puffs_vnops.c revision 1.181 1 /* $NetBSD: puffs_vnops.c,v 1.181 2014/03/24 13:42:40 hannken 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 1.181 2014/03/24 13:42:40 hannken 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_fcntl(void *);
77 int puffs_vnop_ioctl(void *);
78 int puffs_vnop_inactive(void *);
79 int puffs_vnop_print(void *);
80 int puffs_vnop_pathconf(void *);
81 int puffs_vnop_advlock(void *);
82 int puffs_vnop_strategy(void *);
83 int puffs_vnop_bmap(void *);
84 int puffs_vnop_mmap(void *);
85 int puffs_vnop_getpages(void *);
86 int puffs_vnop_abortop(void *);
87 int puffs_vnop_getextattr(void *);
88 int puffs_vnop_setextattr(void *);
89 int puffs_vnop_listextattr(void *);
90 int puffs_vnop_deleteextattr(void *);
91
92 int puffs_vnop_spec_read(void *);
93 int puffs_vnop_spec_write(void *);
94 int puffs_vnop_fifo_read(void *);
95 int puffs_vnop_fifo_write(void *);
96
97 int puffs_vnop_checkop(void *);
98
99 #define puffs_vnop_lock genfs_lock
100 #define puffs_vnop_unlock genfs_unlock
101 #define puffs_vnop_islocked genfs_islocked
102
103 int (**puffs_vnodeop_p)(void *);
104 const struct vnodeopv_entry_desc puffs_vnodeop_entries[] = {
105 { &vop_default_desc, vn_default_error },
106 { &vop_lookup_desc, puffs_vnop_lookup }, /* REAL lookup */
107 { &vop_create_desc, puffs_vnop_checkop }, /* create */
108 { &vop_mknod_desc, puffs_vnop_checkop }, /* mknod */
109 { &vop_open_desc, puffs_vnop_open }, /* REAL open */
110 { &vop_close_desc, puffs_vnop_checkop }, /* close */
111 { &vop_access_desc, puffs_vnop_access }, /* REAL access */
112 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */
113 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */
114 { &vop_read_desc, puffs_vnop_checkop }, /* read */
115 { &vop_write_desc, puffs_vnop_checkop }, /* write */
116 { &vop_fsync_desc, puffs_vnop_fsync }, /* REAL fsync */
117 { &vop_seek_desc, puffs_vnop_checkop }, /* seek */
118 { &vop_remove_desc, puffs_vnop_checkop }, /* remove */
119 { &vop_link_desc, puffs_vnop_checkop }, /* link */
120 { &vop_rename_desc, puffs_vnop_checkop }, /* rename */
121 { &vop_mkdir_desc, puffs_vnop_checkop }, /* mkdir */
122 { &vop_rmdir_desc, puffs_vnop_checkop }, /* rmdir */
123 { &vop_symlink_desc, puffs_vnop_checkop }, /* symlink */
124 { &vop_readdir_desc, puffs_vnop_checkop }, /* readdir */
125 { &vop_readlink_desc, puffs_vnop_checkop }, /* readlink */
126 { &vop_getpages_desc, puffs_vnop_checkop }, /* getpages */
127 { &vop_putpages_desc, genfs_putpages }, /* REAL putpages */
128 { &vop_pathconf_desc, puffs_vnop_checkop }, /* pathconf */
129 { &vop_advlock_desc, puffs_vnop_advlock }, /* advlock */
130 { &vop_strategy_desc, puffs_vnop_strategy }, /* REAL strategy */
131 { &vop_revoke_desc, genfs_revoke }, /* REAL revoke */
132 { &vop_abortop_desc, puffs_vnop_abortop }, /* REAL abortop */
133 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
134 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
135 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
136 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
137 { &vop_bmap_desc, puffs_vnop_bmap }, /* REAL bmap */
138 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
139 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
140 { &vop_bwrite_desc, genfs_nullop }, /* REAL bwrite */
141 { &vop_mmap_desc, puffs_vnop_mmap }, /* REAL mmap */
142 { &vop_poll_desc, puffs_vnop_poll }, /* REAL poll */
143 { &vop_getextattr_desc, puffs_vnop_getextattr }, /* getextattr */
144 { &vop_setextattr_desc, puffs_vnop_setextattr }, /* setextattr */
145 { &vop_listextattr_desc, puffs_vnop_listextattr }, /* listextattr */
146 { &vop_deleteextattr_desc, puffs_vnop_deleteextattr },/* deleteextattr */
147 #if 0
148 { &vop_openextattr_desc, puffs_vnop_checkop }, /* openextattr */
149 { &vop_closeextattr_desc, puffs_vnop_checkop }, /* closeextattr */
150 #endif
151 { &vop_kqfilter_desc, genfs_eopnotsupp }, /* kqfilter XXX */
152 { NULL, NULL }
153 };
154 const struct vnodeopv_desc puffs_vnodeop_opv_desc =
155 { &puffs_vnodeop_p, puffs_vnodeop_entries };
156
157
158 int (**puffs_specop_p)(void *);
159 const struct vnodeopv_entry_desc puffs_specop_entries[] = {
160 { &vop_default_desc, vn_default_error },
161 { &vop_lookup_desc, spec_lookup }, /* lookup, ENOTDIR */
162 { &vop_create_desc, spec_create }, /* genfs_badop */
163 { &vop_mknod_desc, spec_mknod }, /* genfs_badop */
164 { &vop_open_desc, spec_open }, /* spec_open */
165 { &vop_close_desc, spec_close }, /* spec_close */
166 { &vop_access_desc, puffs_vnop_checkop }, /* access */
167 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */
168 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */
169 { &vop_read_desc, puffs_vnop_spec_read }, /* update, read */
170 { &vop_write_desc, puffs_vnop_spec_write }, /* update, write */
171 { &vop_ioctl_desc, spec_ioctl }, /* spec_ioctl */
172 { &vop_fcntl_desc, genfs_fcntl }, /* dummy */
173 { &vop_poll_desc, spec_poll }, /* spec_poll */
174 { &vop_kqfilter_desc, spec_kqfilter }, /* spec_kqfilter */
175 { &vop_revoke_desc, spec_revoke }, /* genfs_revoke */
176 { &vop_mmap_desc, spec_mmap }, /* spec_mmap */
177 { &vop_fsync_desc, spec_fsync }, /* vflushbuf */
178 { &vop_seek_desc, spec_seek }, /* genfs_nullop */
179 { &vop_remove_desc, spec_remove }, /* genfs_badop */
180 { &vop_link_desc, spec_link }, /* genfs_badop */
181 { &vop_rename_desc, spec_rename }, /* genfs_badop */
182 { &vop_mkdir_desc, spec_mkdir }, /* genfs_badop */
183 { &vop_rmdir_desc, spec_rmdir }, /* genfs_badop */
184 { &vop_symlink_desc, spec_symlink }, /* genfs_badop */
185 { &vop_readdir_desc, spec_readdir }, /* genfs_badop */
186 { &vop_readlink_desc, spec_readlink }, /* genfs_badop */
187 { &vop_abortop_desc, spec_abortop }, /* genfs_badop */
188 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
189 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
190 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
191 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
192 { &vop_bmap_desc, spec_bmap }, /* dummy */
193 { &vop_strategy_desc, spec_strategy }, /* dev strategy */
194 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
195 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
196 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
197 { &vop_advlock_desc, spec_advlock }, /* lf_advlock */
198 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
199 { &vop_getpages_desc, spec_getpages }, /* genfs_getpages */
200 { &vop_putpages_desc, spec_putpages }, /* genfs_putpages */
201 { &vop_getextattr_desc, puffs_vnop_checkop }, /* getextattr */
202 { &vop_setextattr_desc, puffs_vnop_checkop }, /* setextattr */
203 { &vop_listextattr_desc, puffs_vnop_checkop }, /* listextattr */
204 { &vop_deleteextattr_desc, puffs_vnop_checkop },/* deleteextattr */
205 #if 0
206 { &vop_openextattr_desc, _openextattr }, /* openextattr */
207 { &vop_closeextattr_desc, _closeextattr }, /* closeextattr */
208 #endif
209 { NULL, NULL }
210 };
211 const struct vnodeopv_desc puffs_specop_opv_desc =
212 { &puffs_specop_p, puffs_specop_entries };
213
214
215 int (**puffs_fifoop_p)(void *);
216 const struct vnodeopv_entry_desc puffs_fifoop_entries[] = {
217 { &vop_default_desc, vn_default_error },
218 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup, ENOTDIR */
219 { &vop_create_desc, vn_fifo_bypass }, /* genfs_badop */
220 { &vop_mknod_desc, vn_fifo_bypass }, /* genfs_badop */
221 { &vop_open_desc, vn_fifo_bypass }, /* open */
222 { &vop_close_desc, vn_fifo_bypass }, /* close */
223 { &vop_access_desc, puffs_vnop_checkop }, /* access */
224 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */
225 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */
226 { &vop_read_desc, puffs_vnop_fifo_read }, /* read, update */
227 { &vop_write_desc, puffs_vnop_fifo_write }, /* write, update */
228 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */
229 { &vop_fcntl_desc, genfs_fcntl }, /* dummy */
230 { &vop_poll_desc, vn_fifo_bypass }, /* poll */
231 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */
232 { &vop_revoke_desc, vn_fifo_bypass }, /* genfs_revoke */
233 { &vop_mmap_desc, vn_fifo_bypass }, /* genfs_badop */
234 { &vop_fsync_desc, vn_fifo_bypass }, /* genfs_nullop*/
235 { &vop_seek_desc, vn_fifo_bypass }, /* genfs_badop */
236 { &vop_remove_desc, vn_fifo_bypass }, /* genfs_badop */
237 { &vop_link_desc, vn_fifo_bypass }, /* genfs_badop */
238 { &vop_rename_desc, vn_fifo_bypass }, /* genfs_badop */
239 { &vop_mkdir_desc, vn_fifo_bypass }, /* genfs_badop */
240 { &vop_rmdir_desc, vn_fifo_bypass }, /* genfs_badop */
241 { &vop_symlink_desc, vn_fifo_bypass }, /* genfs_badop */
242 { &vop_readdir_desc, vn_fifo_bypass }, /* genfs_badop */
243 { &vop_readlink_desc, vn_fifo_bypass }, /* genfs_badop */
244 { &vop_abortop_desc, vn_fifo_bypass }, /* genfs_badop */
245 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
246 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
247 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
248 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
249 { &vop_bmap_desc, vn_fifo_bypass }, /* dummy */
250 { &vop_strategy_desc, vn_fifo_bypass }, /* genfs_badop */
251 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
252 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
253 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */
254 { &vop_advlock_desc, vn_fifo_bypass }, /* genfs_einval */
255 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
256 { &vop_putpages_desc, vn_fifo_bypass }, /* genfs_null_putpages*/
257 #if 0
258 { &vop_openextattr_desc, _openextattr }, /* openextattr */
259 { &vop_closeextattr_desc, _closeextattr }, /* closeextattr */
260 #endif
261 { &vop_getextattr_desc, puffs_vnop_checkop }, /* getextattr */
262 { &vop_setextattr_desc, puffs_vnop_checkop }, /* setextattr */
263 { &vop_listextattr_desc, puffs_vnop_checkop }, /* listextattr */
264 { &vop_deleteextattr_desc, puffs_vnop_checkop }, /* deleteextattr */
265 { NULL, NULL }
266 };
267 const struct vnodeopv_desc puffs_fifoop_opv_desc =
268 { &puffs_fifoop_p, puffs_fifoop_entries };
269
270
271 /* "real" vnode operations */
272 int (**puffs_msgop_p)(void *);
273 const struct vnodeopv_entry_desc puffs_msgop_entries[] = {
274 { &vop_default_desc, vn_default_error },
275 { &vop_create_desc, puffs_vnop_create }, /* create */
276 { &vop_mknod_desc, puffs_vnop_mknod }, /* mknod */
277 { &vop_open_desc, puffs_vnop_open }, /* open */
278 { &vop_close_desc, puffs_vnop_close }, /* close */
279 { &vop_access_desc, puffs_vnop_access }, /* access */
280 { &vop_getattr_desc, puffs_vnop_getattr }, /* getattr */
281 { &vop_setattr_desc, puffs_vnop_setattr }, /* setattr */
282 { &vop_read_desc, puffs_vnop_read }, /* read */
283 { &vop_write_desc, puffs_vnop_write }, /* write */
284 { &vop_seek_desc, puffs_vnop_seek }, /* seek */
285 { &vop_remove_desc, puffs_vnop_remove }, /* remove */
286 { &vop_link_desc, puffs_vnop_link }, /* link */
287 { &vop_rename_desc, puffs_vnop_rename }, /* rename */
288 { &vop_mkdir_desc, puffs_vnop_mkdir }, /* mkdir */
289 { &vop_rmdir_desc, puffs_vnop_rmdir }, /* rmdir */
290 { &vop_symlink_desc, puffs_vnop_symlink }, /* symlink */
291 { &vop_readdir_desc, puffs_vnop_readdir }, /* readdir */
292 { &vop_readlink_desc, puffs_vnop_readlink }, /* readlink */
293 { &vop_print_desc, puffs_vnop_print }, /* print */
294 { &vop_islocked_desc, puffs_vnop_islocked }, /* islocked */
295 { &vop_pathconf_desc, puffs_vnop_pathconf }, /* pathconf */
296 { &vop_getpages_desc, puffs_vnop_getpages }, /* getpages */
297 { NULL, NULL }
298 };
299 const struct vnodeopv_desc puffs_msgop_opv_desc =
300 { &puffs_msgop_p, puffs_msgop_entries };
301
302 /*
303 * for dosetattr / update_va
304 */
305 #define SETATTR_CHSIZE 0x01
306 #define SETATTR_ASYNC 0x02
307
308 #define ERROUT(err) \
309 do { \
310 error = err; \
311 goto out; \
312 } while (/*CONSTCOND*/0)
313
314 /*
315 * This is a generic vnode operation handler. It checks if the necessary
316 * operations for the called vnode operation are implemented by userspace
317 * and either returns a dummy return value or proceeds to call the real
318 * vnode operation from puffs_msgop_v.
319 *
320 * XXX: this should described elsewhere and autogenerated, the complexity
321 * of the vnode operations vectors and their interrelationships is also
322 * getting a bit out of hand. Another problem is that we need this same
323 * information in the fs server code, so keeping the two in sync manually
324 * is not a viable (long term) plan.
