFFS_MSGSTRUCT_MAX)
2744 panic("puffs_strategy: wildly inappropriate buf bcount %d",
2745 bp->b_bcount);
2746 #endif
2747
2748 /*
2749 * See explanation for the necessity of a FAF in puffs_fsync.
2750 *
2751 * Also, do FAF in case we're suspending.
2752 * See puffs_vfsops.c:pageflush()
2753 */
2754 if (BUF_ISWRITE(bp)) {
2755 mutex_enter(vp->v_interlock);
2756 if (vdead_check(vp, VDEAD_NOWAIT) != 0)
2757 dofaf = 1;
2758 if (pn->pn_stat & PNODE_FAF)
2759 dofaf = 1;
2760 mutex_exit(vp->v_interlock);
2761 }
2762
2763 cansleep = (curlwp == uvm.pagedaemon_lwp || dofaf) ? 0 : 1;
2764
2765 KASSERT(curlwp != uvm.pagedaemon_lwp || dofaf || BIOASYNC(bp));
2766
2767 /* allocate transport structure */
2768 tomove = PUFFS_TOMOVE(bp->b_bcount, pmp);
2769 argsize = sizeof(struct puffs_vnmsg_rw);
2770 error = puffs_msgmem_alloc(argsize + tomove, &park_rw,
2771 (void *)&rw_msg, cansleep);
2772 if (error)
2773 goto out;
2774 RWARGS(rw_msg, 0, tomove, bp->b_blkno << DEV_BSHIFT, FSCRED);
2775
2776 /* 2x2 cases: read/write, faf/nofaf */
2777 if (BUF_ISREAD(bp)) {
2778 puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2779 PUFFS_VN_READ, VPTOPNC(vp));
2780 puffs_msg_setdelta(park_rw, tomove);
2781 if (BIOASYNC(bp)) {
2782 puffs_msg_setcall(park_rw,
2783 puffs_parkdone_asyncbioread, bp);
2784 puffs_msg_enqueue(pmp, park_rw);
2785 dobiodone = 0;
2786 } else {
2787 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data,
2788 NULL, error);
2789 error = checkerr(pmp, error, __func__);
2790 if (error)
2791 goto out;
2792
2793 if (rw_msg->pvnr_resid > tomove) {
2794 puffs_senderr(pmp, PUFFS_ERR_READ,
2795 E2BIG, "resid grew", VPTOPNC(vp));
2796 ERROUT(EPROTO);
2797 }
2798
2799 moved = tomove - rw_msg->pvnr_resid;
2800
2801 (void)memcpy(bp->b_data, rw_msg->pvnr_data, moved);
2802 bp->b_resid = bp->b_bcount - moved;
2803 }
2804 } else {
2805 puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2806 PUFFS_VN_WRITE, VPTOPNC(vp));
2807 /*
2808 * make pages read-only before we write them if we want
2809 * write caching info
2810 */
2811 if (PUFFS_WCACHEINFO(pmp)) {
2812 struct uvm_object *uobj = &vp->v_uobj;
2813 int npages = (bp->b_bcount + PAGE_SIZE-1) >> PAGE_SHIFT;
2814 struct vm_page *vmp;
2815 int i;
2816
2817 for (i = 0; i < npages; i++) {
2818 vmp= uvm_pageratop((vaddr_t)bp->b_data
2819 + (i << PAGE_SHIFT));
2820 DPRINTF(("puffs_strategy: write-protecting "
2821 "vp %p page %p, offset %" PRId64"\n",
2822 vp, vmp, vmp->offset));
2823 mutex_enter(uobj->vmobjlock);
2824 vmp->flags |= PG_RDONLY;
2825 pmap_page_protect(vmp, VM_PROT_READ);
2826 mutex_exit(uobj->vmobjlock);
2827 }
2828 }
2829
2830 (void)memcpy(&rw_msg->pvnr_data, bp->b_data, tomove);
2831 if (dofaf) {
2832 puffs_msg_setfaf(park_rw);
2833 } else if (BIOASYNC(bp)) {
2834 puffs_msg_setcall(park_rw,
2835 puffs_parkdone_asyncbiowrite, bp);
2836 dobiodone = 0;
2837 }
2838
2839 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data, NULL, error);
2840
2841 if (dobiodone == 0)
2842 goto out;
2843
2844 error = checkerr(pmp, error, __func__);
2845 if (error)
2846 goto out;
2847
2848 if (rw_msg->pvnr_resid > tomove) {
2849 puffs_senderr(pmp, PUFFS_ERR_WRITE,
2850 E2BIG, "resid grew", VPTOPNC(vp));
2851 ERROUT(EPROTO);
2852 }
2853
2854 /*
2855 * FAF moved everything. Frankly, we don't
2856 * really have a choice.
2857 */
2858 if (dofaf && error == 0)
2859 moved = tomove;
2860 else
2861 moved = tomove - rw_msg->pvnr_resid;
2862
2863 bp->b_resid = bp->b_bcount - moved;
2864 if (bp->b_resid != 0) {
2865 ERROUT(EIO);
2866 }
2867 }
2868
2869 out:
2870 if (park_rw)
2871 puffs_msgmem_release(park_rw);
2872
2873 if (error)
2874 bp->b_error = error;
2875
2876 if (error || dobiodone)
2877 biodone(bp);
2878
2879 return error;
2880 }
2881
2882 int
2883 puffs_vnop_mmap(void *v)
2884 {
2885 struct vop_mmap_args /* {
2886 const struct vnodeop_desc *a_desc;
2887 struct vnode *a_vp;
2888 vm_prot_t a_prot;
2889 kauth_cred_t a_cred;
2890 } */ *ap = v;
2891 PUFFS_MSG_VARS(vn, mmap);
2892 struct vnode *vp = ap->a_vp;
2893 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2894 int error;
2895
2896 if (!PUFFS_USE_PAGECACHE(pmp))
2897 return genfs_eopnotsupp(v);
2898
2899 if (EXISTSOP(pmp, MMAP)) {
2900 PUFFS_MSG_ALLOC(vn, mmap);
2901 mmap_msg->pvnr_prot = ap->a_prot;
2902 puffs_credcvt(&mmap_msg->pvnr_cred, ap->a_cred);
2903 puffs_msg_setinfo(park_mmap, PUFFSOP_VN,
2904 PUFFS_VN_MMAP, VPTOPNC(vp));
2905
2906 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mmap, vp->v_data, NULL, error);
2907 error = checkerr(pmp, error, __func__);
2908 PUFFS_MSG_RELEASE(mmap);
2909 } else {
2910 error = genfs_mmap(v);
2911 }
2912
2913 return error;
2914 }
2915
2916
2917 /*
2918 * The rest don't get a free trip to userspace and back, they
2919 * have to stay within the kernel.
2920 */
2921
2922 /*
2923 * bmap doesn't really make any sense for puffs, so just 1:1 map it.
2924 * well, maybe somehow, somewhere, some day ....
2925 */
2926 int
2927 puffs_vnop_bmap(void *v)
2928 {
2929 struct vop_bmap_args /* {
2930 const struct vnodeop_desc *a_desc;
2931 struct vnode *a_vp;
2932 daddr_t a_bn;
2933 struct vnode **a_vpp;
2934 daddr_t *a_bnp;
2935 int *a_runp;
2936 } */ *ap = v;
2937 struct puffs_mount *pmp;
2938
2939 pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2940
2941 if (ap->a_vpp)
2942 *ap->a_vpp = ap->a_vp;
2943 if (ap->a_bnp)
2944 *ap->a_bnp = ap->a_bn;
2945 if (ap->a_runp)
2946 *ap->a_runp
2947 = (PUFFS_TOMOVE(pmp->pmp_msg_maxsize, pmp)>>DEV_BSHIFT) - 1;
2948
2949 return 0;
2950 }
2951
2952 /*
2953 * Handle getpages faults in puffs. We let genfs_getpages() do most
2954 * of the dirty work, but we come in this route to do accounting tasks.