325 */
326
327 /* not supported, handle locking protocol */
328 #define CHECKOP_NOTSUPP(op) \
329 case VOP_##op##_DESCOFFSET: \
330 if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0) \
331 return genfs_eopnotsupp(v); \
332 break
333
334 /* always succeed, no locking */
335 #define CHECKOP_SUCCESS(op) \
336 case VOP_##op##_DESCOFFSET: \
337 if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0) \
338 return 0; \
339 break
340
341 int
342 puffs_vnop_checkop(void *v)
343 {
344 struct vop_generic_args /* {
345 struct vnodeop_desc *a_desc;
346 spooky mystery contents;
347 } */ *ap = v;
348 struct vnodeop_desc *desc = ap->a_desc;
349 struct puffs_mount *pmp;
350 struct vnode *vp;
351 int offset, rv;
352
353 offset = ap->a_desc->vdesc_vp_offsets[0];
354 #ifdef DIAGNOSTIC
355 if (offset == VDESC_NO_OFFSET)
356 panic("puffs_checkop: no vnode, why did you call me?");
357 #endif
358 vp = *VOPARG_OFFSETTO(struct vnode **, offset, ap);
359 pmp = MPTOPUFFSMP(vp->v_mount);
360
361 DPRINTF_VERBOSE(("checkop call %s (%d), vp %p\n",
362 ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp));
363
364 if (!ALLOPS(pmp)) {
365 switch (desc->vdesc_offset) {
366 CHECKOP_NOTSUPP(CREATE);
367 CHECKOP_NOTSUPP(MKNOD);
368 CHECKOP_NOTSUPP(GETATTR);
369 CHECKOP_NOTSUPP(SETATTR);
370 CHECKOP_NOTSUPP(READ);
371 CHECKOP_NOTSUPP(WRITE);
372 CHECKOP_NOTSUPP(FCNTL);
373 CHECKOP_NOTSUPP(IOCTL);
374 CHECKOP_NOTSUPP(REMOVE);
375 CHECKOP_NOTSUPP(LINK);
376 CHECKOP_NOTSUPP(RENAME);
377 CHECKOP_NOTSUPP(MKDIR);
378 CHECKOP_NOTSUPP(RMDIR);
379 CHECKOP_NOTSUPP(SYMLINK);
380 CHECKOP_NOTSUPP(READDIR);
381 CHECKOP_NOTSUPP(READLINK);
382 CHECKOP_NOTSUPP(PRINT);
383 CHECKOP_NOTSUPP(PATHCONF);
384 CHECKOP_NOTSUPP(GETEXTATTR);
385 CHECKOP_NOTSUPP(SETEXTATTR);
386 CHECKOP_NOTSUPP(LISTEXTATTR);
387 CHECKOP_NOTSUPP(DELETEEXTATTR);
388
389 CHECKOP_SUCCESS(ACCESS);
390 CHECKOP_SUCCESS(CLOSE);
391 CHECKOP_SUCCESS(SEEK);
392
393 case VOP_GETPAGES_DESCOFFSET:
394 if (!EXISTSOP(pmp, READ))
395 return genfs_eopnotsupp(v);
396 break;
397
398 default:
399 panic("puffs_checkop: unhandled vnop %d",
400 desc->vdesc_offset);
401 }
402 }
403
404 rv = VOCALL(puffs_msgop_p, ap->a_desc->vdesc_offset, v);
405
406 DPRINTF_VERBOSE(("checkop return %s (%d), vp %p: %d\n",
407 ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp, rv));
408
409 return rv;
410 }
411
412 static int callremove(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
413 struct componentname *);
414 static int callrmdir(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
415 struct componentname *);
416 static void callinactive(struct puffs_mount *, puffs_cookie_t, int);
417 static void callreclaim(struct puffs_mount *, puffs_cookie_t, int);
418 static int flushvncache(struct vnode *, off_t, off_t, bool);
419 static void update_va(struct vnode *, struct vattr *, struct vattr *,
420 struct timespec *, struct timespec *, int);
421 static void update_parent(struct vnode *, struct vnode *);
422
423
424 #define PUFFS_ABORT_LOOKUP 1
425 #define PUFFS_ABORT_CREATE 2
426 #define PUFFS_ABORT_MKNOD 3
427 #define PUFFS_ABORT_MKDIR 4
428 #define PUFFS_ABORT_SYMLINK 5
429
430 /*
431 * Press the pani^Wabort button! Kernel resource allocation failed.
432 */
433 static void
434 puffs_abortbutton(struct puffs_mount *pmp, int what,
435 puffs_cookie_t dck, puffs_cookie_t ck, struct componentname *cnp)
436 {
437
438 switch (what) {
439 case PUFFS_ABORT_CREATE:
440 case PUFFS_ABORT_MKNOD:
441 case PUFFS_ABORT_SYMLINK:
442 callremove(pmp, dck, ck, cnp);
443 break;
444 case PUFFS_ABORT_MKDIR:
445 callrmdir(pmp, dck, ck, cnp);
446 break;
447 }
448
449 callinactive(pmp, ck, 0);
450 callreclaim(pmp, ck, 1);
451 }
452
453 /*
454 * Begin vnode operations.
455 *
456 * A word from the keymaster about locks: generally we don't want
457 * to use the vnode locks at all: it creates an ugly dependency between
458 * the userlandia file server and the kernel. But we'll play along with
459 * the kernel vnode locks for now. However, even currently we attempt
460 * to release locks as early as possible. This is possible for some
461 * operations which a) don't need a locked vnode after the userspace op
462 * and b) return with the vnode unlocked. Theoretically we could
463 * unlock-do op-lock for others and order the graph in userspace, but I
464 * don't want to think of the consequences for the time being.
465 */
466
467 #define TTL_TO_TIMEOUT(ts) \
468 (hardclock_ticks + (ts->tv_sec * hz) + (ts->tv_nsec * hz / 1000000000))
469 #define TTL_VALID(ts) \
470 ((ts != NULL) && !((ts->tv_sec == 0) && (ts->tv_nsec == 0)))
471 #define TIMED_OUT(expire) \
472 ((int)((unsigned int)hardclock_ticks - (unsigned int)expire) > 0)
473 int
474 puffs_vnop_lookup(void *v)
475 {
476 struct vop_lookup_v2_args /* {
477 const struct vnodeop_desc *a_desc;
478 struct vnode *a_dvp;
479 struct vnode **a_vpp;
480 struct componentname *a_cnp;
481 } */ *ap = v;
482 PUFFS_MSG_VARS(vn, lookup);
483 struct puffs_mount *pmp;
484 struct componentname *cnp;
485 struct vnode *vp, *dvp, *cvp;
486 struct puffs_node *dpn, *cpn;
487 int isdot;
488 int error;
489
490 pmp = MPTOPUFFSMP(ap->a_dvp->v_mount);
491 cnp = ap->a_cnp;
492 dvp = ap->a_dvp;
493 cvp = NULL;
494 cpn = NULL;
495 *ap->a_vpp = NULL;
496
497 /* r/o fs? we check create later to handle EEXIST */
498 if ((cnp->cn_flags & ISLASTCN)
499 && (dvp->v_mount->mnt_flag & MNT_RDONLY)
500 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
501 return EROFS;
502
503 isdot = cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.';
504
505 DPRINTF(("puffs_lookup: \"%s\", parent vnode %p, op: %x\n",
506 cnp->cn_nameptr, dvp, cnp->cn_nameiop));
507
508 /*
509 * If dotdot cache is enabled, add reference to .. and return.
510 */
511 if (PUFFS_USE_DOTDOTCACHE(pmp) && (cnp->cn_flags & ISDOTDOT)) {
512 vp = VPTOPP(ap->a_dvp)->pn_parent;
513 vref(vp);
514
515 *ap->a_vpp = vp;
516 return 0;
517 }
518
519 /*
520 * Check if someone fed it into the cache
521 */
522 if (!isdot && PUFFS_USE_NAMECACHE(pmp)) {
523 int found, iswhiteout;
524
525 found = cache_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
526 cnp->cn_nameiop, cnp->cn_flags,
527 &iswhiteout, ap->a_vpp);
528 if (iswhiteout) {
529 cnp->cn_flags |= ISWHITEOUT;
530 }
531
532 if (found && *ap->a_vpp != NULLVP && PUFFS_USE_FS_TTL(pmp)) {
533 cvp = *ap->a_vpp;
534 cpn = VPTOPP(cvp);
535
536 if (TIMED_OUT(cpn->pn_cn_timeout)) {
537 cache_purge(cvp);
538 /*
539 * cached vnode (cvp) is still referenced
540 * so that we can reuse it upon a new
541 * successful lookup.
542 */
543 *ap->a_vpp = NULL;
544 found = 0;
545 }
546 }
547
548 /*
549 * Do not use negative caching, since the filesystem
550 * provides no TTL for it.
551 */
552 if (found && *ap->a_vpp == NULLVP && PUFFS_USE_FS_TTL(pmp))
553 found = 0;
554
555 if (found) {
556 return *ap->a_vpp == NULLVP ? ENOENT : 0;
557 }
558
559 /*
560 * This is what would have been left in ERROR before
561 * the rearrangement of cache_lookup(). What with all
562 * the macros, I am not sure if this is a dead value
563 * below or not.
564 */
565 error = -1;
566 }
567
568 if (isdot) {
569 /* deal with rename lookup semantics */
570 if (cnp->cn_nameiop == RENAME && (cnp->cn_flags & ISLASTCN))
571 return EISDIR;
572
573 vp = ap->a_dvp;
574 vref(vp);
575 *ap->a_vpp = vp;
576 return 0;
577 }
578
579 if (cvp != NULL) {
580 if (vn_lock(cvp, LK_EXCLUSIVE) != 0) {
581 vrele(cvp);
582 cvp = NULL;
583 } else
584 mutex_enter(&cpn->pn_sizemtx);
585 }
586
587 PUFFS_MSG_ALLOC(vn, lookup);
588 puffs_makecn(&lookup_msg->pvnr_cn, &lookup_msg->pvnr_cn_cred,
589 cnp, PUFFS_USE_FULLPNBUF(pmp));
590
591 if (cnp->cn_flags & ISDOTDOT)
592 VOP_UNLOCK(dvp);
593
594 puffs_msg_setinfo(park_lookup, PUFFSOP_VN,
595 PUFFS_VN_LOOKUP, VPTOPNC(dvp));
596 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_lookup, dvp->v_data, NULL, error);
597 DPRINTF(("puffs_lookup: return of the userspace, part %d\n", error));
598
599 /*
600 * In case of error, there is no new vnode to play with, so be
601 * happy with the NULL value given to vpp in the beginning.
602 * Also, check if this really was an error or the target was not
603 * present. Either treat it as a non-error for CREATE/RENAME or
604 * enter the component into the negative name cache (if desired).
605 */
606 if (error) {
607 error = checkerr(pmp, error, __func__);
608 if (error == ENOENT) {
609 /* don't allow to create files on r/o fs */
610 if ((dvp->v_mount->mnt_flag & MNT_RDONLY)
611 && cnp->cn_nameiop == CREATE) {
612 error = EROFS;
613
614 /* adjust values if we are creating */
615 } else if ((cnp->cn_flags & ISLASTCN)
616 && (cnp->cn_nameiop == CREATE
617 || cnp->cn_nameiop == RENAME)) {
618 error = EJUSTRETURN;
619
620 /* save negative cache entry */
621 } else {
622 if (PUFFS_USE_NAMECACHE(pmp) &&
623 !PUFFS_USE_FS_TTL(pmp))
624 cache_enter(dvp, NULL, cnp->cn_nameptr,
625 cnp->cn_namelen, cnp->cn_flags);
626 }
627 }
628 goto out;
629 }
630
631 /*
632 * Check that we don't get our parent node back, that would cause
633 * a pretty obvious deadlock.
634 */
635 dpn = dvp->v_data;
636 if (lookup_msg->pvnr_newnode == dpn->pn_cookie) {
637 puffs_senderr(pmp, PUFFS_ERR_LOOKUP, EINVAL,
638 "lookup produced parent cookie", lookup_msg->pvnr_newnode);
639 error = EPROTO;
640 goto out;
641 }
642
643 /*
644 * Check if we looked up the cached vnode
645 */
646 vp = NULL;
647 if (cvp && (VPTOPP(cvp)->pn_cookie == lookup_msg->pvnr_newnode)) {
648 int grace;
649
650 /*
651 * Bump grace time of this node so that it does not get
652 * reclaimed too fast. We try to increase a bit more the
653 * lifetime of busiest * nodes - with some limits.
654 */
655 grace = 10 * puffs_sopreq_expire_timeout;
656 cpn->pn_cn_grace = hardclock_ticks + grace;
657 vp = cvp;
658 }
659
660 /*
661 * No cached vnode available, or the cached vnode does not
662 * match the userland cookie anymore: is the node known?
663 */
664 if (vp == NULL) {
665 error = puffs_cookie2vnode(pmp, lookup_msg->pvnr_newnode,
666 1, 1, &vp);
667 }
668
669 if (error == PUFFS_NOSUCHCOOKIE) {
670 error = puffs_getvnode(dvp->v_mount,
671 lookup_msg->pvnr_newnode, lookup_msg->pvnr_vtype,
672 lookup_msg->pvnr_size, lookup_msg->pvnr_rdev, &vp);
673 if (error) {
674 puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP,
675 VPTOPNC(dvp), lookup_msg->pvnr_newnode,
676 ap->a_cnp);
677 goto out;
678 }
679
680 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
681 } else if (error) {
682 puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP, VPTOPNC(dvp),
683 lookup_msg->pvnr_newnode, ap->a_cnp);
684 goto out;
685 }
686
687 /*
688 * Update cache and TTL
689 */
690 if (PUFFS_USE_FS_TTL(pmp)) {
691 struct timespec *va_ttl = &lookup_msg->pvnr_va_ttl;
692 struct timespec *cn_ttl = &lookup_msg->pvnr_cn_ttl;
693 update_va(vp, NULL, &lookup_msg->pvnr_va,
694 va_ttl, cn_ttl, SETATTR_CHSIZE);
695 }
696
697 KASSERT(lookup_msg->pvnr_newnode == VPTOPP(vp)->pn_cookie);
698 *ap->a_vpp = vp;
699
700 if (PUFFS_USE_NAMECACHE(pmp))
701 cache_enter(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
702 cnp->cn_flags);
703
704 /* XXX */
705 if ((lookup_msg->pvnr_cn.pkcn_flags & REQUIREDIR) == 0)
706 cnp->cn_flags &= ~REQUIREDIR;
707 if (lookup_msg->pvnr_cn.pkcn_consume)
708 cnp->cn_consume = MIN(lookup_msg->pvnr_cn.pkcn_consume,
709 strlen(cnp->cn_nameptr) - cnp->cn_namelen);
710
711 VPTOPP(vp)->pn_nlookup++;
712
713 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
714 (VPTOPP(vp)->pn_parent != dvp))
715 update_parent(vp, dvp);
716
717 out:
718 if (cvp != NULL) {
719 mutex_exit(&cpn->pn_sizemtx);
720
721 if (error || (cvp != vp))
722 vput(cvp);
723 }
724 if (error == 0)
725 VOP_UNLOCK(*ap->a_vpp);
726
727 if (cnp->cn_flags & ISDOTDOT)
728 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
729
730 DPRINTF(("puffs_lookup: returning %d %p\n", error, *ap->a_vpp));
731 PUFFS_MSG_RELEASE(lookup);
732 return error;
733 }
734
735 #define REFPN_AND_UNLOCKVP(a, b) \
736 do { \
737 mutex_enter(&b->pn_mtx); \
738 puffs_referencenode(b); \
739 mutex_exit(&b->pn_mtx); \
740 VOP_UNLOCK(a); \
741 } while (/*CONSTCOND*/0)
742
743 #define REFPN(b) \
744 do { \
745 mutex_enter(&b->pn_mtx); \
746 puffs_referencenode(b); \
747 mutex_exit(&b->pn_mtx); \
748 } while (/*CONSTCOND*/0)
749
750 #define RELEPN_AND_VP(a, b) \
751 do { \
752 puffs_releasenode(b); \
753 vrele(a); \
754 } while (/*CONSTCOND*/0)
755
756 int
757 puffs_vnop_create(void *v)
758 {
759 struct vop_create_v3_args /* {
760 const struct vnodeop_desc *a_desc;
761 struct vnode *a_dvp;
762 struct vnode **a_vpp;
763 struct componentname *a_cnp;
764 struct vattr *a_vap;
765 } */ *ap = v;
766 PUFFS_MSG_VARS(vn, create);
767 struct vnode *dvp = ap->a_dvp;
768 struct puffs_node *dpn = VPTOPP(dvp);
769 struct componentname *cnp = ap->a_cnp;
770 struct mount *mp = dvp->v_mount;
771 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
772 int error;
773
774 DPRINTF(("puffs_create: dvp %p, cnp: %s\n",
775 dvp, ap->a_cnp->cn_nameptr));
776
777 PUFFS_MSG_ALLOC(vn, create);
778 puffs_makecn(&create_msg->pvnr_cn, &create_msg->pvnr_cn_cred,
779 cnp, PUFFS_USE_FULLPNBUF(pmp));
780 create_msg->pvnr_va = *ap->a_vap;
781 puffs_msg_setinfo(park_create, PUFFSOP_VN,
782 PUFFS_VN_CREATE, VPTOPNC(dvp));
783 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_create, dvp->v_data, NULL, error);
784
785 error = checkerr(pmp, error, __func__);
786 if (error)
787 goto out;
788
789 error = puffs_newnode(mp, dvp, ap->a_vpp,
790 create_msg->pvnr_newnode, cnp, ap->a_vap->va_type, 0);
791 if (error) {
792 puffs_abortbutton(pmp, PUFFS_ABORT_CREATE, dpn->pn_cookie,
793 create_msg->pvnr_newnode, cnp);
794 goto out;
795 }
796
797 if (PUFFS_USE_FS_TTL(pmp)) {
798 struct timespec *va_ttl = &create_msg->pvnr_va_ttl;
799 struct timespec *cn_ttl = &create_msg->pvnr_cn_ttl;
800 struct vattr *rvap = &create_msg->pvnr_va;
801
802 update_va(*ap->a_vpp, NULL, rvap,
803 va_ttl, cn_ttl, SETATTR_CHSIZE);
804 }
805
806 VPTOPP(*ap->a_vpp)->pn_nlookup++;
807
808 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
809 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
810 update_parent(*ap->a_vpp, dvp);
811
812 out:
813 DPRINTF(("puffs_create: return %d\n", error));
814 PUFFS_MSG_RELEASE(create);
815 return error;
816 }
817
818 int
819 puffs_vnop_mknod(void *v)
820 {
821 struct vop_mknod_v3_args /* {
822 const struct vnodeop_desc *a_desc;
823 struct vnode *a_dvp;
824 struct vnode **a_vpp;
825 struct componentname *a_cnp;
826 struct vattr *a_vap;
827 } */ *ap = v;
828 PUFFS_MSG_VARS(vn, mknod);
829 struct vnode *dvp = ap->a_dvp;
830 struct puffs_node *dpn = VPTOPP(dvp);
831 struct componentname *cnp = ap->a_cnp;
832 struct mount *mp = dvp->v_mount;
833 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
834 int error;
835
836 PUFFS_MSG_ALLOC(vn, mknod);
837 puffs_makecn(&mknod_msg->pvnr_cn, &mknod_msg->pvnr_cn_cred,
838 cnp, PUFFS_USE_FULLPNBUF(pmp));
839 mknod_msg->pvnr_va = *ap->a_vap;
840 puffs_msg_setinfo(park_mknod, PUFFSOP_VN,
841 PUFFS_VN_MKNOD, VPTOPNC(dvp));
842
843 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mknod, dvp->v_data, NULL, error);
844
845 error = checkerr(pmp, error, __func__);
846 if (error)
847 goto out;
848
849 error = puffs_newnode(mp, dvp, ap->a_vpp,
850 mknod_msg->pvnr_newnode, cnp, ap->a_vap->va_type,
851 ap->a_vap->va_rdev);
852 if (error) {
853 puffs_abortbutton(pmp, PUFFS_ABORT_MKNOD, dpn->pn_cookie,
854 mknod_msg->pvnr_newnode, cnp);
855 goto out;
856 }
857
858 if (PUFFS_USE_FS_TTL(pmp)) {
859 struct timespec *va_ttl = &mknod_msg->pvnr_va_ttl;
860 struct timespec *cn_ttl = &mknod_msg->pvnr_cn_ttl;
861 struct vattr *rvap = &mknod_msg->pvnr_va;
862
863 update_va(*ap->a_vpp, NULL, rvap,
864 va_ttl, cn_ttl, SETATTR_CHSIZE);
865 }
866
867 VPTOPP(*ap->a_vpp)->pn_nlookup++;
868
869 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
870 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
871 update_parent(*ap->a_vpp, dvp);
872
873 out:
874 PUFFS_MSG_RELEASE(mknod);
875 return error;
876 }
877
878 int
879 puffs_vnop_open(void *v)
880 {
881 struct vop_open_args /* {
882 const struct vnodeop_desc *a_desc;
883 struct vnode *a_vp;
884 int a_mode;
885 kauth_cred_t a_cred;
886 } */ *ap = v;
887 PUFFS_MSG_VARS(vn, open);
888 struct vnode *vp = ap->a_vp;
889 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
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 out:
911 DPRINTF(("puffs_open: returning %d\n", error));
912 PUFFS_MSG_RELEASE(open);
913 return error;
914 }
915
916 int
917 puffs_vnop_close(void *v)
918 {
919 struct vop_close_args /* {
920 const struct vnodeop_desc *a_desc;
921 struct vnode *a_vp;
922 int a_fflag;
923 kauth_cred_t a_cred;
924 } */ *ap = v;
925 PUFFS_MSG_VARS(vn, close);
926 struct vnode *vp = ap->a_vp;
927 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
928
929 PUFFS_MSG_ALLOC(vn, close);
930 puffs_msg_setfaf(park_close);
931 close_msg->pvnr_fflag = ap->a_fflag;
932 puffs_credcvt(&close_msg->pvnr_cred, ap->a_cred);
933 puffs_msg_setinfo(park_close, PUFFSOP_VN,
934 PUFFS_VN_CLOSE, VPTOPNC(vp));
935
936 puffs_msg_enqueue(pmp, park_close);
937 PUFFS_MSG_RELEASE(close);
938 return 0;
939 }
940
941 int
942 puffs_vnop_access(void *v)
943 {
944 struct vop_access_args /* {
945 const struct vnodeop_desc *a_desc;
946 struct vnode *a_vp;
947 int a_mode;
948 kauth_cred_t a_cred;
949 } */ *ap = v;
950 PUFFS_MSG_VARS(vn, access);
951 struct vnode *vp = ap->a_vp;
952 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
953 int mode = ap->a_mode;
954 int error;
955
956 if (mode & VWRITE) {
957 switch (vp->v_type) {
958 case VDIR:
959 case VLNK:
960 case VREG:
961 if ((vp->v_mount->mnt_flag & MNT_RDONLY)
962 || !EXISTSOP(pmp, WRITE))
963 return EROFS;
964 break;
965 default:
966 break;
967 }
968 }
969
970 if (!EXISTSOP(pmp, ACCESS))
971 return 0;
972
973 PUFFS_MSG_ALLOC(vn, access);
974 access_msg->pvnr_mode = ap->a_mode;
975 puffs_credcvt(&access_msg->pvnr_cred, ap->a_cred);
976 puffs_msg_setinfo(park_access, PUFFSOP_VN,
977 PUFFS_VN_ACCESS, VPTOPNC(vp));
978
979 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_access, vp->v_data, NULL, error);
980 error = checkerr(pmp, error, __func__);
981 PUFFS_MSG_RELEASE(access);
982
983 return error;
984 }
985
986 static void
987 update_va(struct vnode *vp, struct vattr *vap, struct vattr *rvap,
988 struct timespec *va_ttl, struct timespec *cn_ttl, int flags)
989 {
990 struct puffs_node *pn = VPTOPP(vp);
991
992 if (TTL_VALID(cn_ttl)) {
993 pn->pn_cn_timeout = TTL_TO_TIMEOUT(cn_ttl);
994 pn->pn_cn_grace = MAX(pn->pn_cn_timeout, pn->pn_cn_grace);
995 }
996
997 /*
998 * Don't listen to the file server regarding special device
999 * size info, the file server doesn't know anything about them.
1000 */
1001 if (vp->v_type == VBLK || vp->v_type == VCHR)
1002 rvap->va_size = vp->v_size;
1003
1004 /* Ditto for blocksize (ufs comment: this doesn't belong here) */
1005 if (vp->v_type == VBLK)
1006 rvap->va_blocksize = BLKDEV_IOSIZE;
1007 else if (vp->v_type == VCHR)
1008 rvap->va_blocksize = MAXBSIZE;
1009
1010 if (vap != NULL) {
1011 (void) memcpy(vap, rvap, sizeof(struct vattr));
1012 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1013
1014 if (pn->pn_stat & PNODE_METACACHE_ATIME)
1015 vap->va_atime = pn->pn_mc_atime;
1016 if (pn->pn_stat & PNODE_METACACHE_CTIME)
1017 vap->va_ctime = pn->pn_mc_ctime;
1018 if (pn->pn_stat & PNODE_METACACHE_MTIME)
1019 vap->va_mtime = pn->pn_mc_mtime;
1020 if (pn->pn_stat & PNODE_METACACHE_SIZE)
1021 vap->va_size = pn->pn_mc_size;
1022 }
1023
1024 if (!(pn->pn_stat & PNODE_METACACHE_SIZE) && (flags & SETATTR_CHSIZE)) {
1025 if (rvap->va_size != VNOVAL
1026 && vp->v_type != VBLK && vp->v_type != VCHR) {
1027 uvm_vnp_setsize(vp, rvap->va_size);
1028 pn->pn_serversize = rvap->va_size;
1029 }
1030 }
1031
1032 if ((va_ttl != NULL) && TTL_VALID(va_ttl)) {
1033 if (pn->pn_va_cache == NULL)
1034 pn->pn_va_cache = pool_get(&puffs_vapool, PR_WAITOK);
1035
1036 (void)memcpy(pn->pn_va_cache, rvap, sizeof(*rvap));
1037
1038 pn->pn_va_timeout = TTL_TO_TIMEOUT(va_ttl);
1039 }
1040 }
1041
1042 static void
1043 update_parent(struct vnode *vp, struct vnode *dvp)
1044 {
1045 struct puffs_node *pn = VPTOPP(vp);
1046
1047 if (pn->pn_parent != NULL) {
1048 KASSERT(pn->pn_parent != dvp);
1049 vrele(pn->pn_parent);
1050 }
1051
1052 vref(dvp);
1053 pn->pn_parent = dvp;
1054 }
1055
1056 int
1057 puffs_vnop_getattr(void *v)
1058 {
1059 struct vop_getattr_args /* {
1060 const struct vnodeop_desc *a_desc;
1061 struct vnode *a_vp;
1062 struct vattr *a_vap;
1063 kauth_cred_t a_cred;
1064 } */ *ap = v;
1065 PUFFS_MSG_VARS(vn, getattr);
1066 struct vnode *vp = ap->a_vp;
1067 struct mount *mp = vp->v_mount;
1068 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1069 struct vattr *vap, *rvap;
1070 struct puffs_node *pn = VPTOPP(vp);
1071 struct timespec *va_ttl = NULL;
1072 int error = 0;
1073
1074 /*
1075 * A lock is required so that we do not race with
1076 * setattr, write and fsync when changing vp->v_size.
1077 * This is critical, since setting a stall smaler value
1078 * triggers a file truncate in uvm_vnp_setsize(), which
1079 * most of the time means data corruption (a chunk of
1080 * data is replaced by zeroes). This can be removed if
1081 * we decide one day that VOP_GETATTR must operate on
1082 * a locked vnode.
1083 *
1084 * XXX Should be useless now that VOP_GETATTR has been
1085 * fixed to always require a shared lock at least.
1086 */
1087 mutex_enter(&pn->pn_sizemtx);
1088
1089 REFPN(pn);
1090 vap = ap->a_vap;
1091
1092 if (PUFFS_USE_FS_TTL(pmp)) {
1093 if (!TIMED_OUT(pn->pn_va_timeout)) {
1094 update_va(vp, vap, pn->pn_va_cache,
1095 NULL, NULL, SETATTR_CHSIZE);
1096 goto out2;
1097 }
1098 }
1099
1100 PUFFS_MSG_ALLOC(vn, getattr);
1101 vattr_null(&getattr_msg->pvnr_va);
1102 puffs_credcvt(&getattr_msg->pvnr_cred, ap->a_cred);
1103 puffs_msg_setinfo(park_getattr, PUFFSOP_VN,
1104 PUFFS_VN_GETATTR, VPTOPNC(vp));
1105
1106 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getattr, vp->v_data, NULL, error);
1107 error = checkerr(pmp, error, __func__);
1108 if (error)
1109 goto out;
1110
1111 rvap = &getattr_msg->pvnr_va;
1112
1113 if (PUFFS_USE_FS_TTL(pmp))
1114 va_ttl = &getattr_msg->pvnr_va_ttl;
1115
1116 update_va(vp, vap, rvap, va_ttl, NULL, SETATTR_CHSIZE);
1117
1118 out:
1119 PUFFS_MSG_RELEASE(getattr);
1120
1121 out2:
1122 puffs_releasenode(pn);
1123
1124 mutex_exit(&pn->pn_sizemtx);
1125
1126 return error;
1127 }
1128
1129 static int
1130 dosetattr(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, int flags)
1131 {
1132 PUFFS_MSG_VARS(vn, setattr);
1133 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1134 struct puffs_node *pn = vp->v_data;
1135 int error = 0;
1136
1137 KASSERT(!(flags & SETATTR_CHSIZE) || mutex_owned(&pn->pn_sizemtx));
1138
1139 if ((vp->v_mount->mnt_flag & MNT_RDONLY) &&
1140 (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL
1141 || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL
1142 || vap->va_mode != (mode_t)VNOVAL))
1143 return EROFS;
1144
1145 if ((vp->v_mount->mnt_flag & MNT_RDONLY)
1146 && vp->v_type == VREG && vap->va_size != VNOVAL)
1147 return EROFS;
1148
1149 /*
1150 * Flush metacache first. If we are called with some explicit
1151 * parameters, treat them as information overriding metacache
1152 * information.
1153 */
1154 if (pn->pn_stat & PNODE_METACACHE_MASK) {
1155 if ((pn->pn_stat & PNODE_METACACHE_ATIME)
1156 && vap->va_atime.tv_sec == VNOVAL)
1157 vap->va_atime = pn->pn_mc_atime;
1158 if ((pn->pn_stat & PNODE_METACACHE_CTIME)
1159 && vap->va_ctime.tv_sec == VNOVAL)
1160 vap->va_ctime = pn->pn_mc_ctime;
1161 if ((pn->pn_stat & PNODE_METACACHE_MTIME)
1162 && vap->va_mtime.tv_sec == VNOVAL)
1163 vap->va_mtime = pn->pn_mc_mtime;
1164 if ((pn->pn_stat & PNODE_METACACHE_SIZE)
1165 && vap->va_size == VNOVAL)
1166 vap->va_size = pn->pn_mc_size;
1167
1168 pn->pn_stat &= ~PNODE_METACACHE_MASK;
1169 }
1170
1171 /*
1172 * Flush attribute cache so that another thread do
1173 * not get a stale value during the operation.