2955 * If the user server has specified functions for cache notifications
2956 * about reads and/or writes, we record which type of operation we got,
2957 * for which page range, and proceed to issue a FAF notification to the
2958 * server about it.
2959 */
2960 int
2961 puffs_vnop_getpages(void *v)
2962 {
2963 struct vop_getpages_args /* {
2964 const struct vnodeop_desc *a_desc;
2965 struct vnode *a_vp;
2966 voff_t a_offset;
2967 struct vm_page **a_m;
2968 int *a_count;
2969 int a_centeridx;
2970 vm_prot_t a_access_type;
2971 int a_advice;
2972 int a_flags;
2973 } */ *ap = v;
2974 struct puffs_mount *pmp;
2975 struct puffs_node *pn;
2976 struct vnode *vp;
2977 struct vm_page **pgs;
2978 struct puffs_cacheinfo *pcinfo = NULL;
2979 struct puffs_cacherun *pcrun;
2980 void *parkmem = NULL;
2981 size_t runsizes;
2982 int i, npages, si, streakon;
2983 int error, locked, write;
2984
2985 pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2986 npages = *ap->a_count;
2987 pgs = ap->a_m;
2988 vp = ap->a_vp;
2989 pn = vp->v_data;
2990 locked = (ap->a_flags & PGO_LOCKED) != 0;
2991 write = (ap->a_access_type & VM_PROT_WRITE) != 0;
2992
2993 /* ccg xnaht - gets Wuninitialized wrong */
2994 pcrun = NULL;
2995 runsizes = 0;
2996
2997 /*
2998 * Check that we aren't trying to fault in pages which our file
2999 * server doesn't know about. This happens if we extend a file by
3000 * skipping some pages and later try to fault in pages which
3001 * are between pn_serversize and vp_size. This check optimizes
3002 * away the common case where a file is being extended.
3003 */
3004 if (ap->a_offset >= pn->pn_serversize && ap->a_offset < vp->v_size) {
3005 struct vattr va;
3006
3007 /* try again later when we can block */
3008 if (locked)
3009 ERROUT(EBUSY);
3010
3011 mutex_exit(vp->v_interlock);
3012 vattr_null(&va);
3013 va.va_size = vp->v_size;
3014 error = dosetattr(vp, &va, FSCRED, 0);
3015 if (error)
3016 ERROUT(error);
3017 mutex_enter(vp->v_interlock);
3018 }
3019
3020 if (write && PUFFS_WCACHEINFO(pmp)) {
3021 #ifdef notnowjohn
3022 /* allocate worst-case memory */
3023 runsizes = ((npages / 2) + 1) * sizeof(struct puffs_cacherun);
3024 KASSERT(curlwp != uvm.pagedaemon_lwp || locked);
3025 pcinfo = kmem_zalloc(sizeof(struct puffs_cacheinfo) + runsize,
3026 locked ? KM_NOSLEEP : KM_SLEEP);
3027
3028 /*
3029 * can't block if we're locked and can't mess up caching
3030 * information for fs server. so come back later, please
3031 */
3032 if (pcinfo == NULL)
3033 ERROUT(ENOMEM);
3034
3035 parkmem = puffs_park_alloc(locked == 0);
3036 if (parkmem == NULL)
3037 ERROUT(ENOMEM);
3038
3039 pcrun = pcinfo->pcache_runs;
3040 #else
3041 (void)parkmem;
3042 #endif
3043 }
3044
3045 error = genfs_getpages(v);
3046 if (error)
3047 goto out;
3048
3049 if (PUFFS_WCACHEINFO(pmp) == 0)
3050 goto out;
3051
3052 /*
3053 * Let's see whose fault it was and inform the user server of
3054 * possibly read/written pages. Map pages from read faults
3055 * strictly read-only, since otherwise we might miss info on
3056 * when the page is actually write-faulted to.