1174 */
1175 if (PUFFS_USE_FS_TTL(pmp))
1176 pn->pn_va_timeout = 0;
1177
1178 PUFFS_MSG_ALLOC(vn, setattr);
1179 (void)memcpy(&setattr_msg->pvnr_va, vap, sizeof(struct vattr));
1180 puffs_credcvt(&setattr_msg->pvnr_cred, cred);
1181 puffs_msg_setinfo(park_setattr, PUFFSOP_VN,
1182 PUFFS_VN_SETATTR, VPTOPNC(vp));
1183 if (flags & SETATTR_ASYNC)
1184 puffs_msg_setfaf(park_setattr);
1185
1186 puffs_msg_enqueue(pmp, park_setattr);
1187 if ((flags & SETATTR_ASYNC) == 0)
1188 error = puffs_msg_wait2(pmp, park_setattr, vp->v_data, NULL);
1189
1190 if ((error == 0) && PUFFS_USE_FS_TTL(pmp)) {
1191 struct timespec *va_ttl = &setattr_msg->pvnr_va_ttl;
1192 struct vattr *rvap = &setattr_msg->pvnr_va;
1193
1194 update_va(vp, NULL, rvap, va_ttl, NULL, flags);
1195 }
1196
1197 PUFFS_MSG_RELEASE(setattr);
1198 if ((flags & SETATTR_ASYNC) == 0) {
1199 error = checkerr(pmp, error, __func__);
1200 if (error)
1201 return error;
1202 } else {
1203 error = 0;
1204 }
1205
1206 if (vap->va_size != VNOVAL) {
1207 pn->pn_serversize = vap->va_size;
1208 if (flags & SETATTR_CHSIZE)
1209 uvm_vnp_setsize(vp, vap->va_size);
1210 }
1211
1212 return 0;
1213 }
1214
1215 int
1216 puffs_vnop_setattr(void *v)
1217 {
1218 struct vop_getattr_args /* {
1219 const struct vnodeop_desc *a_desc;
1220 struct vnode *a_vp;
1221 struct vattr *a_vap;
1222 kauth_cred_t a_cred;
1223 } */ *ap = v;
1224 struct puffs_node *pn = ap->a_vp->v_data;
1225 int error;
1226
1227 mutex_enter(&pn->pn_sizemtx);
1228 error = dosetattr(ap->a_vp, ap->a_vap, ap->a_cred, SETATTR_CHSIZE);
1229 mutex_exit(&pn->pn_sizemtx);
1230
1231 return error;
1232 }
1233
1234 static __inline int
1235 doinact(struct puffs_mount *pmp, int iaflag)
1236 {
1237
1238 if (EXISTSOP(pmp, INACTIVE))
1239 if (pmp->pmp_flags & PUFFS_KFLAG_IAONDEMAND)
1240 if (iaflag || ALLOPS(pmp))
1241 return 1;
1242 else
1243 return 0;
1244 else
1245 return 1;
1246 else
1247 return 0;
1248 }
1249
1250 static void
1251 callinactive(struct puffs_mount *pmp, puffs_cookie_t ck, int iaflag)
1252 {
1253 PUFFS_MSG_VARS(vn, inactive);
1254
1255 if (doinact(pmp, iaflag)) {
1256 PUFFS_MSG_ALLOC(vn, inactive);
1257 puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1258 PUFFS_VN_INACTIVE, ck);
1259 PUFFS_MSG_ENQUEUEWAIT_NOERROR(pmp, park_inactive);
1260 PUFFS_MSG_RELEASE(inactive);
1261 }
1262 }
1263
1264 /* XXX: callinactive can't setback */
1265 int
1266 puffs_vnop_inactive(void *v)
1267 {
1268 struct vop_inactive_args /* {
1269 const struct vnodeop_desc *a_desc;
1270 struct vnode *a_vp;
1271 } */ *ap = v;
1272 PUFFS_MSG_VARS(vn, inactive);
1273 struct vnode *vp = ap->a_vp;
1274 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1275 struct puffs_node *pnode;
1276 bool recycle = false;
1277
1278 pnode = vp->v_data;
1279 mutex_enter(&pnode->pn_sizemtx);
1280
1281 if (doinact(pmp, pnode->pn_stat & PNODE_DOINACT)) {
1282 flushvncache(vp, 0, 0, false);
1283 PUFFS_MSG_ALLOC(vn, inactive);
1284 puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1285 PUFFS_VN_INACTIVE, VPTOPNC(vp));
1286 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_inactive, vp->v_data,
1287 NULL);
1288 PUFFS_MSG_RELEASE(inactive);
1289 }
1290 pnode->pn_stat &= ~PNODE_DOINACT;
1291
1292 /*
1293 * file server thinks it's gone? then don't be afraid care,
1294 * node's life was already all it would ever be
1295 */
1296 if (pnode->pn_stat & PNODE_NOREFS) {
1297 pnode->pn_stat |= PNODE_DYING;
1298 recycle = true;
1299 }
1300
1301 /*
1302 * Handle node TTL.
1303 * If grace has already timed out, make it reclaimed.
1304 * Otherwise, we queue its expiration by sop thread, so
1305 * that it does not remain for ages in the freelist,
1306 * holding memory in userspace, while we will have
1307 * to look it up again anyway.
1308 */
1309 if (PUFFS_USE_FS_TTL(pmp) && !(vp->v_vflag & VV_ROOT) && !recycle) {
1310 bool incache = !TIMED_OUT(pnode->pn_cn_timeout);
1311 bool ingrace = !TIMED_OUT(pnode->pn_cn_grace);
1312 bool reclaimqueued = pnode->pn_stat & PNODE_SOPEXP;
1313
1314 if (!incache && !ingrace && !reclaimqueued) {
1315 pnode->pn_stat |= PNODE_DYING;
1316 recycle = true;
1317 }
1318
1319 if (!recycle && !reclaimqueued) {
1320 struct puffs_sopreq *psopr;
1321 int at = MAX(pnode->pn_cn_grace, pnode->pn_cn_timeout);
1322
1323 KASSERT(curlwp != uvm.pagedaemon_lwp);
1324 psopr = kmem_alloc(sizeof(*psopr), KM_SLEEP);
1325 psopr->psopr_ck = VPTOPNC(pnode->pn_vp);
1326 psopr->psopr_sopreq = PUFFS_SOPREQ_EXPIRE;
1327 psopr->psopr_at = at;
1328
1329 mutex_enter(&pmp->pmp_sopmtx);
1330
1331 /*
1332 * If thread has disapeared, just give up. The
1333 * fs is being unmounted and the node will be
1334 * be reclaimed anyway.
1335 *
1336 * Otherwise, we queue the request but do not
1337 * immediatly signal the thread, as the node
1338 * has not been expired yet.
1339 */
1340 if (pmp->pmp_sopthrcount == 0) {
1341 kmem_free(psopr, sizeof(*psopr));
1342 } else {
1343 TAILQ_INSERT_TAIL(&pmp->pmp_sopnodereqs,
1344 psopr, psopr_entries);
1345 pnode->pn_stat |= PNODE_SOPEXP;
1346 }
1347
1348 mutex_exit(&pmp->pmp_sopmtx);
1349 }
1350 }
1351
1352 *ap->a_recycle = recycle;
1353
1354 mutex_exit(&pnode->pn_sizemtx);
1355 VOP_UNLOCK(vp);
1356
1357 return 0;
1358 }
1359
1360 static void
1361 callreclaim(struct puffs_mount *pmp, puffs_cookie_t ck, int nlookup)
1362 {
1363 PUFFS_MSG_VARS(vn, reclaim);
1364
1365 if (!EXISTSOP(pmp, RECLAIM))
1366 return;
1367
1368 PUFFS_MSG_ALLOC(vn, reclaim);
1369 reclaim_msg->pvnr_nlookup = nlookup;
1370 puffs_msg_setfaf(park_reclaim);
1371 puffs_msg_setinfo(park_reclaim, PUFFSOP_VN, PUFFS_VN_RECLAIM, ck);
1372
1373 puffs_msg_enqueue(pmp, park_reclaim);
1374 PUFFS_MSG_RELEASE(reclaim);
1375 return;
1376 }
1377
1378 /*
1379 * always FAF, we don't really care if the server wants to fail to
1380 * reclaim the node or not
1381 */
1382 int
1383 puffs_vnop_reclaim(void *v)
1384 {
1385 struct vop_reclaim_args /* {
1386 const struct vnodeop_desc *a_desc;
1387 struct vnode *a_vp;
1388 } */ *ap = v;
1389 struct vnode *vp = ap->a_vp;
1390 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1391 struct puffs_node *pnode = vp->v_data;
1392 bool notifyserver = true;
1393
1394 /*
1395 * first things first: check if someone is trying to reclaim the
1396 * root vnode. do not allow that to travel to userspace.
1397 * Note that we don't need to take the lock similarly to
1398 * puffs_root(), since there is only one of us.
1399 */
1400 if (vp->v_vflag & VV_ROOT) {
1401 mutex_enter(&pmp->pmp_lock);
1402 KASSERT(pmp->pmp_root != NULL);
1403 pmp->pmp_root = NULL;
1404 mutex_exit(&pmp->pmp_lock);
1405 notifyserver = false;
1406 }
1407
1408 /*
1409 * purge info from kernel before issueing FAF, since we
1410 * don't really know when we'll get around to it after
1411 * that and someone might race us into node creation
1412 */
1413 mutex_enter(&pmp->pmp_lock);
1414 LIST_REMOVE(pnode, pn_hashent);
1415 if (PUFFS_USE_NAMECACHE(pmp))
1416 cache_purge(vp);
1417 mutex_exit(&pmp->pmp_lock);
1418
1419 if (notifyserver) {
1420 int nlookup = VPTOPP(vp)->pn_nlookup;
1421
1422 callreclaim(MPTOPUFFSMP(vp->v_mount), VPTOPNC(vp), nlookup);
1423 }
1424
1425 if (PUFFS_USE_DOTDOTCACHE(pmp)) {
1426 if (__predict_true(VPTOPP(vp)->pn_parent != NULL))
1427 vrele(VPTOPP(vp)->pn_parent);
1428 else
1429 KASSERT(vp->v_vflag & VV_ROOT);
1430 }
1431
1432 puffs_putvnode(vp);
1433 vp->v_data = NULL;
1434
1435 return 0;
1436 }
1437
1438 #define CSIZE sizeof(**ap->a_cookies)
1439 int
1440 puffs_vnop_readdir(void *v)
1441 {
1442 struct vop_readdir_args /* {
1443 const struct vnodeop_desc *a_desc;
1444 struct vnode *a_vp;
1445 struct uio *a_uio;
1446 kauth_cred_t a_cred;
1447 int *a_eofflag;
1448 off_t **a_cookies;
1449 int *a_ncookies;
1450 } */ *ap = v;
1451 PUFFS_MSG_VARS(vn, readdir);
1452 struct vnode *vp = ap->a_vp;
1453 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1454 size_t argsize, tomove, cookiemem, cookiesmax;
1455 struct uio *uio = ap->a_uio;
1456 size_t howmuch, resid;
1457 int error;
1458
1459 /*
1460 * ok, so we need: resid + cookiemem = maxreq
1461 * => resid + cookiesize * (resid/minsize) = maxreq
1462 * => resid + cookiesize/minsize * resid = maxreq
1463 * => (cookiesize/minsize + 1) * resid = maxreq
1464 * => resid = maxreq / (cookiesize/minsize + 1)
1465 *
1466 * Since cookiesize <= minsize and we're not very big on floats,
1467 * we approximate that to be 1. Therefore:
1468 *
1469 * resid = maxreq / 2;
1470 *
1471 * Well, at least we didn't have to use differential equations
1472 * or the Gram-Schmidt process.
1473 *
1474 * (yes, I'm very afraid of this)
1475 */
1476 KASSERT(CSIZE <= _DIRENT_MINSIZE((struct dirent *)0));
1477
1478 if (ap->a_cookies) {
1479 KASSERT(ap->a_ncookies != NULL);
1480 if (pmp->pmp_args.pa_fhsize == 0)
1481 return EOPNOTSUPP;
1482 resid = PUFFS_TOMOVE(uio->uio_resid, pmp) / 2;
1483 cookiesmax = resid/_DIRENT_MINSIZE((struct dirent *)0);
1484 cookiemem = ALIGN(cookiesmax*CSIZE); /* play safe */
1485 } else {
1486 resid = PUFFS_TOMOVE(uio->uio_resid, pmp);
1487 cookiesmax = 0;
1488 cookiemem = 0;
1489 }
1490
1491 argsize = sizeof(struct puffs_vnmsg_readdir);
1492 tomove = resid + cookiemem;
1493 puffs_msgmem_alloc(argsize + tomove, &park_readdir,
1494 (void *)&readdir_msg, 1);
1495
1496 puffs_credcvt(&readdir_msg->pvnr_cred, ap->a_cred);
1497 readdir_msg->pvnr_offset = uio->uio_offset;
1498 readdir_msg->pvnr_resid = resid;
1499 readdir_msg->pvnr_ncookies = cookiesmax;
1500 readdir_msg->pvnr_eofflag = 0;
1501 readdir_msg->pvnr_dentoff = cookiemem;
1502 puffs_msg_setinfo(park_readdir, PUFFSOP_VN,
1503 PUFFS_VN_READDIR, VPTOPNC(vp));
1504 puffs_msg_setdelta(park_readdir, tomove);
1505
1506 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readdir, vp->v_data, NULL, error);
1507 error = checkerr(pmp, error, __func__);
1508 if (error)
1509 goto out;
1510
1511 /* userspace is cheating? */
1512 if (readdir_msg->pvnr_resid > resid) {
1513 puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1514 "resid grew", VPTOPNC(vp));
1515 ERROUT(EPROTO);
1516 }
1517 if (readdir_msg->pvnr_ncookies > cookiesmax) {
1518 puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1519 "too many cookies", VPTOPNC(vp));
1520 ERROUT(EPROTO);
1521 }
1522
1523 /* check eof */
1524 if (readdir_msg->pvnr_eofflag)
1525 *ap->a_eofflag = 1;
1526
1527 /* bouncy-wouncy with the directory data */
1528 howmuch = resid - readdir_msg->pvnr_resid;
1529
1530 /* force eof if no data was returned (getcwd() needs this) */
1531 if (howmuch == 0) {
1532 *ap->a_eofflag = 1;
1533 goto out;
1534 }
1535
1536 error = uiomove(readdir_msg->pvnr_data + cookiemem, howmuch, uio);
1537 if (error)
1538 goto out;
1539
1540 /* provide cookies to caller if so desired */
1541 if (ap->a_cookies) {
1542 KASSERT(curlwp != uvm.pagedaemon_lwp);
1543 *ap->a_cookies = malloc(readdir_msg->pvnr_ncookies*CSIZE,
1544 M_TEMP, M_WAITOK);
1545 *ap->a_ncookies = readdir_msg->pvnr_ncookies;
1546 memcpy(*ap->a_cookies, readdir_msg->pvnr_data,
1547 *ap->a_ncookies*CSIZE);
1548 }
1549
1550 /* next readdir starts here */
1551 uio->uio_offset = readdir_msg->pvnr_offset;
1552
1553 out:
1554 puffs_msgmem_release(park_readdir);
1555 return error;
1556 }
1557 #undef CSIZE
1558
1559 /*
1560 * poll works by consuming the bitmask in pn_revents. If there are
1561 * events available, poll returns immediately. If not, it issues a
1562 * poll to userspace, selrecords itself and returns with no available
1563 * events. When the file server returns, it executes puffs_parkdone_poll(),
1564 * where available events are added to the bitmask. selnotify() is
1565 * then also executed by that function causing us to enter here again
1566 * and hopefully find the missing bits (unless someone got them first,
1567 * in which case it starts all over again).
1568 */
1569 int
1570 puffs_vnop_poll(void *v)
1571 {
1572 struct vop_poll_args /* {
1573 const struct vnodeop_desc *a_desc;
1574 struct vnode *a_vp;
1575 int a_events;
1576 } */ *ap = v;
1577 PUFFS_MSG_VARS(vn, poll);
1578 struct vnode *vp = ap->a_vp;
1579 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1580 struct puffs_node *pn = vp->v_data;
1581 int events;
1582
1583 if (EXISTSOP(pmp, POLL)) {
1584 mutex_enter(&pn->pn_mtx);
1585 events = pn->pn_revents & ap->a_events;
1586 if (events & ap->a_events) {
1587 pn->pn_revents &= ~ap->a_events;
1588 mutex_exit(&pn->pn_mtx);
1589
1590 return events;
1591 } else {
1592 puffs_referencenode(pn);
1593 mutex_exit(&pn->pn_mtx);
1594
1595 PUFFS_MSG_ALLOC(vn, poll);
1596 poll_msg->pvnr_events = ap->a_events;
1597 puffs_msg_setinfo(park_poll, PUFFSOP_VN,
1598 PUFFS_VN_POLL, VPTOPNC(vp));
1599 puffs_msg_setcall(park_poll, puffs_parkdone_poll, pn);
1600 selrecord(curlwp, &pn->pn_sel);
1601
1602 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_poll,
1603 vp->v_data, NULL);
1604 PUFFS_MSG_RELEASE(poll);
1605
1606 return 0;
1607 }
1608 } else {
1609 return genfs_poll(v);
1610 }
1611 }
1612
1613 static int
1614 flushvncache(struct vnode *vp, off_t offlo, off_t offhi, bool wait)
1615 {
1616 struct puffs_node *pn = VPTOPP(vp);
1617 struct vattr va;
1618 int pflags, error;
1619
1620 /* flush out information from our metacache, see vop_setattr */
1621 if (pn->pn_stat & PNODE_METACACHE_MASK
1622 && (pn->pn_stat & PNODE_DYING) == 0) {
1623 vattr_null(&va);
1624 error = dosetattr(vp, &va, FSCRED,
1625 SETATTR_CHSIZE | (wait ? 0 : SETATTR_ASYNC));
1626 if (error)
1627 return error;
1628 }
1629
1630 /*
1631 * flush pages to avoid being overly dirty
1632 */
1633 pflags = PGO_CLEANIT;
1634 if (wait)
1635 pflags |= PGO_SYNCIO;
1636
1637 mutex_enter(vp->v_interlock);
1638 return VOP_PUTPAGES(vp, trunc_page(offlo), round_page(offhi), pflags);
1639 }
1640
1641 int
1642 puffs_vnop_fsync(void *v)
1643 {
1644 struct vop_fsync_args /* {
1645 const struct vnodeop_desc *a_desc;
1646 struct vnode *a_vp;
1647 kauth_cred_t a_cred;
1648 int a_flags;
1649 off_t a_offlo;
1650 off_t a_offhi;
1651 } */ *ap = v;
1652 PUFFS_MSG_VARS(vn, fsync);
1653 struct vnode *vp;
1654 struct puffs_node *pn;
1655 struct puffs_mount *pmp;
1656 int error, dofaf;
1657
1658 vp = ap->a_vp;
1659 KASSERT(vp != NULL);
1660 pn = VPTOPP(vp);
1661 KASSERT(pn != NULL);
1662 pmp = MPTOPUFFSMP(vp->v_mount);
1663 if (ap->a_flags & FSYNC_WAIT) {
1664 mutex_enter(&pn->pn_sizemtx);
1665 } else {
1666 if (mutex_tryenter(&pn->pn_sizemtx) == 0)
1667 return EDEADLK;
1668 }
1669
1670 error = flushvncache(vp, ap->a_offlo, ap->a_offhi,
1671 (ap->a_flags & FSYNC_WAIT) == FSYNC_WAIT);
1672 if (error)
1673 goto out;
1674
1675 /*
1676 * HELLO! We exit already here if the user server does not
1677 * support fsync OR if we should call fsync for a node which
1678 * has references neither in the kernel or the fs server.
1679 * Otherwise we continue to issue fsync() forward.
1680 */
1681 error = 0;
1682 if (!EXISTSOP(pmp, FSYNC) || (pn->pn_stat & PNODE_DYING))
1683 goto out;
1684
1685 dofaf = (ap->a_flags & FSYNC_WAIT) == 0 || ap->a_flags == FSYNC_LAZY;
1686 /*
1687 * We abuse VXLOCK to mean "vnode is going to die", so we issue
1688 * only FAFs for those. Otherwise there's a danger of deadlock,
1689 * since the execution context here might be the user server
1690 * doing some operation on another fs, which in turn caused a
1691 * vnode to be reclaimed from the freelist for this fs.
1692 */
1693 if (dofaf == 0) {
1694 mutex_enter(vp->v_interlock);
1695 if (vdead_check(vp, VDEAD_NOWAIT) != 0)
1696 dofaf = 1;
1697 mutex_exit(vp->v_interlock);
1698 }
1699
1700 PUFFS_MSG_ALLOC(vn, fsync);
1701 if (dofaf)
1702 puffs_msg_setfaf(park_fsync);
1703
1704 puffs_credcvt(&fsync_msg->pvnr_cred, ap->a_cred);
1705 fsync_msg->pvnr_flags = ap->a_flags;
1706 fsync_msg->pvnr_offlo = ap->a_offlo;
1707 fsync_msg->pvnr_offhi = ap->a_offhi;
1708 puffs_msg_setinfo(park_fsync, PUFFSOP_VN,
1709 PUFFS_VN_FSYNC, VPTOPNC(vp));
1710
1711 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fsync, vp->v_data, NULL, error);
1712 PUFFS_MSG_RELEASE(fsync);
1713
1714 error = checkerr(pmp, error, __func__);
1715
1716 out:
1717 mutex_exit(&pn->pn_sizemtx);
1718 return error;
1719 }
1720
1721 int
1722 puffs_vnop_seek(void *v)
1723 {
1724 struct vop_seek_args /* {
1725 const struct vnodeop_desc *a_desc;
1726 struct vnode *a_vp;
1727 off_t a_oldoff;
1728 off_t a_newoff;
1729 kauth_cred_t a_cred;
1730 } */ *ap = v;
1731 PUFFS_MSG_VARS(vn, seek);
1732 struct vnode *vp = ap->a_vp;
1733 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1734 int error;
1735
1736 PUFFS_MSG_ALLOC(vn, seek);
1737 seek_msg->pvnr_oldoff = ap->a_oldoff;
1738 seek_msg->pvnr_newoff = ap->a_newoff;
1739 puffs_credcvt(&seek_msg->pvnr_cred, ap->a_cred);
1740 puffs_msg_setinfo(park_seek, PUFFSOP_VN,
1741 PUFFS_VN_SEEK, VPTOPNC(vp));
1742
1743 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_seek, vp->v_data, NULL, error);
1744 PUFFS_MSG_RELEASE(seek);
1745 return checkerr(pmp, error, __func__);
1746 }
1747
1748 static int
1749 callremove(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1750 struct componentname *cnp)
1751 {
1752 PUFFS_MSG_VARS(vn, remove);
1753 int error;
1754
1755 PUFFS_MSG_ALLOC(vn, remove);
1756 remove_msg->pvnr_cookie_targ = ck;
1757 puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1758 cnp, PUFFS_USE_FULLPNBUF(pmp));
1759 puffs_msg_setinfo(park_remove, PUFFSOP_VN, PUFFS_VN_REMOVE, dck);
1760
1761 PUFFS_MSG_ENQUEUEWAIT(pmp, park_remove, error);
1762 PUFFS_MSG_RELEASE(remove);
1763
1764 return checkerr(pmp, error, __func__);
1765 }
1766
1767 /*
1768 * XXX: can't use callremove now because can't catch setbacks with
1769 * it due to lack of a pnode argument.
1770 */
1771 int
1772 puffs_vnop_remove(void *v)
1773 {
1774 struct vop_remove_args /* {
1775 const struct vnodeop_desc *a_desc;
1776 struct vnode *a_dvp;
1777 struct vnode *a_vp;
1778 struct componentname *a_cnp;
1779 } */ *ap = v;
1780 PUFFS_MSG_VARS(vn, remove);
1781 struct vnode *dvp = ap->a_dvp;
1782 struct vnode *vp = ap->a_vp;
1783 struct puffs_node *dpn = VPTOPP(dvp);
1784 struct puffs_node *pn = VPTOPP(vp);
1785 struct componentname *cnp = ap->a_cnp;
1786 struct mount *mp = dvp->v_mount;
1787 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1788 int error;
1789
1790 PUFFS_MSG_ALLOC(vn, remove);
1791 remove_msg->pvnr_cookie_targ = VPTOPNC(vp);
1792 puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1793 cnp, PUFFS_USE_FULLPNBUF(pmp));
1794 puffs_msg_setinfo(park_remove, PUFFSOP_VN,
1795 PUFFS_VN_REMOVE, VPTOPNC(dvp));
1796
1797 puffs_msg_enqueue(pmp, park_remove);
1798 REFPN_AND_UNLOCKVP(dvp, dpn);
1799 if (dvp == vp)
1800 REFPN(pn);
1801 else
1802 REFPN_AND_UNLOCKVP(vp, pn);
1803 error = puffs_msg_wait2(pmp, park_remove, dpn, pn);
1804
1805 PUFFS_MSG_RELEASE(remove);
1806
1807 RELEPN_AND_VP(dvp, dpn);
1808 RELEPN_AND_VP(vp, pn);
1809
1810 error = checkerr(pmp, error, __func__);
1811 return error;
1812 }
1813
1814 int
1815 puffs_vnop_mkdir(void *v)
1816 {
1817 struct vop_mkdir_v3_args /* {
1818 const struct vnodeop_desc *a_desc;
1819 struct vnode *a_dvp;
1820 struct vnode **a_vpp;
1821 struct componentname *a_cnp;
1822 struct vattr *a_vap;
1823 } */ *ap = v;
1824 PUFFS_MSG_VARS(vn, mkdir);
1825 struct vnode *dvp = ap->a_dvp;
1826 struct puffs_node *dpn = VPTOPP(dvp);
1827 struct componentname *cnp = ap->a_cnp;
1828 struct mount *mp = dvp->v_mount;
1829 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1830 int error;
1831
1832 PUFFS_MSG_ALLOC(vn, mkdir);
1833 puffs_makecn(&mkdir_msg->pvnr_cn, &mkdir_msg->pvnr_cn_cred,
1834 cnp, PUFFS_USE_FULLPNBUF(pmp));
1835 mkdir_msg->pvnr_va = *ap->a_vap;
1836 puffs_msg_setinfo(park_mkdir, PUFFSOP_VN,
1837 PUFFS_VN_MKDIR, VPTOPNC(dvp));
1838
1839 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mkdir, dvp->v_data, NULL, error);
1840
1841 error = checkerr(pmp, error, __func__);
1842 if (error)
1843 goto out;
1844
1845 error = puffs_newnode(mp, dvp, ap->a_vpp,
1846 mkdir_msg->pvnr_newnode, cnp, VDIR, 0);
1847 if (error) {
1848 puffs_abortbutton(pmp, PUFFS_ABORT_MKDIR, dpn->pn_cookie,
1849 mkdir_msg->pvnr_newnode, cnp);
1850 goto out;
1851 }
1852
1853 if (PUFFS_USE_FS_TTL(pmp)) {
1854 struct timespec *va_ttl = &mkdir_msg->pvnr_va_ttl;
1855 struct timespec *cn_ttl = &mkdir_msg->pvnr_cn_ttl;
1856 struct vattr *rvap = &mkdir_msg->pvnr_va;
1857
1858 update_va(*ap->a_vpp, NULL, rvap,
1859 va_ttl, cn_ttl, SETATTR_CHSIZE);
1860 }
1861
1862 VPTOPP(*ap->a_vpp)->pn_nlookup++;
1863
1864 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
1865 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
1866 update_parent(*ap->a_vpp, dvp);
1867
1868 out:
1869 PUFFS_MSG_RELEASE(mkdir);
1870 return error;
1871 }
1872
1873 static int
1874 callrmdir(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1875 struct componentname *cnp)
1876 {
1877 PUFFS_MSG_VARS(vn, rmdir);
1878 int error;
1879
1880 PUFFS_MSG_ALLOC(vn, rmdir);
1881 rmdir_msg->pvnr_cookie_targ = ck;
1882 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1883 cnp, PUFFS_USE_FULLPNBUF(pmp));
1884 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN, PUFFS_VN_RMDIR, dck);
1885
1886 PUFFS_MSG_ENQUEUEWAIT(pmp, park_rmdir, error);
1887 PUFFS_MSG_RELEASE(rmdir);
1888
1889 return checkerr(pmp, error, __func__);
1890 }
1891
1892 int
1893 puffs_vnop_rmdir(void *v)
1894 {
1895 struct vop_rmdir_args /* {
1896 const struct vnodeop_desc *a_desc;
1897 struct vnode *a_dvp;
1898 struct vnode *a_vp;
1899 struct componentname *a_cnp;
1900 } */ *ap = v;
1901 PUFFS_MSG_VARS(vn, rmdir);
1902 struct vnode *dvp = ap->a_dvp;
1903 struct vnode *vp = ap->a_vp;
1904 struct puffs_node *dpn = VPTOPP(dvp);
1905 struct puffs_node *pn = VPTOPP(vp);
1906 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1907 struct componentname *cnp = ap->a_cnp;
1908 int error;
1909
1910 PUFFS_MSG_ALLOC(vn, rmdir);
1911 rmdir_msg->pvnr_cookie_targ = VPTOPNC(vp);
1912 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1913 cnp, PUFFS_USE_FULLPNBUF(pmp));
1914 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN,
1915 PUFFS_VN_RMDIR, VPTOPNC(dvp));
1916
1917 puffs_msg_enqueue(pmp, park_rmdir);
1918 REFPN_AND_UNLOCKVP(dvp, dpn);
1919 REFPN_AND_UNLOCKVP(vp, pn);
1920 error = puffs_msg_wait2(pmp, park_rmdir, dpn, pn);
1921
1922 PUFFS_MSG_RELEASE(rmdir);
1923
1924 /* XXX: some call cache_purge() *for both vnodes* here, investigate */
1925 RELEPN_AND_VP(dvp, dpn);
1926 RELEPN_AND_VP(vp, pn);
1927
1928 return error;
1929 }
1930
1931 int
1932 puffs_vnop_link(void *v)
1933 {
1934 struct vop_link_args /* {
1935 const struct vnodeop_desc *a_desc;
1936 struct vnode *a_dvp;
1937 struct vnode *a_vp;
1938 struct componentname *a_cnp;
1939 } */ *ap = v;
1940 PUFFS_MSG_VARS(vn, link);
1941 struct vnode *dvp = ap->a_dvp;
1942 struct vnode *vp = ap->a_vp;
1943 struct puffs_node *dpn = VPTOPP(dvp);
1944 struct puffs_node *pn = VPTOPP(vp);
1945 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1946 struct componentname *cnp = ap->a_cnp;
1947 int error;
1948
1949 PUFFS_MSG_ALLOC(vn, link);
1950 link_msg->pvnr_cookie_targ = VPTOPNC(vp);
1951 puffs_makecn(&link_msg->pvnr_cn, &link_msg->pvnr_cn_cred,
1952 cnp, PUFFS_USE_FULLPNBUF(pmp));
1953 puffs_msg_setinfo(park_link, PUFFSOP_VN,
1954 PUFFS_VN_LINK, VPTOPNC(dvp));
1955
1956 puffs_msg_enqueue(pmp, park_link);
1957 REFPN_AND_UNLOCKVP(dvp, dpn);
1958 REFPN(pn);
1959 error = puffs_msg_wait2(pmp, park_link, dpn, pn);
1960
1961 PUFFS_MSG_RELEASE(link);
1962
1963 error = checkerr(pmp, error, __func__);
1964
1965 /*
1966 * XXX: stay in touch with the cache. I don't like this, but
1967 * don't have a better solution either. See also puffs_rename().