3057 */
3058 if (!locked)
3059 mutex_enter(vp->v_uobj.vmobjlock);
3060 for (i = 0, si = 0, streakon = 0; i < npages; i++) {
3061 if (pgs[i] == NULL || pgs[i] == PGO_DONTCARE) {
3062 if (streakon && write) {
3063 streakon = 0;
3064 pcrun[si].pcache_runend
3065 = trunc_page(pgs[i]->offset) + PAGE_MASK;
3066 si++;
3067 }
3068 continue;
3069 }
3070 if (streakon == 0 && write) {
3071 streakon = 1;
3072 pcrun[si].pcache_runstart = pgs[i]->offset;
3073 }
3074
3075 if (!write)
3076 pgs[i]->flags |= PG_RDONLY;
3077 }
3078 /* was the last page part of our streak? */
3079 if (streakon) {
3080 pcrun[si].pcache_runend
3081 = trunc_page(pgs[i-1]->offset) + PAGE_MASK;
3082 si++;
3083 }
3084 if (!locked)
3085 mutex_exit(vp->v_uobj.vmobjlock);
3086
3087 KASSERT(si <= (npages / 2) + 1);
3088
3089 #ifdef notnowjohn
3090 /* send results to userspace */
3091 if (write)
3092 puffs_cacheop(pmp, parkmem, pcinfo,
3093 sizeof(struct puffs_cacheinfo) + runsizes, VPTOPNC(vp));
3094 #endif
3095
3096 out:
3097 if (error) {
3098 if (pcinfo != NULL)
3099 kmem_free(pcinfo,
3100 sizeof(struct puffs_cacheinfo) + runsizes);
3101 #ifdef notnowjohn
3102 if (parkmem != NULL)
3103 puffs_park_release(parkmem, 1);
3104 #endif
3105 }
3106
3107 return error;
3108 }
3109
3110 /*
3111 * Extended attribute support.
3112 */
3113
3114 int
3115 puffs_vnop_getextattr(void *v)
3116 {
3117 struct vop_getextattr_args /*
3118 struct vnode *a_vp;
3119 int a_attrnamespace;
3120 const char *a_name;
3121 struct uio *a_uio;
3122 size_t *a_size;
3123 kauth_cred_t a_cred;
3124 }; */ *ap = v;
3125 PUFFS_MSG_VARS(vn, getextattr);
3126 struct vnode *vp = ap->a_vp;
3127 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3128 int attrnamespace = ap->a_attrnamespace;
3129 const char *name = ap->a_name;
3130 struct uio *uio = ap->a_uio;
3131 size_t *sizep = ap->a_size;
3132 size_t tomove, resid;
3133 int error;
3134
3135 if (uio)
3136 resid = uio->uio_resid;
3137 else
3138 resid = 0;
3139
3140 tomove = PUFFS_TOMOVE(resid, pmp);
3141 if (tomove != resid) {
3142 error = E2BIG;
3143 goto out;
3144 }
3145
3146 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_getextattr) + tomove,
3147 &park_getextattr, (void *)&getextattr_msg, 1);
3148
3149 getextattr_msg->pvnr_attrnamespace = attrnamespace;
3150 strlcpy(getextattr_msg->pvnr_attrname, name,
3151 sizeof(getextattr_msg->pvnr_attrname));
3152 puffs_credcvt(&getextattr_msg->pvnr_cred, ap->a_cred);
3153 if (sizep)
3154 getextattr_msg->pvnr_datasize = 1;
3155 getextattr_msg->pvnr_resid = tomove;
3156
3157 puffs_msg_setinfo(park_getextattr,
3158 PUFFSOP_VN, PUFFS_VN_GETEXTATTR, VPTOPNC(vp));
3159 puffs_msg_setdelta(park_getextattr, tomove);
3160 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getextattr, vp->v_data, NULL, error);
3161
3162 error = checkerr(pmp, error, __func__);
3163 if (error)
3164 goto out;
3165
3166 resid = getextattr_msg->pvnr_resid;
3167 if (resid > tomove) {
3168 puffs_senderr(pmp, PUFFS_ERR_GETEXTATTR, E2BIG,
3169 "resid grew", VPTOPNC(vp));
3170 error = EPROTO;
3171 goto out;
3172 }
3173
3174 if (sizep)
3175 *sizep = getextattr_msg->pvnr_datasize;