1968 */
1969 if (error == 0)
1970 puffs_updatenode(pn, PUFFS_UPDATECTIME, 0);
1971
1972 RELEPN_AND_VP(dvp, dpn);
1973 puffs_releasenode(pn);
1974
1975 return error;
1976 }
1977
1978 int
1979 puffs_vnop_symlink(void *v)
1980 {
1981 struct vop_symlink_v3_args /* {
1982 const struct vnodeop_desc *a_desc;
1983 struct vnode *a_dvp;
1984 struct vnode **a_vpp;
1985 struct componentname *a_cnp;
1986 struct vattr *a_vap;
1987 char *a_target;
1988 } */ *ap = v;
1989 PUFFS_MSG_VARS(vn, symlink);
1990 struct vnode *dvp = ap->a_dvp;
1991 struct puffs_node *dpn = VPTOPP(dvp);
1992 struct mount *mp = dvp->v_mount;
1993 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1994 struct componentname *cnp = ap->a_cnp;
1995 int error;
1996
1997 *ap->a_vpp = NULL;
1998
1999 PUFFS_MSG_ALLOC(vn, symlink);
2000 puffs_makecn(&symlink_msg->pvnr_cn, &symlink_msg->pvnr_cn_cred,
2001 cnp, PUFFS_USE_FULLPNBUF(pmp));
2002 symlink_msg->pvnr_va = *ap->a_vap;
2003 (void)strlcpy(symlink_msg->pvnr_link, ap->a_target,
2004 sizeof(symlink_msg->pvnr_link));
2005 puffs_msg_setinfo(park_symlink, PUFFSOP_VN,
2006 PUFFS_VN_SYMLINK, VPTOPNC(dvp));
2007
2008 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_symlink, dvp->v_data, NULL, error);
2009
2010 error = checkerr(pmp, error, __func__);
2011 if (error)
2012 goto out;
2013
2014 error = puffs_newnode(mp, dvp, ap->a_vpp,
2015 symlink_msg->pvnr_newnode, cnp, VLNK, 0);
2016 if (error) {
2017 puffs_abortbutton(pmp, PUFFS_ABORT_SYMLINK, dpn->pn_cookie,
2018 symlink_msg->pvnr_newnode, cnp);
2019 goto out;
2020 }
2021
2022 if (PUFFS_USE_FS_TTL(pmp)) {
2023 struct timespec *va_ttl = &symlink_msg->pvnr_va_ttl;
2024 struct timespec *cn_ttl = &symlink_msg->pvnr_cn_ttl;
2025 struct vattr *rvap = &symlink_msg->pvnr_va;
2026
2027 update_va(*ap->a_vpp, NULL, rvap,
2028 va_ttl, cn_ttl, SETATTR_CHSIZE);
2029 }
2030
2031 VPTOPP(*ap->a_vpp)->pn_nlookup++;
2032
2033 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2034 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
2035 update_parent(*ap->a_vpp, dvp);
2036
2037 out:
2038 PUFFS_MSG_RELEASE(symlink);
2039
2040 return error;
2041 }
2042
2043 int
2044 puffs_vnop_readlink(void *v)
2045 {
2046 struct vop_readlink_args /* {
2047 const struct vnodeop_desc *a_desc;
2048 struct vnode *a_vp;
2049 struct uio *a_uio;
2050 kauth_cred_t a_cred;
2051 } */ *ap = v;
2052 PUFFS_MSG_VARS(vn, readlink);
2053 struct vnode *vp = ap->a_vp;
2054 struct puffs_mount *pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2055 size_t linklen;
2056 int error;
2057
2058 PUFFS_MSG_ALLOC(vn, readlink);
2059 puffs_credcvt(&readlink_msg->pvnr_cred, ap->a_cred);
2060 linklen = sizeof(readlink_msg->pvnr_link);
2061 readlink_msg->pvnr_linklen = linklen;
2062 puffs_msg_setinfo(park_readlink, PUFFSOP_VN,
2063 PUFFS_VN_READLINK, VPTOPNC(vp));
2064
2065 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readlink, vp->v_data, NULL, error);
2066 error = checkerr(pmp, error, __func__);
2067 if (error)
2068 goto out;
2069
2070 /* bad bad user file server */
2071 if (readlink_msg->pvnr_linklen > linklen) {
2072 puffs_senderr(pmp, PUFFS_ERR_READLINK, E2BIG,
2073 "linklen too big", VPTOPNC(ap->a_vp));
2074 error = EPROTO;
2075 goto out;
2076 }
2077
2078 error = uiomove(&readlink_msg->pvnr_link, readlink_msg->pvnr_linklen,
2079 ap->a_uio);
2080 out:
2081 PUFFS_MSG_RELEASE(readlink);
2082 return error;
2083 }
2084
2085 int
2086 puffs_vnop_rename(void *v)
2087 {
2088 struct vop_rename_args /* {
2089 const struct vnodeop_desc *a_desc;
2090 struct vnode *a_fdvp;
2091 struct vnode *a_fvp;
2092 struct componentname *a_fcnp;
2093 struct vnode *a_tdvp;
2094 struct vnode *a_tvp;
2095 struct componentname *a_tcnp;
2096 } */ *ap = v;
2097 PUFFS_MSG_VARS(vn, rename);
2098 struct vnode *fdvp = ap->a_fdvp, *fvp = ap->a_fvp;
2099 struct vnode *tdvp = ap->a_tdvp, *tvp = ap->a_tvp;
2100 struct puffs_node *fpn = ap->a_fvp->v_data;
2101 struct puffs_mount *pmp = MPTOPUFFSMP(fdvp->v_mount);
2102 int error;
2103 bool doabort = true;
2104
2105 if ((fvp->v_mount != tdvp->v_mount) ||
2106 (tvp && (fvp->v_mount != tvp->v_mount))) {
2107 ERROUT(EXDEV);
2108 }
2109
2110 PUFFS_MSG_ALLOC(vn, rename);
2111 rename_msg->pvnr_cookie_src = VPTOPNC(fvp);
2112 rename_msg->pvnr_cookie_targdir = VPTOPNC(tdvp);
2113 if (tvp)
2114 rename_msg->pvnr_cookie_targ = VPTOPNC(tvp);
2115 else
2116 rename_msg->pvnr_cookie_targ = NULL;
2117 puffs_makecn(&rename_msg->pvnr_cn_src, &rename_msg->pvnr_cn_src_cred,
2118 ap->a_fcnp, PUFFS_USE_FULLPNBUF(pmp));
2119 puffs_makecn(&rename_msg->pvnr_cn_targ, &rename_msg->pvnr_cn_targ_cred,
2120 ap->a_tcnp, PUFFS_USE_FULLPNBUF(pmp));
2121 puffs_msg_setinfo(park_rename, PUFFSOP_VN,
2122 PUFFS_VN_RENAME, VPTOPNC(fdvp));
2123
2124 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rename, fdvp->v_data, NULL, error);
2125 doabort = false;
2126 PUFFS_MSG_RELEASE(rename);
2127 error = checkerr(pmp, error, __func__);
2128
2129 /*
2130 * XXX: stay in touch with the cache. I don't like this, but
2131 * don't have a better solution either. See also puffs_link().
2132 */
2133 if (error == 0) {
2134 puffs_updatenode(fpn, PUFFS_UPDATECTIME, 0);
2135
2136 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2137 (VPTOPP(fvp)->pn_parent != tdvp))
2138 update_parent(fvp, tdvp);
2139 }
2140
2141
2142 out:
2143 if (doabort)
2144 VOP_ABORTOP(tdvp, ap->a_tcnp);
2145 if (tvp != NULL)
2146 vput(tvp);
2147 if (tdvp == tvp)
2148 vrele(tdvp);
2149 else
2150 vput(tdvp);
2151
2152 if (doabort)
2153 VOP_ABORTOP(fdvp, ap->a_fcnp);
2154 vrele(fdvp);
2155 vrele(fvp);
2156
2157 return error;
2158 }
2159
2160 #define RWARGS(cont, iofl, move, offset, creds) \
2161 (cont)->pvnr_ioflag = (iofl); \
2162 (cont)->pvnr_resid = (move); \
2163 (cont)->pvnr_offset = (offset); \
2164 puffs_credcvt(&(cont)->pvnr_cred, creds)
2165
2166 int
2167 puffs_vnop_read(void *v)
2168 {
2169 struct vop_read_args /* {
2170 const struct vnodeop_desc *a_desc;
2171 struct vnode *a_vp;
2172 struct uio *a_uio;
2173 int a_ioflag;
2174 kauth_cred_t a_cred;
2175 } */ *ap = v;
2176 PUFFS_MSG_VARS(vn, read);
2177 struct vnode *vp = ap->a_vp;
2178 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2179 struct uio *uio = ap->a_uio;
2180 size_t tomove, argsize;
2181 vsize_t bytelen;
2182 int error;
2183
2184 read_msg = NULL;
2185 error = 0;
2186
2187 /* std sanity */
2188 if (uio->uio_resid == 0)
2189 return 0;
2190 if (uio->uio_offset < 0)
2191 return EINVAL;
2192
2193 if (vp->v_type == VREG && PUFFS_USE_PAGECACHE(pmp)) {
2194 const int advice = IO_ADV_DECODE(ap->a_ioflag);
2195
2196 while (uio->uio_resid > 0) {
2197 if (vp->v_size <= uio->uio_offset) {
2198 break;
2199 }
2200 bytelen = MIN(uio->uio_resid,
2201 vp->v_size - uio->uio_offset);
2202 if (bytelen == 0)
2203 break;
2204
2205 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
2206 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
2207 if (error)
2208 break;
2209 }
2210
2211 if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
2212 puffs_updatenode(VPTOPP(vp), PUFFS_UPDATEATIME, 0);
2213 } else {
2214 /*
2215 * in case it's not a regular file or we're operating
2216 * uncached, do read in the old-fashioned style,
2217 * i.e. explicit read operations
2218 */
2219
2220 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2221 argsize = sizeof(struct puffs_vnmsg_read);
2222 puffs_msgmem_alloc(argsize + tomove, &park_read,
2223 (void *)&read_msg, 1);
2224
2225 error = 0;
2226 while (uio->uio_resid > 0) {
2227 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2228 memset(read_msg, 0, argsize); /* XXX: touser KASSERT */
2229 RWARGS(read_msg, ap->a_ioflag, tomove,
2230 uio->uio_offset, ap->a_cred);
2231 puffs_msg_setinfo(park_read, PUFFSOP_VN,
2232 PUFFS_VN_READ, VPTOPNC(vp));
2233 puffs_msg_setdelta(park_read, tomove);
2234
2235 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_read, vp->v_data,
2236 NULL, error);
2237 error = checkerr(pmp, error, __func__);
2238 if (error)
2239 break;
2240
2241 if (read_msg->pvnr_resid > tomove) {
2242 puffs_senderr(pmp, PUFFS_ERR_READ,
2243 E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2244 error = EPROTO;
2245 break;
2246 }
2247
2248 error = uiomove(read_msg->pvnr_data,
2249 tomove - read_msg->pvnr_resid, uio);
2250
2251 /*
2252 * in case the file is out of juice, resid from
2253 * userspace is != 0. and the error-case is
2254 * quite obvious
2255 */
2256 if (error || read_msg->pvnr_resid)
2257 break;
2258 }
2259
2260 puffs_msgmem_release(park_read);
2261 }
2262
2263 return error;
2264 }
2265
2266 /*
2267 * XXX: in case of a failure, this leaves uio in a bad state.
2268 * We could theoretically copy the uio and iovecs and "replay"
2269 * them the right amount after the userspace trip, but don't
2270 * bother for now.
2271 */
2272 int
2273 puffs_vnop_write(void *v)
2274 {
2275 struct vop_write_args /* {
2276 const struct vnodeop_desc *a_desc;
2277 struct vnode *a_vp;
2278 struct uio *a_uio;
2279 int a_ioflag;
2280 kauth_cred_t a_cred;
2281 } */ *ap = v;
2282 PUFFS_MSG_VARS(vn, write);
2283 struct vnode *vp = ap->a_vp;
2284 struct puffs_node *pn = VPTOPP(vp);
2285 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2286 struct uio *uio = ap->a_uio;
2287 size_t tomove, argsize;
2288 off_t oldoff, newoff, origoff;
2289 vsize_t bytelen;
2290 int error, uflags;
2291 int ubcflags;
2292
2293 error = uflags = 0;
2294 write_msg = NULL;
2295
2296 mutex_enter(&pn->pn_sizemtx);
2297
2298 if (vp->v_type == VREG && PUFFS_USE_PAGECACHE(pmp)) {
2299 ubcflags = UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp);
2300
2301 /*
2302 * userspace *should* be allowed to control this,
2303 * but with UBC it's a bit unclear how to handle it
2304 */
2305 if (ap->a_ioflag & IO_APPEND)
2306 uio->uio_offset = vp->v_size;
2307
2308 origoff = uio->uio_offset;
2309 while (uio->uio_resid > 0) {
2310 if (vp->v_mount->mnt_flag & MNT_RELATIME)
2311 uflags |= PUFFS_UPDATEATIME;
2312 uflags |= PUFFS_UPDATECTIME;
2313 uflags |= PUFFS_UPDATEMTIME;
2314 oldoff = uio->uio_offset;
2315 bytelen = uio->uio_resid;
2316
2317 newoff = oldoff + bytelen;
2318 if (vp->v_size < newoff) {
2319 uvm_vnp_setwritesize(vp, newoff);
2320 }
2321 error = ubc_uiomove(&vp->v_uobj, uio, bytelen,
2322 UVM_ADV_RANDOM, ubcflags);
2323
2324 /*
2325 * In case of a ubc_uiomove() error,
2326 * opt to not extend the file at all and
2327 * return an error. Otherwise, if we attempt
2328 * to clear the memory we couldn't fault to,
2329 * we might generate a kernel page fault.
2330 */
2331 if (vp->v_size < newoff) {
2332 if (error == 0) {
2333 uflags |= PUFFS_UPDATESIZE;
2334 uvm_vnp_setsize(vp, newoff);
2335 } else {
2336 uvm_vnp_setwritesize(vp, vp->v_size);
2337 }
2338 }
2339 if (error)
2340 break;
2341
2342 /*
2343 * If we're writing large files, flush to file server
2344 * every 64k. Otherwise we can very easily exhaust
2345 * kernel and user memory, as the file server cannot
2346 * really keep up with our writing speed.
2347 *
2348 * Note: this does *NOT* honor MNT_ASYNC, because
2349 * that gives userland too much say in the kernel.
2350 */
2351 if (oldoff >> 16 != uio->uio_offset >> 16) {
2352 mutex_enter(vp->v_interlock);
2353 error = VOP_PUTPAGES(vp, oldoff & ~0xffff,
2354 uio->uio_offset & ~0xffff,
2355 PGO_CLEANIT | PGO_SYNCIO);
2356 if (error)
2357 break;
2358 }
2359 }
2360
2361 /* synchronous I/O? */
2362 if (error == 0 && ap->a_ioflag & IO_SYNC) {
2363 mutex_enter(vp->v_interlock);
2364 error = VOP_PUTPAGES(vp, trunc_page(origoff),
2365 round_page(uio->uio_offset),
2366 PGO_CLEANIT | PGO_SYNCIO);
2367
2368 /* write through page cache? */
2369 } else if (error == 0 && pmp->pmp_flags & PUFFS_KFLAG_WTCACHE) {
2370 mutex_enter(vp->v_interlock);
2371 error = VOP_PUTPAGES(vp, trunc_page(origoff),
2372 round_page(uio->uio_offset), PGO_CLEANIT);
2373 }
2374
2375 puffs_updatenode(VPTOPP(vp), uflags, vp->v_size);
2376 } else {
2377 /* tomove is non-increasing */
2378 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2379 argsize = sizeof(struct puffs_vnmsg_write) + tomove;
2380 puffs_msgmem_alloc(argsize, &park_write, (void *)&write_msg,1);
2381
2382 while (uio->uio_resid > 0) {
2383 /* move data to buffer */
2384 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2385 memset(write_msg, 0, argsize); /* XXX: touser KASSERT */
2386 RWARGS(write_msg, ap->a_ioflag, tomove,
2387 uio->uio_offset, ap->a_cred);
2388 error = uiomove(write_msg->pvnr_data, tomove, uio);
2389 if (error)
2390 break;
2391
2392 /* move buffer to userspace */
2393 puffs_msg_setinfo(park_write, PUFFSOP_VN,
2394 PUFFS_VN_WRITE, VPTOPNC(vp));
2395 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_write, vp->v_data,
2396 NULL, error);
2397 error = checkerr(pmp, error, __func__);
2398 if (error)
2399 break;
2400
2401 if (write_msg->pvnr_resid > tomove) {
2402 puffs_senderr(pmp, PUFFS_ERR_WRITE,
2403 E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2404 error = EPROTO;
2405 break;
2406 }
2407
2408 /* adjust file size */
2409 if (vp->v_size < uio->uio_offset)
2410 uvm_vnp_setsize(vp, uio->uio_offset);
2411
2412 /* didn't move everything? bad userspace. bail */
2413 if (write_msg->pvnr_resid != 0) {
2414 error = EIO;
2415 break;
2416 }
2417 }
2418 puffs_msgmem_release(park_write);
2419 }
2420
2421 mutex_exit(&pn->pn_sizemtx);
2422 return error;
2423 }
2424
2425 int
2426 puffs_vnop_print(void *v)
2427 {
2428 struct vop_print_args /* {
2429 struct vnode *a_vp;
2430 } */ *ap = v;
2431 PUFFS_MSG_VARS(vn, print);
2432 struct vnode *vp = ap->a_vp;
2433 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2434 struct puffs_node *pn = vp->v_data;
2435
2436 /* kernel portion */
2437 printf("tag VT_PUFFS, vnode %p, puffs node: %p,\n"
2438 "\tuserspace cookie: %p", vp, pn, pn->pn_cookie);
2439 if (vp->v_type == VFIFO)
2440 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
2441 printf("\n");
2442
2443 /* userspace portion */
2444 if (EXISTSOP(pmp, PRINT)) {
2445 PUFFS_MSG_ALLOC(vn, print);
2446 puffs_msg_setinfo(park_print, PUFFSOP_VN,
2447 PUFFS_VN_PRINT, VPTOPNC(vp));
2448 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_print, vp->v_data,
2449 NULL);
2450 PUFFS_MSG_RELEASE(print);
2451 }
2452
2453 return 0;
2454 }
2455
2456 int
2457 puffs_vnop_pathconf(void *v)
2458 {
2459 struct vop_pathconf_args /* {
2460 const struct vnodeop_desc *a_desc;
2461 struct vnode *a_vp;
2462 int a_name;
2463 register_t *a_retval;
2464 } */ *ap = v;
2465 PUFFS_MSG_VARS(vn, pathconf);
2466 struct vnode *vp = ap->a_vp;
2467 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2468 int error;
2469
2470 PUFFS_MSG_ALLOC(vn, pathconf);
2471 pathconf_msg->pvnr_name = ap->a_name;
2472 puffs_msg_setinfo(park_pathconf, PUFFSOP_VN,
2473 PUFFS_VN_PATHCONF, VPTOPNC(vp));
2474 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_pathconf, vp->v_data, NULL, error);
2475 error = checkerr(pmp, error, __func__);
2476 if (!error)
2477 *ap->a_retval = pathconf_msg->pvnr_retval;
2478 PUFFS_MSG_RELEASE(pathconf);
2479
2480 return error;
2481 }
2482
2483 int
2484 puffs_vnop_advlock(void *v)
2485 {
2486 struct vop_advlock_args /* {
2487 const struct vnodeop_desc *a_desc;
2488 struct vnode *a_vp;
2489 void *a_id;
2490 int a_op;
2491 struct flock *a_fl;
2492 int a_flags;
2493 } */ *ap = v;
2494 PUFFS_MSG_VARS(vn, advlock);
2495 struct vnode *vp = ap->a_vp;
2496 struct puffs_node *pn = VPTOPP(vp);
2497 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2498 int error;
2499
2500 if (!EXISTSOP(pmp, ADVLOCK))
2501 return lf_advlock(ap, &pn->pn_lockf, vp->v_size);
2502
2503 PUFFS_MSG_ALLOC(vn, advlock);
2504 (void)memcpy(&advlock_msg->pvnr_fl, ap->a_fl,
2505 sizeof(advlock_msg->pvnr_fl));
2506 advlock_msg->pvnr_id = ap->a_id;
2507 advlock_msg->pvnr_op = ap->a_op;
2508 advlock_msg->pvnr_flags = ap->a_flags;
2509 puffs_msg_setinfo(park_advlock, PUFFSOP_VN,
2510 PUFFS_VN_ADVLOCK, VPTOPNC(vp));
2511 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_advlock, vp->v_data, NULL, error);
2512 error = checkerr(pmp, error, __func__);
2513 PUFFS_MSG_RELEASE(advlock);
2514
2515 return error;
2516 }
2517
2518 int
2519 puffs_vnop_abortop(void *v)
2520 {
2521 struct vop_abortop_args /* {
2522 struct vnode *a_dvp;
2523 struct componentname *a_cnp;
2524 }; */ *ap = v;
2525 PUFFS_MSG_VARS(vn, abortop);
2526 struct vnode *dvp = ap->a_dvp;
2527 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2528 struct componentname *cnp = ap->a_cnp;
2529
2530 if (EXISTSOP(pmp, ABORTOP)) {
2531 PUFFS_MSG_ALLOC(vn, abortop);
2532 puffs_makecn(&abortop_msg->pvnr_cn, &abortop_msg->pvnr_cn_cred,
2533 cnp, PUFFS_USE_FULLPNBUF(pmp));
2534 puffs_msg_setfaf(park_abortop);
2535 puffs_msg_setinfo(park_abortop, PUFFSOP_VN,
2536 PUFFS_VN_ABORTOP, VPTOPNC(dvp));
2537
2538 puffs_msg_enqueue(pmp, park_abortop);
2539 PUFFS_MSG_RELEASE(abortop);
2540 }
2541
2542 return genfs_abortop(v);
2543 }
2544
2545 #define BIOASYNC(bp) (bp->b_flags & B_ASYNC)
2546
2547 /*
2548 * This maps itself to PUFFS_VN_READ/WRITE for data transfer.
2549 */
2550 int
2551 puffs_vnop_strategy(void *v)
2552 {
2553 struct vop_strategy_args /* {
2554 const struct vnodeop_desc *a_desc;
2555 struct vnode *a_vp;
2556 struct buf *a_bp;
2557 } */ *ap = v;
2558 PUFFS_MSG_VARS(vn, rw);
2559 struct vnode *vp = ap->a_vp;
2560 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2561 struct puffs_node *pn;
2562 struct buf *bp;
2563 size_t argsize;
2564 size_t tomove, moved;
2565 int error, dofaf, cansleep, dobiodone;
2566
2567 pmp = MPTOPUFFSMP(vp->v_mount);
2568 bp = ap->a_bp;
2569 error = 0;
2570 dofaf = 0;
2571 cansleep = 0;
2572 pn = VPTOPP(vp);
2573 park_rw = NULL; /* explicit */
2574 dobiodone = 1;
2575
2576 if ((BUF_ISREAD(bp) && !EXISTSOP(pmp, READ))
2577 || (BUF_ISWRITE(bp) && !EXISTSOP(pmp, WRITE)))
2578 ERROUT(EOPNOTSUPP);
2579
2580 /*
2581 * Short-circuit optimization: don't flush buffer in between
2582 * VOP_INACTIVE and VOP_RECLAIM in case the node has no references.
2583 */
2584 if (pn->pn_stat & PNODE_DYING) {
2585 KASSERT(BUF_ISWRITE(bp));
2586 bp->b_resid = 0;
2587 goto out;
2588 }
2589
2590 #ifdef DIAGNOSTIC
2591 if (bp->b_bcount > pmp->pmp_msg_maxsize - PUFFS_MSGSTRUCT_MAX)
2592 panic("puffs_strategy: wildly inappropriate buf bcount %d",
2593 bp->b_bcount);
2594 #endif
2595
2596 /*
2597 * See explanation for the necessity of a FAF in puffs_fsync.
2598 *
2599 * Also, do FAF in case we're suspending.
2600 * See puffs_vfsops.c:pageflush()
2601 */
2602 if (BUF_ISWRITE(bp)) {
2603 mutex_enter(vp->v_interlock);
2604 if (vdead_check(vp, VDEAD_NOWAIT) != 0)
2605 dofaf = 1;
2606 if (pn->pn_stat & PNODE_FAF)
2607 dofaf = 1;
2608 mutex_exit(vp->v_interlock);
2609 }
2610
2611 cansleep = (curlwp == uvm.pagedaemon_lwp || dofaf) ? 0 : 1;
2612
2613 KASSERT(curlwp != uvm.pagedaemon_lwp || dofaf || BIOASYNC(bp));
2614
2615 /* allocate transport structure */
2616 tomove = PUFFS_TOMOVE(bp->b_bcount, pmp);
2617 argsize = sizeof(struct puffs_vnmsg_rw);
2618 error = puffs_msgmem_alloc(argsize + tomove, &park_rw,
2619 (void *)&rw_msg, cansleep);
2620 if (error)
2621 goto out;
2622 RWARGS(rw_msg, 0, tomove, bp->b_blkno << DEV_BSHIFT, FSCRED);
2623
2624 /* 2x2 cases: read/write, faf/nofaf */
2625 if (BUF_ISREAD(bp)) {
2626 puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2627 PUFFS_VN_READ, VPTOPNC(vp));
2628 puffs_msg_setdelta(park_rw, tomove);
2629 if (BIOASYNC(bp)) {
2630 puffs_msg_setcall(park_rw,
2631 puffs_parkdone_asyncbioread, bp);
2632 puffs_msg_enqueue(pmp, park_rw);
2633 dobiodone = 0;
2634 } else {
2635 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data,
2636 NULL, error);
2637 error = checkerr(pmp, error, __func__);
2638 if (error)
2639 goto out;
2640
2641 if (rw_msg->pvnr_resid > tomove) {
2642 puffs_senderr(pmp, PUFFS_ERR_READ,
2643 E2BIG, "resid grew", VPTOPNC(vp));
2644 ERROUT(EPROTO);
2645 }
2646
2647 moved = tomove - rw_msg->pvnr_resid;
2648
2649 (void)memcpy(bp->b_data, rw_msg->pvnr_data, moved);
2650 bp->b_resid = bp->b_bcount - moved;
2651 }
2652 } else {
2653 puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2654 PUFFS_VN_WRITE, VPTOPNC(vp));
2655 /*
2656 * make pages read-only before we write them if we want
2657 * write caching info
2658 */
2659 if (PUFFS_WCACHEINFO(pmp)) {
2660 struct uvm_object *uobj = &vp->v_uobj;
2661 int npages = (bp->b_bcount + PAGE_SIZE-1) >> PAGE_SHIFT;
2662 struct vm_page *vmp;
2663 int i;
2664
2665 for (i = 0; i < npages; i++) {
2666 vmp= uvm_pageratop((vaddr_t)bp->b_data
2667 + (i << PAGE_SHIFT));
2668 DPRINTF(("puffs_strategy: write-protecting "
2669 "vp %p page %p, offset %" PRId64"\n",
2670 vp, vmp, vmp->offset));
2671 mutex_enter(uobj->vmobjlock);
2672 vmp->flags |= PG_RDONLY;
2673 pmap_page_protect(vmp, VM_PROT_READ);
2674 mutex_exit(uobj->vmobjlock);
2675 }
2676 }
2677
2678 (void)memcpy(&rw_msg->pvnr_data, bp->b_data, tomove);
2679 if (dofaf) {
2680 puffs_msg_setfaf(park_rw);
2681 } else if (BIOASYNC(bp)) {
2682 puffs_msg_setcall(park_rw,
2683 puffs_parkdone_asyncbiowrite, bp);
2684 dobiodone = 0;
2685 }
2686
2687 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data, NULL, error);
2688
2689 if (dobiodone == 0)
2690 goto out;
2691
2692 /*
2693 * XXXXXXXX: wrong, but kernel can't survive strategy
2694 * failure currently. Here, have one more X: X.
2695 */
2696 if (error != ENOMEM)
2697 error = 0;
2698
2699 error = checkerr(pmp, error, __func__);
2700 if (error)
2701 goto out;
2702
2703 if (rw_msg->pvnr_resid > tomove) {
2704 puffs_senderr(pmp, PUFFS_ERR_WRITE,
2705 E2BIG, "resid grew", VPTOPNC(vp));
2706 ERROUT(EPROTO);
2707 }
2708
2709 /*
2710 * FAF moved everything. Frankly, we don't
2711 * really have a choice.
2712 */
2713 if (dofaf && error == 0)
2714 moved = tomove;
2715 else
2716 moved = tomove - rw_msg->pvnr_resid;
2717
2718 bp->b_resid = bp->b_bcount - moved;
2719 if (bp->b_resid != 0) {
2720 ERROUT(EIO);
2721 }
2722 }
2723
2724 out:
2725 if (park_rw)
2726 puffs_msgmem_release(park_rw);
2727
2728 if (error)
2729 bp->b_error = error;
2730
2731 if (error || dobiodone)
2732 biodone(bp);
2733
2734 return error;
2735 }
2736
2737 int
2738 puffs_vnop_mmap(void *v)
2739 {
2740 struct vop_mmap_args /* {
2741 const struct vnodeop_desc *a_desc;
2742 struct vnode *a_vp;
2743 vm_prot_t a_prot;
2744 kauth_cred_t a_cred;
2745 } */ *ap = v;
2746 PUFFS_MSG_VARS(vn, mmap);
2747 struct vnode *vp = ap->a_vp;
2748 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2749 int error;
2750
2751 if (!PUFFS_USE_PAGECACHE(pmp))
2752 return genfs_eopnotsupp(v);
2753
2754 if (EXISTSOP(pmp, MMAP)) {
2755 PUFFS_MSG_ALLOC(vn, mmap);
2756 mmap_msg->pvnr_prot = ap->a_prot;
2757 puffs_credcvt(&mmap_msg->pvnr_cred, ap->a_cred);
2758 puffs_msg_setinfo(park_mmap, PUFFSOP_VN,
2759 PUFFS_VN_MMAP, VPTOPNC(vp));
2760
2761 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mmap, vp->v_data, NULL, error);
2762 error = checkerr(pmp, error, __func__);
2763 PUFFS_MSG_RELEASE(mmap);
2764 } else {
2765 error = genfs_mmap(v);
2766 }
2767
2768 return error;
2769 }
2770
2771
2772 /*
2773 * The rest don't get a free trip to userspace and back, they
2774 * have to stay within the kernel.
2775 */
2776
2777 /*
2778 * bmap doesn't really make any sense for puffs, so just 1:1 map it.
2779 * well, maybe somehow, somewhere, some day ....
2780 */
2781 int
2782 puffs_vnop_bmap(void *v)
2783 {
2784 struct vop_bmap_args /* {
2785 const struct vnodeop_desc *a_desc;
2786 struct vnode *a_vp;
2787 daddr_t a_bn;
2788 struct vnode **a_vpp;
2789 daddr_t *a_bnp;
2790 int *a_runp;
2791 } */ *ap = v;
2792 struct puffs_mount *pmp;
2793
2794 pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2795
2796 if (ap->a_vpp)
2797 *ap->a_vpp = ap->a_vp;
2798 if (ap->a_bnp)
2799 *ap->a_bnp = ap->a_bn;
2800 if (ap->a_runp)
2801 *ap->a_runp
2802 = (PUFFS_TOMOVE(pmp->pmp_msg_maxsize, pmp)>>DEV_BSHIFT) - 1;
2803
2804 return 0;
2805 }
2806
2807 /*
2808 * Handle getpages faults in puffs. We let genfs_getpages() do most
2809 * of the dirty work, but we come in this route to do accounting tasks.
2810 * If the user server has specified functions for cache notifications
2811 * about reads and/or writes, we record which type of operation we got,
2812 * for which page range, and proceed to issue a FAF notification to the
2813 * server about it.