3176 if (uio)
3177 error = uiomove(getextattr_msg->pvnr_data, tomove - resid, uio);
3178
3179 out:
3180 PUFFS_MSG_RELEASE(getextattr);
3181 return error;
3182 }
3183
3184 int
3185 puffs_vnop_setextattr(void *v)
3186 {
3187 struct vop_setextattr_args /* {
3188 struct vnode *a_vp;
3189 int a_attrnamespace;
3190 const char *a_name;
3191 struct uio *a_uio;
3192 kauth_cred_t a_cred;
3193 }; */ *ap = v;
3194 PUFFS_MSG_VARS(vn, setextattr);
3195 struct vnode *vp = ap->a_vp;
3196 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3197 int attrnamespace = ap->a_attrnamespace;
3198 const char *name = ap->a_name;
3199 struct uio *uio = ap->a_uio;
3200 size_t tomove, resid;
3201 int error;
3202
3203 if (uio)
3204 resid = uio->uio_resid;
3205 else
3206 resid = 0;
3207
3208 tomove = PUFFS_TOMOVE(resid, pmp);
3209 if (tomove != resid) {
3210 error = E2BIG;
3211 goto out;
3212 }
3213
3214 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_setextattr) + tomove,
3215 &park_setextattr, (void *)&setextattr_msg, 1);
3216
3217 setextattr_msg->pvnr_attrnamespace = attrnamespace;
3218 strlcpy(setextattr_msg->pvnr_attrname, name,
3219 sizeof(setextattr_msg->pvnr_attrname));
3220 puffs_credcvt(&setextattr_msg->pvnr_cred, ap->a_cred);
3221 setextattr_msg->pvnr_resid = tomove;
3222
3223 if (uio) {
3224 error = uiomove(setextattr_msg->pvnr_data, tomove, uio);
3225 if (error)
3226 goto out;
3227 }
3228
3229 puffs_msg_setinfo(park_setextattr,
3230 PUFFSOP_VN, PUFFS_VN_SETEXTATTR, VPTOPNC(vp));
3231 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_setextattr, vp->v_data, NULL, error);
3232
3233 error = checkerr(pmp, error, __func__);
3234 if (error)
3235 goto out;
3236
3237 if (setextattr_msg->pvnr_resid != 0)
3238 error = EIO;
3239
3240 out:
3241 PUFFS_MSG_RELEASE(setextattr);
3242
3243 return error;
3244 }
3245
3246 int
3247 puffs_vnop_listextattr(void *v)
3248 {
3249 struct vop_listextattr_args /* {
3250 struct vnode *a_vp;
3251 int a_attrnamespace;
3252 struct uio *a_uio;
3253 size_t *a_size;
3254 int a_flag,
3255 kauth_cred_t a_cred;
3256 }; */ *ap = v;
3257 PUFFS_MSG_VARS(vn, listextattr);
3258 struct vnode *vp = ap->a_vp;
3259 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3260 int attrnamespace = ap->a_attrnamespace;
3261 struct uio *uio = ap->a_uio;
3262 size_t *sizep = ap->a_size;
3263 int flag = ap->a_flag;
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_listextattr) + tomove,
3279 &park_listextattr, (void *)&listextattr_msg, 1);
3280
3281 listextattr_msg->pvnr_attrnamespace = attrnamespace;
3282 listextattr_msg->pvnr_flag = flag;
3283 puffs_credcvt(&listextattr_msg->pvnr_cred, ap->a_cred);
3284 listextattr_msg->pvnr_resid = tomove;
3285 if (sizep)
3286 listextattr_msg->pvnr_datasize = 1;
3287
3288 puffs_msg_setinfo(park_listextattr,
3289 PUFFSOP_VN, PUFFS_VN_LISTEXTATTR, VPTOPNC(vp));
3290 puffs_msg_setdelta(park_listextattr, tomove);
3291 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_listextattr, vp->v_data, NULL, error);
3292
3293 error = checkerr(pmp, error, __func__);
3294 if (error)
3295 goto out;
3296