2814 */
2815 int
2816 puffs_vnop_getpages(void *v)
2817 {
2818 struct vop_getpages_args /* {
2819 const struct vnodeop_desc *a_desc;
2820 struct vnode *a_vp;
2821 voff_t a_offset;
2822 struct vm_page **a_m;
2823 int *a_count;
2824 int a_centeridx;
2825 vm_prot_t a_access_type;
2826 int a_advice;
2827 int a_flags;
2828 } */ *ap = v;
2829 struct puffs_mount *pmp;
2830 struct puffs_node *pn;
2831 struct vnode *vp;
2832 struct vm_page **pgs;
2833 struct puffs_cacheinfo *pcinfo = NULL;
2834 struct puffs_cacherun *pcrun;
2835 void *parkmem = NULL;
2836 size_t runsizes;
2837 int i, npages, si, streakon;
2838 int error, locked, write;
2839
2840 pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2841 npages = *ap->a_count;
2842 pgs = ap->a_m;
2843 vp = ap->a_vp;
2844 pn = vp->v_data;
2845 locked = (ap->a_flags & PGO_LOCKED) != 0;
2846 write = (ap->a_access_type & VM_PROT_WRITE) != 0;
2847
2848 /* ccg xnaht - gets Wuninitialized wrong */
2849 pcrun = NULL;
2850 runsizes = 0;
2851
2852 /*
2853 * Check that we aren't trying to fault in pages which our file
2854 * server doesn't know about. This happens if we extend a file by
2855 * skipping some pages and later try to fault in pages which
2856 * are between pn_serversize and vp_size. This check optimizes
2857 * away the common case where a file is being extended.
2858 */
2859 if (ap->a_offset >= pn->pn_serversize && ap->a_offset < vp->v_size) {
2860 struct vattr va;
2861
2862 /* try again later when we can block */
2863 if (locked)
2864 ERROUT(EBUSY);
2865
2866 mutex_exit(vp->v_interlock);
2867 vattr_null(&va);
2868 va.va_size = vp->v_size;
2869 error = dosetattr(vp, &va, FSCRED, 0);
2870 if (error)
2871 ERROUT(error);
2872 mutex_enter(vp->v_interlock);
2873 }
2874
2875 if (write && PUFFS_WCACHEINFO(pmp)) {
2876 #ifdef notnowjohn
2877 /* allocate worst-case memory */
2878 runsizes = ((npages / 2) + 1) * sizeof(struct puffs_cacherun);
2879 KASSERT(curlwp != uvm.pagedaemon_lwp || locked);
2880 pcinfo = kmem_zalloc(sizeof(struct puffs_cacheinfo) + runsize,
2881 locked ? KM_NOSLEEP : KM_SLEEP);
2882
2883 /*
2884 * can't block if we're locked and can't mess up caching
2885 * information for fs server. so come back later, please
2886 */
2887 if (pcinfo == NULL)
2888 ERROUT(ENOMEM);
2889
2890 parkmem = puffs_park_alloc(locked == 0);
2891 if (parkmem == NULL)
2892 ERROUT(ENOMEM);
2893
2894 pcrun = pcinfo->pcache_runs;
2895 #else
2896 (void)parkmem;
2897 #endif
2898 }
2899
2900 error = genfs_getpages(v);
2901 if (error)
2902 goto out;
2903
2904 if (PUFFS_WCACHEINFO(pmp) == 0)
2905 goto out;
2906
2907 /*
2908 * Let's see whose fault it was and inform the user server of
2909 * possibly read/written pages. Map pages from read faults
2910 * strictly read-only, since otherwise we might miss info on
2911 * when the page is actually write-faulted to.
2912 */
2913 if (!locked)
2914 mutex_enter(vp->v_uobj.vmobjlock);
2915 for (i = 0, si = 0, streakon = 0; i < npages; i++) {
2916 if (pgs[i] == NULL || pgs[i] == PGO_DONTCARE) {
2917 if (streakon && write) {
2918 streakon = 0;
2919 pcrun[si].pcache_runend
2920 = trunc_page(pgs[i]->offset) + PAGE_MASK;
2921 si++;
2922 }
2923 continue;
2924 }
2925 if (streakon == 0 && write) {
2926 streakon = 1;
2927 pcrun[si].pcache_runstart = pgs[i]->offset;
2928 }
2929
2930 if (!write)
2931 pgs[i]->flags |= PG_RDONLY;
2932 }
2933 /* was the last page part of our streak? */
2934 if (streakon) {
2935 pcrun[si].pcache_runend
2936 = trunc_page(pgs[i-1]->offset) + PAGE_MASK;
2937 si++;
2938 }
2939 if (!locked)
2940 mutex_exit(vp->v_uobj.vmobjlock);
2941
2942 KASSERT(si <= (npages / 2) + 1);
2943
2944 #ifdef notnowjohn
2945 /* send results to userspace */
2946 if (write)
2947 puffs_cacheop(pmp, parkmem, pcinfo,
2948 sizeof(struct puffs_cacheinfo) + runsizes, VPTOPNC(vp));
2949 #endif
2950
2951 out:
2952 if (error) {
2953 if (pcinfo != NULL)
2954 kmem_free(pcinfo,
2955 sizeof(struct puffs_cacheinfo) + runsizes);
2956 #ifdef notnowjohn
2957 if (parkmem != NULL)
2958 puffs_park_release(parkmem, 1);
2959 #endif
2960 }
2961
2962 return error;
2963 }
2964
2965 /*
2966 * Extended attribute support.
2967 */
2968
2969 int
2970 puffs_vnop_getextattr(void *v)
2971 {
2972 struct vop_getextattr_args /*
2973 struct vnode *a_vp;
2974 int a_attrnamespace;
2975 const char *a_name;
2976 struct uio *a_uio;
2977 size_t *a_size;
2978 kauth_cred_t a_cred;
2979 }; */ *ap = v;
2980 PUFFS_MSG_VARS(vn, getextattr);
2981 struct vnode *vp = ap->a_vp;
2982 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2983 int attrnamespace = ap->a_attrnamespace;
2984 const char *name = ap->a_name;
2985 struct uio *uio = ap->a_uio;
2986 size_t *sizep = ap->a_size;
2987 size_t tomove, resid;
2988 int error;
2989
2990 if (uio)
2991 resid = uio->uio_resid;
2992 else
2993 resid = 0;
2994
2995 tomove = PUFFS_TOMOVE(resid, pmp);
2996 if (tomove != resid) {
2997 error = E2BIG;
2998 goto out;
2999 }
3000
3001 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_getextattr) + tomove,
3002 &park_getextattr, (void *)&getextattr_msg, 1);
3003
3004 getextattr_msg->pvnr_attrnamespace = attrnamespace;
3005 strlcpy(getextattr_msg->pvnr_attrname, name,
3006 sizeof(getextattr_msg->pvnr_attrname));
3007 puffs_credcvt(&getextattr_msg->pvnr_cred, ap->a_cred);
3008 if (sizep)
3009 getextattr_msg->pvnr_datasize = 1;
3010 getextattr_msg->pvnr_resid = tomove;
3011
3012 puffs_msg_setinfo(park_getextattr,
3013 PUFFSOP_VN, PUFFS_VN_GETEXTATTR, VPTOPNC(vp));
3014 puffs_msg_setdelta(park_getextattr, tomove);
3015 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getextattr, vp->v_data, NULL, error);
3016
3017 error = checkerr(pmp, error, __func__);
3018 if (error)
3019 goto out;
3020
3021 resid = getextattr_msg->pvnr_resid;
3022 if (resid > tomove) {
3023 puffs_senderr(pmp, PUFFS_ERR_GETEXTATTR, E2BIG,
3024 "resid grew", VPTOPNC(vp));
3025 error = EPROTO;
3026 goto out;
3027 }
3028
3029 if (sizep)
3030 *sizep = getextattr_msg->pvnr_datasize;
3031 if (uio)
3032 error = uiomove(getextattr_msg->pvnr_data, tomove - resid, uio);
3033
3034 out:
3035 PUFFS_MSG_RELEASE(getextattr);
3036 return error;
3037 }
3038
3039 int
3040 puffs_vnop_setextattr(void *v)
3041 {
3042 struct vop_setextattr_args /* {
3043 struct vnode *a_vp;
3044 int a_attrnamespace;
3045 const char *a_name;
3046 struct uio *a_uio;
3047 kauth_cred_t a_cred;
3048 }; */ *ap = v;
3049 PUFFS_MSG_VARS(vn, setextattr);
3050 struct vnode *vp = ap->a_vp;
3051 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3052 int attrnamespace = ap->a_attrnamespace;
3053 const char *name = ap->a_name;
3054 struct uio *uio = ap->a_uio;
3055 size_t tomove, resid;
3056 int error;
3057
3058 if (uio)
3059 resid = uio->uio_resid;
3060 else
3061 resid = 0;
3062
3063 tomove = PUFFS_TOMOVE(resid, pmp);
3064 if (tomove != resid) {
3065 error = E2BIG;
3066 goto out;
3067 }
3068
3069 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_setextattr) + tomove,
3070 &park_setextattr, (void *)&setextattr_msg, 1);
3071
3072 setextattr_msg->pvnr_attrnamespace = attrnamespace;
3073 strlcpy(setextattr_msg->pvnr_attrname, name,
3074 sizeof(setextattr_msg->pvnr_attrname));
3075 puffs_credcvt(&setextattr_msg->pvnr_cred, ap->a_cred);
3076 setextattr_msg->pvnr_resid = tomove;
3077
3078 if (uio) {
3079 error = uiomove(setextattr_msg->pvnr_data, tomove, uio);
3080 if (error)
3081 goto out;
3082 }
3083
3084 puffs_msg_setinfo(park_setextattr,
3085 PUFFSOP_VN, PUFFS_VN_SETEXTATTR, VPTOPNC(vp));
3086 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_setextattr, vp->v_data, NULL, error);
3087
3088 error = checkerr(pmp, error, __func__);
3089 if (error)
3090 goto out;
3091
3092 if (setextattr_msg->pvnr_resid != 0)
3093 error = EIO;
3094
3095 out:
3096 PUFFS_MSG_RELEASE(setextattr);
3097
3098 return error;
3099 }
3100
3101 int
3102 puffs_vnop_listextattr(void *v)
3103 {
3104 struct vop_listextattr_args /* {
3105 struct vnode *a_vp;
3106 int a_attrnamespace;
3107 struct uio *a_uio;
3108 size_t *a_size;
3109 int a_flag,
3110 kauth_cred_t a_cred;
3111 }; */ *ap = v;
3112 PUFFS_MSG_VARS(vn, listextattr);
3113 struct vnode *vp = ap->a_vp;
3114 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3115 int attrnamespace = ap->a_attrnamespace;
3116 struct uio *uio = ap->a_uio;
3117 size_t *sizep = ap->a_size;
3118 int flag = ap->a_flag;
3119 size_t tomove, resid;
3120 int error;
3121
3122 if (uio)
3123 resid = uio->uio_resid;
3124 else
3125 resid = 0;
3126
3127 tomove = PUFFS_TOMOVE(resid, pmp);
3128 if (tomove != resid) {
3129 error = E2BIG;
3130 goto out;
3131 }
3132
3133 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_listextattr) + tomove,
3134 &park_listextattr, (void *)&listextattr_msg, 1);
3135
3136 listextattr_msg->pvnr_attrnamespace = attrnamespace;
3137 listextattr_msg->pvnr_flag = flag;
3138 puffs_credcvt(&listextattr_msg->pvnr_cred, ap->a_cred);
3139 listextattr_msg->pvnr_resid = tomove;
3140 if (sizep)
3141 listextattr_msg->pvnr_datasize = 1;
3142
3143 puffs_msg_setinfo(park_listextattr,
3144 PUFFSOP_VN, PUFFS_VN_LISTEXTATTR, VPTOPNC(vp));
3145 puffs_msg_setdelta(park_listextattr, tomove);
3146 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_listextattr, vp->v_data, NULL, error);
3147
3148 error = checkerr(pmp, error, __func__);
3149 if (error)
3150 goto out;
3151
3152 resid = listextattr_msg->pvnr_resid;
3153 if (resid > tomove) {
3154 puffs_senderr(pmp, PUFFS_ERR_LISTEXTATTR, E2BIG,
3155 "resid grew", VPTOPNC(vp));
3156 error = EPROTO;
3157 goto out;
3158 }
3159
3160 if (sizep)
3161 *sizep = listextattr_msg->pvnr_datasize;
3162 if (uio)
3163 error = uiomove(listextattr_msg->pvnr_data, tomove-resid, uio);
3164
3165 out:
3166 PUFFS_MSG_RELEASE(listextattr);
3167 return error;
3168 }
3169
3170 int
3171 puffs_vnop_deleteextattr(void *v)
3172 {
3173 struct vop_deleteextattr_args /* {
3174 struct vnode *a_vp;
3175 int a_attrnamespace;
3176 const char *a_name;
3177 kauth_cred_t a_cred;
3178 }; */ *ap = v;
3179 PUFFS_MSG_VARS(vn, deleteextattr);
3180 struct vnode *vp = ap->a_vp;
3181 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3182 int attrnamespace = ap->a_attrnamespace;
3183 const char *name = ap->a_name;
3184 int error;
3185
3186 PUFFS_MSG_ALLOC(vn, deleteextattr);
3187 deleteextattr_msg->pvnr_attrnamespace = attrnamespace;
3188 strlcpy(deleteextattr_msg->pvnr_attrname, name,
3189 sizeof(deleteextattr_msg->pvnr_attrname));
3190 puffs_credcvt(&deleteextattr_msg->pvnr_cred, ap->a_cred);
3191
3192 puffs_msg_setinfo(park_deleteextattr,
3193 PUFFSOP_VN, PUFFS_VN_DELETEEXTATTR, VPTOPNC(vp));
3194 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_deleteextattr,
3195 vp->v_data, NULL, error);
3196
3197 error = checkerr(pmp, error, __func__);
3198
3199 PUFFS_MSG_RELEASE(deleteextattr);
3200 return error;
3201 }
3202
3203 /*
3204 * spec & fifo. These call the miscfs spec and fifo vectors, but issue
3205 * FAF update information for the puffs node first.
3206 */
3207 int
3208 puffs_vnop_spec_read(void *v)
3209 {
3210 struct vop_read_args /* {
3211 const struct vnodeop_desc *a_desc;
3212 struct vnode *a_vp;
3213 struct uio *a_uio;
3214 int a_ioflag;
3215 kauth_cred_t a_cred;
3216 } */ *ap = v;
3217
3218 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3219 return VOCALL(spec_vnodeop_p, VOFFSET(vop_read), v);
3220 }
3221
3222 int
3223 puffs_vnop_spec_write(void *v)
3224 {
3225 struct vop_write_args /* {
3226 const struct vnodeop_desc *a_desc;
3227 struct vnode *a_vp;
3228 struct uio *a_uio;
3229 int a_ioflag;
3230 kauth_cred_t a_cred;
3231 } */ *ap = v;
3232
3233 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3234 return VOCALL(spec_vnodeop_p, VOFFSET(vop_write), v);
3235 }
3236
3237 int
3238 puffs_vnop_fifo_read(void *v)
3239 {
3240 struct vop_read_args /* {
3241 const struct vnodeop_desc *a_desc;
3242 struct vnode *a_vp;
3243 struct uio *a_uio;
3244 int a_ioflag;
3245 kauth_cred_t a_cred;
3246 } */ *ap = v;
3247
3248 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3249 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), v);
3250 }
3251
3252 int
3253 puffs_vnop_fifo_write(void *v)
3254 {
3255 struct vop_write_args /* {
3256 const struct vnodeop_desc *a_desc;
3257 struct vnode *a_vp;
3258 struct uio *a_uio;
3259 int a_ioflag;
3260 kauth_cred_t a_cred;
3261 } */ *ap = v;
3262
3263 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3264 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), v);
3265 }
3266