3297 resid = listextattr_msg->pvnr_resid;
3298 if (resid > tomove) {
3299 puffs_senderr(pmp, PUFFS_ERR_LISTEXTATTR, E2BIG,
3300 "resid grew", VPTOPNC(vp));
3301 error = EPROTO;
3302 goto out;
3303 }
3304
3305 if (sizep)
3306 *sizep = listextattr_msg->pvnr_datasize;
3307 if (uio)
3308 error = uiomove(listextattr_msg->pvnr_data, tomove-resid, uio);
3309
3310 out:
3311 PUFFS_MSG_RELEASE(listextattr);
3312 return error;
3313 }
3314
3315 int
3316 puffs_vnop_deleteextattr(void *v)
3317 {
3318 struct vop_deleteextattr_args /* {
3319 struct vnode *a_vp;
3320 int a_attrnamespace;
3321 const char *a_name;
3322 kauth_cred_t a_cred;
3323 }; */ *ap = v;
3324 PUFFS_MSG_VARS(vn, deleteextattr);
3325 struct vnode *vp = ap->a_vp;
3326 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3327 int attrnamespace = ap->a_attrnamespace;
3328 const char *name = ap->a_name;
3329 int error;
3330
3331 PUFFS_MSG_ALLOC(vn, deleteextattr);
3332 deleteextattr_msg->pvnr_attrnamespace = attrnamespace;
3333 strlcpy(deleteextattr_msg->pvnr_attrname, name,
3334 sizeof(deleteextattr_msg->pvnr_attrname));
3335 puffs_credcvt(&deleteextattr_msg->pvnr_cred, ap->a_cred);
3336
3337 puffs_msg_setinfo(park_deleteextattr,
3338 PUFFSOP_VN, PUFFS_VN_DELETEEXTATTR, VPTOPNC(vp));
3339 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_deleteextattr,
3340 vp->v_data, NULL, error);
3341
3342 error = checkerr(pmp, error, __func__);
3343
3344 PUFFS_MSG_RELEASE(deleteextattr);
3345 return error;
3346 }
3347
3348 /*
3349 * spec & fifo. These call the miscfs spec and fifo vectors, but issue
3350 * FAF update information for the puffs node first.
3351 */
3352 int
3353 puffs_vnop_spec_read(void *v)
3354 {
3355 struct vop_read_args /* {
3356 const struct vnodeop_desc *a_desc;
3357 struct vnode *a_vp;
3358 struct uio *a_uio;
3359 int a_ioflag;
3360 kauth_cred_t a_cred;
3361 } */ *ap = v;
3362
3363 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3364 return VOCALL(spec_vnodeop_p, VOFFSET(vop_read), v);
3365 }
3366
3367 int
3368 puffs_vnop_spec_write(void *v)
3369 {
3370 struct vop_write_args /* {
3371 const struct vnodeop_desc *a_desc;
3372 struct vnode *a_vp;
3373 struct uio *a_uio;
3374 int a_ioflag;
3375 kauth_cred_t a_cred;
3376 } */ *ap = v;
3377
3378 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3379 return VOCALL(spec_vnodeop_p, VOFFSET(vop_write), v);
3380 }
3381
3382 int
3383 puffs_vnop_fifo_read(void *v)
3384 {
3385 struct vop_read_args /* {
3386 const struct vnodeop_desc *a_desc;
3387 struct vnode *a_vp;
3388 struct uio *a_uio;
3389 int a_ioflag;
3390 kauth_cred_t a_cred;
3391 } */ *ap = v;
3392
3393 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3394 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), v);
3395 }
3396
3397 int
3398 puffs_vnop_fifo_write(void *v)
3399 {
3400 struct vop_write_args /* {
3401 const struct vnodeop_desc *a_desc;
3402 struct vnode *a_vp;
3403 struct uio *a_uio;
3404 int a_ioflag;
3405 kauth_cred_t a_cred;
3406 } */ *ap = v;
3407
3408 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3409 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), v);
3410 }
3411